Clustering High Needs/Complex Patients Using Latent Class Analysis

• Activity Number: 215 - Contributed Poster Presentations: Section on Statistical Learning and Data Science
• Type: Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Meghan Hatfield* and Jodi McCloskey and Connie Uratsu and Richard Grant
• Companies: Kaiser Permanente and Kaiser Permanente and Kaiser Permanente and Kaiser Permanente
• Keywords: Latent Class Analysis; Complex Patients; Electronic Medical Records; Clustering; Patient Profiles

Abstract

Medically complex patients generate disproportionate costs to health systems. Latent class analysis (LCA) is a useful tool to identify specific subgroups within this heterogenous population for targeted management. We used LCA to identify distinct patient profiles among the top 3.0% (N=104,869) medically complex adults in Kaiser Permanente Northern California using multi-domain EHR data. We clustered on 107 binary variables including past healthcare utilization history, socio-demographics, health behaviors, procedures, medications, care contacts, abnormal laboratory results, and orders for durable medical equipment. We determined the appropriate number of classes using model fit statistics (Log-likelihood; Akaike information criteria; Bayes information criteria), class separation statistics (odds of correct classification ratio; average posterior class probability), and clinical review. We identified 7 patient profiles. We discuss the differences in these profiles and how clustering may be a useful tool to identify underlying subgroups of patients for more targeted care.