Prediction Model of Long-term Survival After Esophageal Cancer Surgery

Objective: We aimed to develop prediction models for estimating the long-term survival in patients who have undergone surgery for esophageal cancer. Background: Few prediction models have been developed for the long-term survival in esophageal cancer patients. Methods: This nationwide Swedish population-based cohort study included 1542 patients who survived for >= 90 days after esophageal cancer surgery between 1987 and 2010, with follow-up until 2016. Risk prediction models for 1-, 3-, and 5-year all-cause mortality and 3- and 5-year disease-specific mortality were developed using logistic regression. Candidate predictors were established and readily identifiable prognostic factors. The performance of the models was assessed by the area under receiver-operating characteristic curve (AUC) with interquartile range (IQR) using bootstrap cross-validation and risk calibration. Results: Predictors included in all models were age, sex, pathological tumor stage, tumor histology, and resection margin status. The models also included various additional predictors depending on the outcome, that is, education level, neoadjuvant therapy, reoperation (within 30 d of primary surgery) and comorbidity (Charlson comorbidity index). The AUC statistics after cross-validation were 0.71 (IQR 0.69-0.74) for 1-year, 0.77 (IQR 0.75-0.80) for 3-year, and 0.78 (IQR 0.76-0.81) for 5-year all-cause mortality. The corresponding values were 0.76 (IQR 0.74-0.79) for 3-year and 0.77 (IQR 0.71-0.83) for 5-year disease-specific mortality. All models showed good agreement between the observed and predicted risks. Conclusions: These models showed good performance for predicting long-term survival after esophageal cancer surgery and may thus be useful for patients in planning their lives and to guide the postoperative treatment and follow-up.