← Back

Blood sugar

Everything in the library studied for it, medical and holistic together. Within each kind, what was shown to help comes first, strongest evidence first. What was tested and not shown to help stays on the list.

Strong several reviews agree · Moderate one review or consistent trials · Limited one or a few small studies · Untested no study in people, listed with what it rests on

Medication28 Peptides and SARMs1 Supplements36 Practices6 Daily life14 Food and drink16 Diets23 Devices and wearables2 Lab markers45

Medication

MetforminStrong

Preventing type 2 diabetes

Shown to lower type 2 diabetes incidence about 31% in adults with raised blood sugar, less than the 58% from lifestyle change.

The numbers and sources

Effect: Over an average 2.8 years, diabetes incidence was 11.0, 7.8 and 4.8 cases per 100 person-years on placebo, metformin and lifestyle change. Metformin lowered incidence by 31% (95% CI 17 to 43) and lifestyle change by 58% (95% CI 48 to 66) against placebo. Preventing one case over 3 years took 13.9 people on metformin or 6.9 in the lifestyle program [1]. Over about 21 years of follow-up, incidence stayed lower in the original metformin group than placebo (HR 0.83, 95% CI 0.74 to 0.93), with median diabetes-free survival 2.5 years longer [2]. In people with metabolic syndrome, metformin 1,700 mg a day lowered type 2 diabetes incidence by 80% against placebo over about 3 years [3]. Pooling RCTs in high-risk adults, metformin lowered diabetes incidence (OR 0.77, 95% CI 0.67 to 0.88), including in prediabetes (OR 0.75, 95% CI 0.66 to 0.86). Added to a lifestyle program, it did not lower incidence more than the same program alone [4]. A second meta-analysis of 18 RCTs found a risk ratio of 0.65 (95% CI 0.55 to 0.77), lowest in impaired fasting glucose (RR 0.38) [5].
Population: 3,234 adults without diabetes but with raised fasting and post-load glucose, mean age 51, mean BMI 34.0, 68% women [1]; 3,195 of the same participants over follow-up [2]; 1,442 adults with metabolic syndrome [3]; RCTs in metformin-naive high-risk adults [4]; 18 RCTs in prediabetes [5].
Sources: [1], [2], [3], [4], [5]
Caveats: Lifestyle change lowered incidence more than metformin [1][2]. After the first 3 years, placebo was stopped and metformin continued open-label [2]. The effect weakened after metformin was stopped [4]. [5] reported moderate heterogeneity.

MetforminStrongStudied in women

Insulin resistance and blood sugar in PCOS

Shown to lower insulin resistance and fasting glucose slightly against placebo in women with PCOS. Added to lifestyle change, one review found no extra effect.

Compared with placebo [34][31]; lifestyle change with or without placebo [33]; combined oral contraceptive pills [35]

The numbers and sources

Effect: Against placebo, metformin lowered HOMA-IR by 0.50 (95% CI 0.09 to 0.91) and fasting glucose by 0.13 mmol/L (95% CI 0.07 to 0.19), both moderate certainty [34]. A second review also found lower fasting glucose than placebo (SMD -0.23, 95% CI -0.40 to -0.06) [31]. Lifestyle change plus metformin did not differ from lifestyle change with or without placebo on surrogate markers of insulin resistance or fasting glucose after 6 months [33]. Against combined oral contraceptive pills, metformin lowered insulin by 27.12 pmol/L (95% CI 13.59 to 40.65) [35].
Population: RCTs in women with PCOS published through July 2022 [34]; 24 RCTs with 564 women given metformin [31]; 12 RCTs with 608 women [33]; 36 RCTs in women with PCOS [35].
Sources: [34], [31], [33], [35]
Caveats: The changes are small. The add-on comparison rests on small, short trials [33].

OrforglipronStrong

Blood sugar in type 2 diabetes

Shown to lower HbA1c about 1.2 to 1.9 points in adults with type 2 diabetes, more than placebo, oral semaglutide or dapagliflozin. Not US approved for diabetes.

The numbers and sources

Effect: In early type 2 diabetes treated with diet and exercise, HbA1c fell 1.24 to 1.48 points across doses from a baseline of 8.0% at 40 weeks, against 0.41 on placebo, a difference of -1.07 points on 36 mg (95% CI -1.33 to -0.81) [7]. In people taking metformin, HbA1c fell 1.91 points on orforglipron 36 mg and 1.47 on oral semaglutide 14 mg at 52 weeks, a difference of -0.44 (95% CI -0.62 to -0.26) [8]. Against dapagliflozin it fell 1.56 points on 36 mg and 0.81 points, a difference of -0.75 (95% CI -0.96 to -0.55) [9]. Added to insulin glargine, it fell 1.82 points on 36 mg and 0.79 on placebo, a difference of -1.03 (95% CI -1.28 to -0.77) [10]. Meta-analyses reported pooled HbA1c reductions of 1.07 points [5] and 1.29 points [4].
Population: 559 adults with early type 2 diabetes [7]; 1,698 adults on metformin [8]; 962 adults on metformin [9]; 546 adults on insulin glargine [10].
Sources: [7], [8], [9], [10], [4], [5]
Caveats: Type 2 diabetes is not an approved use in the US. The label covers weight management only [6]. The comparisons with oral semaglutide and dapagliflozin were open-label [8][9]. ACHIEVE-1, ACHIEVE-2 and ACHIEVE-3 were funded by Eli Lilly [7][8][9]. HbA1c is a blood marker, not a measure of diabetes complications.

SGLT2 inhibitorsStrong

Blood sugar in type 2 diabetes

Shown to lower HbA1c by about 0.66 points more than placebo in adults with type 2 diabetes, and about as much as other diabetes drugs.

The numbers and sources

Effect: Against placebo in 45 studies, SGLT2 inhibitors lowered HbA1c by 0.66 percentage points (95% CI 0.58 to 0.73). Against other diabetes drugs in 13 studies HbA1c was 0.06 points lower (95% CI 0.18 lower to 0.05 higher), and the risk of low blood sugar was similar [14]. In a network meta-analysis of trials of at least 24 weeks, canagliflozin, dapagliflozin and empagliflozin each lowered HbA1c and fasting glucose against placebo. Canagliflozin 300 mg lowered HbA1c by 0.2 points (95% CI 0.1 to 0.3) more than dapagliflozin 10 mg or empagliflozin 25 mg [15].
Population: 11,232 participants in placebo-controlled studies and 5,175 in active-comparator studies [14]; 23,997 participants in 38 trials [15].
Sources: [14], [15]
Caveats: Most trials in the first review were rated at high risk of bias for missing data [14]. The glucose-lowering effect depends on kidney filtration and shrinks as kidney function falls.

SulfonylureasStrong

Blood sugar in type 2 diabetes

Shown to lower HbA1c by about 1.5 points more than placebo as single therapy in type 2 diabetes, with no added drop at higher doses.

The numbers and sources

Effect: Across nine trials, sulfonylurea monotherapy lowered HbA1c by 1.51 percentage points more than placebo (95% CI 1.25 to 1.78). Added to other oral treatment it lowered HbA1c by 1.62 points (95% CI 1.0 to 2.24), and added to insulin by 0.46 points (95% CI 0.24 to 0.69). Higher doses did not lower HbA1c more than lower doses [1]. Added to maximal metformin, the drug classes studied lowered HbA1c by a similar 0.64 to 0.97 points against placebo [2]. In a 14-week label trial, glimepiride 1 mg, 4 mg and 8 mg lowered HbA1c by 1.2, 1.8 and 1.8 points against placebo, and in a 22-week trial by 1.1 points [3]. Over 10 years, intensive treatment with a sulfonylurea or insulin held HbA1c at 7.0% against 7.9% on diet [4].
Population: 31 trials with a median length of 16 weeks [1]; 11,198 participants in 27 trials of drugs added to metformin [2]; 304 and 249 adults in two glimepiride trials [3]; 3,867 newly diagnosed patients [4].
Sources: [1], [2], [3], [4]
Caveats: Most trials were short. Trial length, drug and duration of diabetes contributed to differences between trials [1].

TopiramateStrong

Weight loss

Shown to lower weight by about 5.3 kg more than placebo in adults with overweight or obesity, and 3.4 kg in adults with type 2 diabetes, with more adverse events.

Compared with placebo [7][8][3]

The numbers and sources

Effect: Across 10 trials of at least 16 weeks, topiramate lowered weight 5.34 kg more than placebo (95% CI -6.12 to -4.56), and 6.58 kg in trials longer than 28 weeks at 96 to 200 mg a day. Withdrawal for adverse events was more likely (OR 1.94, 95% CI 1.64 to 2.29) [7]. In adults with obesity and type 2 diabetes, topiramate alone lowered weight 3.4 kg more than placebo (95% CI -3.79 to -3.04) and HbA1c by 0.4 points (95% CI -0.58 to -0.32), with more serious adverse events (RR 1.69, 95% CI 1.00 to 2.87) [8]. In the label's migraine trials, mean weight fell 2% at 50 mg, 3% at 100 mg and 4% at 200 mg, against 0% on placebo [3].
Population: 3,320 adults with overweight or obesity in 10 trials [7]; adults with obesity and type 2 diabetes in 5 trials [8]; adults in migraine trials [3].
Sources: [7], [8], [3]
Caveats: Topiramate alone is not approved for weight loss. All but one diabetes trial had high risk of bias [8].

ARBs (angiotensin receptor blockers)Moderate

Preventing type 2 diabetes

Shown to lower new type 2 diabetes by about 14 to 15%, with 33.1% against 36.8% developing it over 5 years in one valsartan trial.

The numbers and sources

Effect: In adults with impaired glucose tolerance, diabetes developed in 33.1% on valsartan and 36.8% on placebo over a median 5 years (HR 0.86, 95% CI 0.80 to 0.92). Valsartan did not lower cardiovascular events (14.5% against 14.8%) [18]. Across randomized trials of ARBs, new diabetes was less frequent than in controls (RR 0.85, 95% CI 0.78 to 0.93) [6].
Population: 9,306 adults with impaired glucose tolerance and cardiovascular disease or risk factors [18]; 147,020 participants in 37 trials [6].
Sources: [18], [6]
Caveats: The trial evidence is for valsartan up to 160 mg a day, given alongside lifestyle modification [18].

DulaglutideModerateVery common

Blood sugar control in type 2 diabetes

Shown to lower HbA1c by about 1.0 to 1.2 points against almost no change on placebo in adults with type 2 diabetes, with 62% reaching under 7% against 22%.

Compared with placebo [1]; sitagliptin, metformin, exenatide and insulin glargine [1]; other diabetes drugs in a network meta-analysis [2]

The numbers and sources

How common: Very common: more than 1 in 10 people.
Effect: Added to metformin, HbA1c changed by -1.0 points on 0.75 mg, -1.2 on 1.5 mg, -0.6 on sitagliptin and +0.1 on placebo at 26 weeks. HbA1c under 7% was reached by 56%, 62%, 39% and 22%, and fasting glucose fell 35, 41, 18 and 9 mg/dL. In other placebo-controlled trials, dulaglutide lowered HbA1c more than placebo when added to glimepiride, to metformin with pioglitazone, and to SGLT2 inhibitors [1]. A network meta-analysis of 453 trials found that specific GLP-1 receptor agonists added to metformin, with insulin regimens, produced the largest HbA1c reductions [2].
Population: 972 adults on metformin in one trial and other placebo-controlled trials described in the label [1]; 453 trials of 21 glucose-lowering treatments [2].
Sources: [1], [2]
Caveats: The trials were funded by the manufacturer. The network meta-analysis abstract does not give a dulaglutide estimate [2].

InsulinModerate

Eye, kidney and nerve complications in type 1 diabetes

Shown to cut new eye disease by about 76%, kidney protein leakage by 39% to 54% and nerve damage by 60% with intensive insulin treatment in type 1 diabetes.

The numbers and sources

Effect: In people with type 1 diabetes and no eye disease, intensive insulin treatment lowered the risk of developing retinopathy by 76% (95% CI 62 to 85) against conventional treatment. In people with mild retinopathy it slowed progression by 54% (95% CI 39 to 66). Across both groups it lowered microalbuminuria by 39% (95% CI 21 to 52), albuminuria by 54% (95% CI 19 to 74) and clinical neuropathy by 60% (95% CI 38 to 74) [1]. A Cochrane review of 12 trials found retinopathy developed in 6.2% on intensive control against 23.2% on conventional control (RR 0.27, 95% CI 0.18 to 0.42), nephropathy in 16.3% against 28.4% (RR 0.56, 95% CI 0.46 to 0.68) and neuropathy in 4.9% against 13.9% (RR 0.35, 95% CI 0.23 to 0.53) [2].
Population: 1,441 people with type 1 diabetes, assigned to an insulin pump or three or more daily injections, or to one or two daily injections [1]; 2,230 patients in 12 trials followed for 1 to 6.5 years [2].
Sources: [1], [2]
Caveats: The trials compared more intensive with less intensive insulin treatment. Neither arm went without insulin. The largest trial dominates the review [2]. Once complications were present, the effect was weaker for eye disease and uncertain for kidney disease [2].

InsulinModerate

Blood sugar and complications in type 2 diabetes

Shown to lower HbA1c in type 2 diabetes, to 7.0% against 7.9% on diet over 10 years in one trial, with 25% fewer small-vessel complications.

The numbers and sources

Effect: In newly diagnosed type 2 diabetes, intensive treatment with insulin or a sulfonylurea held HbA1c at 7.0% against 7.9% on diet over 10 years, with no difference between agents. The risk of any diabetes-related endpoint was 12% lower (95% CI 1 to 21) and of microvascular endpoints 25% lower (95% CI 7 to 40) [3]. Added to metformin, glargine kept HbA1c below 7.0% for longer than glimepiride or sitagliptin. An HbA1c of 7.0% or higher was reached at 26.5 per 100 participant-years on glargine, 26.1 on liraglutide, 30.4 on glimepiride and 38.1 on sitagliptin [4]. In a 52-week label trial of glargine added to oral drugs, HbA1c fell by 0.5 points from a baseline of 9.0% [5]. In a network meta-analysis, drug classes added to metformin were associated with similar HbA1c levels [6].
Population: 3,867 newly diagnosed patients, median age 54 [3]; 5,047 participants on metformin followed a mean of 5.0 years [4]; 570 adults in the label trial [5]; 301 trials of glucose-lowering drugs [6].
Sources: [3], [4], [5], [6]
Caveats: The 10-year trial pooled insulin with sulfonylureas in its intensive group, and found no drop in diabetes-related death or death from any cause [3]. A meta-analysis of 20 trials found no effect of insulin on any clinical outcome in type 2 diabetes [7].

InsulinModerate

Long-acting analogs compared with NPH insulin

Shown to give HbA1c similar to NPH insulin with fewer people having low blood sugar, while differences in severe episodes were small.

The numbers and sources

Effect: In type 2 diabetes, changes in HbA1c were comparable between long-acting analogs and NPH insulin, and glargine and detemir each lowered confirmed and nighttime hypoglycemia. For severe hypoglycemia the risk ratio was 0.68 for glargine (95% CI 0.46 to 1.01, absolute reduction 1.2%) and 0.45 for detemir (95% CI 0.17 to 1.20) [10]. In type 1 diabetes, severe hypoglycemia occurred in 8.5% on detemir against 11.5% on NPH (RR 0.69, 95% CI 0.52 to 0.92) and in 10.2% on glargine against 12.5% on NPH (RR 0.84, 95% CI 0.67 to 1.04). HbA1c did not differ between detemir and NPH (mean difference 0.01%, 95% CI -0.1 to 0.1) [11]. In label trials in type 2 diabetes, the HbA1c difference between glargine and NPH ranged from 0.1 points lower to 0.2 points higher [5].
Population: 24 trials in type 2 diabetes lasting 24 weeks to 5 years [10]; 8,784 participants with type 1 diabetes in 26 trials, 21% of them children [11]; 2,105 adults in three label trials [5].
Sources: [10], [11], [5]
Caveats: All of the type 2 trials had unclear or high risk of bias, and certainty for severe hypoglycemia was low or very low [10]. No trial compared degludec with NPH insulin [10][11].

InsulinModerate

Degludec compared with glargine

Shown to give the same HbA1c as glargine with fewer low blood sugar episodes, including severe episodes in 4.9% against 6.6% in one large trial.

The numbers and sources

Effect: In type 2 diabetes at high cardiovascular risk, HbA1c at 24 months was 7.5% in both groups. Severe hypoglycemia occurred in 4.9% on degludec against 6.6% on glargine (rate ratio 0.60) [9]. In type 1 diabetes, symptomatic hypoglycemia during the maintenance period was lower on degludec (rate ratio 0.89, 95% CI 0.85 to 0.94), and severe hypoglycemia occurred in 10.3% against 17.1% [12]. In insulin-treated type 2 diabetes, 22.5% on degludec against 31.6% on glargine had a symptomatic episode, while severe hypoglycemia did not differ (1.6% against 2.4%) [13]. A Cochrane review of four trials in type 1 diabetes found no evidence of a clinically relevant difference between degludec and glargine on its main outcomes [11].
Population: 7,637 patients with type 2 diabetes [9]; 501 patients with type 1 diabetes and a risk factor for hypoglycemia [12]; 721 patients with type 2 diabetes and a risk factor for hypoglycemia [13]; four trials in type 1 diabetes [11].
Sources: [9], [12], [13], [11]
Caveats: The three trials were funded by the maker of degludec, and two enrolled only people at raised risk of hypoglycemia [12][13]. The Cochrane review did not find a relevant difference [11].

MetforminModerate

Blood sugar in type 2 diabetes

Shown to lower HbA1c about 1.1 points more than placebo in people with diabetes, and more at higher doses.

The numbers and sources

Effect: Across 35 trials of at least 12 weeks, metformin alone lowered HbA1c by 1.12 points more than placebo (95% CI 0.92 to 1.32). Added to other oral diabetes drugs it lowered HbA1c 0.95 points more than placebo (95% CI 0.77 to 1.13), and added to insulin 0.60 points more than insulin alone (95% CI 0.30 to 0.91). Higher doses lowered HbA1c more than lower doses, with no significant rise in side effects [6]. A Cochrane review of 29 trials found a strong benefit for HbA1c against placebo and against diet [7].
Population: People with diabetes in trials of at least 12 weeks [6]; 5,259 people with type 2 diabetes [7].
Sources: [6], [7]
Caveats: Heterogeneity was high (I² 77 to 80%) [6]. The Cochrane abstract gives no pooled effect size, so the grade rests on one review with numbers. These are people with diabetes. For people without it, see diabetes prevention above.

RetatrutideGrey marketModerate

Blood sugar in type 2 diabetes

Shown to lower HbA1c about 0.9 to 1.1 points more than placebo in adults with type 2 diabetes, with no severe low blood sugar reported.

The numbers and sources

Effect: At 40 weeks, HbA1c changed by -1.69% on 4 mg, -1.86% on 9 mg and -1.94% on 12 mg, against -0.81% on placebo. Differences from placebo were -0.88 (95% CI -1.18 to -0.59), -1.04 (95% CI -1.32 to -0.76) and -1.12 (95% CI -1.39 to -0.85) points. No severe hypoglycemia was reported [6]. A meta-analysis of 8 studies found HbA1c fell 1.04% (95% CI -1.42 to -0.67). Doses of 8 mg or less lowered it more (-1.39%) than doses of 12 mg or more (-0.65%) [7]. Pooling 3 RCTs in adults with obesity, with or without diabetes, HbA1c fell 0.91% more than on placebo [3].
Population: 537 adults with type 2 diabetes, baseline HbA1c 7.9%, mean diabetes duration 2.5 years [6]; 8 studies in people with diabetes, obesity or both, framed around chronic kidney disease [7]; 3 RCTs with 878 adults [3].
Sources: [6], [7], [3]
Caveats: The 40-week trial tested retatrutide as a single drug rather than added to others [6]. The abstract of [7] does not say how many participants had kidney disease. [3] mixed people with and without diabetes.

SGLT2 inhibitorsModerate

Type 1 diabetes, added to insulin

Shown to lower HbA1c by about 0.38 points when added to insulin in adults with type 1 diabetes, while about doubling the risk of ketoacidosis.

The numbers and sources

Effect: Added to insulin, SGLT2 inhibitors lowered HbA1c by 0.38 points, raised time in range by 8.6 percentage points, and lowered insulin use by 4.7 units a day and body weight by 2.5 kg. The risk of diabetic ketoacidosis rose (RR 2.19, 95% CI 1.16 to 4.17) [19]. The US labels state that the drugs are not for blood sugar control in type 1 diabetes and may raise the risk of ketoacidosis in these patients [20].
Population: Adults with type 1 diabetes in a meta-analysis of 90 studies of two drug classes [19].
Sources: [19], [20]
Caveats: Not an approved use in the US. The review pooled trials with other study designs. See the ketoacidosis block below.

DulaglutideLimited

Higher doses of 3 mg and 4.5 mg

One trial found 4.5 mg lowered HbA1c 0.24 points more and weight 1.6 kg more than 1.5 mg over 36 weeks, with more nausea and vomiting.

Compared with dulaglutide 1.5 mg [3]

The numbers and sources

Effect: At 36 weeks, HbA1c fell 1.77 points on 4.5 mg and 1.54 on 1.5 mg (difference -0.24, P under 0.001). The 3 mg dose beat 1.5 mg only in the on-treatment analysis (-0.17 points). Weight fell 4.6 kg on 4.5 mg and 3.0 kg on 1.5 mg. Nausea occurred in 13.4%, 15.6% and 16.4%, and vomiting in 5.6%, 8.3% and 9.3%, on 1.5, 3 and 4.5 mg [3].
Population: 1,842 adults with type 2 diabetes on metformin, mean HbA1c 8.6% and BMI 34.2, over 52 weeks [3].
Sources: [3]
Caveats: One trial with no placebo arm, funded by the manufacturer.
Why limited: This rests on one trial, of 1,842 people. Strength counts independent studies that agree, so one trial is limited at any size.

DulaglutideLimitedStudied in children

Blood sugar in children and teenagers with type 2 diabetes

One trial in youths with type 2 diabetes found dulaglutide lowered HbA1c by 0.9 points while it rose 0.6 on placebo, with no difference in BMI.

