Predicting health-related quality of life for patients with gastroesophageal cancer

Background: Gastroesophageal cancer has a poor prognosis, and treatment significantly impacts health-related quality of life (HRQoL). Accurate prediction of HRQoL changes after treatment can support shared decision-making. This study aimed to develop and validate HRQoL prediction models for patients with gastroesophageal cancer using established risk-prediction models and a newly proposed sequential score model. Methods: HRQoL data came from the Prospective Observational Cohort Study of Esophageal-Gastric Cancer Patients registry, linked to the Netherlands Cancer Registry. The EORTC QLQ-C30 functioning scales were used as outcomes. Risk-prediction models, based on logistic elastic-net regression, estimated the probability of meaningful HRQoL deterioration at 3, 6, and 12 months post-treatment. The sequential score model, using XGBoost regression, predicted the next HRQoL score at any time. Calibration curves and integrated calibration index (ICI) assessed predictive performance, with Brier scores and AUC for risk-prediction models and root mean squared error plus Out-of-Sample r^² for sequential models. Results: Risk-prediction models showed strong performance (ICI: 0.03–0.08; Brier score: 0.09–0.17; AUC: 0.79–0.87) for predicting significant deterioration in Summary Score, Physical Functioning, and Fatigue, with good calibration. Sequential score models explained up to 40% of the variance in HRQoL scores. Conclusion: Both models effectively predicted HRQoL in gastroesophageal cancer patients, demonstrating potential to enhance patient care and information sharing through accurate prediction of HRQoL outcomes.