A. N. Kothari1,2, S. A. Brownlee2, C. Fischer1, P. C. Kuo1,2, G. J. Abood1 1Loyola University Medical Center,Department Of Surgery,Maywood, IL, USA 2Loyola University Medical Center,One:MAP Division Of Clinical Informatics And Analytics,Maywood, IL, USA
Introduction: Residential zip code can provide insight into patient socioeconomic status and community factors that influence perioperative care. Disparity in socioeconomic status has been well demonstrated to effect operative morbidity and mortality, but the role geography plays in outcomes of oncologic surgeries has yet to be established. The objective of this study was to determine the ability for residential zip code to predict postoperative outcomes in patients undergoing elective oncologic resections.
Methods: We conducted a retrospective cohort review using the Healthcare Cost and Utilization Project State Inpatient databases for Florida, Iowa, New York, and Washington. Included were patients that underwent open elective resection for the following malignancies: pancreas, colon, esophagus, stomach, and liver. The primary outcome was major inpatient morbidity or mortality. Regression-based predictive models were constructed using data from 2009 to 2012 (training set), with automated feature selection used to optimize fit. Zip code was added as a forced in variable when not selected for inclusion. Model performance was measured using in-sample data and validated using 2013 data.
Results: 12,088 patients met inclusion criteria. Composite event rate: 19.9% (range across procedures: 18.7% – 21.1%). Event rate varied by residential zip code with a median event rate of 19.8 % (0.0 – 33.3%). Best fit regression model, without the inclusion of zip code, had an accuracy of 64.9% for predicting the primary outcome on the validation cohort. Zip code alone predicted the primary outcome in the validation cohort with an accuracy of 68.0%. Adding zip code to the best fit regression model increased the accuracy of prediction to 72.3%. Comparison of model performances is shown in Table.
Conclusion: Residential zip code can act as an independent predictor of postoperative outcomes. In addition, the inclusion of zip code can improve the performance of conventional models in the prediction of inpatient outcomes. Inclusion of residential zip code can offer an important adjunct for measuring risk-adjusted outcomes and identifying high risk geographic areas in order to optimize preoperative risk counseling.