Compared with placebo [5]

The numbers and sources

Effect: At 26 weeks, HbA1c rose 0.6 points on placebo and fell 0.6 on 0.75 mg and 0.9 on 1.5 mg. HbA1c under 7% was reached by 51% on dulaglutide against 14% on placebo. Fasting glucose rose 17.1 mg/dL on placebo and fell 18.9 mg/dL on dulaglutide. BMI change did not differ [5].
Population: 154 youths aged 10 to under 18 with type 2 diabetes, on lifestyle changes alone or with metformin, with or without basal insulin [5].
Sources: [5]
Caveats: One small trial, funded by the manufacturer. Gastrointestinal adverse events were more frequent than on placebo [5].

InsulinLimited

Delaying type 2 diabetes in people with raised blood sugar

One trial found new diabetes in 30% on insulin glargine against 35% on standard care, measured about 3 months after treatment stopped.

The numbers and sources

Effect: Among 1,456 participants without diabetes at the start, new diabetes was diagnosed about 3 months after therapy was stopped in 30% on glargine against 35% on standard care (OR 0.80, 95% CI 0.64 to 1.00) [8].
Population: 1,456 people with impaired fasting glucose or impaired glucose tolerance within a trial of 12,537, followed a median 6.2 years [8].
Sources: [8]
Caveats: Not an approved use. The upper confidence limit reached 1.00. In the same trial, glargine raised severe hypoglycemia and weight [8].

LiraglutideLimited

Type 2 diabetes onset in prediabetes

One trial found fewer diabetes diagnoses on liraglutide than placebo over 3 years in adults with prediabetes and obesity, 2% against 6%.

The numbers and sources

Effect: By week 160, diabetes was diagnosed on treatment in 2% on liraglutide and 6% on placebo, a hazard ratio of 0.21 (95% CI 0.13 to 0.34). Weight changed by -6.1% on liraglutide and -1.9% on placebo at 3 years [8].
Population: 2,254 adults with prediabetes and a BMI of 30 or more, or 27 or more with a comorbidity [8].
Sources: [8]
Caveats: One trial. Half of participants finished the 160 weeks, and people who withdrew were not followed after stopping. Funded by Novo Nordisk [8].
Why limited: This rests on one trial, of 2,254 people. Strength counts independent studies that agree, so one trial is limited at any size.

Naltrexone-bupropionLimited

Weight and blood sugar in type 2 diabetes

One trial found more weight loss, 5.0% against 1.8%, and a larger HbA1c drop than placebo in adults with type 2 diabetes.

The numbers and sources

Effect: Weight changed by -5.0% on the combination and -1.8% on placebo at 56 weeks. HbA1c fell 0.6 points against 0.1 points, and 44.1% against 26.3% reached an HbA1c under 7% [5].
Population: 505 adults with type 2 diabetes and overweight or obesity, mean HbA1c 8.0% [5].
Sources: [5]
Caveats: One trial.

OrlistatLimited

Type 2 diabetes onset

One trial found 37% fewer diabetes cases over 4 years in adults with obesity, explained by those with impaired glucose tolerance.

The numbers and sources

Effect: After 4 years, cumulative diabetes incidence was 9.0% on placebo and 6.2% on orlistat, a risk reduction of 37.3%. The effect was explained by the subgroup with impaired glucose tolerance. Weight loss was 5.8 kg on orlistat and 3.0 kg on placebo [4]. A meta-analysis of long-term trials also reported a lower incidence of diabetes on orlistat [1].
Population: 3,305 adults with BMI 30 or more, 21% with impaired glucose tolerance, over 4 years [4].
Sources: [4], [1]
Caveats: 52% on orlistat and 34% on placebo completed treatment. When dropouts' baseline weights were carried forward, weight loss was 3.6 kg against 1.4 kg [4].

SulfonylureasLimited

Eye, kidney and other small-vessel complications

One trial found 25% fewer small-vessel complications over 10 years with intensive sulfonylurea or insulin treatment than with diet in newly diagnosed type 2 diabetes.

The numbers and sources

Effect: Against diet, intensive treatment with a sulfonylurea or insulin lowered any diabetes-related endpoint by 12% (95% CI 1 to 21) and microvascular endpoints by 25% (95% CI 7 to 40), including the need for retinal laser treatment. Results did not differ between chlorpropamide, glibenclamide and insulin. Diabetes-related death (10% lower, 95% CI -11 to 27) and death from any cause (6% lower, 95% CI -10 to 20) did not differ significantly [4].
Population: 3,867 newly diagnosed patients with type 2 diabetes, median age 54, followed over 10 years [4].
Sources: [4]
Caveats: The trial pooled sulfonylureas with insulin in its intensive group, and used chlorpropamide, glibenclamide and glipizide. It did not find fewer large-vessel events [4].
Why limited: This rests on one trial, of 3,867 people. Strength counts independent studies that agree, so one trial is limited at any size.

SulfonylureasLimitedVery common

Neonatal diabetes from potassium channel mutations

One study found 90% of patients with diabetes from Kir6.2 mutations stopped insulin on a sulfonylurea, with HbA1c falling from 8.1% to 6.4%.

The numbers and sources

How common: Very common: more than 1 in 10 people.
Effect: Of 49 patients with diabetes caused by Kir6.2 mutations, 44 (90%) stopped insulin after starting a sulfonylurea. HbA1c improved in everyone who switched, from 8.1% before treatment to 6.4% after 12 weeks, and the improvement was sustained at one year [11].
Population: 49 consecutive patients with Kir6.2 (KCNJ11) mutations [11].
Sources: [11]
Caveats: Not an approved use. An uncontrolled study in a rare genetic form of diabetes.

InsulinModerate

Rapid-acting analogs compared with regular human insulin

Mixed results: rapid-acting analogs lowered HbA1c 0.15 points more than regular human insulin in type 1 diabetes, with no clear difference in type 2 diabetes.

The numbers and sources

Effect: In adults with type 1 diabetes, HbA1c was 0.15 points lower on rapid-acting analogs (95% CI 0.1 to 0.2), and severe hypoglycemia did not differ (OR 0.89, 95% CI 0.71 to 1.12) [14]. In adults with type 2 diabetes, HbA1c change did not differ (mean difference -0.03%, 95% CI -0.16 to 0.09), and no trial showed a clear difference in severe hypoglycemia [15].
Population: 2,693 participants with type 1 diabetes in 9 trials of 24 to 52 weeks [14]; 2,751 participants with type 2 diabetes in 10 trials of 24 to 104 weeks [15].
Sources: [14], [15]
Caveats: None of the trials was blinded, and the evidence was rated low or very low quality. No trial measured long-term complications or mortality as a main outcome [14][15].

SulfonylureasModerate

Keeping blood sugar controlled over years

Not shown to hold blood sugar control as long as comparators: 34% failed on glyburide alone by 5 years against 21% on metformin and 15% on rosiglitazone.

The numbers and sources

Effect: As first treatment, the cumulative incidence of monotherapy failure at 5 years was 34% with glyburide, 21% with metformin and 15% with rosiglitazone [9]. Added to metformin, an HbA1c of 7.0% or higher was reached at 30.4 per 100 participant-years on glimepiride, against 26.5 on glargine, 26.1 on liraglutide and 38.1 on sitagliptin [10].
Population: 4,360 patients with recently diagnosed type 2 diabetes treated a median 4.0 years [9]; 5,047 participants on metformin followed a mean 5.0 years [10].
Sources: [9], [10]
Caveats: Glimepiride held control longer than sitagliptin in the second trial [10]. Both trials compared active drugs, with no placebo group.

SulfonylureasModerateStudied in pregnancy

Gestational diabetes

Not shown to match insulin in gestational diabetes: glyburide was followed by heavier babies and about twice the newborn low blood sugar.

The numbers and sources

Effect: Against insulin, glibenclamide (glyburide) was associated with higher birth weight (mean difference 109 g, 95% CI 35.9 to 181), more macrosomia (RR 2.62, 95% CI 1.35 to 5.08) and more neonatal hypoglycemia (RR 2.04, 95% CI 1.30 to 3.20). Against glibenclamide, metformin was associated with less maternal weight gain, lower birth weight (209 g lower) and less macrosomia (RR 0.33, 95% CI 0.13 to 0.81). Treatment failure was higher with metformin than with glibenclamide [12].
Population: 2,509 women with gestational diabetes needing drug treatment, in 15 articles [12].
Sources: [12]
Caveats: Not an approved use. Short-term outcomes only.

ACE inhibitorsLimited

Preventing type 2 diabetes

One trial found ramipril did not lower new diabetes or death over 3 years in adults with raised blood sugar, 18.1% against 19.5% on placebo.

The numbers and sources

Effect: Diabetes or death occurred in 18.1% on ramipril and 19.5% on placebo (HR 0.91, 95% CI 0.81 to 1.03). More people on ramipril returned to normal blood sugar (HR 1.16, 95% CI 1.07 to 1.27). Glucose 2 hours after a glucose load was lower on ramipril (135.1 against 140.5 mg/dL), and fasting glucose did not differ significantly [16].
Population: 5,269 adults without cardiovascular disease with impaired fasting glucose or impaired glucose tolerance, median 3 years [16].
Sources: [16]
Caveats: One trial, of ramipril up to 15 mg a day.
Why limited: This rests on one trial, of 5,269 people. Strength counts independent studies that agree, so one trial is limited at any size.

DulaglutideLimited

Compared with semaglutide

One head-to-head trial found semaglutide lowered HbA1c about 0.4 points more and weight 2.3 to 3.6 kg more than dulaglutide at matched dose levels over 40 weeks.

Compared with semaglutide [4]

The numbers and sources

Effect: HbA1c fell 1.4 points on dulaglutide 1.5 mg and 1.8 on semaglutide 1.0 mg (difference -0.41), and 1.1 on dulaglutide 0.75 mg against 1.5 on semaglutide 0.5 mg (difference -0.40). Weight fell 3.0 kg against 6.5 kg (difference -3.55 kg) and 2.3 kg against 4.6 kg. Gastrointestinal disorders occurred in 48% on dulaglutide 1.5 mg and 44% on semaglutide 1.0 mg [4].
Population: 1,201 adults with type 2 diabetes on metformin in 16 countries [4].
Sources: [4]
Caveats: Open label, funded by the maker of semaglutide. Dulaglutide still lowered HbA1c and weight from baseline in both arms [4].
Why limited: This rests on one trial, of 1,201 people. Strength counts independent studies that agree, so one trial is limited at any size.

Peptides and SARMs

MOTS-cGrey marketUntested

Insulin sensitivity and blood sugar

Not studied in people. In mice, MOTS-c treatment prevented insulin resistance brought on by aging and by a high-fat diet.

No study in people. This rests on animal studies, how it acts in the body.

The numbers and sources

Effect: MOTS-c treatment in mice prevented age-dependent and high-fat-diet-induced insulin resistance [9]. A 2023 review describes MOTS-c as acting mainly through the AICAR-AMPK pathway, with effects on the expression of genes such as GLUT4, a glucose transporter [11].
Population: Mice [9]. Mechanisms reviewed in [11].
Sources: [9], [11], [10]
Caveats: In people, only the MOTS-c the body makes has been measured. In one study, body mass index and HOMA-IR, a measure of insulin resistance, each predicted higher circulating MOTS-c [10]. That is an association, not a test of taking it. FDA found no human exposure data for MOTS-c by any route [3].

Supplements

Apple cider vinegarStrong

Blood sugar

Shown to lower fasting blood sugar and HbA1c in adults with type 2 diabetes. In apparently healthy participants, fasting glucose rose instead.

The numbers and sources

Effect: In 7 controlled trials in type 2 diabetes, ACV lowered fasting blood sugar by 21.9 mg/dL (95% CI -29.19 to -14.67) and HbA1c by 1.53 (95% CI -2.65 to -0.41), with no change in HOMA-IR [1]. Across 9 studies in adults, ACV lowered fasting glucose by 7.97 mg/dL (95% CI -13.74 to -2.21) and HbA1c by 0.50 (95% CI -0.90 to -0.09), with no change in fasting insulin or HOMA-IR [2]. A review of 6 studies in 317 people with type 2 diabetes found lower fasting glucose and HbA1c with vinegar [3].
Population: Adults with type 2 diabetes [1][3]. Adults with and without diabetes [2].
Sources: [1], [2], [3]
Caveats: Heterogeneity for HbA1c was high (I² 91%) [2]. Vinegar content varied across studies and samples were small [3]. In apparently healthy participants, fasting glucose rose [2].

BerberineStrong

Blood sugar in type 2 diabetes

Shown to lower fasting glucose and HbA1c in adults with type 2 diabetes, with HbA1c about 0.6 points lower. Many trials were small and low quality.

The numbers and sources

Effect: Across 37 trials (3,048 patients), berberine lowered fasting glucose by 0.82 mmol/L (95% CI -0.95 to -0.70), HbA1c by 0.63% (95% CI -0.72 to -0.53) and 2-hour glucose by 1.16 mmol/L, with larger effects at higher baseline glucose [1]. Against placebo in 20 trials (1,761 participants), it lowered fasting glucose by 0.52 mmol/L (95% CI -0.72 to -0.33) and HOMA-IR by 0.85 [2].
Population: Adults with type 2 diabetes [1]. Adults with and without diabetes [2].
Sources: [1], [2]
Caveats: Effects on fasting glucose were larger in people with diabetes and in Asian participants [2]. Many trials were small and of low methodological quality [3].

Black seed oilStrong

Blood sugar

Shown to lower fasting glucose by about 10 mg/dL and HbA1c by about 0.6 points in adults.

The numbers and sources

Effect: In 17 RCTs, fasting glucose fell 9.93 mg/dL (95% CI -13.44 to -6.41), post-meal glucose 14.79 mg/dL and HbA1c 0.57 percentage points (95% CI -0.77 to -0.37). Oil lowered fasting glucose more than powder [3]. The 2025 review also reported improved fasting glucose, HbA1c and post-meal glucose [2].
Population: Adults in clinical trials [2][3].
Sources: [3], [2]
Caveats: The authors call for longer, stronger trials [3].

ChromiumStrong

Blood sugar in type 2 diabetes

Shown to lower HbA1c by about 0.6 points and lower fasting glucose in people with type 2 diabetes, with larger effects in poorer-quality trials.

The numbers and sources

Effect: HbA1c fell 0.6 points (95% CI -0.9 to -0.2) and fasting glucose 1.0 mmol/L (95% CI -1.4 to -0.5) on chromium [1]. A 2014 analysis found HbA1c -0.55 points (95% CI -0.88 to -0.22) and fasting glucose -1.15 mmol/L [2]. A 2020 analysis found HbA1c -0.71 points (95% CI -1.19 to -0.23) and fasting glucose -19.0 mg/dL (95% CI -36.2 to -1.9) [3].
Population: 41 trials, almost half poor quality [1]; 25 trials in diabetes [2]; 28 trials in type 2 diabetes [3].
Sources: [1], [2], [3], [4]
Caveats: Sources disagree on what this means. Larger effects came more often from poor-quality trials [1]. Heterogeneity for fasting glucose was 99.8% [3]. A 2016 review found few of 20 trials reached clinical treatment goals, with HbA1c falling 0.5 points or more in 5 of 14 [4]. FDA ranked the evidence for preventing insulin resistance or diabetes at its lowest level for a qualified claim [5].

FenugreekStrong

Blood sugar in diabetes

Shown to lower fasting glucose by about 1 mmol/L and HbA1c by about 0.85% in people with diabetes, only at medium or high doses.

The numbers and sources

Effect: Across 10 trials, fenugreek seed changed fasting glucose by -0.96 mmol/L (95% CI -1.52 to -0.40, I² 80%), 2-hour glucose by -2.19 mmol/L and HbA1c by -0.85% (95% CI -1.49 to -0.22, 3 trials). Significant effects appeared only at medium or high doses in people with diabetes [1]. A 2023 meta-analysis of 10 RCTs (706 participants) with type 2 diabetes or prediabetes also reported lower fasting glucose, 2-hour glucose and HbA1c, and no change in HOMA-IR [2].
Population: People with and without diabetes [1]; type 2 diabetes and prediabetes [2].
Sources: [1], [2]
Caveats: Most trials in [1] were of low methodological quality. The 2023 abstract does not report effect sizes [2].

InositolStrong

Gestational diabetes prevention

Shown to lower gestational diabetes about 44% to 47% with myo-inositol in pregnant women. One trial in women with PCOS found no change in a combined outcome.

The numbers and sources

Effect: Myo-inositol lowered gestational diabetes (RR 0.53, 95% CI 0.31 to 0.90), hypertensive disorders of pregnancy (RR 0.34, 95% CI 0.19 to 0.61) and preterm birth (RR 0.35, 95% CI 0.17 to 0.70). Cesarean section did not differ [4]. A 2026 meta-analysis of 12 RCTs (4,765 women) found a 44% lower risk of gestational diabetes, with no significant difference in hypertensive disorders, cesarean delivery, macrosomia, preterm birth or newborn low blood sugar [5]. In one Italian trial in overweight women, gestational diabetes occurred in 11.6% on myo-inositol and 27.4% on placebo (OR 0.33, 95% CI 0.15 to 0.70) [6]. In a Dutch trial in pregnant women with PCOS, 2 g of myo-inositol twice a day did not lower a composite of gestational diabetes, preeclampsia or preterm birth (25.0% against 26.8%, RR 0.93, 95% CI 0.68 to 1.28) [7].
Population: 1,319 women in 7 trials, 6 in Italy and 1 in Ireland [4]; 4,765 pregnant women or women planning pregnancy, without diabetes, in 12 RCTs [5]; 220 overweight, non-obese pregnant women [6]; 464 pregnant women with PCOS at 13 hospitals [7].
Sources: [4], [5], [6], [7]
Caveats: The Cochrane authors rated the evidence low to very low certainty. No trial reported perinatal death [4]. The two reviews agree on gestational diabetes and differ on high blood pressure and preterm birth [4][5]. The overweight trial was open-label [6]. The PCOS trial reported a composite outcome, and the abstract does not give gestational diabetes alone [7].

TaurineStrong

Blood sugar and blood lipids

Shown to lower fasting glucose about 6 mg/dL and triglycerides about 14 to 18 mg/dL, mostly in people with obesity, diabetes or metabolic syndrome. HDL did not change.

The numbers and sources

Effect: Fasting glucose fell by 5.9 mg/dL (95% CI -10.7 to -1.0) and triglycerides by 18.3 mg/dL (95% CI -25.6 to -11.0). HDL cholesterol did not change [4]. A second meta-analysis found fasting glucose 5.9 mg/dL lower, HbA1c 0.21% lower, triglycerides 14.4 mg/dL lower and LDL cholesterol 5.1 mg/dL lower [5].
Population: 25 RCTs, 1,024 participants [4]; 34 RCTs [5].
Sources: [4], [5]
Caveats: Most trials were short and enrolled people with obesity, diabetes or metabolic syndrome. Neither review measured hard outcomes such as heart attacks or diabetes onset.

TurmericStrong

Blood sugar

Shown to lower fasting glucose about 6 to 10 mg/dL and HbA1c about 0.3 points in adults with prediabetes or type 2 diabetes.

The numbers and sources

Effect: Across 34 RCTs in adults with prediabetes or type 2 diabetes, curcumin or turmeric lowered fasting glucose by 10.15 mg/dL (95% CI -12.59 to -7.72), HbA1c by 0.32% (95% CI -0.43 to -0.21) and HOMA-IR by 0.46 against control [6]. A meta-analysis of 63 trials drawn from 13 earlier reviews (3,706 participants) found fasting glucose 6.30 mg/dL lower (95% CI -9.33 to -3.27) and HbA1c 0.31% lower (95% CI -0.57 to -0.05) [7]. With piperine added, 15 RCTs (1,020 participants) showed fasting glucose 5.89 mg/dL lower, while HbA1c, fasting insulin and HOMA-IR did not differ significantly [8].
Population: Adults with prediabetes or type 2 diabetes [6][7]; adults, many overweight with metabolic disorders [8].
Sources: [6], [7], [8], [9]
Caveats: Heterogeneity was significant [6] and high [7]. GRADE certainty was low for fasting glucose and HbA1c in the piperine review [8]. In a network meta-analysis of 24 RCTs in prediabetes, turmeric extract was linked to fewer new cases of type 2 diabetes (RR 0.05, 95% CI 0.00 to 0.78), with low certainty and short trials [9].

AstaxanthinModerate

Blood sugar and blood lipids

Shown to lower fasting glucose by about 16 mg/dL and triglycerides in people with prediabetes or type 2 diabetes. Reviews disagree on LDL and total cholesterol.

The numbers and sources

Effect: In prediabetes and type 2 diabetes, astaxanthin lowered fasting glucose by 16.1 mg/dL (95% CI -29.0 to -3.3), HbA1c by 0.34 points, triglycerides by 20.9 mg/dL, total cholesterol by 12.2 mg/dL and LDL by 9.4 mg/dL, and raised HDL by 3.0 mg/dL. HOMA-IR, weight and BMI did not change [4]. At 6 to 20 mg a day, a second meta-analysis found higher HDL and lower triglycerides, with no change in LDL or total cholesterol [5]. In adults at risk of metabolic syndrome, effects on total cholesterol and systolic pressure were marginal, and LDL fell [6].
Population: 403 patients in 9 RCTs [4]; 8 studies with 11 results [5]; 321 adults in 7 RCTs [6].
Sources: [4], [5], [6]
Caveats: The reviews disagree on LDL and total cholesterol. All pooled small trials.

BerberineModerate

Compared with metformin and other diabetes drugs

Shown to lower blood sugar about as much as metformin and other oral diabetes drugs in adults with type 2 diabetes, in mostly low-quality trials.

The numbers and sources

Effect: In 14 trials (1,068 participants), berberine did not give better glucose control than metformin, glipizide or rosiglitazone, and added to the same drug it gave better control than the drug alone [3]. A 27-trial meta-analysis (2,569 patients) found no significant difference between berberine and oral diabetes drugs [4]. In 36 adults with new type 2 diabetes, 0.5 g of berberine or metformin three times a day for 3 months gave similar glucose lowering, with HbA1c falling from 9.5% to 7.5% on berberine [5]. In 89 women with PCOS, berberine reduced waist-to-hip ratio and LDL more than metformin over 3 months [6].
Population: Adults with type 2 diabetes [3][4][5]. Chinese women with PCOS and insulin resistance [6].
Sources: [3], [4], [5], [6]
Caveats: Methodological quality was generally low [3]. The 36-person comparison was a pilot [5]. An integrity review rated every PCOS trial it found at high risk of compromised integrity [7].

Beta-glucanModerate

Blood sugar after meals

Shown to cut the blood sugar rise after a carbohydrate meal by 23% when oat beta-glucan is added, only at higher molecular weights. These were single-meal tests.

