Predicting Patient Cost Blooms: A Longitudinal Population-Based Study

• Activity Number: 686
• Type: Topic Contributed
• Date/Time: Thursday, August 4, 2016 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Suzanne Tamang* and Jean-Raymond Betterton and Lester Mackey and Lucas Jansen and Arnold Milstein and Henrik Sorensen and Lars Pedersen and Nigam Shah
• Companies: Stanford University and Stanford University and Stanford University and Stanford University and Stanford University and Aarhus University and Aarhus University and Stanford University
• Keywords: population health ; classification ; risk-adjustment ; high cost patients ; healthcare

Abstract

Accurate healthcare cost prediction tools are necessary to align rewards appropriately, transparently, and efficiently in any value-based care delivery system. However, scant progress has been made in improvement of cost prediction tools in over a decade. We sought to improve on classification of high cost patients, as measured by per capita annual healthcare spending. Using data for the entire population of Western Denmark between 2004 and 2011, we analyzed high-cost spending trends and based on one year of prior data, classified high cost patients in the subsequent year. Our key contributions are new insights into multiyear high-cost trends and the development of an enhanced model for proactively identifying "cost blooms" namely patients move from a lower decile to the top decile of per capita population health spending between consecutive years. Our best models for cost bloom prediction achieved a 30% relative improvement in predictive power over a standard diagnosis-based tool. We expect our study to inform healthcare leaders, payers and providers, who need better strategies to proactively identify high cost patients.