Table 1.
Summary of findings for effects of long chain omega-3 and α-linolenic acid (omega-3) on primary outcomes. High compared with low long chain omega-3 (LCn3) and α-linolenic acid (ALA) for prevention or treatment of diabetes
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect (95% CI) |
No of participants (studies) |
Certainty of evidence (GRADE)† |
Comments | |
|---|---|---|---|---|---|---|
| Risk with low omega-3 fats | Risk with high omega-3 fats | |||||
| Diagnosis of type 2 diabetes mellitus—LCn3 | 37 per 1000 | 37 (32 to 44) per 1000 | RR 1.00 (0.85 to 1.17) | 58 643 (17 RCTs) | ⨁⨁⨁◯ MODERATEa,b,c,d | LCn3 probably has little or no effect on diagnosis of type 2 diabetes. Downgraded once for imprecision |
| Diagnosis of type 2 diabetes mellitus—ALA | 14 per 1000 | 9 (5 to 19) per 1000 | RR 0.68 (0.33 to 1.39) | 18 243 (2 RCTs) | ⨁◯◯◯ VERY LOWb,e,f | Effect of ALA on diabetes diagnosis is uncertain as evidence is of very low quality. Downgraded once for inconsistency and twice for imprecision |
| Diagnosis of impaired glucose tolerance—LCn3 | Not pooled | Not pooled | Not pooled | (0 RCTs) | - | No RCTs assessed effects of LCn3 on diagnosis of impaired glucose tolerance |
| Diagnosis of impaired glucose tolerance—ALA | Not pooled | Not pooled | Not pooled | (0 RCTs) | - | No RCTs assessed effects of ALA on diagnosis of impaired glucose tolerance |
| Glycated haemoglobin (HbA1c, %)—LCn3 | Mean HbA1c 6.5% | Mean HbA1c 0.02% lower (0.07% lower to 0.04% higher) | - | 32 798 (16 RCTs) | ⨁⨁⨁◯ MODERATEb,g | LCn3 probably has little or no effect on glycated haemoglobin. Downgraded once for risk of bias |
| Glycated haemoglobin (HbA1c, %)—ALA | Mean HbA1c 7.0% | Mean HbA1c 0.01% higher (0.43% lower to 0.45% higher) | - | 178 (3 RCTs) | ⨁⨁◯◯ LOWg,h,i | ALA may have little or no effect on glycated haemoglobin. Downgraded once each for imprecision and risk of bias |
| Plasma glucose, fasting (mmol/L)—LCn3 | Mean plasma glucose 6.2 mmol/L | Mean plasma glucose 0.04 (0.02 to 0.07) mmol/L higher | - | 35 156 (33 RCTs) | ⨁⨁◯◯ LOWj,k | LCn3 may have little or no effect on fasting plasma glucose. Downgraded once each for risk of bias and publication bias |
| Plasma glucose, fasting (mmol/L —ALA | Mean plasma glucose 6.2 mmol/L | Mean plasma glucose 0.07 mmol/L lower (0.16 lower to 0.02 higher) | - | 648 (7 RCTs) | ⨁⨁⨁◯ MODERATEl,m | ALA probably has little or no effect on fasting plasma glucose. Downgraded once for risk of bias |
| Fasting insulin (pmol/L)—LCn3 | Mean fasting insulin 98 pmol/L | Mean fasting insulin 1.02 pmol/L higher (4.34 lower to 6.37 higher) | - | 2077 (17 RCTs) | ⨁⨁◯◯ LOWn,p | LCn3 may have little or no effect on fasting insulin. Downgraded once each for risk of bias and imprecision |
| Fasting insulin (pmol/L)—ALA | Mean fasting insulin 80 pmol/L | Mean fasting insulin 5.3 pmol/L higher (4.68 lower to 15.27 higher) | - | 469 (6 RCTs) | ⨁⨁◯◯ LOWo,p | ALA may increase fasting insulin. Downgraded once each for imprecision and risk of bias |
| HOMA-IR—LCn3 | Mean HOMA-IR 4.6 | Mean HOMA-IR 0.06 higher (0.21 lower to 0.33 higher) | - | 1064 (13 RCTs) | ⨁⨁⨁◯ MODERATEp,q,r | LCn3 probably has little or no effect on HOMA-IR. Downgraded once for imprecision |
| HOMA-IR—ALA | Mean HOMA-IR 3.4 | Mean HOMA-IR 0.1 higher (0.5 lower to 0.7 higher) | - | 294 (3 RCTs) | ⨁⨁◯◯ LOWs,t | ALA may have little or no effect on HOMA-IR. Downgraded once for imprecision and once for risk of bias and publication bias combined |
HOMA-IR=homoeostatic model assessment for insulin resistance; RCT=randomised controlled trial; RR=risk ratio.