The numbers and sources

Effect: Across 103 trial comparisons (538 people), adding oat beta-glucan to carbohydrate meals cut the glucose rise (area under the curve) by 23% and the glucose peak by 28%. Insulin fell by 22%. Effects grew with dose and appeared only at molecular weights above 300 kg/mol. Results were similar with and without diabetes [5].
Population: Acute crossover feeding trials [5].
Sources: [5]
Caveats: Single-meal tests, not long-term blood sugar control. Two authors were employees of INQUIS Clinical Research, and 28 of the comparisons were industry funded [5].

ChromiumModerateStudied in women

Insulin resistance

Shown to lower HOMA-IR and fasting insulin slightly in overweight adults and women with PCOS. Fasting glucose and HbA1c did not change significantly.

The numbers and sources

Effect: In 20 trials in overweight and obese adults, HOMA-IR fell 0.26 (95% CI -0.48 to -0.03) and fasting insulin 12.55 pmol/L (95% CI -23.62 to -1.47). Fasting glucose (-1.60 mg/dL, 95% CI -3.28 to 0.07) and HbA1c (-0.05 points, 95% CI -0.19 to 0.07) did not change significantly [6]. In 10 RCTs in 683 women with polycystic ovary syndrome, chromium lowered fasting insulin (p 0.01) and raised the insulin sensitivity index QUICKI (p 0.02) against placebo. Against metformin, HOMA-IR fell more on chromium (p < 0.00001) [7].
Population: Overweight and obese adults, with trial sizes of 1 to 137 [6]; women with polycystic ovary syndrome [7].
Sources: [6], [7]
Caveats: The two reviews studied different populations. The PCOS abstract gives p-values without effect sizes [7]. The overweight analysis abstract calls the fasting glucose fall significant though its interval crosses zero [6]. In 2005 FDA ranked the evidence for chromium picolinate and insulin resistance at its lowest level for a qualified claim [5].

CLAModerate

CLA combined with exercise

Shown to lower body fat and insulin resistance more than exercise alone, mainly in women, people with a BMI of 25 or more and after 4 weeks.

The numbers and sources

Effect: In 20 trials of CLA plus exercise against exercise alone, body fat fell more (SMD -0.42, 95% CI -0.70 to -0.14, I² 65%) and insulin resistance fell more (SMD -0.25, 95% CI -0.44 to -0.06). Body weight, exercise performance and blood lipids did not differ [5].
Population: 18 RCTs and 2 crossover trials [5].
Sources: [5]
Caveats: Effects were seen mainly in subgroups with BMI of 25 or more, women and interventions longer than 4 weeks [5].

GarlicModerate

Blood sugar

Shown to lower fasting glucose, and HbA1c by about 0.5 points, more in people with type 2 diabetes. Insulin did not change.

The numbers and sources

Effect: In 9 RCTs (768 patients with type 2 diabetes), fasting glucose was lower with garlic at 1 to 2, 3 to 4, 12 and 24 weeks. Fructosamine and HbA1c were lower at 12 and 24 weeks. Complications did not differ [5]. A 2026 GRADE-assessed meta-analysis of 27 RCTs (1,619 adults) found fasting glucose fell 2.76 mg/dL (95% CI -5.11 to -0.41) and HbA1c 0.53 percentage points (95% CI -0.86 to -0.20). Insulin and HOMA-IR did not change significantly [6]. In 8 trials in type 2 diabetes, fasting glucose fell 12.41 mg/dL and HbA1c 0.5 points [7]. The 108-trial review found fasting glucose 2.77 mg/dL lower [4].
Population: Adults with type 2 diabetes [5][7]. Adults in 27 RCTs [6] and 108 trials [4].
Sources: [4], [5], [6], [7]
Caveats: The reviews pooled different populations, and the fasting glucose drop in the type 2 diabetes review [7] is more than four times that in the all-adult reviews. The pooled standardized differences in [5] are very large (SMD -21.02 at 24 weeks), which points to small, varied trials. Heterogeneity in [7] was very high (I² 99.2% for fasting glucose), and its authors rate the included trials as low quality.

GlucomannanModerate

Fasting blood sugar

Shown to lower fasting glucose about 7 mg/dL in adults. In insulin resistance, the result lost significance under another analysis.

The numbers and sources

Effect: Fasting blood glucose was 7.44 mg/dL lower with glucomannan (95% CI -14.16 to -0.72). People with impaired glucose metabolism did not benefit to the same degree [1]. In 11 trials in adults with insulin resistance-related conditions, fasting glucose was 7.02 mg/dL lower (95% CI -13.56 to -0.49) [3].
Population: 14 studies with 531 participants [1]. Adults with insulin resistance-related conditions in 11 trials [3].
Sources: [1], [3]
Caveats: The two reviews studied different populations. The result in [3] lost statistical significance under an alternative correlation assumption, and its authors call for more high-quality trials.

Grape seed extractModerate

Blood sugar

Shown to lower fasting glucose slightly in adults across 50 trials, with no change in HbA1c.

The numbers and sources

Effect: Across 50 trials, grape seed extract lowered fasting plasma glucose (WMD -2.01, 95% CI -3.14 to -0.86). HbA1c did not change [5].
Population: Adults in RCTs [5].
Sources: [5]
Caveats: The abstract does not state the units or the number of trials for each outcome [5].

Green tea extractModerate

Blood sugar

Shown to lower fasting glucose, HbA1c and insulin resistance modestly with EGCG supplements in adults, with HbA1c down about 0.18 points.

The numbers and sources

Effect: In a 2025 meta-analysis of 41 RCTs, EGCG supplements lowered fasting glucose, HbA1c and HOMA-IR by modest amounts. HbA1c fell 0.18 percentage points (95% CI -0.35 to -0.02), and fasting insulin did not change significantly (95% CI -1.46 to 0.47) [6].
Population: Adults in 41 RCTs of EGCG [6].
Sources: [6]
Caveats: The authors describe the HbA1c and insulin effects as modest and possibly not clinically meaningful for most people [6].

InositolModerateStudied in women

Hormones and blood sugar in PCOS

Shown to lower free and total testosterone and glucose and raise SHBG in women with PCOS. Metformin did better on hirsutism.

The numbers and sources

Effect: Against placebo, inositol lowered BMI (MD -0.45), free testosterone (MD -0.41), total testosterone (MD -20.4) and glucose (MD -3.14), and raised sex-hormone-binding globulin (MD 32.1) [1]. The guideline review found benefits for some metabolic measures and none for others. Metformin did better on waist-hip ratio and hirsutism [3].
Population: 1,691 women in 26 trials [1]; 2,230 women in 30 trials [3].
Sources: [1], [3]
Caveats: Units are as reported in the abstract. The guideline review rated BMI evidence very uncertain [3].

InositolModerate

Weight, blood sugar and lipids

Shown to lower BMI slightly, about 0.6 kg/m², and glucose, triglycerides and blood pressure in adults across several conditions, at low certainty for some.

The numbers and sources

Effect: In 18 RCTs (898 participants) of at least 4 weeks, inositol lowered BMI by 0.57 kg/m² (95% CI -1.10 to -0.03), waist circumference by 2.36 cm, glucose by 7.25 mg/dL, HOMA-IR by 1.21, triglycerides by 29.80 mg/dL, LDL cholesterol by 5.15 mg/dL, and systolic and diastolic pressure by 5.34 and 6.12 mm Hg [11].
Population: Adults in RCTs across several conditions [11].
Sources: [11]
Caveats: Certainty was low to very low for BMI, waist and blood pressure, and moderate for glucose, insulin and HOMA-IR. Heterogeneity was high for most outcomes, and the authors note a high risk of bias in the trials [11].

Milk thistleModerate

Blood sugar in type 2 diabetes

Shown to lower fasting glucose about 27 mg/dL and HbA1c about 1 point in adults with type 2 diabetes, in low quality trials.

The numbers and sources

Effect: In 5 trials (270 patients), silymarin lowered fasting blood glucose by 26.86 mg/dL (95% CI -35.42 to -18.30) and HbA1c by 1.07 points (95% CI -1.73 to -0.40). Lipids did not change [9]. A 2025 network meta-analysis of 17 studies (1,337 people with type 2 diabetes) found silymarin lowered fasting glucose, HOMA-IR and HbA1c against placebo (P < 0.05) [10]. Across 6 RCTs (673 participants), HOMA-IR fell by 2.29 (95% CI -4.55 to -0.03) with no change in fasting insulin. The fall was seen in type 2 diabetes but not in fatty liver disease [11].
Population: Adults with type 2 diabetes [9][10]; adults with type 2 diabetes, alcoholic cirrhosis or fatty liver disease in Iran, Egypt and Italy [11].
Sources: [9], [10], [11]
Caveats: The authors of [9] rated the evidence as low quality, with high heterogeneity. The network meta-analysis abstract reports p-values for silymarin against placebo without effect sizes [10]. The authors of [11] called the evidence limited.

Vitamin DModerate

Type 2 diabetes prevention

Shown to lower new type 2 diabetes about 15% in adults with prediabetes. In older adults not selected for prediabetes, VITAL found no significant difference.

The numbers and sources

Effect: In 3 trials of adults with prediabetes, vitamin D lowered new-onset diabetes by 15% (HR 0.85, 95% CI 0.75 to 0.96), a 3-year absolute risk reduction of 3.3%. Return to normal glucose regulation rose (rate ratio 1.30, 95% CI 1.16 to 1.46) [28]. In VITAL, among 22,220 older US adults not selected for prediabetes, type 2 diabetes incidence was 3.98 per 1,000 person-years on 2,000 IU a day and 4.37 on placebo (HR 0.91, 95% CI 0.76 to 1.09). Pooled with 3 earlier trials, the HR was 0.89 (95% CI 0.80 to 0.99) [29].
Population: Adults with prediabetes in trials of 20,000 IU a week, 4,000 IU a day or eldecalcitol [28]; US adults with mean age 67.2 over a median 5.3 years [29].
Sources: [28], [29]
Caveats: The two pooled analyses share most of their prediabetes trials. Results in prediabetes do not apply to the general population [28], where VITAL found no significant difference [29].

Vitamin DModerateStudied in women

Blood sugar and insulin in PCOS

Shown in one review to lower fasting glucose and insulin slightly against placebo in women with PCOS. An earlier review found changes mainly with other nutrients.

Compared with placebo [71][72]

The numbers and sources

Effect: Across 13 RCTs (691 women), vitamin D lowered fasting glucose by 2.91 mg/dL (95% CI 1.04 to 4.78) and insulin by 1.98 µIU/mL (95% CI 0.64 to 3.32) more than placebo, with lower triglycerides, total cholesterol and LDL cholesterol [71]. An earlier review of 11 RCTs (601 women) found lower fasting glucose and HOMA-IR when vitamin D was given with other micronutrients, and lower HOMA-IR at doses under 4,000 IU a day [72].
Population: Women aged 18 and older with PCOS in 13 RCTs [71]; 601 women with PCOS in 11 RCTs [72].
Sources: [71], [72]
Caveats: The changes are small. The earlier review's abstract gives no effect size and does not report a significant effect for vitamin D alone at all doses [72].

ZincModerate

Blood sugar

Shown to lower fasting glucose about 14 mg/dL and HbA1c about 0.55% in people with and without diabetes.

The numbers and sources

Effect: Across 32 placebo-controlled interventions (1,700 participants, 14 countries), zinc lowered fasting glucose by 14.15 mg/dL (95% CI -17.36 to -10.93), 2-hour glucose by 36.85 mg/dL, HbA1c by 0.55% (95% CI -0.84 to -0.27) and hs-CRP by 1.31 mg/L [3]. Across 18 RCTs (1,023 participants) with diabetes, gestational diabetes or prediabetes, zinc lowered serum insulin (MD -2.50, 95% CI -4.69 to -0.31) and HOMA-IR (MD -1.10, 95% CI -2.05 to -0.15), and LDL cholesterol rose slightly (MD 3.46, 95% CI 1.48 to 5.43) [4].
Population: People with and without diabetes [3]; people with diabetes or prediabetes [4].
Sources: [3], [4]
Caveats: The fasting glucose effect depended on diabetes status and on the zinc formulation [3].

Fiber supplementsLimited

Blood sugar

Limited evidence: psyllium lowered fasting glucose and HbA1c in type 2 diabetes but not in normal blood sugar. Other fiber results were small or inconsistent.

The numbers and sources

Effect: In people treated for type 2 diabetes, psyllium taken before meals lowered fasting glucose by 37.0 mg/dL and HbA1c by 0.97%. There was no glucose lowering in people with normal blood sugar and a modest effect in prediabetes [5]. Across 12 RCTs in overweight and obese adults, isolated soluble fiber lowered fasting glucose by 0.17 mmol/L (95% CI -0.28 to -0.06) and fasting insulin by 15.88 pmol/L [6]. A 2026 meta-analysis of 11 glucomannan trials in adults with insulin resistance-related conditions found fasting glucose 7.02 mg/dL lower (95% CI -13.56 to -0.49), an effect that lost significance under an alternative correlation assumption [7]. In a 4-week trial of 131 adults, inulin lowered 1-hour and 2-hour glucose on a glucose tolerance test in overweight or obese participants but not in healthy-weight ones, and fructooligosaccharides did not change glucose in either group [8].
Population: 35 psyllium RCTs across three groups [5]; 609 overweight and obese adults [6]; 11 glucomannan trials [7]; 131 adults in China [8].
Sources: [5], [6], [7], [8]
Caveats: Reference [5] searched clinical records held by Procter & Gamble, a psyllium supplement maker, as well as published studies, and its abstract does not report confidence intervals. The glucomannan result was not robust to the analysis method [7]. In the inulin trial, fasting insulin rose [8].

Fisetin and quercetin (senolytics)LimitedStudied in men

Inflammation and metabolic markers (fisetin alone)

Small trials found lower inflammation and insulin resistance markers with 200 mg a day of fisetin in men with obesity, most with training.

The numbers and sources

Effect: Over 12 weeks, IL-6 and TNF-α fell in the fisetin-only group, and fasting glucose, insulin and HOMA-IR fell in all active groups, most in training plus fisetin [11]. A second report from the same research group found weight loss in the fisetin-only group compared with placebo, and the largest drops in asprosin and MCP-1 with training plus fisetin [12].
Population: 44 men with obesity at 200 mg a day [11]; 60 sedentary men with obesity at 200 mg a day [12].
Sources: [11], [12]
Caveats: Both reports come from one research group with similar four-arm designs. Neither measured senescent cells.

Alpha-lipoic acidModerate

Blood sugar

Mixed results: one large review found lower fasting glucose and HbA1c, while a review of overweight adults and neuropathy trials found no change.

The numbers and sources

Effect: Across 63 trials, fasting glucose fell 5.28 mg/dL (95% CI -7.21 to -3.35), HbA1c 0.40% (95% CI -0.66 to -0.14), fasting insulin 1.70 mU/mL and HOMA-IR 0.74 against control [8]. In 11 trials of 704 overweight or obese adults, fasting glucose (SMD 0.13, 95% CI -0.16 to 0.41) and HOMA-IR (SMD -0.23, 95% CI -0.60 to 0.15) did not differ from control [9]. In diabetic neuropathy trials, HbA1c did not change [5].
Population: 63 randomized trials [8]; adults with a BMI of 25 or more [9]; people with diabetic neuropathy [5].
Sources: [8], [9], [5]
Caveats: The reviews disagree. The overweight review rated 8 of its 11 trials at low risk of bias [9].

GinsengModerate

Blood sugar

Mixed results: one review found fasting glucose about 0.3 mmol/L lower, and a larger review found no significant effect on blood sugar.

The numbers and sources

Effect: Across 16 trials of 30 days or longer, ginseng lowered fasting blood glucose by 0.31 mmol/L (95% CI -0.59 to -0.03). Fasting insulin, HbA1c and HOMA-IR did not change overall [5]. A 2025 meta-analysis of 70 studies (4,506 participants) on cardiovascular risk factors found no significant effect on the glycemic profile, with a dose-response association for fasting glucose and fasting insulin [6].
Population: 770 people with and without diabetes [5]. 4,506 participants in 70 studies [6].
Sources: [5], [6]
Caveats: The two reviews disagree on fasting glucose. Two thirds of trials in [5] ran under 12 weeks, and most participants already had well-controlled glucose [5].

MagnesiumModerate

Blood sugar and insulin resistance

Mixed results: most reviews found lower insulin resistance but disagreed on fasting glucose, and one review in people with diabetes found no significant change.

The numbers and sources

Effect: One meta-analysis found magnesium lowered HOMA-IR (WMD -0.67, 95% CI -1.20 to -0.14) but not fasting glucose (WMD -0.20 mmol/L, 95% CI -0.45 to 0.05) or HbA1c overall. Trials of 4 months or longer showed lower fasting glucose and HOMA-IR [3]. A 2021 meta-analysis of double-blind trials reported lower fasting glucose in people with diabetes and improved glucose and insulin-sensitivity markers in people at high risk [4]. A 2026 meta-analysis of 78 RCTs found lower HOMA-IR (WMD -0.48, 95% CI -0.79 to -0.18), fasting glucose (WMD -3.60 mg/dL) and HbA1c (WMD -0.15%) [5]. A 2026 meta-analysis of 15 RCTs (1,085 people with diabetes or prediabetes) found no significant change in insulin (MD -1.73) or HOMA-IR (MD -0.74, 95% CI -1.58 to 0.09). Baseline insulin and HOMA-IR modified the effect [6].
Population: People with and without diabetes [3][5]; people with or at high risk of diabetes [4][6].
Sources: [3], [4], [5], [6]
Caveats: The sources disagree on fasting glucose and on HOMA-IR in people with diabetes. The 2021 abstract does not report effect sizes [4]. The authors of [5] rated certainty as variable across outcomes.

ResveratrolModerate

Blood sugar and insulin sensitivity

Mixed results: one review found lower glucose and insulin resistance in people with diabetes, while a later review found no effect. No effect without diabetes.

The numbers and sources

Effect: Across 11 RCTs, resveratrol reduced fasting glucose, insulin, HbA1c and HOMA insulin resistance in participants with diabetes. No significant effect on any glycemic measure was found in participants without diabetes [1]. A 2026 umbrella review pooling 10 meta-analyses in type 2 diabetes found no significant effect on fasting glucose, HbA1c, insulin or HOMA insulin resistance [2].
Population: 388 participants in 11 RCTs [1]. Meta-analyses of trials in type 2 diabetes [2].
Sources: [1], [2]
Caveats: The two sources disagree on people with diabetes [1][2]. The pooled effects in people without diabetes did not vary by BMI, dose, duration or trial quality [1]. Heterogeneity was significant in [2].

ChromiumModerate

Blood sugar in people without diabetes

Not shown to change glucose or lipid metabolism in people without diabetes.

The numbers and sources

Effect: Chromium did not change glucose or lipid metabolism in people without diabetes [1].
Population: Trials of people without diabetes within a 41-trial review [1].
Sources: [1]
Caveats: The review notes no consensus on how to measure chromium status [1].

Garcinia cambogiaModerate

Blood sugar and insulin

Not shown to change fasting blood sugar or insulin in adults. Subgroup results pointed in different directions.

The numbers and sources

Effect: Across 9 publications with 444 participants, garcinia did not change fasting blood sugar (1.02 mg/dL, 95% CI -1.29 to 3.33) or insulin (-0.12 mU/L, 95% CI -1.50 to 1.25) compared with control. In subgroups, insulin rose in women, insulin and fasting blood sugar rose in those with BMI of 30 or more, and insulin fell in trials longer than 8 weeks [6].
Population: Adults in RCTs [6].
Sources: [6]
Caveats: Subgroup results point in different directions and come from few trials.

NMNModerate

Insulin sensitivity and metabolic markers

Not shown to change weight, glucose, HbA1c or lipids across trials. One small trial found better insulin sensitivity in women with prediabetes.

The numbers and sources

Effect: In women with prediabetes and overweight or obesity, insulin-stimulated glucose disposal measured by clamp rose after 10 weeks of NMN and did not change on placebo [7]. Across 15 trials, NMN did not change body weight, BMI, fasting glucose, HbA1c or lipids. HOMA-IR showed a downward trend that was not significant [8]. In the 60-day dose trial, HOMA-IR did not differ from placebo [1].
Population: Postmenopausal women with prediabetes [7]; 15 RCTs at 250 to 2,000 mg a day for 14 days to 24 weeks [8]; 80 middle-aged adults [1].
Sources: [7], [8], [1]
Caveats: The clamp result comes from a single small trial in one population [7]. The pooled trials were short [8].

ReishiModerate

Blood sugar, cholesterol and blood pressure

Not shown to change HbA1c, fasting glucose, cholesterol or blood pressure against placebo, with one spore oil trial finding lower lipids.

The numbers and sources

Effect: In a Cochrane review of 5 trials (398 participants), 1.4 to 3 g a day for 12 to 16 weeks did not change HbA1c (WMD -0.10%, 95% CI -1.05 to 0.85), total cholesterol, LDL or BMI against placebo. Fasting glucose, blood pressure and triglycerides did not differ. Post-meal glucose results were inconsistent [2]. A 2025 meta-analysis of 17 RCTs (971 participants) at 200 to 11,200 mg a day for 1 to 24 weeks found no effect on blood pressure, fasting glucose or the lipid profile. BMI fell by 0.43 (95% CI 0.10 to 0.77) and heart rate by 3.92 beats a minute (95% CI 0.40 to 7.45) [3]. In 110 adults with dyslipidemia, a spore oil extract for 12 weeks lowered total cholesterol, triglycerides and LDL and raised HDL compared with placebo. The abstract gives no effect sizes [4].
Population: Adults with type 2 diabetes in the analyzed trials [2]. Healthy, at-risk and chronically ill adults [3]. Adults with dyslipidemia [4].
Sources: [2], [3], [4]
Caveats: Data in [2] came from 3 small trials, and risk of bias was unclear in 4 of 5 [2]. The 2025 review rated the certainty of evidence very low for every outcome [3]. The spore oil trial tested a different product and disagrees with both reviews on lipids [4].

Beta-glucanLimited

Long-term blood sugar (HbA1c)

One trial found 6 g a day of oat beta-glucan in bread did not change HbA1c over 16 weeks in adults at risk of type 2 diabetes.

