Comments on: Prenatal acetaminophen and autism: a meta-analytic signal driven by a single, mis-extracted estimate

Dear Editor,

We read with great interest the systematic review and meta-analysis by Oliveira et al.,⁽¹⁾ which reported an 18% increase in the odds of autism spectrum disorder (ASD) following prenatal acetaminophen exposure. Because this conclusion may influence clinical practice and public health recommendations,⁽²⁾ several methodological aspects deserve clarification.

First, the pooled estimate for ASD diagnosis appears to be driven almost entirely by a single study contributing 95.7% of the statistical weight.⁽¹⁾ However, the effect estimate entered for Ahlqvist et al. (OR 1.18) cannot be identified in the original publication.⁽³⁾ The authors reported a population-based hazard ratio (HR) of 1.05 and, more importantly, a full-sibling comparison HR of 0.98 for ASD, but no odds ratio of 1.18.⁽³⁾ Consequently, the estimate underlying the pooled result is not reproducible from the published data.

Second, this discrepancy changes the interpretation of the evidence. Ahlqvist et al. concluded that the modest association observed in conventional analyses was largely explained by shared familial factors, whereas the sibling-comparison analysis showed no association between prenatal acetaminophen exposure and ASD (HR 0.98).⁽³⁾ Therefore, the principal conclusion of the meta-analysis appears to rely on an effect estimate that neither corresponds to the published ASD results nor reflects the interpretation of the study contributing nearly all of the statistical weight.

Additional methodological issues further reduce confidence in the summary estimate. The abstract reports a 95% confidence interval that differs from the corresponding forest plot for the same pooled effect.⁽¹⁾ Moreover, odds ratios, hazard ratios, and incidence rate ratios were combined without any reported transformation or justification. Several extracted estimates are also difficult to verify because the cited references include cohort-profile publications that do not report ASD effect estimates,^(4,5) while other estimates appear to have been visually extracted from figures in a previous meta-analysis⁶ or attributed to a study that investigated cryptorchidism and hypospadias rather than ASD.^(6,7) These approaches limit reproducibility and may introduce dependence if cohorts from a previous meta-analysis are subsequently analyzed as independent studies.

Finally, the review recommends minimizing acetaminophen prescribing during pregnancy despite the substantial risk of residual confounding by indication and shared familial factors, biases that sibling-comparison studies consistently attenuate toward the null.^(3-8) This recommendation is not supported by the presented evidence and is inconsistent with current guidance from major international professional societies.^(9-10)

Given the potential implications of this review for clinical practice and public health, clarification of the extracted data and reanalysis using transparent and reproducible effect estimates would strengthen confidence in the reported conclusions.

Data availability

The research data are described in the letter presented.