To the Editor In a recent epidemiological study, Gregory and colleagues1 concluded that induction or augmentation of labor during childbirth is associated with increased odds of autism in children. Unfortunately, several aspects of the design and interpretation of the findings were overlooked. We articulate them next.
(1) The primary outcome (autism) is poorly defined since the specific designation for children with multiple exceptionalities was “at the discretion of school personnel” because of lack of guidelines. This raises the possibility of response misclassification likely biasing the results. (2) The exposed (labor induction/augmentation) and the unexposed (no induction/augmented) groups are not comparable, thereby violating one of the fundamental principles of cohort study designs.2 The unexposed group included patients who delivered after spontaneous labor as well patients who delivered by repeat cesarean prior to the onset of labor. The cesarean delivery rate in North Carolina during the study years 1990 to 1998 was 22% to 23% and the repeat cesarean delivery rate was estimated to be approximately 10% of the total number of deliveries.3 Therefore, approximately 62 504 women who had repeat cesarean delivery were included in the unexposed group. This translates to a significant percentage (approximately 14%) of patients in the unexposed group who may never have been exposed to labor. Therefore, the reference group was at a clear advantage, thus having less chance for autism (selection bias). This bias would have likely led to an overestimation of the associations reported by Gregory et al. (3) The potential inclusion of more than 1 pregnancy per woman violates the statistical assumption of independence of observations, potentially leading to incorrect statistical and biologic inferences. This is very important given the heritable nature of autism.4 (4) A Cox proportional hazards regression that takes account of the changing profile of autism risks by the child’s age (at diagnosis) is more appropriate than a logistic regression model. (5) The conclusions leave room for bias due to residual confounding because several important confounders were not adjusted. (6) Our concerns regarding the heterogeneity of the primary exposure and our argument that there was no association between labor/augmentation and autism when the correct diagnostic criteria for autism were applied for the last 3 birth years of the study (1996-1998) have been espoused elsewhere.5