The numbers and sources

Effect: In 194 adults at risk of type 2 diabetes, bread enriched to 6 g of oat beta-glucan a day for 16 weeks did not change HbA1c compared with whole-grain wheat bread (difference -0.01%, 95% CI -0.03 to 0.06). Fasting glucose, insulin and LDL cholesterol (-0.11 mmol/L, 95% CI -0.27 to 0.05) did not differ either [6].
Population: Adults aged about 58 with a mean BMI of 32 in Germany, Norway and Sweden [6].
Sources: [6]
Caveats: One trial, run as an effectiveness trial in real-life conditions, against whole-grain wheat bread, not a fiber-free control [6].

Nicotinamide ribosideLimitedStudied in men

Insulin sensitivity and metabolism

One trial found no change in insulin sensitivity, energy expenditure or body composition in men with obesity and insulin resistance.

The numbers and sources

Effect: Over 12 weeks at 2,000 mg a day, NR did not improve insulin sensitivity, glucose production or glucose disposal measured by clamp. Resting energy expenditure, fat oxidation and body composition did not change [7].
Population: 40 sedentary men with obesity and insulin resistance, aged 40 to 70 [7].
Sources: [7]
Caveats: One investigator-initiated trial, in men only [7].

DandelionUntested

Blood sugar

Not studied in people. In mice on a high-fat diet a leaf extract lowered fasting glucose and insulin, and root extract blocked starch-digesting enzymes in the lab.

No study in people. This rests on animal studies, lab studies.

The numbers and sources

Effect: In mice fed a high-fat diet, dandelion leaf extract lowered fasting glucose, insulin and HOMA-IR insulin resistance compared with the high-fat diet alone [5]. A water extract of dandelion root inhibited the enzymes alpha-glucosidase and alpha-amylase, and eased insulin resistance in liver cells grown in the lab [7].
Population: C57BL/6 mice [5]. Enzyme assays and HepG2 liver cells [7].
Sources: [5], [7]
Caveats: No trial of dandelion and blood sugar in people was located. Animal and cell results do not show an effect in people.

Practices

YogaModerate

Stress hormones, heart rate and blood fats, against active controls

Shown in one meta-analysis to lower cortisol, ambulatory systolic pressure, resting heart rate, fasting glucose and LDL cholesterol more than active control activities.

Compared with active control interventions [11]

The numbers and sources

Effect: Across 42 trials, practices including yoga postures were associated with lower evening and waking cortisol, ambulatory systolic blood pressure, resting heart rate, fasting blood glucose, cholesterol and LDL cholesterol, and higher high-frequency heart rate variability, compared with active controls [11].
Population: Participants in 42 randomized trials in various populations [11].
Sources: [11]
Caveats: The abstract gives no effect sizes and describes the interventions as heterogeneous. Some included mindfulness-based stress reduction alongside postures [11].

YogaModerate

Blood sugar and blood fats in type 2 diabetes

Shown in one meta-analysis to lower HbA1c by 0.47 points and fasting glucose by 24 mg/dL in adults with type 2 diabetes, with high heterogeneity.

Compared with control groups including usual care or standard treatment [12]

The numbers and sources

Effect: Pooled differences were -23.72 mg/dL for fasting blood glucose (95% CI -37.78 to -9.65, I² 82%), -0.47% for HbA1c (95% CI -0.87 to -0.07, I² 82%), -17.38 mg/dL for glucose after meals, -18.50 mg/dL for total cholesterol, +4.30 mg/dL for HDL cholesterol and -12.95 mg/dL for LDL cholesterol. Triglycerides did not differ (-12.57 mg/dL, 95% CI -29.91 to 4.76) [12].
Population: 864 adults with type 2 diabetes in 12 trials [12].
Sources: [12]
Caveats: The authors note limited methods and heterogeneity, and long-term effects were not studied [12].

YogaModerateStudied in women

Cycles, androgens and blood sugar in PCOS

Shown in one small meta-analysis to lower menstrual irregularity, signs of excess androgens, fasting glucose and insulin resistance in women with PCOS, at low strength of evidence.

Compared with controls the source abstract does not describe [21]; conventional physical activity, routine care or no intervention [22]

The numbers and sources

Effect: Pooling 2 RCTs, yoga lowered menstrual irregularity (MD -0.41, 95% CI -0.74 to -0.08), clinical hyperandrogenism (MD -0.70, 95% CI -1.15 to -0.26), fasting glucose (MD -0.22 mmol/L, 95% CI -0.44 to -0.01), fasting insulin (MD -28.21 pmol/L, 95% CI -43.79 to -12.63) and HOMA-IR (MD -0.86, 95% CI -1.29 to -0.43) [21]. A 2026 review of trials with 258 participants found fasting glucose fell 4.50 mg/dL more than in controls in one trial (95% CI 2.39 to 6.61) and did not differ significantly in another. In one trial, testosterone did not change significantly and LH fell [22].
Population: 515 women with PCOS in 11 studies, 2 RCTs pooled [21]; 258 women with PCOS in RCTs [22].
Sources: [21], [22]
Caveats: The authors of [21] rated the strength of evidence low. [22] could not pool its trials, and three of its five reports came from one trial.

Cold water immersionLimited

Insulin sensitivity

Limited evidence: insulin sensitivity rose over a season of cold water swimming in women and lean swimmers, in one uncontrolled study of 30 people.

The numbers and sources

Effect: Among 30 people swimming in cold water at least twice a week for 6 months, insulin sensitivity increased and insulin secretion and resistance decreased from the start to the middle of the season in women and in lean participants [26]. A review of 104 studies concluded that cold water immersion seems to reduce insulin resistance and improve insulin sensitivity [25].
Population: 30 healthy adults with a mean age of 50 [26]. 104 studies of voluntary cold water exposure [25].
Sources: [26], [25]
Caveats: The change appeared in subgroups, and swimming is also exercise. The review notes most studies were small and often in one sex [25].

GroundingLimited

Blood chemistry and thyroid tests

One set of small experiments found grounding during night rest lowered serum iron, free T3 and several electrolytes and raised free T4, TSH and ferritin.

Compared with a group not in contact with the earth [9]

The numbers and sources

Effect: Earthing during night rest lowered serum iron, ionized calcium and inorganic phosphorus, and lowered urinary calcium and phosphorus. It decreased free T3 and increased free T4 and TSH. Continuous earthing decreased blood glucose in patients with diabetes. Earthing decreased sodium, potassium, magnesium, total protein and albumin, and increased transferrin, ferritin and globulins [9].
Population: Five experiments of 12 to 84 people each [9].
Sources: [9]
Caveats: The abstract gives no sizes, no durations for each experiment and no clinical meaning for the changes. Nobody uses grounding to shift these tests, and the direction is neither a benefit nor a harm that the authors define. One author is a co-author of a review of grounding's proposed benefits [10].

Infrared saunaLimited

Blood sugar after a meal in type 2 diabetes

One trial found a single 40-minute infrared sauna session did not improve blood sugar handling in 12 people with type 2 diabetes. The glucose rise was larger after heat.

The numbers and sources

Effect: The incremental area under the glucose curve during a 2-hour oral glucose tolerance test was 17.7 mmol/L/120 min after the 60°C condition and 14.8 after the 21°C condition (P < 0.001). Insulin concentrations (P = 0.93) and insulin sensitivity index (P = 0.67) did not differ [24].
Population: 12 people with type 2 diabetes, mean age 69, in a randomized crossover [24].
Sources: [24]
Caveats: A single session in 12 people. Repeated use was not tested.

Daily life

Physical activityStrong

Type 2 diabetes

Linked to 26% less type 2 diabetes at 150 minutes a week of moderate activity in pooled cohorts, with further but smaller drops at higher amounts.

Compared with inactive individuals [5]; the lowest activity group [6]; inactive people [3]; insufficiently active people [4]

The numbers and sources

Effect: At 11.25 MET-hours a week, about 150 minutes of moderate activity, type 2 diabetes risk was 26% lower (95% CI 20% to 31%). Twice that amount was linked to 36% lower risk and 60 MET-hours a week to 53% lower [5]. High against low leisure-time activity had an RR of 0.74 (95% CI 0.70 to 0.79), and walking 0.85, with steeper drops at low activity levels [6]. A third review found 26% lower risk at 150 minutes a week (RR 0.74) [3]. In another, raising activity from 9,000 to 12,000 MET-minutes a week lowered risk by only 0.6% more [4].
Population: 28 prospective studies [5]. 81 studies [6]. 3 studies of diabetes [3]. 55 studies of diabetes [4].
Sources: [5], [6], [3], [4]
Caveats: Observational. Part of the link may run through body weight [6]. Heterogeneity was high for leisure-time activity (I² 84%) [6].

Sitting timeStrong

Type 2 diabetes

Linked to more type 2 diabetes with more sitting or TV time in pooled cohorts, with a steady rise and no threshold.

Compared with people with less sedentary time, with physical activity held equal [1]; each hour less of sitting or TV a day [3]; each 2 hours less of TV a day [6]

The numbers and sources

Effect: Sedentary time was linked to new type 2 diabetes (HR 1.910, 95% CI 1.642 to 2.222) [1]. Each hour of total sitting carried 1% more risk (RR 1.01, 95% CI 1.00 to 1.01) and each hour of TV 9% more (RR 1.09, 95% CI 1.07 to 1.12), both linear [3]. Each 2 hours of TV was linked to 20% more type 2 diabetes (RR 1.20, 95% CI 1.14 to 1.27) [6].
Population: 14 studies [1]. 34 studies [3]. 175,938 people with 6,428 cases in 4 cohorts [6].
Sources: [1], [3], [6]
Caveats: Observational. Body weight, diet and snacking while watching TV are likely confounders.

Sitting timeStrong

Blood sugar after meals when sitting is broken up

Shown to lower blood sugar and insulin after meals when long sitting was broken up with short activity breaks, in two meta-analyses of short trials.

Compared with continuous sitting [7]; prolonged sitting without breaks [8]

The numbers and sources

Effect: Activity breaks lowered post-meal glucose (SMD -0.54, 95% CI -0.70 to -0.37) and insulin (SMD -0.56, 95% CI -0.74 to -0.38), with larger glucose effects at higher BMI. Triglycerides fell slightly (SMD -0.26), with possible publication bias. Against one continuous exercise bout of matched energy, breaks lowered glucose slightly more (SMD -0.26) [7]. A second review found lower glucose (d -0.36, 95% CI -0.50 to -0.21) and insulin (d -0.37), and no change in triglycerides (d 0.06, 95% CI -0.15 to 0.26) [8].
Population: 37 studies in adults who were metabolically healthy or impaired [7]. 20 studies in apparently healthy people [8].
Sources: [7], [8]
Caveats: Most trials lasted a single day in a laboratory. Post-meal readings are not fasting glucose or HbA1c, and a day's effect does not show long-term change. The two reviews disagree on triglycerides.

Sleep durationStrong

Type 2 diabetes

Linked to more type 2 diabetes at both short and long sleep in pooled cohorts, a U shape with the lowest risk at 7 to 8 hours a night.

Compared with people sleeping 7 hours a day [5]; normal sleepers [3]; people with normal sleep duration [4]

The numbers and sources

Effect: Against 7 hours, each hour shorter among people sleeping under 7 hours was linked to 9% more type 2 diabetes (RR 1.09, 95% CI 1.04 to 1.15), and each hour longer among long sleepers to 14% more (RR 1.14, 95% CI 1.03 to 1.26) [5]. Short sleepers had a 37% higher risk (RR 1.37, 95% CI 1.22 to 1.53) [3]. Long sleepers had a 26% higher risk (RR 1.26, 95% CI 1.11 to 1.43) [4].
Population: 482,502 participants with 18,443 cases, follow-up 2.5 to 16 years [5]. 5,172,710 participants in 153 studies [3]. 5,134,036 participants in 137 cohorts [4].
Sources: [5], [3], [4]
Caveats: Observational. Confounding by weight, activity and mood is likely, and reverse causation is possible because early metabolic disease and its symptoms can disturb sleep.

Strength trainingStrong

Type 2 diabetes

Linked to 17% less type 2 diabetes with muscle-strengthening activity in pooled cohorts, with most of the drop by 60 minutes a week.

Compared with people doing no muscle-strengthening activity [1]; lower amounts of resistance exercise [4]; men doing no weight training [5]

The numbers and sources

Effect: Muscle-strengthening activity was linked to lower diabetes risk (RR 0.83, 95% CI 0.77 to 0.89), an L shape with a large drop up to 60 minutes a week [1]. A second review found an inverse, nonlinear link between resistance exercise and diabetes [4]. In US men, 150 minutes a week or more of weight training was linked to 34% lower risk (95% CI 7% to 54%), and combined with the same amount of aerobic exercise to 59% lower [5].
Population: 202,486 adults with 9,548 cases in 5 cohorts [1]. 81 studies [4]. 32,002 men followed 18 years, with 2,278 cases [5].
Sources: [1], [4], [5]
Caveats: Observational. Certainty was rated low [1]. People who lift weights may differ in diet and body weight.

Strength trainingStrong

HbA1c and fasting glucose in type 2 diabetes

Shown to lower HbA1c by about 0.4% to 0.6% with resistance training in type 2 diabetes in pooled trials, more at high intensity, without a clear fasting glucose change.

Compared with control groups in randomized trials [6][7][8]; aerobic training [6]

The numbers and sources

Effect: Resistance training lowered HbA1c by 0.39% against controls (95% CI -0.60 to -0.18), more when strength rose more, and did not differ from aerobic training [6]. High-intensity training lowered HbA1c by 0.61% and low to moderate intensity by 0.23%. Fasting blood glucose fell only as a nonsignificant trend across 15 trials [7]. A third review found a 0.57% fall (95% CI -1.14% to -0.01%) [8].
Population: 1,172 adults with type 2 diabetes in 20 trials [6]. 962 patients in 24 trials, 443 in the fasting glucose analysis [7]. Patients with type 2 diabetes in trials of at least 12 weeks [8].
Sources: [6], [7], [8]
Caveats: Heterogeneity was high (I² 69.2% [6], 92.5% [8]).

Physical activityModerate

HbA1c in people with type 2 diabetes

Shown to lower HbA1c by 0.67% with structured exercise training in type 2 diabetes, and by 0.89% with more than 150 minutes a week.

Compared with control groups in randomized trials [7]

The numbers and sources

Effect: Structured exercise training lowered HbA1c by 0.67% (95% CI -0.84% to -0.49%) against control groups. Aerobic training lowered it 0.73%, resistance training 0.57% and both combined 0.51%. More than 150 minutes a week lowered it 0.89%, and 150 minutes or less 0.36%. Activity advice lowered HbA1c only when combined with diet advice (-0.58%) [7].
Population: 8,538 patients with type 2 diabetes in 47 randomized trials of at least 12 weeks [7].
Sources: [7]
Caveats: Heterogeneity was very high (I² 91.3%) [7].

Physical activityModerateStudied in women

Insulin resistance and weight in PCOS, in trials

Shown in trials to lower fasting insulin and insulin resistance slightly in women with PCOS. Fasting glucose did not change, at very low to low quality.

Compared with usual care or control groups [27]; before exercise, including trials with no control group [28]

The numbers and sources

Effect: Against control, exercise lowered fasting insulin (MD -2.44 µIU/mL, 95% CI -4.24 to -0.64) and HOMA-IR (MD -0.57, 95% CI -0.99 to -0.14), both very low quality, and LDL cholesterol and triglycerides. Waist fell 2.62 cm and BMI 1.02 kg/m2 after the programs. Fasting glucose and blood pressure did not change. Exercise added to diet did not differ from diet alone [27]. Vigorous exercise was linked to a 36.2% fall in HOMA-IR, a 4.2% fall in waist circumference and a 24.2% rise in VO2peak [28].
Population: 18 trials in women with PCOS [27]; 777 women in 19 studies, including uncontrolled trials, of 6 to 26 weeks [28].
Sources: [27], [28]
Caveats: Many effects were modest with wide intervals and shifted when single trials were added or removed [27]. [28] pooled controlled and uncontrolled studies.

Sitting timeModerate

Long-term programs to cut sitting

Shown to cut sitting by about an hour a day and lower HbA1c 0.17%, body fat and waist size in trials of people with chronic conditions.

Compared with control interventions [9]; control offices whose workers were told to behave as usual [10]

The numbers and sources

Effect: Lifestyle programs cut sedentary time by 64 minutes a day (95% CI -91 to -38). HbA1c fell 0.17% (95% CI -0.30 to -0.04), body fat 0.66 percentage points and waist circumference 1.52 cm against control groups [9]. In office workers, a workplace program cut sitting by 71 minutes per 8-hour workday at 1 month and 48 minutes at 3 months, and body fat was 0.61 percentage points lower at 3 months [10].
Population: 18 randomized trials in people with overweight, type 2 diabetes, heart, neurological or musculoskeletal disease [9]. 317 office workers in 19 offices, cluster randomized [10].
Sources: [9], [10]
Caveats: Programs often added counseling and activity, so the effect of less sitting alone cannot be separated out. Follow-up was months, not years.

Time in natureModerate

Type 2 diabetes

Linked to 28% less type 2 diabetes with more greenspace exposure in one meta-analysis of mostly observational studies.

Compared with people with lower greenspace exposure [2]

The numbers and sources

Effect: Greenspace exposure was linked to a lower risk of type 2 diabetes (0.72, 95% CI 0.61 to 0.85) [2].
Population: Studies of greenspace exposure within 103 observational and 40 interventional studies [2].
Sources: [2]
Caveats: Observational. Wealthier, more active people tend to live near green space, which could explain the link.

Daily stepsModerate

Type 2 diabetes

Mixed results: one meta-analysis linked 7,000 steps a day to 14% less type 2 diabetes than 2,000, while an earlier review found inconsistent results for blood sugar outcomes.

Compared with 2,000 steps a day [4]; lower step counts at baseline, as each study compared them [7]

The numbers and sources

Effect: Against 2,000 steps a day, 7,000 was linked to 14% lower risk of type 2 diabetes (HR 0.86, 95% CI 0.74 to 0.99), with a linear pattern [4]. An earlier systematic review of 8 studies of diabetes, insulin sensitivity, fasting glucose and HbA1c reported inconsistent findings, which it put down partly to differences in markers, analysis and samples [7].
Population: 4 studies in the diabetes meta-analysis [4]. Over 30,000 adults in 17 prospective studies, 8 on blood sugar outcomes with 3 months to 5 years of follow-up [7].
Sources: [4], [7]
Caveats: Observational and few studies. The upper confidence limit sits close to no difference [4]. Weight, diet and early disease may affect both steps and diabetes risk.

Daily stepsModerate

HbA1c in people with type 2 diabetes using step counters

Mixed results: two meta-analyses found no change in HbA1c with step counters in type 2 diabetes, and a later one of 24 trials found a 0.22% drop.

Compared with control groups in randomized trials [10]; control groups in controlled trials [12]; control participants [13]

The numbers and sources

Effect: Step counter use did not change HbA1c (WMD 0.02%, 95% CI -0.08 to 0.13), with or without a goal [10]. A review of 12 trials found no differences in HbA1c, BMI, blood pressure or lipids [12]. A review of 24 randomized trials found HbA1c fell 0.22% (95% CI -0.4 to -0.05, I² 77%) and triglycerides fell 13.11 mg/dL (95% CI -25.21 to -1.02) against control participants [13].
Population: 1,423 participants in 10 trials [10]. 1,458 participants in 12 trials [12]. 1,969 patients in 24 trials [13].
Sources: [10], [12], [13]
Caveats: All participants had type 2 diabetes. Heterogeneity was high in the positive review [13].

Sleep durationModerate

Sleep extension and blood pressure, glucose and insulin

Mixed results: pooled sleep extension trials found no overall change in blood pressure, fasting glucose or insulin resistance. Blood pressure fell in people with raised pressure.

Compared with control or baseline conditions the source does not describe [14]; control or baseline conditions the source does not describe [15]

The numbers and sources

Effect: Sleep extension lengthened total sleep by 51 minutes (95% CI 39 to 63) and did not change blood pressure overall. In people with pre-hypertension or stage 1 hypertension, 24-hour systolic pressure fell 7.8 mmHg (95% CI -10.6 to -4.9) and diastolic pressure 4.2 mmHg. Pooled effects on fasting glucose and insulin resistance were not significant [14]. A systematic review of 7 extension studies reported better direct and indirect measures of insulin sensitivity, less appetite and less desire for sweet and salty foods [15].
Population: 13 studies from 11 trials, 3 days to 6 weeks, adults without a sleep disorder [14]. 138 participants in 7 studies, mostly short sleepers [15].
Sources: [14], [15]
Caveats: Small and short trials. The blood pressure fall is a subgroup result. The review of 7 studies gave no pooled estimates [15].

SmokingModerate

Type 2 diabetes after quitting

Mixed results: in pooled cohorts, diabetes risk was higher in the first 5 years after quitting than in current or never smokers, then fell with time.

Compared with never smokers [18]

The numbers and sources

Effect: Against never smokers, the RR of type 2 diabetes was 1.54 (95% CI 1.36 to 1.74) in people who had quit under 5 years before, 1.18 at 5 to 9 years and 1.11 (95% CI 1.02 to 1.20) at 10 years or more. Current smokers had an RR of 1.37 (95% CI 1.33 to 1.42) [18].
Population: 88 prospective studies with 5,898,795 participants and 295,446 cases. 10 studies with 1,086,608 participants for time since quitting [18].
Sources: [18]
Caveats: Observational. Weight gain after quitting may explain the early rise. The review gives no direct comparison of recent quitters with people who kept smoking.

Food and drink

DairyStrong

Type 2 diabetes

Linked to 18% lower type 2 diabetes per daily serving of yogurt in pooled cohorts. Total dairy showed no link in one review and 3% lower risk in another.

Compared with one fewer serving a day [10]; 50 g a day less yogurt or 200 g a day less total dairy, in dose-response models [9]

The numbers and sources

Effect: Across 14 cohorts, each daily serving of total dairy was not linked to type 2 diabetes (RR 0.98, 95% CI 0.96 to 1.01), and each daily serving of yogurt was linked to lower risk (RR 0.82, 95% CI 0.70 to 0.96). In three US cohorts, neither low-fat nor high-fat dairy was appreciably linked [10]. A later review linked 3% lower risk to each extra 200 g a day of total dairy and 7% lower risk to each extra 50 g a day of yogurt [9].
Population: 459,790 participants and 35,863 cases in 14 cohorts, including 194,458 US adults followed up to 30 years [10]. Cohort studies [9].
Sources: [10], [9]
Caveats: Observational. People who eat yogurt may differ in other habits. The reviews pooled different cohorts and dose units.

