Baseline imbalance in factors that are strongly related to outcome measures can cause bias in the intervention effect estimate. This can happen through chance alone, but imbalance may also arise through non-randomized (unconcealed) allocation of interventions. Sometimes trial authors may exclude some randomized individuals, causing imbalance in participant characteristics in the different intervention groups. Sequence generation, lack of allocation concealment or exclusion of participants should each be addressed using the specific entries for these in the tool. If further inexplicable baseline imbalance is observed that is sufficient to lead to important exaggeration of effect estimates, then it should be noted. Tests of baseline imbalance have no value in truly randomized trials, but very small P values could suggest bias in the intervention allocation.
Example (of high risk of bias): A trial of captopril vs conventional anti-hypertensive had small but highly significant imbalances in height, weight, systolic and diastolic BP: P=10-4 to 10-18 (Hansson 1999). Such an imbalance suggests failure of randomization (which was by sealed envelopes) at some centres (Peto 1999).