Table 2.
Agreement of treatment effects reported in RCD studies and subsequent randomized trial evidence
| Analysis | No of treatment comparisons | Summary ROR (95% CI) | I2 (%; 95% CI) |
|---|---|---|---|
| Main analysis | |||
| Random effect models to combine RCTs | 16 | 1.31 (1.03 to 1.65) | 0 (0 to 45) |
| Sensitivity analyses | |||
| Fixed effect models to combine RCTs | 16 | 1.34 (1.09 to 1.63) | 0 (0 to 45) |
| Exclusion of RCTs with high risk of bias | 14 | 1.21 (0.92 to 1.59) | 0 (0 to 47) |
| Exclusion of RCTs with high treatment crossover rates or with asymmetric crossover | 15 | 1.34 (1.05 to 1.70) | 0 (0 to 46) |
| RCTs that had low treatment crossover rates | 12 | 1.27 (0.92 to 1.76) | 0 (0 to 51) |
| Exclusion of RCTs with frequent non-initiation of randomized treatment | 16 | 1.31 (1.04 to 1.65) | 0 (0 to 45) |
| Exclusion of RCTs with age differences >2 SD | 16 | 1.28 (1.01 to 1.62) | 0 (0 to 45) |
| Two subquestions instead of main question | 15 | 1.33 (1.05 to 1.68) | 0 (0 to 46) |
| Exclusion of treatment comparison with RCTs that are also pertinent for another treatment comparison | 15 | 1.28 (1.01 to 1.62) | 0 (0 to 46) |
| RCTs that were identified in available systematic reviews | 10 | 1.21 (0.88 to 1.65) | 0 (0 to 53) |
| RCD studies with low risk of bias except for “bias due to confounding”* | 11 | 1.20 (0.89 to 1.62) | 0 (0 to 51) |
RCD=routinely collected health data; RCT=randomized controlled trial; ROR=relative odds ratio; SD=standard deviation.
*All RCD studies were deemed to have moderate risk of bias due to confounding.