Introduction: Women with breast cancer experience psychological distress, and resilience-strengthening psychosocial support may improve their quality of life (QoL). Identifying those at risk of low QoL is challenging. This study evaluated the cost-effectiveness of a machine learning-based QoL predictor to support clinical decision-making regarding psychosocial support (sample size: 660). Methods: A decision tree cost–utility model was developed to compare four decision-making strategies in offering psychosocial support: the clinician alone, the QoL predictor alone, the clinician supported by the predictor, and no prediction with no psychosocial support. QoL after one year was used as a proxy for resilience. Costs, health outcomes, and net monetary benefits (NMBs) were estimated using a one-year time horizon. Incremental cost-effectiveness ratios (ICERs) were calculated and dominance assessed. A societal scenario analysis incorporated productivity losses. A probabilistic sensitivity analysis generated cost-effectiveness acceptability curves. Results: Clinicians supported by the QoL predictor produced the highest NMB (EUR 16,349) and the greatest quality-adjusted life year (QALY) gain (0.759), with an ICER of EUR 22,892 compared with the next least costly strategy. Clinician-only prediction and predictor-only approaches were dominated or extendedly dominated. Under the societal perspective, all strategies produced negative NMB values due to productivity losses, but the overall ranking remained unchanged. The probabilistic sensitivity analysis showed that the combined clinician and predictor strategy had a 69% probability of being cost-effective at a willingness to pay threshold of EUR 30,000. Conclusions: Combining clinician judgement with the machine learning-based QoL predictor improved the targeting of psychosocial support and was the most cost-effective strategy. Further prospective and comparative studies are needed to confirm its long-term effectiveness and cost-effectiveness in clinical practice.
Hakkarainen et al. (2026) studied this question.