Comments on “A Cross‐Sectional Study Reveals Hormonal Signatures of Smoking‐Related Infertility”

Choudhury et al. present a cross-sectional analysis of hormonal signatures in smoking-related male infertility, applying inverse probability weighting, augmented inverse probability weighting, and mediation analysis [1]. The authors should be commended for tackling an important clinical question with modern causal inference tools and for integrating multiple reproductive hormones with a composite semen quality index. Our comments focus on methodological aspects that affect the strength and interpretation of the reported causal and mediational claims. First, the study is explicitly cross-sectional, with smoking status, hormones, and semen parameters measured at a single time point. Yet the manuscript frequently describes “causal effects” and a primary “mediating” role for luteinising hormone. Causal effect estimation and mediation require not only control for confounding but also a clear temporal ordering of exposure, mediator, and outcome, and the absence of important unmeasured confounding for each component pathway [2, 3]. In cross-sectional designs, mediators and outcomes are observed concurrently, and reverse causation cannot be ruled out, which is known to bias mediated effect estimates when underlying processes are longitudinal [4]. The analytic framework used is technically appropriate, but in our view, the results are better described as adjusted associations and exploratory pathway decompositions rather than definitive evidence of hormonal mediation. Second, the application of inverse probability of treatment weighting and augmented inverse probability weighting is a strength, reflecting best practice in adjustment for measured confounding [2, 5]. However, with 104 men and 21 baseline covariates reported as entering the propensity score model, there is a risk of model instability and extreme or highly variable weights, particularly if some covariates are rare or highly correlated [5, 6]. The main text presents a balance for a smaller subset of covariates and does not list all variables included in the propensity model, which limits readers' ability to assess the positivity and exchangeability assumptions. Greater transparency about covariate selection and weight diagnostics would help quantify the robustness of the estimated average treatment effects. Third, the paper describes mediation analyses with four hormones in parallel using bootstrapping; however, it is unclear whether these models themselves incorporated the inverse probability weights or the augmented outcome models. Doubly robust estimators provide protection against model misspecification for total effects, but this property does not automatically extend to complex mediation decompositions unless carefully constructed [3]. Presenting the indirect effects as “doubly robust causal mediation” may therefore overstate their identification, and we suggest more cautious wording that reflects the observational and cross-sectional nature of the data. Finally, the principal component analysis used to derive a Sperm Quality Index is an innovative way to summarize multiple semen parameters. Nonetheless, there appear to be minor inconsistencies between the reported eigenvalue and the percentage of variance explained for the first component, which makes independent replication difficult. Clear reporting of PCA procedures and component statistics is important for reproducibility and interpretation [7]. Choudhury et al. contribute valuable data on the relation between smoking, reproductive hormones, and semen quality in an infertile clinic population. We suggest that the findings be interpreted as carefully adjusted cross-sectional associations that suggest, rather than prove, hormonal pathways linking smoking to semen quality. Future longitudinal studies with repeated hormonal measurements, explicit causal diagrams, and pre-specified mediation strategies would further strengthen causal inference in this important field. Ankur Sharma: conceptualization, writing – review and editing, writing – original draft. Sushma Narsing Katkuri: conceptualization, writing – original draft, writing – review and editing. Varshini Vadhithala: conceptualization, writing – original draft, writing – review and editing. Arun Kumar: writing – original draft, writing – review and editing. Sushma Verma: writing – original draft, writing – review and editing. Dhanya Dedeepya: validation, writing – review and editing. The authors have nothing to report. The authors received no specific funding for this work. The authors have nothing to report. The authors have nothing to report. The authors declare no conflicts of interest. The authors have nothing to report.