Risk prediction scores for postoperative mortality after esophagectomy: validation of different models

Abstract

Background Different prediction models for operative mortality after esophagectomy have been developed. The aim of this study is to independently validate prediction models from Philadelphia, Rotterdam, Munich, and the ASA. Methods The scores were validated using logistic regression models in two cohorts of patients undergoing esophagectomy for cancer from Switzerland (n = 170) and Australia (n = 176). Results All scores except ASA were significantly higher in the Australian cohort. There was no significant difference in 30-day mortality or in-hospital death between groups. The Philadelphia and Rotterdam scores had a significant predictive value for 30-day mortality (p = 0.001) and in-hospital death (p = 0.003) in the pooled cohort, but only the Philadelphia score had a significant prediction value for 30-day mortality in both cohorts. Neither score showed any predictive value for in-hospital death in Australians but were highly significant in the Swiss cohort. ASA showed only a significant predictive value for 30-day mortality in the Swiss. For in-hospital death, ASA was a significant predictor in the pooled and Swiss cohorts. The Munich score did not have any significant predictive value whatsoever. Conclusion None of the scores can be applied generally. A better overall predictive score or specific prediction scores for each country should be developed.