Table 1 Methods to reduce the risk of selection bias in non-double-blind clinical trials*
Techniques | Rationale | Effect on risk of selection bias |
|---|---|---|
Use blinded recruiters | If recruiters are blind to previous trial allocations, they will be unable to predict the next allocation | Risk of selection bias will be eliminated provided the blinding is maintained |
Use simple (unrestricted) randomisation | Recruiters cannot guess the next allocation with any degree of accuracy | Risk of selection bias will be eliminated |
If restricted randomisation is used, do not stratify by site of recruitment | The probability of the allocation will depend on previous allocations at other sites, which recruiters are unlikely to have access to, making an accurate guess more difficult | Risk of selection bias will be reduced, but not necessarily eliminated |
When randomisation is stratified by site, avoid permuted blocks | Permuted blocks stratified by site will maximise the probability of correctly guessing the next allocation. Using alternative randomisation methods will reduce the probability of correctly guessing the next allocation | Risk of selection bias will be reduced, but not necessarily eliminated |
When restricted randomisation is used, stratify by prognostic covariates as well | There is typically less variation in prognoses for patients with the same covariate pattern, making it more difficult for investigators to identify patients with a specific prognosis to enrol into the trial when their preferred treatment is more likely | Risk of selection bias will be reduced, but not necessarily eliminated |