Improving Surgical Risk Prediction and Decision Making among Patients with Cirrhosis

PROJECT SUMMARY Patients with cirrhosis have increased surgical risk relative to the general population Several risk factors have been established to predict cirrhosis surgical risk. These are reflected in the primary clinical tools used for risk prediction—the Model for End-stage Liver Disease-sodium (MELD-Na), Child-Turcotte-Pugh (CTP) score, and the Mayo surgical risk score—which rely on age, cirrhosis severity, ASA physical status score, and etiology of liver disease. However, significant heterogeneity in post-operative mortality by surgery type (e.g., cardiac versus orthopedic) suggests that these tools are inadequate. The literature on cirrhosis surgical risk prediction is further limited by: 1) single-center designs with small sample sizes, 2) lack of granular data for risk prediction, 3) evidence of poor prediction score calibration, 4) lack of key stakeholder involvement to inform real-world implementation of prediction tools, and 5) no incorporation of decision analysis methods to compare surgery to non-operative management. The impact of these shortcomings is that many patients with cirrhosis are denied necessary surgery due to overestimates of risk, and others receive surgery with inaccurate prognostic counseling or inadequate consideration of non-operative options. Granular, population-level data are needed to address the above gaps. By using national Veterans Health Administration (VHA) and University of Pennsylvania Hospital System (UPHS) data, we hypothesize that we will be able to create and implement an accurate, well-calibrated cirrhosis surgical risk calculator with broad clinical utility. The primary aims of this proposal are as follows: Aim 1 – derive, internally validate, and externally validate cirrhosis surgical risk models for short- and intermediate-term post-operative mortality among diverse patients with cirrhosis.; Aim 2 – create a web application for surgical risk prediction informed by key stakeholder input.; Aim 3 – use Markov modeling to compare operative to non-operative management pathways and determine optimal clinical decisions for a common clinical scenario: acute cholecystitis. This proposal will foster Dr. Nadim Mahmud's development as an independent, NIH-funded clinical researcher with a focus on improving risk prediction for patients with chronic liver diseases, as well as specific expertise in advanced prediction modeling, qualitative methods, and decision analysis. This will be facilitated through a comprehensive mentorship plan consisting of: 1) biweekly to monthly meetings with his mentorship team, 2) formal coursework in advanced prediction modeling, qualitative research methods, and decision analysis through the Center for Clinical Epidemiology and Biostatistics (CCEB), Wharton School, Department of Health Policy Research (HPR), Operations, Information, and Decisions Department (OIDD), and Department of Statistics (STAT) at the University of Pennsylvania, 3) structured research workshops and national conferences, and 4) conception, development, and submission of future grants during the latter portion of the award period to further explore issues related to surgical risk prediction among patients with cirrhosis.