Olive oilStrong

Type 2 diabetes

Linked to 16% to 22% lower type 2 diabetes at higher olive oil intake, in two pooled analyses of cohorts.

Compared with the lowest intake category [8]; 25 g a day less, in a dose-response model [3]

The numbers and sources

Effect: The highest against the lowest intake was linked to lower type 2 diabetes risk (RR 0.84, 95% CI 0.77 to 0.92), with a nonlinear pattern [8]. Each extra 25 g a day was linked to 22% lower risk (RR 0.78, 95% CI 0.69 to 0.87), with no evidence of heterogeneity [3].
Population: 4 cohorts with 15,784 cases [8]. 680,239 participants and 13,389 cases [3].
Sources: [8], [3]
Caveats: Observational, and the two analyses share cohorts.

CoffeeModerate

Type 2 diabetes

Linked to lower type 2 diabetes, 33% lower at 6 cups a day than no coffee, with decaf too. A genetic study did not support cause.

Compared with people drinking no coffee or rarely [7]

The numbers and sources

Effect: Against no or rare coffee, the relative risk of type 2 diabetes was 0.92 (95% CI 0.90 to 0.94) at 1 cup a day, 0.79 (95% CI 0.75 to 0.83) at 3 cups and 0.67 (95% CI 0.61 to 0.74) at 6 cups. Each extra daily cup was linked to a relative risk of 0.91 (95% CI 0.89 to 0.94) for caffeinated coffee and 0.94 (95% CI 0.91 to 0.98) for decaffeinated coffee [7]. In a Mendelian randomization study, genetic variants linked to 29% higher coffee intake were not convincingly linked to type 2 diabetes, obesity or glucose, though high intake was linked to lower diabetes risk in the same people observationally [8].
Population: 1,109,272 participants and 45,335 cases in 28 prospective studies followed 10 months to 20 years [7]. 93,179 Danish adults, plus up to 78,021 in a genetic consortium [8].
Sources: [7], [8]
Caveats: Observational cohorts. The genetic study points away from a causal effect, though genetic tools for coffee intake have low power [5].

Dark chocolateModerate

Blood vessel function and insulin resistance

Shown to improve blood vessel widening and lower insulin resistance in short trials of chocolate or cocoa, in one meta-analysis rated low to moderate quality.

Compared with control conditions the abstract does not describe [3]

The numbers and sources

Effect: Flow-mediated dilation rose 1.34% (95% CI 1.00 to 1.68) with regular intake and 3.19% after a single dose. HOMA-IR fell 0.67 (95% CI 0.36 to 0.98) through lower serum insulin. Changes in LDL cholesterol (-0.07 mmol/L, 95% CI -0.13 to 0.00) and HDL cholesterol were of borderline significance [3].
Population: 1,297 participants in 42 trials of up to 18 weeks [3].
Sources: [3]
Caveats: The authors rated the evidence low to moderate quality and called for larger, longer, independently funded trials [3].

Fermented foodsModerate

Blood sugar

Shown to lower fasting glucose slightly with kimchi and kefir, in two meta-analyses. Probiotic yogurt did not differ from ordinary yogurt.

Compared with control groups the abstract does not describe [7][9]; conventional yogurt [10]

The numbers and sources

Effect: Kefir lowered fasting glucose (MD -8.46 mg/dL) and HOMA-IR (MD -1.71) [7]. Fermented kimchi lowered fasting glucose (WMD -1.93 mg/dL, 95% CI -3.82 to -0.03) [9]. Probiotic yogurt did not change HbA1c, fasting glucose, fasting insulin or HOMA-IR against mostly conventional yogurt in people with type 2 diabetes or obesity [10].
Population: 24 studies of kefir [7]. 205 participants in 5 kimchi trials [9]. 472 people with type 2 diabetes or obesity in 9 trials [10].
Sources: [7], [9], [10]
Caveats: Small trials of single foods. The yogurt review compared two fermented foods with each other, not fermented food with none [10].

NutsModerate

Blood sugar in type 2 diabetes

Shown to lower HbA1c by 0.07 percentage points and fasting glucose slightly in people with type 2 diabetes, in one meta-analysis of short trials.

Compared with diets with the same calories without tree nuts [12]

The numbers and sources

Effect: A median 56 g a day of tree nuts lowered HbA1c (MD -0.07%, 95% CI -0.10 to -0.03) and fasting glucose (MD -0.15 mmol/L, 95% CI -0.27 to -0.02). Fasting insulin and HOMA-IR did not change significantly [12].
Population: 450 people with diabetes in 12 randomized trials of 3 weeks or more [12].
Sources: [12]
Caveats: The reviewers describe most trials as short and of poor quality [12].

Olive oilModerate

Blood sugar in people with type 2 diabetes

Shown to lower HbA1c by 0.27 points and fasting glucose by 0.44 mmol/L in people with type 2 diabetes, in pooled trials against control groups.

Compared with control groups in the trials pooled [8]

The numbers and sources

Effect: In people with type 2 diabetes, olive oil lowered HbA1c (mean difference -0.27%, 95% CI -0.37 to -0.17) and fasting plasma glucose (-0.44 mmol/L, 95% CI -0.66 to -0.22) more than in control groups [8].
Population: People with type 2 diabetes in trials among the 29 pooled [8].
Sources: [8]
Caveats: The abstract does not describe the control groups. The authors note the finding applies to olive oil as a food, not its single components [8].

Whole grainsModerate

Type 2 diabetes

Linked to 32% lower type 2 diabetes per three daily servings of whole grains in pooled cohorts. Brown rice was linked to 11% lower risk.

Compared with three fewer servings a day, in a dose-response model [6]; the lowest brown rice intake [7]

The numbers and sources

Effect: Per extra 3 servings a day, the relative risk of type 2 diabetes was 0.68 (95% CI 0.58 to 0.81, I² 82%) for whole grains and 0.95 (95% CI 0.88 to 1.04) for refined grains. Whole grain bread, whole grain cereals, wheat bran and brown rice were each linked to lower risk, in few studies [6]. Brown rice at the highest intake was linked to lower risk (RR 0.89, 95% CI 0.81 to 0.97), and each 50 g a day serving to 13% lower risk [7].
Population: 16 cohort studies [6]. 3 estimates covering 197,228 participants and 10,507 cases for brown rice [7].
Sources: [6], [7]
Caveats: Observational, with high heterogeneity (I² 82%) [6]. The brown rice estimate rests on 3 estimates [7].

GrapefruitLimited

Insulin resistance

One trial of 91 obese adults found lower insulin 2 hours after a glucose drink with fresh grapefruit before meals than with placebo.

Compared with placebo capsules with apple juice [3]

The numbers and sources

Effect: After 12 weeks, 2-hour post-glucose insulin fell significantly in the fresh grapefruit group compared with placebo [3].
Population: 91 obese adults for 12 weeks [3].
Sources: [3]
Caveats: One small trial, and the abstract gives no size for the change in insulin.

Green teaLimited

Type 2 diabetes

Limited evidence: one Japanese cohort linked 6 or more cups a day to 33% lower odds of type 2 diabetes than under 1 cup a week.

Compared with people drinking under 1 cup a week [6]

The numbers and sources

Effect: At 6 or more cups a day, the odds of new diabetes over 5 years were lower (OR 0.67, 95% CI 0.47 to 0.94). Black and oolong tea were not linked to diabetes [6].
Population: 17,413 Japanese adults aged 40 to 65 [6].
Sources: [6]
Caveats: Diabetes was self-reported and follow-up was low, at 49% of the original study population [6].

EggsModerate

Type 2 diabetes

Mixed results: one pooled analysis linked an egg a day to 42% more type 2 diabetes, while a larger update found no overall link, except in US cohorts.

Compared with under one egg a week or never [7]; one fewer egg a day [9]

The numbers and sources

Effect: At 1 or more eggs a day against fewer than 1 a week, type 2 diabetes was higher (HR 1.42, 95% CI 1.09 to 1.86) [7]. Across 16 cohorts, each egg a day was not significantly linked to type 2 diabetes overall (RR 1.07, 95% CI 0.99 to 1.15). It was linked in US studies (RR 1.18, 95% CI 1.10 to 1.27) and not in European (0.99) or Asian (0.82) ones [9].
Population: 22 cohorts [7]. 589,559 participants and 41,248 cases in 16 cohorts [9].
Sources: [7], [9]
Caveats: Observational, with high heterogeneity (I² 69.8%) [9].

Green teaModerate

Blood sugar in trials

Mixed results: pooled trials of green tea and extracts found small falls in fasting glucose and HbA1c, while a trial of a catechin-rich drink found no difference.

Compared with control groups in randomized trials [7]; a green tea drink with 96.3 mg of catechins a day [8]

The numbers and sources

Effect: Across 17 trials, green tea lowered fasting glucose by 0.09 mmol/L (95% CI 0.03 to 0.15) and HbA1c by 0.30% (95% CI 0.22 to 0.37), and fasting insulin in higher-quality trials [7]. In people with type 2 diabetes, a drink with 582.8 mg of catechins a day for 12 weeks showed no apparent difference in glucose or HbA1c from a low-catechin drink, except in people on insulin-releasing drugs [8].
Population: 1,133 participants in 17 randomized trials of green tea or green tea extract [7]. 43 adults with type 2 diabetes not on insulin [8].
Sources: [7], [8]
Caveats: The pooled trials included extracts, and the abstract gives no separate result for the drink [7].

Non-sugar sweetenersModerate

Blood sugar

Mixed results: sweeteners did not raise blood sugar after intake in pooled trials, while against water, sweetened drinks raised HbA1c 0.21 percentage points in one review.

Compared with blood sugar before intake, in the same people [10]; no or lower sweetener intake [8]; water [5]

The numbers and sources

Effect: Across 29 trials of aspartame, saccharin, steviosides and sucralose, blood glucose did not rise after intake and gradually declined over the observation period [10]. In 2 small studies of 52 adults, sweeteners lowered fasting glucose by 0.16 mmol/L (95% CI -0.26 to -0.06) against no sweetener, with very low to low certainty [8]. Replacing water with sweetened drinks was associated with higher HbA1c (0.21 percentage points, 95% CI 0.02 to 0.40) and lower systolic blood pressure (-2.63 mmHg) [5].
Population: 741 participants in 29 randomized trials [10]. 52 adults in 2 studies [8]. 974 adults in 9 trials of sweetened drinks against water [5].
Sources: [10], [8], [5]
Caveats: The glucose trials measured short-term responses [10]. Each finding rests on few or small trials.

Whole grainsModerate

Blood sugar in trials

Shown to lower the rise in blood sugar after a meal against refined foods. Longer trials found no change in fasting glucose, and one review found HbA1c fell.

Compared with similar refined foods [12]; refined grains or placebo [9]

The numbers and sources

Effect: In 14 single-meal trials, whole grain foods reduced the 2-hour glucose response by 29.71 mmol·min/L (95% CI -43.57 to -15.85) and the insulin response. In 16 trials lasting weeks to months, fasting glucose, fasting insulin and HOMA-IR did not change significantly [12]. Another review found whole grains lowered HbA1c (SMD -0.33, 95% CI -0.61 to -0.04) [9].
Population: Healthy adults in 30 trials [12]. 25 trials of adults with or without cardiovascular risk factors [9].
Sources: [12], [9]
Caveats: Most longer trials enrolled healthy people with normal blood sugar [12].

Fermented foodsLimited

Type 2 diabetes, in cohorts

Mixed results: in cohorts, cheese was linked to slightly less type 2 diabetes and yogurt to a 22% lower risk whose interval included no difference.

Compared with people eating 50 g of cheese or 200 g of yogurt a day less [11]

The numbers and sources

Effect: Per 50 g of cheese a day, type 2 diabetes was lower (RR 0.92, 95% CI 0.86 to 0.99). Per 200 g of yogurt a day, RR was 0.78 (95% CI 0.60 to 1.02), with high heterogeneity (I² 70%) and an inverse curve that flattened at higher intakes [11].
Population: 17 cohort studies [11].
Sources: [11]
Caveats: Observational, open to confounding and reverse causation.

NutsLimited

Type 2 diabetes, in cohorts

Not linked to type 2 diabetes in cohorts once body weight is accounted for. One analysis linked peanut butter to 13% lower risk.

Compared with the lowest intake group [11]; people eating one serving a day less [7]

The numbers and sources

Effect: Total nut intake was not significantly linked to type 2 diabetes in cohort or cross-sectional studies. Peanut butter was linked to lower risk (RR 0.87, 95% CI 0.77 to 0.98), and peanuts and tree nuts were not [11]. Per serving a day, the link with diabetes (RR 0.80) disappeared after adjusting for body mass index (RR 1.03, 95% CI 0.91 to 1.16) [7].
Population: 5 prospective and 3 cross-sectional studies [11]. 18 prospective studies with 12,655 diabetes cases [7].
Sources: [11], [7]
Caveats: Observational. The certainty of evidence was rated very low for every exposure [11].

Diets

Keto dietStrongStudied in women

Weight and insulin resistance in PCOS

Shown in pooled studies to lower weight and insulin resistance in women with PCOS. Against other diets, the reviews disagreed on weight.

Compared with other diets [21]; low-calorie diets [22]; before the diet, in the same women [21][22][23]

The numbers and sources

Effect: Against other diets, ketogenic diets lowered weight 4.98 kg more (95% CI 0.91 to 9.05), BMI 1.65 kg/m2 more and insulin resistance more (MD -1.71, 95% CI -2.98 to -0.43, I² 90%). From before to after the diet, weight fell 10.77 kg and BMI 3.38 kg/m2 [21]. In 2 RCTs against low-calorie diets, body weight did not differ significantly, while BMI, fat mass, glucose and HOMA-IR fell more on the ketogenic diets [22]. A third review of 7 studies found weight, BMI, glucose, insulin, HOMA-IR, LDL cholesterol and triglycerides fell from before to after the diet [23].
Population: Women with PCOS, most with BMI above 25, in 15 studies, 10 pooled [21]; 10 studies including 3 RCTs, 4 cohorts and 2 case series [22]; 7 studies [23].
Sources: [21], [22], [23]
Caveats: Heterogeneity was high for most outcomes [21]. Before-and-after results cannot separate the diet from weight loss or time. A published letter disputed whether the definition of real-world exposure in [21] matched its inclusion criteria [24].

Vegetarian dietStrong

Blood sugar in type 2 diabetes

Shown to lower HbA1c about 0.3 to 0.4 points against comparison diets in adults with type 2 diabetes.

Compared with comparator diets the source does not describe [1]; control diets the source does not describe [2]

The numbers and sources

Effect: In 6 controlled trials (255 participants), vegetarian diets lowered HbA1c by 0.39 points (95% CI -0.62 to -0.15), and fasting glucose did not change significantly [1]. In 9 RCTs (664 participants), HbA1c fell 0.29 points (95% CI -0.45 to -0.12) and fasting glucose 0.56 mmol/L, at moderate certainty, with no change in fasting insulin [2].
Population: Adults with type 2 diabetes, mean age 42.5 in [1]; adults with diabetes in trials of 3 weeks or more [2].
Sources: [1], [2]
Caveats: Both reviews pooled lacto-ovo and vegan diets. Trials were small and short.

DASH dietModerate

Heart disease, stroke and diabetes in cohorts

Linked to about 20% lower rates of cardiovascular disease, coronary heart disease, stroke and diabetes in cohort studies, at low certainty.

Compared with people who followed DASH less closely [7]

The numbers and sources

Effect: In 15 prospective cohorts (942,140 people), closer adherence to DASH was linked to lower cardiovascular disease (RR 0.80, 95% CI 0.76 to 0.85), coronary heart disease (RR 0.79), stroke (RR 0.81) and diabetes (RR 0.82, 95% CI 0.74 to 0.92) [7].
Population: Adults in prospective cohorts, observational [7].
Sources: [7]
Caveats: Observational. Certainty was low for cardiovascular disease incidence [7]. No trial of DASH has measured heart attacks or strokes.

High-protein dietModerate

HbA1c in type 2 diabetes

Shown to lower HbA1c about 0.5% more than standard-protein diets within 6 months in type 2 diabetes, in one meta-analysis of nine small trials.

Compared with diets with less protein

The numbers and sources

Effect: Against diets with 15% to 20% of energy from protein, diets with 25% to 32% lowered weight 2.08 kg more (95% CI 0.90 to 3.25) and HbA1c 0.52% more (95% CI 0.14 to 0.90). Fasting glucose and blood lipids did not differ [8].
Population: 418 adults with type 2 diabetes in 9 trials lasting 4 to 24 weeks [8].
Sources: [8]
Caveats: Two studies had a large influence on the weight and HbA1c results [8]. The indexed abstract prints minus signs as the digit 2. The figures here follow its text, which reports reductions.

Intermittent fastingModerate

Blood sugar

Shown to lower fasting glucose slightly in adults with prediabetes or type 2 diabetes, while HbA1c did not change overall in pooled trials.

Compared with a control diet with no time restriction [10]; unrestricted eating [11]; daily calorie restriction or a control group [12]; standard-of-care nutritional counseling alone [13]

The numbers and sources

Effect: In 10 trials with 599 participants, HbA1c changed by -0.30 percentage points (95% CI -0.63 to 0.02, p 0.06, moderate certainty), by -0.09 points in prediabetes and by -0.43 points in type 2 diabetes (95% CI -0.92 to 0.07). Fasting glucose fell 0.30 mmol/L (95% CI -0.53 to -0.07, moderate certainty) [10]. In adults without diabetes, time-restricted eating without calorie restriction lowered fasting glucose by 2.65 mg/dL (95% CI -3.92 to -1.39) and HOMA-IR by 0.58 [11]. In type 2 diabetes, HbA1c fell 0.91% on an 8-hour window and 0.94% on calorie restriction against control, with no difference between the two [12]. In metabolic syndrome, an 8 to 10 hour window lowered HbA1c by 0.10% (95% CI -0.19% to -0.003%) against standard care [13].
Population: Adults with prediabetes or type 2 diabetes, trials averaging 4 months with 4 to 10 hour windows [10]; 11 trials with 653 adults without diabetes [11]; 75 adults with obesity and type 2 diabetes over 6 months [12]; 108 adults with metabolic syndrome over 3 months [13].
Sources: [10], [11], [12], [13]
Caveats: Wide confidence intervals and small trials limited the pooled evidence, and effects on insulin sensitivity and continuous glucose monitoring were inconclusive [10]. Three authors of [10] work at a hospital partly funded by the Novo Nordisk Foundation.

Keto dietModerate

HbA1c in type 2 diabetes

Shown to lower HbA1c about 0.6 points more than comparison diets at 3 to 6 months in type 2 diabetes, with mixed results at 12 months.

Compared with diets that did not restrict carbohydrates to the level of ketosis

The numbers and sources

Effect: Against guideline diabetes diets, HbA1c fell 0.61% more at 3 months (95% CI 0.40 to 0.82, moderate certainty) and 0.58% more at 6 months (95% CI 0.32 to 0.85, low certainty), with no difference at 12 months. Diabetes medication use fell more for up to 12 months [4]. Across 29 trials, HbA1c fell 0.29% (95% CI 0.14 to 0.44) [7]. At 12 months, four studies showed a difference of 0.01% (95% CI -0.22 to 0.25) and two showed 0.65% lower (95% CI 0.31 to 0.99) [6].
Population: 648 adults with type 2 diabetes in 8 RCTs of 12 weeks or more [4]; 606 adults with pre-diabetes or type 2 diabetes in 8 RCTs of 6 months or more [6]; 29 trials in type 2 diabetes [7].
Sources: [4], [6], [7]
Caveats: The authors of [4] name poor adherence to carbohydrate restriction as a major limitation. The 12-month results disagree between analyses [6]. Review [7] pooled clinical trials, not only randomized ones.

Low-carb dietModerateVery common

HbA1c and remission in type 2 diabetes

Shown to lower HbA1c and raise remission rates at 3 to 6 months in type 2 diabetes, with no clear difference at 12 to 24 months.

Compared with diets with more carbohydrate

The numbers and sources

How common: Very common: more than 1 in 10 people.
Effect: At 6 months, 57% on low-carbohydrate diets reached HbA1c under 6.5% against 31% on control diets (risk difference 0.32, 95% CI 0.17 to 0.47, 8 studies, 264 people, I² 58%). With remission defined as HbA1c under 6.5% without medication, the effect was smaller and not significant. At 12 months, remission data were sparse [6]. Diets under 26% of energy from carbohydrate lowered HbA1c 0.47% more at 3 months (95% CI 0.23 to 0.71) and 0.36% more at 6 months (95% CI 0.09 to 0.62), with no difference at 12 or 24 months. Moderate restriction at 26% to 45% of energy did not differ from higher-carbohydrate diets [7]. Over 1 to 2 years, HbA1c showed little to no difference against balanced-carbohydrate diets [1].
Population: 1,357 adults with type 2 diabetes in 23 trials [6]; 2,412 adults with diabetes in 25 RCTs [7]; adults with type 2 diabetes and overweight [1].
Sources: [1], [6], [7]
Caveats: Remission fell markedly in studies that included people using insulin [6]. Certainty was moderate to low [6], and the authors of [7] note issues with evidence quality.

Low-carb dietModerateStudied in women

Weight and insulin resistance in PCOS

Shown in one meta-analysis of 8 trials to lower BMI and insulin resistance more than a regular diet in women with PCOS.

Compared with a regular diet with about 45% of energy from carbohydrate [13]; before the diet, in the same women [14]

The numbers and sources

Effect: Against a regular diet, low-carbohydrate diets lowered BMI (SMD -1.04, 95% CI -1.38 to -0.70), HOMA-IR (SMD -0.66, 95% CI -1.01 to -0.30), total cholesterol (SMD -0.68) and LDL cholesterol (SMD -0.66, 95% CI -1.30 to -0.02) [13]. A second review of 25 diet studies in PCOS found insulin resistance fell from before to after diets with under 50% of energy from carbohydrate (standardized paired difference -0.86, 95% CI -1.23 to -0.50) [14].
Population: 327 women with PCOS in 8 RCTs [13]; women with PCOS in 25 prospective diet studies [14].
Sources: [13], [14]
Caveats: Trials were small. The second review measured change within each diet group rather than against another diet [14].

Mediterranean dietModerate

Blood pressure, cholesterol and blood sugar markers

Shown to lower blood pressure about 3/2 mmHg and LDL cholesterol and HbA1c slightly in adults, at low to moderate certainty.

