Approach | Description | Advantages | Disadvantages | Recommendation |
---|---|---|---|---|
(A) Standard deviation (SD) units (standardized mean difference; effect size) | The pooled mean difference is presented in standard deviation units | (+) Widely used | (-) Interpretation challenging | Consider complimenting other approaches with this; it is not recommended to use this approach independently. |
(-) Misleading when trial SDs are heterogeneous | ||||
(B) Natural units | Linear transformation of trial data to most familiar scale | (+) Easier to interpret if scale well-known | (-) Few instruments in clinical practice are easy to interpret | Approaches to conversion to natural units include those based on SD units and re-scaling approaches. We suggest the latter. In rare situations when instrument very familiar to front line clinicians seriously consider this presentation |
(C) Relative and absolute dichotomized effects | Obtain proportion above threshold in both groups and calculate relative or absolute binary effect measure | (+) Very familiar to clinical audiences | (-) Involve statistical assumptions that may be questionable | If the minimal important difference is known use this strategy in preference to relying on SD units |
Always seriously consider this option | ||||
(D) Ratio of means | The ratio between the mean responses in the intervention and control group | (+) May be easily interpretable to clinical audience | (-) Not applicable for change scores | Consider as complementing other approaches, particularly the presentation of relative and absolute effects |
(+) Fewer questionable assumptions | (-) Interpretation requires knowledge of control group mean | |||
(E) Minimal important difference units | The pooled mean differences is presented in MID units | (+) May be easily interpretable to clinical audience | (-) Only applicable when minimally important difference is known | Consider as complementing other approaches, particularly the presentation of relative and absolute effects |