Data on participants, interventions, weight loss, and weight- loss maintenance were extracted by two researchers...independently by using predefined data templates. Disagreements were resolved through discussion. For the meta-analysis, data on number of subjects and mean changes with corresponding standard deviations in the intervention and control arm were extracted. Many studies did not report these values. In those cases, standard deviations and mean changes were calculated from other data....
Five studies included two arms (14, 15, 18, 25, 26). Weighted mean differences were calculated between the two groups that showed the most resemblance to other studies in the treatment category. Larsen et al. (15) reported both the isolated and combined effects of a high-protein diet and a low-glycemic-index diet. In the meta-analysis, the isolated main effects of high protein compared with low protein and low glycemic index compared with high glycemic index were included. Due et al (18) reported the effect of two interventions (low fat and the Healthy Eating Pyramid, which is high in MUFAs and has a low glycemic index), and both treatment arms were included and compared with the control group. In the study by Hursel et al (14), the green tea effect was analyzed by comparing the green tea/adequate-protein group with the placebo/adequate-protein group, and the high- protein effect was analyzed by comparing the green tea/high- protein with the green tea/adequate-protein group. The study by Kamphuis et al (25) included two different doses of conjugated linoleic acid (1.8 and 3.6 g) compared with placebo (1.8 and 3.6 g). Both doses were included.
Two reviewers...independently evaluated the individual studies regarding the extent of loss to follow-up and the adequacy of randomization and concealment of allocation, blinding of patients, data collectors, and outcome assessors.
Data synthesis and analysis
The primary outcome was the weighted mean difference in weight change (kg) during the weight-loss maintenance phase between the intervention and control groups. The random-effects model was used to weight and pool the studies within each maintenance category (anti-obesity drug, diet, and exercise). The diet studies were further subdivided into high-protein diet, meal replacement, dietary supplements, and macronutrients other than protein, including low glycemic index, low fat, and eating according to the Healthy Eating Pyramid. Heterogeneity between studies was assessed by the I2 statistic (27), and if this exceeded 50% or was statistically significant, the reasons for heterogeneity were explored by subgroup analyses or meta-regression. Low, moderate, and high heterogeneity were defined according to cutoffs of 25%, 50%, and 75%, respectively (28). To investigate possible publication bias, a funnel plot of the inverse of the SE was inspected visually, and statistical significance was calculated by using Egger's test (29).
A secondary aim was to illustrate weight change after the VLCD or LCD phase and weight-loss maintenance phase within each treatment arm. The random-effects model was used to weight and pool the weight changes within each treatment and control arm during the maintenance period. The mean monthly change was estimated from these two measurements. The statistical analyses were conducted by using Comprehensive Meta Analysis (version 2; Biostat Inc). P values, 0.05 were regarded as statistically significant.