Considerations Regarding Autism Spectrum Disorders and Cardiometabolic Diseases—Reply

Abstract

COMMENT & RESPONSEIn Reply On behalf of our coauthors, we appreciate Jin and Shi taking time to review our article 1 and sharing their thoughts in a Letter to the Editor. We address the 2 concerns raised by the authors in this response.First, the authors raised a concern regarding criteria for data extraction and categorizing outcomes of studies that included children vs adults vs pooled children and adults citing our classification of the studies by Chen et al 2 and Flygare Wallén et al 3 into different subgroups as an example. We entirely agree with the authors on the fact that, when possible, studies that report data on subgroups should be separately analyzed in subgroup sensitivity analyses (eg, children vs adults). All eligible studies that reported results of sensitivity analyses were carefully screened before they were assigned to subgroups based on age (eFigures 2 and 4 1 ) or diagnosis (Figure 2 1 ). Flygare Wallén et al 3 reported the raw data separating the data by age group, allowing us to compute the relative risks for the subgroups of children vs adults. However, Chen et al 2 reported raw data only for the whole sample, which included adolescents and young adults. Even though Chen et al reported hazard ratios regarding development of diabetes for adolescents and young adults separately, the data could not be used as surrogates for relative risks given that they were hazard ratios and were adjusted outcomes derived after controlling for demographic factors. Attempting to calculate relative risks using hazard ratios adjusted for multiple variables is likely to introduce more bias and further deviate the pooled outcomes.Second, the authors suggest adjusting the outcomes for the duration of follow-up. It is redundant to consider the duration of follow-up of the studies when the exposure (ie, autism) is likely to have started approximately at the time of birth with a fixed or small margin of error in time compared to the age of the individual. As such, we cannot think of a method better than grouping by age and including age as a moderator in meta-regression analyses to account for the duration of exposure.We agree with the concern that between-study heterogeneity and unmeasured confounding could lead to bias in any meta-analysis. We have tried our best to adhere to the best practices outlined in the Cochrane handbook, 4 chapters 10.10 and 13.5.