Postoperative mortality after esophagectomy for cancer: development of a preoperative risk prediction model.

BACKGROUND

Surgical resection for the treatment of esophageal cancer remains a high-risk procedure. To develop a model to predict risk of postoperative death, we sought to identify factors associated with postoperative mortality for Medicare patients undergoing esophagectomy for cancer.

METHODS

We evaluated patients in the Surveillance, Epidemiology, and End Results Program (SEER)-Medicare database who underwent esophagectomy for esophageal cancer from 1997 to 2003. Variables evaluated were patient age, race, marital status, sex, tumor stage, Charlson score, and hospital volume. Hospital volume was evaluated in tertiles of even volume groups (low, < .67 cases a year; medium, .68 to 2.33 cases a year; high, > 2.33 cases a year). The primary outcome measure was postoperative mortality, defined as death within 30 days of esophagectomy or death during the hospitalization in which the primary surgical procedure was performed. In-hospital deaths more than 30 days after esophagectomy were included in the outcomes to more accurately estimate the true mortality of this procedure. Multivariable logistic regression analyses were performed to evaluate the relationship between patient and provider characteristics and postoperative mortality. Finally, characteristics identified by the regression analysis were used to generate a simplified, clinically applicable model predicting risk of postoperative mortality in the Medicare population.

RESULTS

A total of 1172 patients underwent esophageal cancer surgery during this study period. Overall postoperative mortality was 14%. Multivariable logistic regression demonstrated that age, Charlson score, and hospital volume were statistically significant predictors of postoperative mortality. The other variables such as race, martial status, sex, and disease stage were not found to be significant. The odds of postoperative mortality at low-volume hospitals were almost twice those at a high-volume hospital. Age greater than 80 increased odds of mortality almost twofold. Similarly, Charlson scores of > or = 2 resulted in more than a 1.5-fold risk of postoperative mortality. Our prediction model using these variables accurately stratified postoperative mortality for this population.

CONCLUSIONS

Postoperative mortality (30-day and in-hospital) remains high after esophagectomy. Age, Charlson score, and hospital volume were identified as independent predictors of postoperative mortality. A simple risk prediction model that uses preoperative clinical data accurately predicted patient postoperative mortality for this SEER-Medicare population.