Predictive Modeling for Observational Studies with Adjustment of Selection Bias

• Activity Number: 612
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Risk Analysis
• Author(s): Zugui Zhang*+
• Companies: Christiana Care Health System
• Keywords: Predictive modeling ; observational studies ; by inverse probability weighting ; reduce treatment selection bias ; performances of models ; bias adjustment

Abstract

Predictive modeling is the process by which a model is developed to best predict the probability of outcomes. Unlike randomized clinical trials, in observational studies, outcomes are usually related to treatment choice or the exposure conditions, which are not controlled or balanced by study design. Predictive modeling for observational studies, while subject to treatment selection bias, can supplement randomized trials with greater numbers, greater generalizability, and more contemporary data which can be regularly updated. Because of their size, representativeness, and reflection of real-world practice, observational databases are becoming major contributors to predictive modeling research. In this study, we explored predictive modeling analysis using data from the Society of Thoracic Surgeons (STS) Database and the American College of Cardiology Foundation (ACCF) National Cardiovascular Data Registry (NCDR). We examined the predictive models with risk factors for 30-day readmission of coronary-artery bypass grafting (CABG) versus percutaneous coronary intervention (PCI). Results were adjusted by inverse probability weighting to reduce treatment selection bias and it is foun