Answers are grounded in this paper and cite their sources.
In polycystic ovary syndrome (PCOS) patients undergoing letrozole/follicle-stimulating hormone (FSH) sequential ovulation induction, does the preferable starting dose of letrozole (2.5 mg vs. 5.0 mg) differ between patients, and can baseline characteristics identify who benefits from which dose? Retrospective single-center cohort of 1,917 women with PCOS (2003 Rotterdam criteria) treated between 2020 and 2024, starting either 2.5 mg/day ( n = 979) or 5.0 mg/day ( n = 938) letrozole in routine care, with up to three cycles. The primary outcome was cumulative clinical pregnancy. We used stabilized inverse probability of treatment weighting and an augmented inverse-probability-weighted estimator for the adjusted risk difference, causal forest and pre-specified continuous interaction tests with patient-level bootstrap inference to examine heterogeneity, and a doubly robust policy tree to summarize candidate dosing patterns. Overlap weighting was used as a sensitivity analysis. Five-fold internal cross-validation was used to assess whether the candidate rule retained value in held-out data. The unadjusted cumulative pregnancy rate was 52.3% (2.5 mg) versus 42.9% (5.0 mg). The adjusted risk difference from the AIPW analysis was + 4.00% points (pp), 95% CI − 2.34 to + 10.35, with the CI including zero. BMI-by-treatment interaction was the only pre-specified interaction retained after Bonferroni correction (OR ratio per 1-SD increase in BMI = 1.53, 95% CI 1.16–2.03; Bonferroni P = 0.01386). In the overlap population, the risk difference was + 6.22 pp (95% CI 1.42–11.03), and the BMI interaction remained in the same direction. The policy tree assigned 2.5 or 5.0 mg across seven terminal nodes and had an in-sample policy value of 57.84%; in five-fold internal cross-validation, its value was 51.18% versus 51.43% for uniform 2.5 mg (detailed in Supplementary Table S2 ). No statistically significant average dose difference was detected after adjustment. The BMI interaction and the policy tree provide exploratory, hypothesis-generating signals rather than a validated dosing rule. The candidate tree did not outperform uniform 2.5 mg in internal validation and should not guide clinical dosing without prospective multicenter evaluation. Not applicable.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: