1073 — Hospitalization, Re-Hospitalization, Mortality, and Utilization Patterns Predicted by Novel High-Risk Patient Latent Subgroups and Complexity Scores

Lead/Presenter: Xinhua Zhao,  COIN - Pittsburgh/Philadelphia
All Authors: Zhao X (Center for Health Equity Research & Promotion, Pittsburgh, PA), Vijan S (Center for Clinical Management Research, Ann Arbor, MI), Maciejewski M (Center of Innovation to Accelerate Discovery and Practice Transformation, Durham, NC) Zulman D (Center for Innovation to Implementation, Palo Alto, CA) Thorpe J (Center for Health Equity Research & Promotion, Pittsburgh, PA) Batten A (Clinical Systems Development & Evaluation, Veterans Health Administration, Seattle, WA) Zhang H (Center for Health Equity Research & Promotion, Pittsburgh, PA) Daniels K (Center for Health Equity Research & Promotion, Pittsburgh, PA) Rosland AM (Center for Health Equity Research & Promotion, Pittsburgh, PA)

Objectives:
The ability to define latent groupings among high-risk complex patients has been demonstrated, but these groups' relevance to tailored interventions depends on their ability to predict distinct health outcomes and utilization patterns over time.

Methods:
We used Mixture Item Response Theory (MixIRT) to define 6 consistent clusters of chronic comorbidities and group-specific patient theta (‘complexity') scores among 934,787 high-risk VA primary care patients. We included 934,787 patients whose predicted probability of hospitalization over 12 months was > = 25% during any week in 2014. We used regression models to assess the associations between subgroup membership and patient theta scores with subsequent rates of all-cause mortality, all-cause acute hospitalizations, and 30-day re-hospitalization, and described VA utilization patterns by group.

Results:
81% of patients were matched with a group at a probability of > = 80%. High-risk patient subgroups were characterized by high levels of Substance Use Disorder (SUD, 11% of patients assigned), Cardiometabolic Conditions (CM, 21%), Mental Health Conditions (MH,14%), Pain and Arthritis (PA, 16%), Cancer (13%), and Chronic Liver Disease (12%). The overall rate of one-year mortality was 8% and VA hospitalization was 25%. 17% of hospitalizations resulted in 30-day readmission. Mortality varied significantly among subgroups (SUD reference; AOR (95%CI): Liver 3.2 (3.0, 3.3), Cancer 2.3 (2.2, 2.4), CM 1.6 (1.5, 1.7), PA 1.06 (1.02, 1.11), and MH 0.7 (0.66, 0.73)). One-year hospitalization rates ranged from 61 per 100 patients (Liver) to 31 per 100 patients (MH) (AIRR with SUD = reference: Liver 1.27, Cancer 0.95, CM 0.88, PA 0.75, MH 0.6, all p < 0.001). Among all hospitalizations (n = 253,933), 30-day re-admission rate ranged from Liver 24% to MH 15%; (AIRR Liver 1.1, SUD reference, Cancer 0.99, CM 0.92, PA 0.82, MH 0.73, all p < 0.001 except Cancer p = 0.69). MixIRT derived patient theta scores independently predicted mortality (AOR 1.16, CI 1.15, 1.17), any hospitalization (AOR 1.17, CI 1.16, 1.17), and ?one > = 30-day re-hospitalization (AOR 1.29, CI 1.28, 1.30). Variations in utilization among groups included primary care use (range 5.4 PCP visits/year for MH to 4.2 for SUD, specialty care use (range 13 visits/year for Cancer to 4.8 for SA), and ED visit rate (range 2.7 visits/year for SA to 1.9 for Cancer).

Implications:
MixIRT models were able to assign high-risk patients to subgroups with distinct hospitalization, re-hospitalization, and utilization profiles, despite coming from a single group previously defined by a homogenously high risk score. High rates of re-hospitalization were seen among groups not typically targeted in re-admission prevention programs (Liver, Substance Use Disorder).

Impacts:
Bundled intervention content and intensity could be tailored to the unique characteristics and risk profiles of each high-risk patient group.