Optimal Conversion of Regression Models to Point-Based Scoring Systems for Clinical Use

• Activity Number: 427
• Type: Contributed
• Date/Time: Wednesday, August 5, 2009 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): John Boscardin*+
• Companies: University of California, San Francisco
• Keywords: cox regression ; logistic regression

Abstract

In developing prognostic models, it is very common to convert regression coefficients into point-based scoring systems often by dividing the coefficients by an arbitrary constant and then rounding to the nearest integers. Different values of the constant lead to different numbers of points being assigned to each factor; the resulting point-based systems may have very different clinical interpretations, even when they are quite similar in their discrimination and calibration. We present two approaches to choosing the scaling constant in a non-arbitrary manner. The first is to choose the scale to maximize the correlation between the predicted values from the regression and point-based system. The second involves defining a Bayesian model where coefficient estimates are constrained to correspond to a point-scoring system. An example is given for prediction of post-hospitalization outcomes.