Clarifying contradictions: Transportability in 17OHP-C trials and preterm birth outcomes using coubly debiased machine learning

Authors

Arti V. Virkud, University of Pennsylvania, Perelman School of Medicine, Department of Biostatistics Epidemiology and Informatics.
Eric Tchetgen Tchetgen, University of Pennsylvania, Perelman School of Medicine, Department of Biostatistics Epidemiology and Informatics.
Enrique F. Schisterman, University of Pennsylvania, Perelman School of Medicine, Department of Biostatistics Epidemiology and Informatics.
Beth Pineles, University of Pennsylvania, Perelman School of Medicine, Department of Biostatistics Epidemiology and Informatics.
Lisa D. Levine, University of Pennsylvania, Perelman School of Medicine, Department of Pregnancy and Perinatal Research Center.
Stephen R. Cole, University of North Carolina at Chapel Hill, Gillings School of Public Health, Department of Epidemiology.
Stefanie N. Hinkle, University of Pennsylvania, Perelman School of Medicine, Department of Biostatistics Epidemiology and Informatics.
Sunni Mumford, University of Pennsylvania, Perelman School of Medicine, Department of Biostatistics Epidemiology and Informatics.
Kristin D. Gerson, University of Pennsylvania, Perelman School of Medicine, Department of Pregnancy and Perinatal Research Center.
Brandie D. Taylor, Advocate Health - MidwestFollow
Sean Blackwell, UTHealth Houston McGovern Medical School.
Alan Peaceman, Northwestern Feinberg School of Medicine.
Samuel Parry, University of Pennsylvania, Perelman School of Medicine, Department of Pregnancy and Perinatal Research Center.
Maria T. Johnson, PCORI Stakeholder Advisory Panel.
Dalynn Willis, PCORI Stakeholder Advisory Panel.
Ellen C. Caniglia, University of Pennsylvania, Perelman School of Medicine, Department of Biostatistics Epidemiology and Informatics.

Recommended Citation

Virkud AV, Tchetgen ET, Schisterman EF, et al. Clarifying Contradictions: Transportability in 17OHP-C Trials and Preterm Birth Outcomes Using Doubly Debiased Machine Learning. Am J Epidemiol. Published online September 24, 2025. doi:10.1093/aje/kwaf202

Abstract

Following the Meis et al. trial that identified a benefit of 17-alpha-hydroxyprogesterone caproate (17OHP-C) in reducing the risk of recurrent preterm birth (PTB) (risk difference (RD) -18.6%, 95% confidence interval (CI): -28.2%, -9.2%), a confirmatory trial (PROLONG) identified no benefit of 17OHP-C (RD: 1.2%, 95% CI: -3.0%, 5.3%). The leading hypothesis is that the difference was due to the heterogeneity in PTB risk. We implemented state-of-the-art methods, using doubly debiased machine learning for transportability to investigate whether the conflicting trial results could be explained by measured differences between trial populations. The estimated RD when transporting the effect in Meis to the PROLONG trial population was -18.6% (95% CI: -55.9%, 8.8%) comparing 17OHP-C to placebo. The estimated RD when transporting PROLONG to Meis was 5.2% (95% CI: -17.3%, 18.1%) comparing 17OHP-C to placebo. Transporting from PROLONG to Meis did not recover the protective effect observed in Meis, which we hypothesize is due to a hidden violation of one or more causal assumptions for transportability, such as the presence of unmeasured effect measure modifiers. Transporting from Meis to PROLONG did not recover the null point estimate observed in PROLONG, though the confidence interval was wide. Future studies should explore effect heterogeneity PTB.

Document Type

Article

PubMed ID

40990932