SUMMARY ABSTRACT Background One compelling strategy for improving patient outcomes while reducing healthcare costs is to focus on veterans that account for the vast majority of poor outcomes, health utilization, and VA spending (i.e., HNHC veterans). However, successfully managing HNHC veterans is challenging because these patients are heterogeneous, each requiring a different management strategy. Veterans who intensely use services are of particular interest, especially those with chronic conditions since 20% of them will experience a hospitalization and readmission within 30 days after discharge. Hospitalization, emergency room visits, and re-hospitalization rates are even higher for socioeconomically disadvantaged populations, minorities, and veterans with disability. However, much of this utilization is preventable and could be averted with better longitudinal care. The VA has increased its efforts in identifying HNHC veterans through development of the Care Assessment Needs (CAN) score and care management programs, but without greater detail enabling tailoring of clinical programs HNHC veteran subgroups, linking these scores to strategies to improve care is difficult. Objectives The objectives of this study are to: (1) apply statistical and machine learning clustering methods to classify HNHC veterans into clinically actionable subgroups based on detailed clinical information extending beyond diagnosis codes, (2) compare the HNHC subgroups to veterans with similar diagnoses who were not HNHC, and (3) describe the characteristics of the HNHC subgroups (i.e., CAN Scores) and changes over time. Methods To achieve these objectives, we will analyze patient-level data from the National Patient Care Database (2013- 2015) using the VA Informatics and Computing Infrastructure (VINCI) platform to develop models that cluster HNHC veterans into subgroups based on demographic, clinical, and social characteristics. We will utilize a combination of statistical (latent class analysis) and machine learning clustering (e.g. k-means clustering) algorithms. Our definition of a HNHC veteran will comprise the highest quartiles of predicted risk of death or acute hospitalization (i.e., CAN score > 75). Subgroups and characteristics to compare HNHC and non-HNHC veterans will be constructed using 3 approaches: 1) cluster veterans who are HNHC in 2014, 2) cluster veterans who are HNHC in 2014 and 2015 (persistently HNHC), and 3) cluster all non-HNHC veterans into subgroups. Anticipated Impacts on Veterans Health Care This project aims to identify clinically actionable subgroups of high-need, high-cost (HNHC) veterans using data-driven techniques rather than expert opinion. We hypothesize that distinct clinical characteristics will define subgroups of HNHC veterans and that these subgroups of veterans likely require different management strategies. Thus, the categorization of HNHC veterans into discrete types of patients will support nurse care managers and primary care clinicians in their selection and delivery of appropriate programs to HNHC veterans and more broadly to help the VA to better identify gaps in its clinical and care management programs
External Links for this Project
Grant Number: I21HX002564-01A1
- Parikh RB, Linn KA, Yan J, Maciejewski ML, Rosland AM, Volpp KG, Groeneveld PW, Navathe AS. A machine learning approach to identify distinct subgroups of veterans at risk for hospitalization or death using administrative and electronic health record data. PLoS ONE. 2021 Feb 19; 16(2):e0247203. [view]
None at this time.
Treatment - Observational, TRL - Development, TRL - Applied/Translational
Best Practices, Guideline Development and Implementation, Healthcare Algorithms, Knowledge Integration
None at this time.