Questions to ask of the methods section
- How was the allocation sequence generated — was it genuinely random, or systematic and predictable?
- Who held the sequence, and could the next assignment be known or guessed before a patient was enrolled?
- Were the groups comparable at baseline on the factors that matter for prognosis?
- If the sequence was predictable, what could an enrolling clinician have done with that knowledge?
Alternating assignment in practice
A trial states that patients were "randomized" by alternating treatment and control as they were admitted. The sequence is predictable, so an enrolling clinician who favors the treatment can hold a high-risk patient until the next control slot and enroll a low-risk patient when the next slot is treatment. The arms differ at baseline, the treatment looks better, and the reported effect is inflated. The flaw is in who entered the study, not in the analysis.
Direction of the distortion
Selection bias does not always inflate the treatment effect. If the clinician steers sicker patients toward the treatment arm, the treatment will look worse than it is. The problem is the loss of comparability, not a fixed direction of error.