Accurately predicting the survival status of osteosarcoma patients is critical for clinical diagnosis and treatment planning. This paper proposes a causal machine learning-based osteosarcoma survival prediction model (Causal-FISurv), which integrates a feature interaction-enhanced survival analysis framework (FI-Surv) with causal inference methods 1 to precisely predict patient survival risks and personalized treatment effects. 2 The method combines a feature interaction attention layer, a causal representation learner 3 , a potential outcome estimator, and a doubly robust estimator to address critical challenges in clinical decision-making. 4 Experimental results demonstrate that the proposed approach outperforms existing methods in terms of the C-index (0.6487) while providing more reliable personalized treatment effect estimation. The doubly robust estimation framework aims to mitigate bias caused by unobserved confounders, though complete elimination of such bias cannot be guaranteed under strong violations of causal assumptions.
Wu et al. (Fri,) studied this question.