Patient or population: people with or without diabetes at baseline; setting: these are long term trials, so participants lived in the community; intervention: higher omega-3 intake; comparison: lower omega-3 intake.
Risk in intervention group (and its 95% CI) is based on assumed risk in comparison group and relative effect of intervention (and its 95% CI). Note that GRADE describes risk and 95% CI without using negative numbers; for example, GRADE states “0.02% lower (0.07 lower to 0.04 higher),” which would normally be described as “–0.02% (−0.07% to 0.04%).”
High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
a. Risk of bias: limiting studies to those suggesting good compliance and those at summary risk of bias both suggested greater protection from higher LCn3 intake, although neither was statistically significant; these protective effects were entirely due to two small trials (Derosa 2016 and Sawada 2016), and removing them removed heterogeneity and suggestion of any effect. Not downgraded.
b. Indirectness: most studies were conducted in industrialised countries, but the data from the 3 RCTs conducted in industrialising countries (Martinez 2014, Wang 2016, Zheng 2016) seem consistent. Not downgraded.
c. Imprecision: despite >55 000 participants and >2300 diagnoses, statistical significance is not suggested; 95% CI includes important benefit and important harm. Downgraded once.
d. Publication bias: funnel plot suggests that studies with fewer cases of diabetes in the higher LCn3 arm may be missing, but statistical tests did not confirm this. Not downgraded.
e. Inconsistency: I2>50%. Downgraded once.
f. Imprecision: >18 000 participants, but only 230 diagnoses reported, leading to wide confidence intervals, including very important benefits and harms. Downgraded twice.
g. Risk of bias: although the main analysis suggests no effect of LCn3 on HbA1c, no included trials were at low summary risk of bias, and those at low risk of bias from compliance suggested a reduction in HbA1c with LCn3; overall effects not clear. Downgraded once.
h. Indirectness: the largest of the included studies was carried out in China, so data from industrialising countries are well represented. Not downgraded.
i. Imprecision: 178 participants included; very wide confidence intervals did not exclude important benefits or harms. Downgraded once.
j. Risk of bias: main analysis suggests that LCn3 increases plasma glucose, whereas limiting to studies at low summary risk of bias or with low risk of bias from compliance suggests reductions in plasma glucose. Downgraded once.
k. Publication bias: although the funnel plot is not definitive, we are aware of potentially important missing data. Downgraded once.
l. Risk of bias: effects in studies at low risk of bias from compliance and in fixed effects analyses are similar to those in the main analysis (suggesting that ALA reduces fasting plasma glucose), but no studies were at low summary risk of bias. Downgraded once.
m. Imprecision: data from 648 participants included; important harms and benefits are excluded from the 95% CI; the effect was not statistically significant. Not downgraded.
n. Risk of bias: main analysis suggests little or no effect of LCn3 on fasting insulin, and this is confirmed by fixed effects and low risk of bias from compliance analyses; however, limiting studies to those at low summary risk of bias suggests that LCn3 increases fasting insulin substantially (mean difference 25.3 (95% CI 4.1 to 46) pmol/L. Downgraded once.
o. Risk of bias: main analysis suggests a small increase in fasting insulin with ALA, confirmed by fixed effects analysis (neither statistically significant), but smaller effects suggested when studies limited to those at low risk of bias from compliance, and no studies were at low summary risk of bias (and no studies were at low risk of bias from allocation concealment). Downgraded once.
p. Imprecision: 95% CI did not exclude important harms. Downgraded once.
q. Risk of bias: main analysis, fixed effects, low risk of compliance, and low summary risk of bias analyses all suggested little or no effect of LCn3 on HOMA-IR. Not downgraded.
r. Publication bias: funnel plot suggests that studies with higher HOMA-IR scores in the LCn3 arm may be missing; however, statistical tests did not confirm this. Not downgraded.
s. Imprecision: 294 participants contributed to this analysis; 95% CI did not exclude important benefit or harm. Downgraded once.
t. Risk of bias and publication bias: no included studies were at low summary risk of bias; the study with data not added to the meta-analysis suggested increased HOMA-IR with greater ALA, which may indicate publication bias. Downgraded once between the two risks.