Intellectual Disability After Gestational Exposure to Antidepressant Drugs: The Confidence Interval as a...

There is growing concern among statisticians that the use of P value dichotomization into significant and nonsignificant outcomes does not appropriately describe research findings; in fact, misconceptions about the very meaning of the P value abound. Many alternate approaches to interpreting data have been suggested. The use of the 95% confidence interval (CI) as a compatibility interval is one such approach. By this approach, the study estimate for an outcome is considered to be the most compatible with the population value for that outcome, and all the values in the 95% CI around that estimate are also considered to be compatible with the population value with decreasing likelihood of compatibility the greater the distance of the value from the study estimate. This concept is explained with the help of a study that examined the risk of intellectual disability (ID) in children who had been gestationally exposed to antidepressant drugs (ADs). A conventional interpretation of the study findings is that, after adjustment for confounders, AD exposure was not associated with an increased risk of ID; this, in fact, is also how the authors expressed their conclusions. However, when the 95% CI in the main analyses, subgroup analyses, and sensitivity analyses in this study are examined as compatibility intervals, it becomes apparent that, even after adjustment for confounders, a very sizeable range of values indicating increased risk is compatible with the population value. In contrast, the range of values indicating decreased or no risk is very small. The implication, therefore, is that adjustment for confounding attenuates the risk, but the risk probably remains elevated. There is uncertainty in this subjective conclusion, but this conclusion is more truthful than a conclusion that a significant risk was rendered nonsignificant by adjustment for confounding.