Model-based estimates of sex differences in peak power and fatigue index in track cyclists using directed acyclic graphs, inverse probability of treatment weighting, and Bayesian modeling.

Objective: The purpose of this study was to estimate sex-related differences in peak power and fatigue index in track cyclists under an explicit causal modeling framework integrating directed acyclic graphs, inverse probability of treatment weighting, and hierarchical Bayesian estimation. Design and methods: A cross-sectional observational study with a quantitative approach was conducted to estimate adjusted sex-related differences in anaerobic performance indicators in track cyclists. Because sex was non-randomized and not experimentally manipulable, the estimates were interpreted as model-based associations under explicit causal assumptions rather than as definitive interventional causal effects. The sample included 21 cyclists, 15 males and 6 females, aged 14-17 years, with at least 3 years of experience in track cycling. Peak power, mean power, relative power outputs, and fatigue index were assessed using a 30-s Wingate-type test performed on the athletes' own bicycles mounted on a Tacx NEO 2T Smart electromagnetic resistance system in a controlled velodrome-based setting. The analysis integrated DAG-informed covariate selection, exploratory IPTW diagnostics, sensitivity analyses, and Bayesian estimation with convergence diagnostics. Results: Male cyclists showed higher peak power (mean [M] = 884.16 W; standard deviation [SD] = 130.76) than female cyclists (M = 626.00 W; SD = 59.93). Similarly, the fatigue index was higher in males (M = 46.23%; SD = 1.21) than in females (M = 44.22%; SD = 0.37). IPTW diagnostics indicated limited covariate balance, particularly for body weight and height, and instability of the unpenalized propensity score model. Sensitivity analyses showed that the sex-related difference in peak power was model-dependent, whereas the fatigue index difference was more consistent across model specifications. Four-chain Bayesian models showed adequate convergence, with R-hat values close to 1.00 and effective sample sizes above 7,700 for the main sex-difference parameters. Conclusions: These findings suggest that sex-related differences in power generation and fatigue responses may be relevant for individualized training planning. However, given the small sample size and the observational cross-sectional design, the results should be interpreted as exploratory model-based estimates.