Compared with no or minimal intervention, and another dietary intervention the abstract does not name [2]; a low-fat diet, a low-carbohydrate diet and the American Diabetes Association diet [7]; control diets the source does not describe [10]

The numbers and sources

Effect: Against no or minimal intervention, systolic blood pressure fell 2.99 mmHg (95% CI -3.45 to -2.53) and diastolic 2.0 mmHg (95% CI -2.29 to -1.71) at moderate quality, with little or no effect on LDL or HDL. Against other diets, LDL fell 0.15 mmol/L (95% CI -0.27 to -0.02) and triglycerides 0.09 mmol/L, with little or no effect on blood pressure [2]. In 9 RCTs (1,337 people at raised risk of diabetes or heart disease), HbA1c fell 0.18 points (95% CI -0.35 to -0.01), LDL 0.10 mmol/L and triglycerides 0.20 mmol/L. Fasting glucose, insulin, insulin resistance and HDL did not change [10]. Over 12 months or more, blood pressure and lipids changed about as much as on comparator diets [7].
Population: Healthy adults and adults at high cardiovascular risk [2]; adults at increased risk of type 2 diabetes and heart disease [10]; adults with overweight or obesity [7].
Sources: [2], [7], [10]
Caveats: The Cochrane review rated the lipid evidence low or very low quality for most comparisons [2]. The changes are small next to drug effects.

Nordic dietModerate

Heart disease, diabetes and death in cohorts

Linked to lower death, heart attack, stroke and type 2 diabetes rates with higher adherence in cohort studies, at low certainty.

Compared with people with the lowest adherence to Nordic dietary patterns, in European cohorts [2]; people with lower adherence to the Nordic dietary pattern [9]; people in the lowest category of Nordic diet adherence [10]

The numbers and sources

Effect: In 13 cohorts (930,153 people), high against low adherence was linked to lower all-cause death (RR 0.78, 95% CI 0.69 to 0.87), cardiovascular death (RR 0.78), stroke (RR 0.88), heart attack (RR 0.80) and type 2 diabetes (RR 0.90, 95% CI 0.82 to 0.99) [10]. In 15 cohorts, total cardiovascular disease was 7% lower (RR 0.93, 95% CI 0.88 to 0.99, I² 88%) [2]. A 2025 meta-analysis of 32 cohorts found 12% lower type 2 diabetes and 8% lower cardiovascular disease [9].
Population: Adults in prospective cohorts, observational [2][9][10].
Sources: [2], [9], [10]
Caveats: Observational. Certainty ranged from very low to low [10], and was low for total cardiovascular disease [2].

Vegan dietModerate

Blood sugar

Shown to lower HbA1c by about 0.18 points against control diets in adults with overweight or type 2 diabetes.

Compared with the participants' habitual diet, unchanged, in 3 trials in US and New Zealand adults, and in 8 trials a National Cholesterol Education Program diet, American Diabetes Association and Korean Diabetes Association diets, a portion-controlled diet, a Mediterranean diet, a low-fat diet, a healthy omnivorous diet or a high-carbohydrate lacto-ovo vegetarian diet [1]

The numbers and sources

Effect: In 11 trials, vegan diets lowered HbA1c by 0.18 points (95% CI -0.29 to -0.07) [1].
Population: Adults with overweight or type 2 diabetes, trials of 12 weeks or more [1].
Sources: [1]
Caveats: One meta-analysis of 11 trials. Differences in energy intake may explain part of the effect [1].

Alternate-day fastingLimited

Blood sugar in insulin-treated type 2 diabetes

One trial found 3 fasting days a week lowered HbA1c by 7.3 mmol/mol against control over 12 weeks in insulin-treated type 2 diabetes.

Compared with eating every day, with no fasting days or daily fasting window

The numbers and sources

Effect: HbA1c fell 7.3 mmol/mol (SD 12.0) on fasting against 0.1 mmol/mol in controls (P 0.012). Eight people on fasting and none in control reached the combined goal of 2% weight loss, 10% lower insulin dose and a 3 mmol/mol HbA1c drop [10].
Population: 46 adults with insulin-treated type 2 diabetes, with dietary counseling and continuous glucose monitoring [10].
Sources: [10]
Caveats: One small 12-week trial, published as a brief report [10].

Carnivore dietLimited

Blood sugar in diabetes

Limited evidence: survey respondents with diabetes reported a median HbA1c fall of 0.4 points and less diabetes medication use.

Compared with before the diet, in the same people [1], [3]

The numbers and sources

Effect: Respondents with diabetes reported a median BMI fall of 4.3 (IQR 1.4 to 7.2), a median HbA1c fall of 0.4% (IQR 0% to 1.7%), and reductions in diabetes medication use given as 84% to 100% [1]. In a German study, both followers who started with prediabetic HbA1c values had lower values on the diet [3].
Population: Survey respondents with diabetes [1]; 24 German followers who shared blood tests from before and during the diet [3].
Sources: [1], [3]
Caveats: Self-reported survey data and a very small before-and-after study, with no control group [1][3].

DASH dietLimited

Blood sugar

Small trials found HbA1c fell about 0.5 points. Pooled trials found no change in fasting glucose.

Compared with usual diets, and for fasting glucose also low-antioxidant and weight-loss diets, behavioral programs alone and usual gestational diabetes care [7]

The numbers and sources

Effect: A meta-analysis of 2 controlled trials (65 participants) found HbA1c fell 0.53 points (95% CI -0.62 to -0.43). Fasting insulin fell slightly, and fasting glucose and insulin resistance did not change [7].
Population: Adults with and without diabetes [7].
Sources: [7]
Caveats: Two small trials. Certainty ranged from very low to moderate across secondary outcomes [7].

Keto dietLimited

Liver fat and insulin sensitivity

One trial found that at equal 10% weight loss, liver insulin sensitivity rose two to three times more than on Mediterranean or very-low-fat diets.

Compared with diets that did not restrict carbohydrates to the level of ketosis

The numbers and sources

Effect: Calories were adjusted so every group lost about 10% of body weight. Muscle insulin sensitivity rose about 50% on all three diets. Liver insulin sensitivity rose two to three times more on the very-low-carbohydrate ketogenic diet (p < 0.001), and liver fat, liver fat production, HbA1c and 24-hour glucose and insulin fell most on it. LDL cholesterol, apolipoprotein B and 24-hour triglycerides did not differ between diets [12].
Population: Adults with obesity, prediabetes and fatty liver. The abstract does not give the number randomized [12].
Sources: [12]
Caveats: One small trial. Food was supplied during weight loss, so it measures the diets eaten as written. It does not apply to people without prediabetes and fatty liver.

Mediterranean dietLimited

Type 2 diabetes onset

One large trial found fewer new diabetes cases on the olive oil Mediterranean diet, and cohorts link closer adherence to lower diabetes risk.

Compared with a control diet of advice on a low-fat diet [11]; people with lower Mediterranean diet adherence scores, per 2 points, in 24 cohorts, and a low-fat diet in the one trial (PREDIMED) [12]; ad libitum Mediterranean diet advice, without calorie reduction or added exercise [13]

The numbers and sources

Effect: In 3,541 PREDIMED participants without diabetes, new diabetes ran at 16.0 cases per 1,000 person-years on the olive oil diet, 18.7 on the nut diet and 23.6 on the low-fat control over a median 4.1 years (HR 0.60, 95% CI 0.43 to 0.85, and HR 0.82, 95% CI 0.61 to 1.10) [11]. In 24 cohorts and 1 RCT (991,878 participants), each 2-point rise in adherence score was linked to 8% lower diabetes risk (HR 0.92, 95% CI 0.90 to 0.94, moderate certainty) [12]. In PREDIMED-Plus, adding a 600 kcal a day reduction, exercise and weight-loss support to a Mediterranean diet cut 6-year diabetes incidence by 31% (95% CI 18% to 41%) against ad libitum Mediterranean advice, 9.5% against 12.0% [13].
Population: Spanish adults aged 55 to 80 at high cardiovascular risk [11]; adults in cohorts, observational [12]; 4,746 adults aged 55 to 75 with metabolic syndrome and overweight or obesity [13].
Sources: [11], [12], [13]
Why limited: This rests on one trial, of 3,541 people. Strength counts independent studies that agree, so one trial is limited at any size.
Caveats: The PREDIMED result is a subgroup analysis, randomization was not stratified by diabetes status and withdrawals were higher in the control group [11]. The nut diet result was not significant [11]. The cohort data are observational [12]. PREDIMED-Plus compared two Mediterranean diets, so it tests calorie reduction and exercise, not the Mediterranean pattern itself [13].

5:2 dietModerate

Blood sugar in type 2 diabetes

Mixed results: HbA1c fell about as much as on daily calorie restriction, rose back above baseline by 24 months, and fell more than on metformin with meal replacement fasting days.

Compared with continuous energy restriction [1], [3], [9]; a daily energy-restricted diet or a routine diet [7]; metformin or empagliflozin [10]

The numbers and sources

Effect: At 12 months, HbA1c fell 0.3% on 5:2 and 0.5% on daily restriction (P .65), meeting the equivalence margin [3]. At 24 months, HbA1c was 0.1% and 0.4% above baseline in the two groups [9]. A 5:2 plan using meal replacements on fasting days lowered HbA1c by 1.9% over 16 weeks, 0.3% more than metformin and 0.4% more than empagliflozin, with weight loss of 9.7 kg against 5.5 and 5.8 kg [10]. In adults with overweight or obesity, fasting glucose and HbA1c did not differ from control [7]. In premenopausal women, fasting insulin and insulin resistance fell slightly more on 5:2 than on daily restriction (both P 0.04) [1].
Population: 137 adults with type 2 diabetes, mean HbA1c 7.3% [3][9]; 405 Chinese adults with overweight or obesity and early type 2 diabetes [10]; adults with overweight or obesity [7]; 107 women [1].
Sources: [1], [3], [7], [9], [10]
Caveats: The meal replacement trial was open-label and lasted 16 weeks with 8 weeks of follow-up [10]. Weight loss was kept at 24 months, yet HbA1c rose in both groups [9].

Nordic dietModerate

Blood sugar and insulin

Mixed results: fasting glucose did not change in adults, while fasting insulin and insulin resistance fell slightly.

Compared with an average Nordic diet with the same calories [4]; control diets the source does not describe [7]

The numbers and sources

Effect: In 6 RCTs (618 participants, 6 to 26 weeks), fasting glucose did not change (-0.05 mmol/L, 95% CI -0.13 to 0.01), while fasting insulin fell 1.12 mU/L and HOMA-IR 0.34 against control diets. The insulin result depended on one study [7]. Insulin sensitivity did not change in SYSDIET [4].
Population: Adults in Nordic countries with cardiometabolic risk factors [4][7].
Sources: [4], [7]
Caveats: Two-hour glucose and the Matsuda index did not change [7].

Paleo dietModerate

Blood sugar and insulin

Mixed results: two reviews found lower fasting insulin, while a review against Mediterranean, diabetes and guideline diets found no difference in glucose, insulin or HbA1c.

Compared with control diets the source does not describe [3]; a control group the source does not describe [5]; the Mediterranean diet, a diabetes diet and a diet following Dutch Health Council advice [6]

The numbers and sources

Effect: Against Mediterranean, diabetes and Dutch guideline diets, fasting glucose (SMD -0.34, 95% CI -0.87 to 0.18), fasting insulin, HOMA-IR and HbA1c (SMD -0.38, 95% CI -0.87 to 0.11) did not differ [6]. In people with metabolic disorders, HOMA-IR fell 0.39 (95% CI 0.08 to 0.70) and fasting insulin 12.17 μU/mL (95% CI 0.08 to 24.26) [5]. In 19 RCTs, fasting insulin fell 1.01 (95% CI 0.57 to 1.45) [3].
Population: 98 people with altered glucose metabolism in 4 studies [6]; adults with metabolic disorders [5]; adults in 19 RCTs [3].
Sources: [3], [5], [6]
Caveats: The null review was small [6]. The reviews used different comparison diets.

Low-fat dietLimitedStudied in women

Diabetes

One large trial found no change in treated diabetes over 8.1 years, and slightly less diabetes needing insulin over about 20 years of follow-up.

Compared with diets with more fat

The numbers and sources

Effect: Treated diabetes was reported by 7.1% of the low-fat group and 7.4% of the comparison group (HR 0.96, 95% CI 0.90 to 1.03) [15]. Over a median of 19.6 years, diabetes requiring insulin was lower (HR 0.87, 95% CI 0.77 to 0.98) [14].
Population: 48,835 generally healthy postmenopausal women [14][15].
Sources: [14], [15]
Caveats: Diabetes was self-reported. The authors write that weight loss, more than macronutrient mix, may be the main predictor of diabetes risk [15].
Why limited: This rests on one trial, of 48,835 people. Strength counts independent studies that agree, so one trial is limited at any size.

One meal a dayLimited

Blood sugar

Limited evidence: one small trial found higher fasting glucose and poorer morning glucose tolerance on one daily meal, and another found lower glucose later in the day.

Compared with eating every day, with no fasting days or daily fasting window

The numbers and sources

Effect: On one meal a day eaten between 4 and 8 p.m., morning fasting glucose was higher, glucose rose more and stayed high longer in a morning glucose tolerance test, and the insulin response was delayed, against three meals. The changes reversed after the diet ended [3]. On one evening meal for 11 days, plasma glucose was lower during the second half of the day [2].
Population: Healthy normal-weight middle-aged men and women in a randomized crossover trial [3]; 11 lean adults [2].
Sources: [2], [3]
Caveats: Neither trial included people with diabetes. Glucose was measured at different times of day in the two trials [2][3].

Alternate-day fastingModerate

Blood sugar in people without diabetes

Not shown to change fasting glucose or insulin resistance against control or daily calorie restriction in adults without diabetes.

Compared with eating every day, with no fasting days or daily fasting window

The numbers and sources

Effect: Fasting glucose (p 0.09) and HOMA-IR (p 0.55) did not differ from control [2]. Fasting glucose, fasting insulin and insulin resistance did not differ from daily calorie restriction at 6 or 12 months [3]. HbA1c did not differ between intermittent fasting, continuous restriction and unrestricted diets [6].
Population: 269 adults [2]; 100 adults with obesity [3]; 99 trials [6].
Sources: [2], [3], [6]
Caveats: Few participants had diabetes in these trials.

Atkins dietModerate

Blood sugar and insulin

Not shown to change blood sugar or insulin more than other diets in adults with overweight in three trials.

Compared with the Zone, Weight Watchers and Ornish diets [8]; a conventional low-calorie, high-carbohydrate, low-fat diet [9]; a high-carbohydrate, high-fibre diet and the high-protein Zone diet [11]

The numbers and sources

Effect: No diet, Atkins included, significantly changed glucose at 1 year [8]. The insulin response to a glucose drink fell on both Atkins and a conventional diet [9]. Insulin fell on all three diets in insulin-resistant women, and was the one marker that did not fall more on Atkins than on a high-carbohydrate diet [11].
Population: Adults with cardiac risk factors [8]; adults with obesity [9]; insulin-resistant women without diabetes [11].
Sources: [8], [9], [11]
Caveats: None of these trials studied people with diabetes.

Devices and wearables

Wrist heart rate and calorie estimatesStrong

Blood pressure, cholesterol and blood sugar

Mixed results: effects of activity trackers on blood pressure, cholesterol and HbA1c across systematic reviews were typically small and often not statistically significant.

The numbers and sources

Effect: Effects of activity trackers on blood pressure, cholesterol and glycosylated hemoglobin were typically small and often non-significant, as were effects on quality of life and pain [18].
Population: Participants in 39 systematic reviews and meta-analyses [18].
Sources: [18]
Caveats: The abstract gives no pooled figures for these outcomes [18].

Continuous glucose monitors in people without diabetesModerate

Blood glucose

Mixed results: pooled studies found lower mean glucose with sensor use in people with prediabetes and no appreciable change in healthy people with normal glucose.

The numbers and sources

Effect: A meta-analysis of studies in people without diabetes found sensor use lowered mean blood glucose against controls (SMD -0.54, 95% CI -1.02 to -0.07, P = 0.03). The authors reported the benefit in prediabetes and no appreciable glycemic benefit in healthy people with normal glucose [10]. A meta-analysis of 25 trials, most of them in diabetes, found sensor-based feedback lowered HbA1c by 0.28% (95% CI 0.15 to 0.42) [14]. In 30 adults with prediabetes and overweight, those who could see their sensor data during 30 days of nutrition therapy increased time in target range (p = 0.02), while a group blinded to the data did not change significantly [15]. In 40 overweight young adults, adding a sensor to diet education improved fasting glucose and HbA1c over 8 weeks (p < 0.05) [16].
Population: 1,074 participants without diabetes in 23 studies [10]. 2,996 participants in 25 trials, 3 of them in adults with obesity and the rest in diabetes [14]. 30 adults with prediabetes [15]. 40 overweight young adults [16].
Sources: [10], [14], [15], [16]
Caveats: Sensors were used alongside diet education or coaching, so trials do not isolate the sensor. Eleven of the 25 trials in one review reported conflicts of interest tied to sensor makers, and heterogeneity was high (I² 88%) [14]. Single trials ran 30 days to 8 weeks with 30 to 40 participants [15][16].

Lab markers

25-hydroxyvitamin DStrong

Type 2 diabetes

Linked to more type 2 diabetes at low levels in cohorts, and vitamin D lowered new diabetes 15% in three pooled trials of adults with prediabetes.

The numbers and sources

Effect: In 21 prospective studies, the highest category of 25-hydroxyvitamin D had a relative risk of type 2 diabetes of 0.62 against the lowest (95% CI 0.54 to 0.70), and each 4 ng/mL (10 nmol/L) higher went with a 4% lower risk (95% CI 3 to 6) [16]. In an individual-participant meta-analysis of 3 trials in adults with prediabetes, vitamin D reduced new diabetes by 15% (HR 0.85, 95% CI 0.75 to 0.96), an absolute reduction of 3.3% over 3 years. Among those assigned to cholecalciferol, participants who held a level of 50 ng/mL or above during the trial had a hazard ratio of 0.24 (95% CI 0.16 to 0.36) against those at 20 to 29 ng/mL [17].
Population: 76,220 participants with 4,996 new cases of type 2 diabetes in 21 studies [16]. Adults with prediabetes in 3 trials of cholecalciferol or eldecalcitol against placebo [17].
Sources: [16], [17]
Caveats: The cohort finding is observational. The comparison by level reached during the trials is not randomized. The trial authors state that results in people with prediabetes do not apply to the general population [17].

Fasting glucoseStrong

Heart disease and stroke

Linked to more coronary heart disease above 100 mg/dL in 102 pooled cohorts, 11% to 17% more below the diabetes range, with no link between 70 and 100 mg/dL.

The numbers and sources

Effect: In 102 prospective studies, fasting glucose was non-linearly related to vascular risk, with no significant association between 3.90 and 5.59 mmol/L (70 to 100 mg/dL). Against that range, hazard ratios for coronary heart disease were 1.07 (95% CI 0.97 to 1.18) below 3.90 mmol/L, 1.11 (1.04 to 1.18) at 5.60 to 6.09 mmol/L and 1.17 (1.08 to 1.26) at 6.10 to 6.99 mmol/L. Diabetes went with a hazard ratio of 2.00 (1.83 to 2.19). In people without known diabetes, adding fasting glucose to conventional risk factors did not significantly improve prediction of vascular disease [8]. In a meta-analysis of 53 cohorts, impaired fasting glucose went with a relative risk of cardiovascular disease of 1.13 by the American definition and 1.26 by the WHO definition, and of stroke of 1.06 and 1.17 [9].
Population: 698,782 people without vascular disease in 102 prospective studies, with 52,765 vascular outcomes [8]. 1,611,339 people in 53 prospective cohort studies, median follow-up 9.5 years [9].
Sources: [8], [9]
Caveats: Observational. In the pooled cohorts, hazard ratios for diabetes did not change appreciably after further adjustment for lipid, inflammatory or kidney markers [8].

Fasting glucoseStrong

Death from any cause

Linked to more deaths above 100 mg/dL and not between 70 and 100 mg/dL, in 97 pooled cohorts of 820,900 people.

The numbers and sources

Effect: In 97 prospective studies, fasting glucose above 100 mg/dL was associated with death and levels of 70 to 100 mg/dL were not. Diabetes went with a hazard ratio for death from any cause of 1.80 (95% CI 1.71 to 1.90) [10]. In an updated meta-analysis, prediabetes by any definition went with a relative risk of death from any cause of 1.13 (95% CI 1.10 to 1.17) in the general population [11]. An earlier meta-analysis put the relative risk at 1.13 for impaired fasting glucose by both the American and WHO definitions [9].
Population: 820,900 people in 97 prospective studies, with 123,205 deaths [10]. 10,069,955 people in 129 studies, median follow-up 9.8 years [11].
Sources: [10], [11], [9]
Caveats: Observational. The updated meta-analysis pooled several definitions of prediabetes, and found impaired glucose tolerance carried a higher risk than impaired fasting glucose [11].

Hemoglobin A1c (HbA1c)Strong

Heart disease and stroke in people without diabetes

Linked to more coronary heart disease from 5.5% upward in adults without diabetes, though adding HbA1c to standard risk factors barely improved risk prediction.

The numbers and sources

Effect: In US adults without diabetes, hazard ratios for coronary heart disease against 5.0 to under 5.5% were 1.23 (95% CI 1.07 to 1.41) at 5.5 to under 6.0%, 1.78 (1.48 to 2.15) at 6.0 to under 6.5% and 1.95 (1.53 to 2.48) at 6.5% or greater, with similar results for stroke [8]. Across 73 prospective studies, the association with cardiovascular disease was approximately J-shaped. Adding HbA1c to conventional risk factors changed the C-index by 0.0018 (95% CI 0.0003 to 0.0033) and gave a net reclassification improvement of 0.42 (-0.63 to 1.48), which the authors called little incremental benefit [9].
Population: 11,092 adults [8]. 294,998 participants without known diabetes or cardiovascular disease in 73 studies, median follow-up 9.9 years, 20,840 cardiovascular events [9].
Sources: [8], [9]
Caveats: Observational. The association weakened somewhat after adjustment for HDL cholesterol and C-reactive protein [9].

Hemoglobin A1c (HbA1c)Strong

Eye, kidney and nerve complications of diabetes

Shown in trials: intensive glucose control that lowered HbA1c about 0.9 points cut kidney events 20% and eye events 13% in type 2 diabetes.

