Twenty eight RCTs (n=13,457) were included in the meta-analysis: 16 (n=7,038) studies of orlistat; seven (n=1,475) studies of sibutramine; and five studies (n=4,944) of rimonabant. Twenty two studies reported funding from a drug manufacturer.
Most studies were of a similar quality. All studies were intention-to-treat, all studies specified eligibility criteria and treatment groups were comparable at baseline. Most studies did not report randomisation process or allocation concealment. Three studies did not report whether they were double blinded. Follow up was 12 to 18 months.
Overall median drop out rates were high and similar: 30% for orlistat; 34% for sibutramine; and 39% for rimonabant. Adverse effects drop-out rates were more heterogeneous and were higher for drug treatments than controls: 7.1% for orlistat (4% for placebo); 9.3% for sibutramine (8.9% for placebo); and 15% for rimonabant (7.2% for placebo).
Patients who took rimonabant and orlistat had a significantly increased risk of drop out due to adverse effects compared to controls (RR 2.00, 95% CI 1.66 to 2.41 and RD 0.07, 95% CI 0.05 to 0.09 for rimonabant and RR 1.59, 95% CI 1.21 to 2.08 and RD 0.03, 95% CI 0.01 to 0.04 for orlistat). Sibutramine did not increase risk. No significant heterogeneity was found between the trials. Number needed to harm was 14 (95% CI 11 to 19) for rimonabant, 39 (95% CI 25 to 83) for orlistat and 500 (not significant) for sibutramine, which indicated that rimonabant was the greatest clinical risk. The most common reasons for withdrawal were patient request, adverse effects and poor compliance. The most common adverse effects that led to withdrawal were gastrointestinal for orlistat (40%) and psychiatric for rimonabant (47%). No results were presented for adverse effects for sibutramine.
Sensitivity analysis indicated that exclusion of poor-quality studies (studies without double blinding) did not significantly effect the results. No publication bias was detected.