A Highly Powered Large Outcomes Randomised Controlled Trial Will Usually, but Not Always, Have Balanced Groups at...

Area of Science:

• Clinical Trials
• Biostatistics
• Epidemiology

Background:

• Randomised trials are crucial for evaluating interventions.
• Effective randomisation ensures comparable groups, minimizing bias.
• Baseline group comparability is essential for accurate outcome assessment.

Purpose of the Study:

• To investigate the frequency of chance imbalances in baseline hazards in large randomised trials.
• To assess if baseline covariates can mitigate detected imbalances.

Main Methods:

• Simulated multiple trials using data from 8215 participants in the ORIGIN trial biomarker study.
• Repeatedly randomised participants and calculated hazard ratios for death between groups.
• Tabulated frequency of hazard ratios < 0.85 or > 1.18 across 1000 randomisations, repeated 1000 times.
• Explored mitigation of imbalances using adjusted Cox models with baseline covariates.

Main Results:

• A hazard ratio < 0.85 or > 1.18 occurred by chance in a median of 3 out of 1000 randomisations (0.3%).
• These imbalances were not evident from baseline data.
• Adjusting for baseline predictors of death did not fully mitigate the identified imbalances.

Conclusions:

• Approximately 1 in 300-400 large, two-arm outcomes trials may experience significant chance imbalances in baseline hazard.
• Such imbalances can potentially misrepresent the true intervention effect.
• Careful consideration of baseline characteristics and potential imbalances is warranted in trial interpretation.