Thirteen RCTs (1,569 participants) were included in the review. Nine RCTs were considered to have a low risk of bias although two of these trials did not report the sequence generation method and seven did not report using any measure to conceal the allocation. Five trials were categorised as having a high risk of bias because they used a per-protocol analysis. There was no evidence of selective outcome reporting. Dietary counselling was considered to be adequate in six trials. The proportion of drop-outs ranged from 13.3% to 84.5%.
Patients assigned to a very-low-carbohydrate ketogenic diet had statistically significant greater weight loss than those assigned to a low fat diet (WMD -0.91kg, 95% CI -1.65 to -0.17; 13 RCTs; 1,415 patients). There was no evidence of significant heterogeneity (Ι²=0%).
Patients assigned to a very-low-carbohydrate ketogenic diet had significantly decreased triacylglycerol (WMD -0.18mmol/L, 95% CI -0.27 to -0.08; 12 RCTs; 1,258 patients) and diastolic blood pressure (WMD -1.43mmHG, 95% CI -2.49 to -0.37; 11 RCTs; 1,298 patients) and significantly increased HDL cholesterol (WMD 0.09mmol/L, 95% CI 0.06 to 0.12; 12 RCTs; 1,257 patients) and LDL cholesterol (WMD 0.12mmol/L, 95% CI 0.04 to 0.2; 12 RCTs; 1,255 patients) compared with patients on a low fat diet. There was no significant heterogeneity for any of these outcomes.
There was no significant difference between treatment groups in changes in systolic blood pressure (11 trials), fasting blood glucose (eight trials), insulin (six trials), HbA1c (four trials) and C-reactive protein (four trials).
Subgroup analyses were generally consistent with the main results, except for the subgroup analysis of trials with 24 months of follow-up (four trials) in which most of the results were no longer statistically significant.
There was no evidence of significant publication bias for the primary outcome.