Table 1.
Meta‐analysis of response rates using a random effect model of five venlafaxine versus placebo studies using different hypothesis about missing data
| Studies | Optimistic bias | ITTLOCF | OC | Attrition = failure | Maximum bias |
|---|---|---|---|---|---|
| Sheehan | 3.35 [2.32; 4.84] | 1.34 [0.96; 1.87] | 1.58 [1.09; 2.28] | 1.52 [0.98; 2.37] | 0.56 [0.41; 0.75] |
| Rudolph | 4.12 [2.71; 6.26] | 1.66 [1.19; 2.32] | 1.71 [1.16; 2.52] | 1.77 [1.13; 2.78] | 0.51 [0.41; 0.65] |
| Mendels | 2.02 [1.54; 2.66] | 1.27 [0.99; 1.63] | 1.31 [1.03; 1.65] | 1.44 [1.08; 1.91] | 0.82 [0.69; 0.98] |
| WXL101497 | 1.67 [1.39; 2.00] | 1.38 [1.15; 1.67] | 1.34 [1.12; 1.60] | 1.38 [1.14; 1.69] | 1.03 [0.87; 1.21] |
| AK130940 | 1.82 [1.53; 2.16] | 1.33 [1.11; 1.59] | 1.31 [1.11; 1.55] | 1.30 [1.06; 1.58] | 0.87 [0.74. 1.02] |
| Total | 2.31 [1.75; 3.06] | 1.36 [1.23; 1.51] | 1.36 [1.23; 1.50] | 1.39 [1.24; 1.57] | 0.74 [0.59; 0.94] |
Note: Data were extracted from our previous meta‐analysis: out of 26 randomized double‐blind trials, five studies on venlafaxine versus placebo had extractable data. Meta‐analyses of response rates using a random effect model were performed under different hypotheses about missing data. Five situations were considered:
• Optimistic bias analysis: non‐assessed patients are recorded as in remission if they belong to the antidepressant group and as having not responded if they belong to the placebo group;
• ITTLOCF: patient status is derived from the LOCF method on continuous outcomes;
• OC: observed case analysis;
• Attrition = failure: non‐assessed patient are recorded as not having responded in both groups;
• Maximum bias: non‐assessed patients are recorded as in remission if they belong to the placebo group and as not having responded if they belong to the antidepressant group.
Results are presented as relative risk. Positive relative risk favours venlafaxine and negative relative risk favours placebo.
This example illustrates the uncertainty that arises from missing data when assessing antidepressant effect, which can vary from a marked superiority of antidepressants over placebo to a superiority of placebo over antidepressants, depending on the imputation method used for missing data.