Confounders in cohort studiesIn a previous post, I have used the Rochon article on cohort studies to evaluate the validity of the Johson and Daviss homebirth study. The initial analysis showed that the study suffers from selection bias. The second article in the series,Reader's guide to critical appraisal of cohort studies: 2. Assessing potential for confounding, gives us more information about the potential sources of selection bias,the confounders.
According to the article:
For a characteristic to be a confounder in a particular study, it must meet two criteria. The first is that it must be related to the outcome in terms of prognosis or susceptibility...The article highlights three questions that must be asked to identify confounders in a cohort study:
The second criterion that defines a confounder is that the distribution of the characteristic is different in the groups being compared.
Has there been a systematic effort to identify and measure potential confounders?Let's ask these three questions about the Johnson and Daviss study.
Is there information on how the potential confounders are distributed between the comparison groups?
What methods are used to assess differences in the distribution of potential confounders?
1. Has there been a systematic effort to identify and measure potential confounders?
The article describes what an effort to identify and measure potential confounders looks like:
Information on the distribution of potential confounders in the intervention and comparison groups is usually provided in the first table of the paper. Confounding is a problem only if these characteristics are unevenly distributed between the intervention and comparison groups.Johnson and Daviss do provide a table at the beginning of their study (Table 1). The table shows us characteristics of the homebirth group and the group of all singleton, vertex births at term in the US in 2000. It does not provide any information about the characteristics in the comparison group derived from out-of-date homebirth studies.
2. Is there information about how the potential confounders are distributed between the comparison groups?
Table 1 shows us that various confounders are distributed quite differently between the two groups that are listed. For example, African Americans make up only 1.3% of the homebirth group, but 14.1% of the comparison group. We know that the neonatal mortality rate for African Americans is 2-3 times higher than for white women. Therefore, this is a very serious confounder.
There are potential confounders that are not addressed in the table. For example, pre-existing medical conditions and pregnancy complications have a profound effect on neonatal mortality. Johnson and Daviss do not provide us with any information about these important confounders.
Of course, they provide no information at all about the comparison group derived from out-of-date homebirth studies and this is a serious omission.
3. What methods are used to assess differences in the distribution of potential confounders?
None. Johnson and Daviss made no effort to assess the distribution of potential confounders. According to Rochon: "Perhaps the most common strategy to identify important imbalances in individual confounders between intervention and comparison groups is to use significance tests such as x2 tests (for dichotomous variables) or t tests (for continuous variables)."
The bottom line is that the limited information that Johnson and Daviss provide shows the presence of serious confounders, but no attempt was made to assess the differences in distribution of these confounders and their impact. Every cohort study must include this analysis, and the fact that it is missing renders the conclusions invalid.
Labels: Johnson and Daviss