Using the Lehman Family of ROC Curves to Evaluate Proxy Measures of Post-Operative Complications

• Activity Number: 635
• Type: Topic Contributed
• Date/Time: Thursday, August 8, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Section for Statistical Programmers and Analysts
• Author(s): Andrew Mullard*+
• Companies: Michigan Surgical Quality Collaborative
• Keywords: hospital profiling ; surgical quality improvement ; clinical risk adjustment ; clustered data ; Lehman family of distributions ; Receiver Operating Characteristic (ROC) curve

Abstract

Multihospital collaborations for surgical quality improvement are often hindered by complex data collection needs. Identification of valid and reliable proxy measures that simplify data collection may improve the feasibility of such efforts. This study uses data from the 52 hospital Michigan Surgical Quality Collaborative to examine use of the Lehman family of Receiver Operating Characteristic (ROC) curves coupled with multivariable Cox regression to evaluate hospital length of stay as a proxy measure for postoperative complications. Analyses can be performed using SAS PROC PHREG for covariate adjustment, model selection, and evaluation of model fit for hospital clustered data. Cox regression provides regression-based covariate adjustments for the functional form of the Lehman ROC curves and estimation of the Area Under the Curve (AUC). This approach supports evaluation of the effect of complications severity on discriminatory accuracy, the effect of covariates on discriminatory accuracy, and the incremental value of candidate proxy measures.