Reasons for and against post-randomisation exclusions
| Issue | Reason to include | Reason to exclude |
| Clinical scenario | ||
| Make recommendations of benefit or harm (based on trial results) relating to a certain patient population | Where there is uncertainty over defining patient populations, it would be a conservative approach to retain all participants. | Retains a defined group of included participants meeting inclusion/exclusion criteria neatly in which the intervention is hypothesised to be the most effective. |
| Disease status may be unclear | Measurement cut-off may not relate to a ‘disease’ state and may be arbitrary. | Measurement cut-offs are commonly used to indicate disease severity although knowing there may be some misclassification. |
| Assessment of safety risks | There is no safety risk to participants after review and therefore treatment and follow-up can continue. | Randomisation was mistakenly done, for example when found not to be diseased. Where safety was compromised the participants should cease remaining treatment and most likely be excluded from analysis. |
| Statistical analysis | ||
| Maintain ITT principles, providing an unbiased treatment effect | Stays true to ITT principle ensuring balance on known and unknown factors between arms when all enrolled and randomised participants are analysed. | The risk of bias from excluding some participants has been shown to be low under certain conditions. |
| The inclusion criteria are subject to measurement error. The relationship between the inclusion criteria and the primary outcome should be considered. | Pragmatically, errors in measurement will occur in routine practice. They may have been considered eligible at the point of enrolment. Include if measurement of the primary outcome is not impacted by measurement error in the inclusion criteria. | Identification of errors in the measurement of disease state and excluding them can prevent underestimation of treatment effects. |
| Effect on statistical power | A larger sample size is retained. | If ineligible participants’ responses to treatment differ from those for eligible participants (eg, less response), the variance of the primary outcome may be increased meaning there may be more statistical power to exclude them. |
| Integrity and transparency | ||
| Justifying the decision to include or exclude | Post-randomisation exclusions may be mistrusted in the scientific community if conflicts of interest or the trial sponsor are shown to have influenced the decision-making. | Post-randomisation exclusions are a common approach in the scientific community and will be accepted when clearly justified. |
*ITT intention-to-treat