The numbers and sources

Effect: In an individual-participant meta-analysis of 4 trials in type 2 diabetes, more intensive control left HbA1c 0.90 percentage points lower (95% CI 0.58 to 1.22) and reduced kidney events (HR 0.80, 95% CI 0.72 to 0.88) and eye events (HR 0.87, 95% CI 0.76 to 1.00), with no reduction in nerve events (HR 0.98, 95% CI 0.87 to 1.09) [11]. In type 1 diabetes, intensive therapy reduced the development of retinopathy by 76% (95% CI 62 to 85) and clinical neuropathy by 60% (95% CI 38 to 74) [12]. In an observational analysis of people with type 2 diabetes, each 1% lower updated mean HbA1c went with 37% fewer microvascular complications (95% CI 33 to 41), with no threshold [13]. In pooled data from nine studies, diabetes-specific retinopathy was rare below 6.0% and its threshold fell at 6.3 to 6.7% [14].
Population: 27,049 adults with type 2 diabetes in 4 trials, median follow-up 5.0 years [11]. 1,441 people with type 1 diabetes followed a mean 6.5 years [12]. 3,642 people with type 2 diabetes [13]. 28,010 people with HbA1c and retinal photographs [14].
Sources: [11], [12], [13], [14]
Caveats: The trials tested treatment strategies, not HbA1c itself. Intensive therapy raised severe hypoglycemia two to threefold in the type 1 diabetes trial [12].

HOMA-IRStrong

Heart disease and other cardiovascular events

Linked to more coronary heart disease in adults without diabetes, 64% more in the highest than the lowest HOMA-IR group across pooled cohorts.

The numbers and sources

Effect: In a meta-analysis of adults without diabetes, the relative risk of coronary heart disease for high against low values was 1.64 (95% CI 1.35 to 2.00) for HOMA-IR, 1.52 (1.31 to 1.76) for fasting glucose and 1.12 (0.92 to 1.37) for fasting insulin. Per standard deviation higher it was 1.46 (1.26 to 1.69) for HOMA-IR, 1.21 (1.13 to 1.30) for glucose and 1.04 (0.96 to 1.12) for insulin [11]. A later meta-analysis of cohorts of healthy adults and adults with metabolic syndrome components found that a higher HOMA-IR went with more non-fatal major cardiovascular events (HR 1.46, 95% CI 1.08 to 1.97) and found no association with fatal ones [12].
Population: 516,325 participants in 65 studies of glucose, insulin or HOMA-IR [11]. 215,878 participants in 38 cohort studies [12].
Sources: [11], [12]
Caveats: Observational. Each study compared its own high and low groups, measured with its own insulin assay, so the pooled groups share no common cut point.

HOMA-IRStrong

High blood pressure

Linked to more new high blood pressure, 43% more in the highest than the lowest HOMA-IR group across pooled cohorts of the general population.

The numbers and sources

Effect: Comparing the highest with the lowest category, the pooled adjusted risk ratio for developing hypertension was 1.43 (95% CI 1.27 to 1.62) for HOMA-IR and 1.54 (1.34 to 1.76) for fasting insulin [14]. A second meta-analysis found a hazard ratio for hypertension of 1.35 (95% CI 1.15 to 1.59) with a higher HOMA-IR [12].
Population: 55,059 participants with 10,230 cases of hypertension in 11 prospective studies [14]. 215,878 participants in 38 cohort studies across all outcomes [12].
Sources: [14], [12]
Caveats: Observational, so confounding by body weight and other factors cannot be ruled out. Each study defined its own highest and lowest groups.

Alanine aminotransferase (ALT)Moderate

Type 2 diabetes

Linked to more type 2 diabetes, 16% higher risk per 5 U/L in pooled cohorts, a result the authors said publication bias may explain.

The numbers and sources

Effect: In a meta-analysis of 17 prospective studies, the relative risk of type 2 diabetes was 1.16 (95% CI 1.08 to 1.25) per 5 IU/L higher ALT and 1.26 (1.14 to 1.41) per standard deviation of log ALT, with high heterogeneity (I2 88%). The association became nonsignificant after correction for publication bias [12]. In a Mendelian randomization study, genetically predicted higher ALT was related to more type 2 diabetes (odds ratio 1.45 per standard unit, 95% CI 1.10 to 1.92) [13].
Population: 60,359 participants and 3,890 new cases of type 2 diabetes in 17 studies [12]. Up to 64,094 people with type 2 diabetes and 607,012 controls in the genetic study [13].
Sources: [12], [13]
Caveats: The cohort studies are observational and their pooled result did not survive correction for publication bias [12]. The genetic study's authors read their findings as support for insulin resistance leading to fatty liver, which in turn raises diabetes risk, so ALT may mark that process and not drive it [13].

Fasting glucoseModerate

Developing type 2 diabetes

Linked to later type 2 diabetes: fasting glucose of 100 to 125 mg/dL went with about 4 times the risk in a Cochrane review of cohort studies.

The numbers and sources

Effect: Against normal glucose levels, the pooled hazard ratio for developing type 2 diabetes was 4.32 (95% CI 2.61 to 7.12) for fasting glucose of 100 to 125 mg/dL and 5.47 (3.50 to 8.54) for 110 to 125 mg/dL. In the 47 studies that reported it, 33% to 59% of people with intermediate hyperglycemia returned to normal levels within 1 to 5 years [6]. In a Dutch cohort, the odds ratio for diabetes in the highest against the lowest fifth of fasting glucose was 9.0 (95% CI 4.4 to 18.5), and the association rose more steeply at higher values [7].
Population: 103 prospective cohort studies, with 8 studies and 9,017 participants behind the first hazard ratio and 9 studies and 2,818 participants behind the second [6]. 1,349 people aged 50 to 75 without diabetes, followed a mean 6.4 years [7].
Sources: [6], [7]
Caveats: Observational. The review rated this evidence as low certainty because many studies did not adjust adequately for confounders and the intervals were wide [6]. Fasting glucose is itself one of the tests that diagnose diabetes, which strengthens the link [7].

Fasting glucoseModerate

Diabetic eye disease

Linked to diabetic eye disease from about 6.4 to 6.8 mmol/L (about 115 to 123 mg/dL) upward, in pooled data on 41,411 people.

The numbers and sources

Effect: In pooled data from nine studies, the prevalence of diabetes-specific retinopathy was low at fasting glucose below 6.0 mmol/L and rose above that level. The threshold fell at 6.4 to 6.8 mmol/L by one method and 6.6 mmol/L by another. The authors suggested the diagnostic level for fasting glucose could be lowered from 7.0 to 6.5 mmol/L [15].
Population: 44,623 people aged 20 to 79 with gradable retinal photographs in nine studies from five countries, 41,411 of them with fasting glucose [15].
Sources: [15]
Caveats: Cross-sectional. WHO states that its diabetes criteria mark a group with more premature death and more microvascular and cardiovascular complications, while noting the limits of the data behind them [5].

Fasting insulin and HOMA-IRModerateStudied in women

Polycystic ovary syndrome

Linked to higher fasting insulin in women with polycystic ovary syndrome than in women without it, in both overweight and non-overweight groups, in one meta-analysis.

The numbers and sources

Effect: In 8 studies, fasting insulin was higher in women with PCOS than in women without it (standardized mean difference 0.62, 95% CI 0.36 to 0.88). Compared within weight groups, it was higher in overweight women with PCOS than in overweight women without it (0.65, 0.19 to 1.12, 4 studies), and in non-overweight women with PCOS than in non-overweight women without it (0.47, 0.11 to 0.83, 5 studies). Study means ran from 13.58 to 28.66 µIU/mL in women with PCOS and from 8.00 to 25.6 µIU/mL in women without it [24]. In 28 studies using the insulin clamp, insulin sensitivity was 27% lower in women with PCOS than in controls (99% CI ±6%), independent of body mass index, and a higher body mass index lowered it a further 15% in PCOS [25].
Population: 12 studies with 450 participants in all, 8 of them on fasting insulin [24]. 28 clamp studies of premenopausal women with PCOS and controls [25].
Sources: [24], [25]
Caveats: Observational, from small studies, and the overweight comparison rests on 4 of them [24]. Fasting insulin rises with body mass index [8], which is why the comparisons within weight groups matter. The clamp is a research procedure, not a fasting insulin test [25]. Insulin assays differ (see Agreement between insulin assays).

FerritinModerate

Type 2 diabetes at high levels

Linked to more type 2 diabetes at higher levels, 73% more in the top fifth than the bottom fifth across pooled cohorts.

The numbers and sources

Effect: In a meta-analysis of prospective studies, the fully adjusted relative risk of type 2 diabetes was 1.73 (95% CI 1.35 to 2.22) comparing extreme fifths of baseline ferritin, with heterogeneous findings between studies (I² 58%). In dose-response analysis, each 5 ng/mL higher ferritin had a relative risk of 1.01 (95% CI 0.99 to 1.02) [10].
Population: 9 prospective studies of ferritin, within a review of 12 studies with 185,462 participants and 11,079 new cases of type 2 diabetes [10].
Sources: [10]
Caveats: Observational, and the authors state that causality is not established. The per-unit estimate did not reach significance [10].

FructosamineModerate

Developing diabetes

Linked to more new diabetes at higher levels, about 5 times the rate above the 95th percentile in a US cohort, a weaker predictor than HbA1c.

The numbers and sources

Effect: Among US adults without diabetes followed for two decades, the hazard ratio for developing diabetes was 4.96 (95% CI 4.36 to 5.64) for fructosamine above the 95th percentile and 6.17 (5.45 to 6.99) for glycated albumin above the 95th percentile. The associations weakened and persisted after adjustment for HbA1c. HbA1c predicted new diabetes more accurately, with C statistics of 0.760 for HbA1c, 0.706 for fructosamine and 0.703 for glycated albumin [5]. In a later sample of the same cohort, the top against the bottom quarter of fructosamine had a hazard ratio of 3.99 (95% CI 1.93 to 8.28), which remained significant after adjustment for HbA1c or fasting glucose [6].
Population: 11,348 adults without diabetes in the Atherosclerosis Risk in Communities Study, examined in 1990 to 1992 [5]. 1,299 participants of the same study without diagnosed diabetes, 119 new cases over a median 3.3 years [6].
Sources: [5], [6]
Caveats: Observational, and both analyses come from one US cohort. In the smaller analysis new diabetes was self-reported [6].

FructosamineModerate

Heart attack, stroke and death from any cause

Linked to more heart attacks and deaths at high levels in US and Swedish cohorts, and to slightly more deaths at the lowest levels.

The numbers and sources

Effect: In a US cohort followed for two decades, elevated fructosamine and glycated albumin were each significantly associated with coronary heart disease, ischemic stroke, heart failure and death after adjustment for cardiovascular risk factors. The associations were similar in size to those for HbA1c and tended to be J-shaped, with raised risk at the lowest levels as well [7]. In a Swedish cohort, the hazard ratio in people above 2.70 mmol/L on that cohort's scale, against its reference range, was 2.88 (95% CI 2.70 to 3.07) for heart attack and 2.31 (2.21 to 2.41) for death from any cause, and the associations weakened but remained after adjustment for glucose [8]. Among Swedish adults without diabetes, the lowest tenth of fructosamine had a hazard ratio for death of 1.20 against the middle eight tenths, falling to 1.11 (95% CI 1.07 to 1.15) after adjustment for six other blood markers [9].
Population: 11,104 adults with and without diabetes, with 1,096 new cases of coronary heart disease, 605 ischemic strokes, 1,432 cases of heart failure and 2,860 deaths [7]. 338,443 people followed 19 years on average, with 21,526 heart attacks and 73,458 deaths [8]. 215,011 people without diabetes, 41,388 deaths over 25 years [9].
Sources: [7], [8], [9]
Caveats: Observational. The Swedish cohort reported fructosamine in mmol/L on a scale about ten times this page's, so its cut points are not placed here. Its authors found that smoking and chronic inflammation at least partly explained the higher mortality at low fructosamine [9]. No source here tested whether lowering fructosamine changes these outcomes, and a laboratory guideline states that treatment goals have not been established [2].

Gamma-glutamyl transferase (GGT)Moderate

Type 2 diabetes

Linked to more type 2 diabetes, 34% higher risk in the top third of GGT across 24 cohorts. A genetic study found no evidence that GGT drives it.

The numbers and sources

Effect: In a meta-analysis of 24 cohort studies, the relative risk of type 2 diabetes comparing the top and bottom thirds of baseline GGT was 1.34 (95% CI 1.27 to 1.42), with a nonlinear dose-response pattern in both men and women and significant heterogeneity between studies [10]. In a Mendelian randomization study, genetically predicted GGT showed no evidence of association with type 2 diabetes (odds ratio 0.92 per standard unit, 95% CI 0.80 to 1.06) [11].
Population: 177,307 participants and 11,155 cases of type 2 diabetes in 24 cohorts [10]. Up to 64,094 people with type 2 diabetes and 607,012 controls in the genetic study [11].
Sources: [10], [11]
Caveats: The cohort finding is observational. The genetic result suggests GGT marks a higher risk without causing it [11].

Hemoglobin A1c (HbA1c)Moderate

Developing diabetes

Linked to future diabetes in a graded way: 25 to 50% of people at 6.0 to 6.5% developed diabetes within 5 years.

The numbers and sources

Effect: A systematic review of 16 studies found annualized diabetes incidence ranged from 0.1% at HbA1c below 5.0% to 54.1% at 6.1% or above. Five-year incidence was under 9% at 5.0 to 5.5%, 9 to 25% at 5.5 to 6.0% and 25 to 50% at 6.0 to 6.5% [7]. In US adults without diabetes, hazard ratios for diagnosed diabetes against a reference of 5.0 to under 5.5% were 0.52 (95% CI 0.40 to 0.69) below 5.0%, 1.86 (1.67 to 2.08) at 5.5 to under 6.0%, 4.48 (3.92 to 5.13) at 6.0 to under 6.5% and 16.47 (14.22 to 19.08) at 6.5% or greater [8].
Population: 16 prospective studies [7]. 11,092 black or white adults without diabetes or cardiovascular disease in the Atherosclerosis Risk in Communities study [8].
Sources: [7], [8]
Caveats: Observational. Risk rose steeply across the whole range from 5.0 to 6.5%, not from a single level [7].

Hemoglobin A1c (HbA1c)Moderate

Death from any cause

Linked to more deaths at both ends: above 6.0% and below 5.0% in people without diabetes, and above 9.0% and below 6.0% in people with diabetes.

The numbers and sources

Effect: A meta-analysis of observational studies found higher death rates from any cause in people without diabetes at HbA1c above 6.0% (HR 1.74, 95% CI 1.38 to 2.20) and below 5.0% (HR 1.19, 95% CI 1.04 to 1.36). In people with diabetes, rates were higher above 9.0% (HR 1.69, 95% CI 1.09 to 2.66) and below 6.0% (HR 1.57, 95% CI 1.14 to 2.17). Results for cardiovascular death were similar [10]. A US cohort of adults without diabetes also found a J-shaped association with death from any cause [8].
Population: 46 observational studies in the meta-analysis, of 74 reviewed [10]. 11,092 adults [8].
Sources: [10], [8]
Caveats: Observational. The excess at low values may reflect illness or conditions that lower HbA1c, not an effect of low blood sugar itself, and the studies cannot separate the two.

High-sensitivity C-reactive protein (hs-CRP)Moderate

Type 2 diabetes

Linked to more type 2 diabetes at higher levels, with a 26% higher risk in a meta-analysis of 22 cohorts.

The numbers and sources

Effect: A meta-analysis of prospective studies found elevated CRP was associated with a higher risk of type 2 diabetes (relative risk 1.26, 95% CI 1.16 to 1.37), with no sign of publication bias [11].
Population: 40,735 participants with 5,753 cases in 22 cohorts [11].
Sources: [11]
Caveats: Observational, and the abstract does not give the size of the CRP difference the estimate refers to [11].

HOMA-IRModerate

Developing type 2 diabetes

Linked to more new type 2 diabetes at higher HOMA-IR, about 87% more in a meta-analysis of cohorts, a weaker link than for fasting glucose.

The numbers and sources

Effect: In a meta-analysis of cohort studies, a higher HOMA-IR went with a hazard ratio of 1.87 (95% CI 1.40 to 2.49) for developing type 2 diabetes [12]. In a Dutch cohort, the odds ratio for type 2 diabetes in the highest against the lowest fifth was 2.8 (95% CI 1.4 to 5.6) for HOMA-IR and 1.9 (0.9 to 3.6) for fasting insulin, against 9.0 (4.4 to 18.5) for fasting glucose and 3.8 (2.0 to 7.2) for HbA1c [13].
Population: 215,878 participants in 38 cohort studies across all outcomes [12]. 1,349 people aged 50 to 75 without diabetes, followed a mean 6.4 years, 152 of whom developed diabetes [13].
Sources: [12], [13]
Caveats: Observational. HOMA-IR contains fasting glucose, which is itself used to diagnose diabetes. A laboratory guideline reports that measures of insulin resistance predicted diabetes or coronary disease only moderately, with no threshold effects [2].

HOMA-IRModerateStudied in women

Polycystic ovary syndrome

Linked to higher HOMA-IR in women with polycystic ovary syndrome, 2.68 in non-obese and 4.38 in obese women with it, against 1.34 and 2.44 without it.

The numbers and sources

Effect: Pooled mean HOMA-IR was 4.38 (95% CI 3.84 to 4.92) in obese women with PCOS, 2.68 (2.16 to 3.20) in non-obese women with PCOS, 2.44 (2.06 to 2.82) in obese women without PCOS and 1.34 (1.06 to 1.63) in non-obese women without it. Every group differed from the others except non-obese women with PCOS and obese women without it [20]. In 28 insulin clamp studies, insulin sensitivity was 27% lower in women with PCOS than in controls (99% CI ±6%), independent of body mass index [21].
Population: 3,037 women in observational studies, in four groups by PCOS and obesity [20]. 28 clamp studies of premenopausal women with PCOS and controls [21].
Sources: [20], [21]
Caveats: Observational. Non-obese women with PCOS and obese women without it had similar values, so body weight is tied to HOMA-IR about as strongly as the condition is [20]. The pooled means come from studies using different insulin assays (see Effect of the insulin assay and the blood sample).

Magnesium levelModerate

Type 2 diabetes

Linked to less type 2 diabetes at higher levels, 36% lower risk in the highest category than the lowest across 4 pooled cohorts.

The numbers and sources

Effect: Comparing the highest with the lowest category of circulating magnesium, the pooled relative risk of type 2 diabetes was 0.64 (95% CI 0.50 to 0.81). The linear association per 0.1 mmol/L (0.24 mg/dL) did not reach significance (RR 0.90, 95% CI 0.81 to 1.002) [8].
Population: 31,284 participants in 4 cohorts, 2,680 cases, average follow-up 8.8 years [8].
Sources: [8]
Caveats: Observational. The authors report that the associations were sensitive to the exclusion of individual studies [8].

Non-HDL cholesterol (non-HDL-C)Moderate

Type 2 diabetes

Linked to more new type 2 diabetes at higher levels, 16% more in the higher than the lower non-HDL cholesterol group across 8 cohort studies.

The numbers and sources

Effect: In a meta-analysis of cohort studies in the general population, the pooled effect size for type 2 diabetes was 1.16 (95% CI 1.079 to 1.251) for higher against lower non-HDL cholesterol. The result did not change with region, share of men, sample size or adjustment for body mass index, hypertension, waist circumference and family history of diabetes [19].
Population: 251,672 participants in 8 cohort studies [19].
Sources: [19]
Caveats: Observational, and the authors describe earlier studies as inconsistent and call for large prospective cohorts to confirm the finding. Each study defined its own higher and lower groups [19].

Selenium levelModerate

Type 2 diabetes at higher levels

Shown in trials to raise type 2 diabetes risk about 11% when selenium supplements were taken, and linked to more diabetes at higher blood selenium.

The numbers and sources

Effect: In a meta-analysis, trials of 200 µg of selenium a day against placebo found more new type 2 diabetes with selenium (RR 1.11, 95% CI 1.01 to 1.22). In the observational studies the risk rose roughly in line with plasma or serum selenium, with a relative risk of 3.6 (1.4 to 9.4) at 140 µg/L against less than 45 µg/L [8]. In one of the trials, 200 µg a day for an average of 7.7 years produced a hazard ratio of 1.55 (1.03 to 2.33). Risk rose across thirds of plasma selenium at the start of the trial, and people in the top third had a hazard ratio of 2.70 (1.30 to 5.61) with supplementation [9]. Authors of a later trial give about 122 µg/L as the plasma selenium of that top third [10].
Population: 5 randomized trials and 50 potentially eligible nonexperimental studies [8]. 1,202 people seen in dermatology clinics in the eastern United States [9].
Sources: [8], [9], [10]
Caveats: Diabetes was a secondary outcome in the 2007 trial, and diagnoses were self-reported and then validated in most participants [9]. The meta-analysis authors describe the relative increase in risk as small [8]. The 122 µg/L figure is taken from a later paper's description of the trial, not from the trial report itself [10].

Sex hormone binding globulin (SHBG)Moderate

Type 2 diabetes

Shown in genetic and cohort studies to go with less type 2 diabetes at higher levels, in both women and men.

The numbers and sources

Effect: In a meta-analysis of prospective studies, women with SHBG above 60 nmol/L had an 80% lower risk of type 2 diabetes than women at or below it (relative risk 0.20, 95% CI 0.12 to 0.30), and men above 28.3 nmol/L had a 52% lower risk than men at or below it (0.48, 0.33 to 0.69) [4]. In a nested case-control study of postmenopausal women, the odds ratios for the second to fourth quarters of SHBG against the lowest were 0.16 (0.08 to 0.33), 0.04 (0.01 to 0.12) and 0.09 (0.03 to 0.21), and the finding was replicated in men (highest against lowest quarter 0.10, 0.03 to 0.36). In the same study's genetic analysis, each standard deviation of genetically higher SHBG had a predicted odds ratio of 0.28 (0.13 to 0.58) in women and 0.29 (0.15 to 0.58) in men [5]. A second genetic study found the SHBG-raising allele of one common variant associated with less type 2 diabetes (odds ratio 0.94, 0.91 to 0.97) [6].
Population: 43 prospective and cross-sectional studies with 6,974 women and 6,427 men [4]. 359 women with new type 2 diabetes and 359 controls, then 170 men with it and 170 controls [5]. 27,657 people with type 2 diabetes and 58,481 controls from 15 studies [6].
Sources: [4], [5], [6]
Caveats: The cohort findings are observational. Genetic estimates describe lifelong differences, not the effect of changing the level later in life.

