Introduction: A number of models have been applied to predict outcomes from esophagectomy. This systematic review aimed to compare their clinical credibility, methodological quality and performance.
Methods: A systematic review of the PubMed, EMBASE and Cochrane databases was performed in October 2012. Model and study quality were appraised using the framework of Minne et al.
Results: Twenty studies were included in total; these were heterogeneous, retrospective and conducted over a number of years; all models were generated via logistic regression. Overall mortality was high, and consequently not representative of current practice. Clinical credibility and methodological quality were variable, with frequent failure to perform internal validation and variable presentation of calibration and discrimination metrics. P-POSSUM demonstrated the best calibration and discrimination for predicting mortality. Other than the Southampton score (which has yet to be externally validated) and the Amsterdam score, no studies had utility in predicting complications.
Conclusion: Whilst a number of models have been developed, adapted or trialled, due to numerous limitations, larger and more contemporary studies are required to develop and validate models further. The role of alternative techniques such as decision tree analysis and artificial neural networks is not known.