Date: January 25th, 2020
SGEM#281: EM Docs Got an AmbuBag
Statistically Significant: Dan Lane
We want to make the SGEM even better and address some of the criticisms from the ClinEpi world about clinicians trying to do critical appraisal. In order to do that we now have a Dr. Dan Lane who has a PhD in Clinical Epidemiology. He will be commenting on each the SGEM episodes.
On this episode of Statistically Significant we are going to discuss the importance of balance of prognostic factors in randomized controlled trials, using the PreVent trial as an example.
Characteristics that indicate when a patient more likely to have an outcome, what we call prognostic factors, need to be accounted for when assessing the effectiveness of a treatment. Without accounting for prognostic factors, the measures of treatment effect can be biased due to observed or unobserved factors amongst patients in each group. Consider if this same study had been conducted as a non-randomized design –clinicians may have decided to ventilate select patients between induction and intubation because they perceived them as more unstable prior to induction. These patients may also be at higher risk for hypoxia during this period for the same reasons the clinicians chose to ventilate them and therefore they would look worse when compared to patients not receiving ventilation if you did not account for these reasons – this is what epidemiologists call an indication bias.
The goal of randomization in clinical trials is to balance patient characteristics between the different groups being investigated in the study. By randomly assigning patients to groups, the sole indication for receiving the treatment is the randomization process. As long as there are enough patients randomized, all known and unknown prognostic factors will be mathematically balanced between the groups. Therefore when talking about the balance of prognostic factors as part of critical appraisal, the key point to realize is there are both known and unknown factors. Although in this study they found some statistical differences between measured prognostic factors at baseline, these are just the prognostic factors that happen to be reported by the investigators. If we trust their randomization process then we can assume that the overall risk of the primary outcome, which includes measured and unmeasured prognostic factors, is mathematically balanced between the groups.
One final point – the use of statistical hypothesis testing to compare prognostic factors is actually inappropriate here because by definition the null hypothesis that the two groups are the same is assumed to be true when the two groups are selected based on randomization. Therefore, any differences between the groups would be due to chance alone and considering them different would be a type 1 error.
- Altman and Bland. Treatment allocation in controlled trials: why randomise? BMJ May 1999
- Sander Greeland. Randomization, statistics, and causal inference. Epidemiology Nov 1990
- Stephen Sean. Baseline Balance and Valid Statistical Analyses: Common Misunderstandings. Applied Clinical Trials. May 2005.