Total testosteroneModerate

Type 2 diabetes

Shown in genetic and cohort studies to run in opposite directions by sex: higher testosterone went with less type 2 diabetes in men and more in women.

The numbers and sources

Effect: In a meta-analysis of 43 studies, men with type 2 diabetes had lower testosterone than controls (mean difference -76.6 ng/dL, 95% CI -99.4 to -53.6) and women with diabetes had higher (6.1 ng/dL, 2.3 to 10.1). In prospective studies, men with higher testosterone, 449.6 to 605.2 ng/dL, had a 42% lower risk (relative risk 0.58, 0.39 to 0.87) [18]. In a genetic study, each standard deviation of genetically higher testosterone raised the odds of type 2 diabetes in women (odds ratio 1.37, 1.22 to 1.53) and lowered them in men (0.86, 0.76 to 0.98) [19]. In a 2-year trial in men with testosterone of 14.0 nmol/L or lower and impaired glucose tolerance or new diabetes, 12% of the testosterone group and 21% of the placebo group had diabetes on a glucose tolerance test at the end (relative risk 0.59, 0.43 to 0.80) [20].
Population: 6,427 men and 6,974 women in 43 prospective and cross-sectional studies [18]. 425,097 UK Biobank participants [19]. 1,007 men aged 50 to 74 with a waist of 95 cm or more, all enrolled in a lifestyle program [20].
Sources: [18], [19], [20]
Caveats: The cohort findings are observational. The trial's authors report that rises in hematocrit might limit treatment and that durability and cardiovascular effects remain to be studied [20].

Uric acid (serum urate)Moderate

Type 2 diabetes

Linked to more type 2 diabetes in pooled cohorts, 17% higher risk per 1 mg/dL, while a genetic study found no association.

The numbers and sources

Effect: In a meta-analysis of 11 cohort studies, each 1 mg/dL higher serum uric acid went with a relative risk of type 2 diabetes of 1.17 (95% CI 1.09 to 1.25), or 1.11 (1.03 to 1.20) after adjustment for publication bias [15]. In a Mendelian randomization study, a genetically higher serum urate of 1 standard deviation was not associated with type 2 diabetes [11].
Population: 42,834 participants and 3,305 new cases of type 2 diabetes over 2.0 to 13.5 years [15]. 26,488 cases and 83,964 controls in the genetic study [11].
Sources: [15], [11]
Caveats: The cohort finding is observational and showed signs of publication bias [15]. The genetic study's authors conclude that lowering serum urate may not reduce cardiometabolic risk [11].

White blood cell count (WBC)Moderate

Type 2 diabetes

Linked to more type 2 diabetes at higher counts, 61% more in the top third than the bottom third, an estimate its authors consider too high.

The numbers and sources

Effect: In a meta-analysis of 20 studies, the relative risk of type 2 diabetes was 1.61 (95% CI 1.45 to 1.79) for the top against the bottom third of white cell count, with large differences between studies (I² 83%). In cohort studies alone it was 1.48 (1.22 to 1.79). In one English cohort the hazard ratio fell from 2.19 to 1.82 after accounting for smoking, family history, activity, education and body size [11].
Population: 8,647 people with type 2 diabetes and 85,040 without, in 20 cross-sectional and prospective studies [11].
Sources: [11]
Caveats: Observational. The authors found publication bias and incomplete control for confounders, and describe the observed association as likely an overestimate [11].

Zinc levelModerate

Type 2 diabetes

Linked to more type 2 diabetes at higher serum or plasma zinc in pooled observational studies (odds ratio 1.64), while higher zinc intake from food went with less.

The numbers and sources

Effect: In a meta-analysis of observational studies, higher serum or plasma zinc was associated with type 2 diabetes (OR 1.64, 95% CI 1.25 to 2.14). Comparing the highest with the lowest zinc intake from diet, the odds ratio was 0.87 (95% CI 0.78 to 0.98), and no association was seen for supplemental or total zinc intake. The authors read the raised blood level as a sign of disturbed zinc handling in diabetes [8].
Population: 16 observational studies of incident diabetes or of people with type 2 diabetes compared with healthy subjects [8].
Sources: [8]
Caveats: Observational, and some of the studies measured zinc in people who already had diabetes, so the direction of the link is not established [8].

Bilirubin (total)Limited

Type 2 diabetes

One genetic study found higher bilirubin went with 42% lower odds of type 2 diabetes per standard deviation in 3,381 adults.

The numbers and sources

Effect: Per standard deviation higher log bilirubin, the observed odds of type 2 diabetes were 25% lower (OR 0.75, 95% CI 0.62 to 0.92). Using a UGT1A1 gene variant that explained 19.5% of the variation in bilirubin, the genetically estimated odds ratio was 0.58 (95% CI 0.39 to 0.84) [9].
Population: 3,381 adults aged 28 to 75 without diabetes at baseline, 210 of whom developed type 2 diabetes over a median of 7.8 years [9].
Sources: [9]
Caveats: One cohort with 210 cases and one gene variant. The same kind of genetic analysis found no effect on heart disease in a larger sample [7].

Blood urea nitrogen (BUN)Limited

Type 2 diabetes

One cohort of 1.3 million US veterans found 23% more new diabetes with BUN above 25 mg/dL than at 25 or below.

The numbers and sources

Effect: In 1,337,452 US veterans without diabetes followed for a median of 4.93 years, a BUN over 25 mg/dL went with more new diabetes than a BUN of 25 mg/dL or less (hazard ratio 1.23, 95% CI 1.21 to 1.25), with eGFR in the model. The association held at every level of eGFR, including 60 mL/min/1.73 m2 or more (hazard ratio 1.27, 95% CI 1.24 to 1.31). Each 10 mg/dL higher BUN went with a hazard ratio of 1.15 (1.14 to 1.16) [8].
Population: 1,337,452 US veterans, with 172,913 new cases of diabetes [8].
Sources: [8]
Caveats: Observational, one cohort of US veterans. The cut point of 25 mg/dL was the study's analysis category, not a reference range [8].

C-peptideLimited

Death in adults without diabetes

Limited evidence: two analyses of one US survey linked the highest quarter of fasting C-peptide to about 70% to 80% more deaths in adults without diabetes.

The numbers and sources

Effect: In 5,153 adults aged 40 to 74 without diabetes, the highest quarter of fasting C-peptide, 0.984 nmol/L (about 2.97 ng/mL) or above, had an adjusted hazard ratio of 1.72 (95% CI 1.34 to 2.21) for death from any cause and 1.60 (1.07 to 2.39) for cardiovascular death against the lowest quarter, 0.418 nmol/L (about 1.26 ng/mL) or below. C-peptide predicted these deaths better than fasting glucose, insulin or indices built from them [7]. In 5,902 adults aged 40 and older from the same survey, the highest quarter had a hazard ratio of 1.80 (1.33 to 2.43) for death from any cause and 3.20 (2.07 to 4.93) for cardiovascular death against the lowest [8].
Population: 5,153 adults without diabetes in the third US National Health and Nutrition Examination Survey of 1988 to 1994, followed for a median of 14.4 years with 513 deaths [7]. 5,902 adults with no history of diabetes in the same survey, mean C-peptide 0.78 nmol/L [8].
Sources: [7], [8]
Caveats: Observational, and both analyses draw on the same survey, so they are not independent. The authors read a high C-peptide in a person without diabetes as a sign of insulin resistance. Neither source tested whether lowering it changes the risk of death [7][8].

C-peptideLimited

Eye disease and low blood sugar in type 1 diabetes

Limited evidence: in one type 1 diabetes trial cohort, higher stimulated C-peptide went with less retinopathy and hypoglycemia, and fasting C-peptide showed weaker links.

The numbers and sources

Effect: In the intensive treatment group of the Diabetes Control and Complications Trial, a 50% higher stimulated C-peptide at entry, such as 0.15 against 0.10 nmol/L, went with a 25% lower risk of sustained retinopathy, an 8.2% lower risk of hypoglycemia, a lower insulin dose and an HbA1c 0.07 percentage points lower. Outcomes changed in a nearly linear way as C-peptide fell to unmeasurable levels, with no threshold. Fasting C-peptide had weaker associations with outcomes than stimulated C-peptide [9].
Population: Participants in the Diabetes Control and Complications Trial with up to 5 years of type 1 diabetes at entry, followed for 7 years [9].
Sources: [9]
Caveats: The main findings are for C-peptide measured after a stimulus, a different quantity from the fasting value on this page. They are unadjusted, come from one trial cohort, and concern people with type 1 diabetes only. The abstract gives no participant count [9].

Free T4 (free thyroxine)Limited

Type 2 diabetes

One cohort found less type 2 diabetes at higher free T4, with a hazard ratio of 0.96 per 1 pmol/L, including inside the reference range.

The numbers and sources

Effect: In the Rotterdam Study, each 1 pmol/L higher free T4 had a hazard ratio for diabetes of 0.96 (95% CI 0.93 to 0.99), and the same within the reference range (0.96, 0.92 to 0.99). Higher TSH went with more diabetes (hazard ratio 1.13 per log unit, 1.08 to 1.18). In people with prediabetes, the hazard ratio for progression was 0.63 (0.48 to 0.82) in the highest third of free T4 against the lowest [5].
Population: 8,452 adults, mean follow-up 7.9 years, 798 new cases of diabetes [5].
Sources: [5]
Caveats: One observational cohort [5].

FructosamineLimited

Eye and kidney complications

One US cohort linked fructosamine above the 95th percentile to 50% more new chronic kidney disease, and higher levels to more retinopathy.

The numbers and sources

Effect: Fructosamine and glycated albumin were strongly associated with retinopathy (P below 0.0001 for trend). The adjusted hazard ratio for chronic kidney disease was 1.50 (95% CI 1.22 to 1.85) for fructosamine above the 95th percentile and 1.48 (1.20 to 1.83) for glycated albumin, compared with people without diabetes below the 75th percentile. Prediction of chronic kidney disease was nearly as strong as with HbA1c, with C statistics of 0.717, 0.717 and 0.726 [5].
Population: 11,348 adults without diabetes and 958 with diagnosed diabetes, followed for two decades [5].
Sources: [5]
Caveats: Observational, from one cohort. A laboratory guideline states that few published studies have convincingly shown a relationship between glycated protein levels and the chronic complications of diabetes [2].

Serum creatinineLimitedStudied in men

Type 2 diabetes at low levels

Limited evidence: one Japanese cohort of 8,570 men linked creatinine of 0.40 to 0.60 mg/dL to about twice the odds of type 2 diabetes over 4 years.

The numbers and sources

Effect: During 4 years, 877 men developed type 2 diabetes. Men with a creatinine of 0.40 to 0.60 mg/dL had an adjusted odds ratio of 1.91 (95% CI 1.44 to 2.54) against men at 0.71 to 0.80 mg/dL. The authors treat low creatinine as a possible stand-in for low skeletal muscle mass [8].
Population: 8,570 Japanese men aged 40 to 55 without diabetes at entry [8].
Sources: [8]
Caveats: One observational cohort of men in one country. Muscle mass was not measured, so the link to muscle is the authors' interpretation [8].

Fasting glucoseStrong

Diabetes and heart disease in trials that treated raised glucose before diabetes

Mixed results: in trials of people with raised glucose, lifestyle change cut new diabetes 58%, while neither it nor insulin aimed at normal fasting glucose cut cardiovascular events.

The numbers and sources

Effect: In people with elevated fasting and post-load glucose, a lifestyle program reduced the incidence of diabetes by 58% (95% CI 48 to 66) and metformin by 31% (95% CI 17 to 43) against placebo over 2.8 years [12]. Over a median 21 years of follow-up of the same participants, neither metformin (HR 1.03, 95% CI 0.78 to 1.37) nor the lifestyle program (HR 1.14, 95% CI 0.87 to 1.50) reduced major cardiovascular events [13]. In a trial that gave insulin glargine to reach a fasting glucose of 95 mg/dL or below, cardiovascular outcomes did not differ from standard care over 6.2 years (HR 1.02, 95% CI 0.94 to 1.11). New diabetes was less frequent (30% against 35%), severe hypoglycemia was more frequent (1.00 against 0.31 per 100 person-years) and weight rose [14].
Population: 3,234 people without diabetes, mean age 51, mean body mass index 34.0 [12][13]. 12,537 people, mean age 63.5, with cardiovascular risk factors plus impaired fasting glucose, impaired glucose tolerance or type 2 diabetes [14].
Sources: [12], [13], [14]
Caveats: A marker that tracks risk is not always one whose lowering changes it. In the long follow-up, all groups were offered a lifestyle program and many used statins and blood pressure drugs, which the authors say may have diluted any effect [13]. The insulin trial included people who already had type 2 diabetes [14].

Fasting insulin and HOMA-IRStrong

Coronary heart disease

Mixed results: two meta-analyses found no clear link between fasting insulin and coronary heart disease and one found 50% more, while HOMA-IR was linked in the one that pooled it.

The numbers and sources

Effect: In 14 Western prospective studies, the odds ratio for coronary heart disease in the top against the bottom third of fasting insulin was 1.12 (95% CI 0.98 to 1.28), and the authors concluded the association is likely to be more modest than previously suspected [9]. In a meta-analysis of adults without diabetes, the relative risk of coronary heart disease for high against low values was 1.12 (95% CI 0.92 to 1.37) for fasting insulin, 1.52 (1.31 to 1.76) for fasting glucose and 1.64 (1.35 to 2.00) for HOMA-IR. Per standard deviation it was 1.04 (0.96 to 1.12) for insulin and 1.46 (1.26 to 1.69) for HOMA-IR [10]. A third meta-analysis found a relative risk of 1.50 (95% CI 1.28 to 1.77) for the highest against the lowest fasting insulin group, and 1.16 (1.10 to 1.22) per 50 pmol/L [11].
Population: 2,649 coronary heart disease cases in 14 studies of fasting insulin [9]. 516,325 participants in 65 studies of glucose, insulin or HOMA-IR [10]. 22,379 people and 1,986 cases in 9 studies [11].
Sources: [9], [10], [11]
Caveats: Observational. Each study compared its own high and low groups, measured with its own insulin assay, so the pooled groups share no common cut point.

Hemoglobin A1c (HbA1c)Strong

Heart attack and death in trials that lowered HbA1c in type 2 diabetes

Mixed results: pooled trials of intensive glucose lowering found 9% fewer major cardiovascular events and no fewer deaths, and one trial aiming below 6.0% found more deaths.

The numbers and sources

Effect: A prospectively planned meta-analysis found that more intensive glucose control reduced major cardiovascular events by 9% (HR 0.91, 95% CI 0.84 to 0.99) and heart attack by 15% (HR 0.85, 95% CI 0.76 to 0.94). Death from any cause (HR 1.04, 95% CI 0.90 to 1.20) and cardiovascular death (HR 1.10, 95% CI 0.84 to 1.42) were not reduced, and major hypoglycemia rose (HR 2.48, 95% CI 1.91 to 3.21) [15]. In one of the pooled trials, a strategy targeting HbA1c below 6.0% reached a median of 6.4% against 7.5% with standard therapy. Deaths were higher in the intensive group (257 against 203, HR 1.22, 95% CI 1.01 to 1.46), and the intensive strategy was stopped after a mean 3.5 years [16].
Population: 27,049 participants with type 2 diabetes and 2,370 major vascular events over 4.4 years [15]. 10,251 patients, mean age 62, with cardiovascular disease or risk factors and a median HbA1c of 8.1% [16].
Sources: [15], [16]
Caveats: A marker that tracks risk is not always one whose lowering reduces it. These trials used the drugs and insulin regimens of their time in older, high-risk people with long-standing diabetes, and do not describe people without diabetes.

Aspartate aminotransferase (AST)Moderate

Type 2 diabetes

Mixed results: 9 pooled cohort studies found no increase in type 2 diabetes with higher AST, while a genetic study found a 25% higher risk.

The numbers and sources

Effect: In a meta-analysis, 9 prospective studies that measured AST gave a relative risk of type 2 diabetes of 1.02 (95% CI 0.99 to 1.04) per standard deviation of log AST, which the authors read as no evidence of increased risk [8]. In a Mendelian randomization study, genetically predicted higher AST was related to more type 2 diabetes (odds ratio 1.25 per standard unit, 95% CI 1.14 to 1.38) [9].
Population: 17 studies with 60,359 participants in the meta-analysis, 9 of which measured AST [8]. Up to 64,094 people with type 2 diabetes and 607,012 controls in the genetic study [9].
Sources: [8], [9]
Caveats: The cohort studies are observational. The genetic study's authors read their findings as support for insulin resistance leading to fatty liver, which in turn raises diabetes risk, so AST may mark that process and not drive it [9].

Fasting insulin and HOMA-IRModerate

Death from any cause

Mixed results: a meta-analysis of adults without diabetes tied the highest HOMA-IR group to 34% more deaths, with a weaker, borderline link for fasting insulin.

The numbers and sources

Effect: Comparing the highest with the lowest category, the pooled adjusted risk ratio for death from any cause was 1.13 (95% CI 1.00 to 1.27, P = 0.058) for fasting insulin and 1.34 (1.11 to 1.62) for HOMA-IR. For cardiovascular death it was 2.11 (1.01 to 4.41) for HOMA-IR in two studies and 1.40 (0.49 to 3.96) for fasting insulin in one [14].
Population: 26,976 adults without diabetes in 7 articles [14].
Sources: [14]
Caveats: Observational. The authors call the findings for cardiovascular death unreliable because so few studies were included [14].

HOMA-IRModerate

Death from any cause

Mixed results: one meta-analysis tied the highest HOMA-IR group to 34% more deaths from any cause, while another found no link with fatal cardiovascular events or cancer deaths.

The numbers and sources

Effect: In adults without diabetes, the pooled adjusted risk ratio for death from any cause comparing the highest with the lowest category was 1.34 (95% CI 1.11 to 1.62) for HOMA-IR and 1.13 (1.00 to 1.27, P = 0.058) for fasting insulin. For cardiovascular death it was 2.11 (1.01 to 4.41) for HOMA-IR in two studies [15]. A later meta-analysis found no association of a higher HOMA-IR with fatal major cardiovascular events or with cancer mortality [12].
Population: 26,976 adults without diabetes in 7 articles [15]. 215,878 participants in 38 cohort studies across all outcomes [12].
Sources: [15], [12]
Caveats: Observational. The authors of the first analysis call the findings for cardiovascular death unreliable because so few studies were included [15].

ProlactinModerate

Type 2 diabetes

Mixed results: higher prolactin within the normal range went with less existing type 2 diabetes and was not tied to new cases in one meta-analysis.

The numbers and sources

Effect: For the top against the bottom quarter of prolactin, the odds of already having type 2 diabetes were 0.52 in men (95% CI 0.35 to 0.77) and 0.46 in women (0.30 to 0.73). For new cases the relative risks were 1.21 in men (0.79 to 1.87) and 0.77 in women (0.48 to 1.22) [9].
Population: 6 studies: cross-sectional studies of 6,670 people and longitudinal studies of 13,203 [9].
Sources: [9]
Caveats: The association appeared only in cross-sectional data, and the authors state that further studies are needed to address whether it is causal [9].

C-peptideLimited

Nerve function in a trial that gave C-peptide in type 1 diabetes

One trial found 52 weeks of long-acting C-peptide in type 1 diabetes did not improve nerve conduction against placebo, while vibration sense improved 25%.

The numbers and sources

Effect: Patients received weekly pegylated C-peptide at 0.8 mg or 2.4 mg, or placebo. Plasma C-peptide rose to 1.8 to 2.2 nmol/L on the low dose and 5.6 to 6.8 nmol/L on the high dose. After 52 weeks, sural nerve conduction velocity had risen by 1.0 m/s on C-peptide and 1.2 m/s on placebo. Vibration perception threshold improved by 25% on C-peptide and was unchanged on placebo (P below 0.001), and the clinical neuropathy score did not change [10].
Population: 250 people with type 1 diabetes and mild to moderate peripheral neuropathy, 144 on C-peptide and 106 on placebo [10].
Sources: [10]
Caveats: One trial, in which the main nerve conduction measure did not differ from placebo. It tested giving C-peptide to people who make little of their own, and says nothing about a person's own C-peptide level [10].

Fasting insulin and HOMA-IRLimited

Developing type 2 diabetes

Limited evidence: in one cohort HOMA-IR predicted type 2 diabetes and fasting insulin alone did not clearly, and both predicted it less strongly than fasting glucose.

The numbers and sources

Effect: In a Dutch cohort, the odds ratio for type 2 diabetes in the highest against the lowest fifth was 1.9 (95% CI 0.9 to 3.6) for fasting insulin and 2.8 (1.4 to 5.6) for HOMA-IR, against 9.0 (4.4 to 18.5) for fasting glucose and 3.8 (2.0 to 7.2) for HbA1c [12]. In a US clinic sample tested for suspected prediabetes or diabetes, the odds of having prediabetes rose across quartiles of fasting insulin, to 5.04 (95% CI 3.12 to 8.12) in the highest quartile (mean 25.9 µIU/mL) against the lowest (mean 4.9 µIU/mL). The authors reported that a fasting insulin above 9.0 µIU/mL identified 80% of people with prediabetes [13].
Population: 1,349 people aged 50 to 75 without diabetes, followed a mean 6.4 years, 152 of whom developed diabetes [12]. 965 patients, mostly white, obese and female, 287 of them with prediabetes [13].
Sources: [12], [13]
Caveats: The clinic study is retrospective and cross-sectional, in people already referred for testing, and its 9.0 µIU/mL figure comes from one assay in one clinic [13]. A US diabetes workgroup states that cutoff values of insulin tied to a known size of diabetes risk have yet to be determined [3].

CortisolLimited

Type 2 diabetes in a genetic study

One genetic study found no association between genetically predicted cortisol and type 2 diabetes.

The numbers and sources

Effect: Genetically predicted cortisol, based on six variants, had an odds ratio for type 2 diabetes of 1.00 per unit of log-transformed cortisol (95% CI 0.96 to 1.04). It was also unrelated to adiposity, blood sugar traits, blood pressure and lipids. Genetically predicted diabetes was not associated with cortisol either [6].
Population: Six variants identified in 12,597 people, applied to genome-wide studies of type 2 diabetes [6].
Sources: [6]
Caveats: One genetic study, whose authors call for replication. Genetic estimates describe lifelong differences, not the effect of a change later in life [6].