Review, replication, and re-analysis of a recent study on the impact of abortion law on maternal mortality in Mexico – Maintaining rigor and research integrity

Abstract

This report provides a review, replication, and re-analysis of an study by Darney et al (Contraception. 2017;95(1):105-111) on the association between maternal mortality ratio (MMR, number of maternal deaths per 100,000 live births) and abortion legislation in Mexico City compared with 31 other states. In their study, Darney et al disputed evidence from our study (Koch et al, BMJ Open 2015;5(2):e006013), which found a null association between abortion laws and MMR, and reported their own conclusion of a negative association between abortion availability and MMR. After replicating their dataset and statistical analysis, we found that the beta coefficient (effect size) for abortion legislation was apparently misinterpreted by Darney et al to support the conclusion that legislation allowing abortion on request in Mexico City was associated with a 22.49-unit decrease in MMR. In fact, their analysis showed the opposite result: after multivariable adjustments, Mexico City was associated with a 22.49-unit increase in MMR compared with the 31 states with restricted access to abortion. Moreover, Darney et al did not report the initial time-adjusted effect size, where the 95% CI supports the hypothesis of null association (beta = 4.17; 95%CI -7.19 to 15.54; p=0.470). In our re-analysis of their data, estimates were highly unstable in multiple regression models, and the initial effect size (4.17) for the association between abortion legislation and the MMR was inflated up to 5-fold (22.49) as a result of a ‘tipping effect’ on coefficients and multicollinearity among covariates. In conclusion, our reanalysis of Darney et al’s study provides evidence of serious methodological flaws, faulty statistical analysis and misinterpretation of regression coefficients, ultimately resulting in an invalid conclusion.