Sixteen randomised trials (7,759 patients), three of which were cross-over, were included. Trial quality was generally moderate-to-poor (median score 2.5 out of 5). None of the studies were double-blind due to the visibly different properties of the comparators. Only 11 trials used intention-to-treat.
Patients treated with biphasic insulin had a greater chance of reaching a haemoglobin A1c target of below 7% than those on basal insulin (OR 1.88, 95% CI 1.38 to 2.55; 10 trials) but was also associated with more hypoglycaemic events (OR 0.34, 95% CI 0 to 0.69; 10 trials) and weight gain (OR 1.0 kg, 95% CI 0.28 to 1.73 kg; 10 trials). There was substantial heterogeneity in all an analyses, but no evidence of a statistically significant difference between large and small trials in the primary analysis.
Use of basal-bolus insulin as a comparator had a substantial impact, as patients treated with basal-bolus insulin had a greater chance of achieving a haemoglobin target of below 7% (OR 1.75, 95% CI 1.11 to 2.77) than those on biphasic insulin. Furthermore, there was no significant difference in incidence of hypoglycaemia or weight gain, although these results were generated by pooling only three trials.
Three trials allowed head-to-head comparison of biphasic versus prandial insulin and four trials compared prandial and basal but the pooled effects of these analyses were non-significant irrespective of outcome.