Table 1 Summary of results by analysis model, trajectory type, treatment effect type and whether proposed trial is the same length as the observational study
Analysis model | Linear or non-linear trajectory in DGM | Proportional or non-proportional treatment effect in DGM | Observational study length relative to trial length | Impact on power | Notes |
|---|---|---|---|---|---|
Random slopes model | Linear | Proportional | Same | None noticeable | Assumptions of sample size calculation are met |
Different | None noticeable | ||||
Non-linear | Proportional | Same | None noticeable | Â | |
Different | Can differ from nominal | Close to nominal when residual error is large relative to between-person variance, but can differ from nominal when residual error is small. Power can be higher or lower than nominal | |||
Non-linear | Non-proportional | Same | Can be very far from nominal | Differing power is partially due to incorrect target treatment effect used in sample size calculation. Power can be higher or lower than nominal | |
Different | Can be very far from nominal | ||||
Free trajectories, free covariancea | Linear | Proportional | Same | Power loss | Power loss is greater when residual error is large relative to between-person variance |
Different | Power loss | ||||
Non-linear | Proportional | Same | Power loss | Power loss is greater when within-person error is large relative to between-person error | |
Different | Can differ from nominal | Power can be higher or lower than nominal | |||
Non-linear | Non-proportional | Same | Some power loss | Â | |
Different | Can be far from nominal | Differing power is partially due to incorrect target treatment effect used in sample size calculation. Power can be higher or lower than nominal |