Unplanned Hospital Admissions Stratification Model With Heart Failure Metrics: A Novel Approach For Measuring Disease Progression
Ravil Bikmetov,Juan Acosta,Zachary Finnel,Julie Santiago,David Priscal,Hymin Zucker,Noah Shreeve,Chris Koscho,Joseph Gartner,David DeCaprio,Gianfranco Bittar,Susana Acosta,Carol McCall
DOI: https://doi.org/10.1016/j.cardfail.2023.10.112
IF: 6.592
2024-01-01
Journal of Cardiac Failure
Abstract:Introduction Heart Failure (HF) is the leading cause of hospitalization for patients over 65, generating the highest rate of 30-day readmissions, accounting for 33% of all Fee for Service Medicare costs. Moreover, many instances of hospitalization for HF patients are considered preventable, making this group a perfect target for preventative care management programs (PCMP) aiming to improve outcomes and reduce costs by directly targeting avoidable and unplanned hospital admissions (UHAs). Hypothesis By leveraging novel Galectin-3 blood tests and HF classification methods with state-of-the-art ML risk stratification techniques, we can improve accuracy in identifying members at higher risk of adverse outcomes with HF. Methods We enable our PCMP team to preemptively address HF by providing them with the most cutting-edge information (e.g., Galectin-3 blood tests, ML risk stratification) on HF predisposition, progression, and overall adverse event risk. Using 21,009 unique outcome-stratified and continuously enrolled members (mean age 73 years, 56% female), we built an XGBoost model to identify members at high risk of having a UHA within 6 months (prevalence 5.24%). Comorbidities include hypertension 77%, obesity 22%, chronic kidney disease 17%, coronary artery disease 27%, and HF 10% with an average Charlson Comorbidity Index (CCI) of 5.21. We use claims data, EHR, and research-based HF features for model training with a 10-fold cross-validation (80/20 split) and 950 features. The key HF model features are a claims-based estimation of "reduced Left Ventricular Ejection Fraction," HF medication adherence, and the HLM System, which seeks to classify HF progression using a person's claims medical history regarding Heart damage (H), Lung involvement (L), and Malfunction (M) of peripheral organs in analogy to TNM classification used in oncology. Results After running the UHA model with HF features for a year, the model identified members at the 90th risk percentile with more developed HF (HLM Score*: 2.40 vs 2.18), higher costs (PMPM***: $2,500 vs $2,227), higher CCI (CCI***: 16 vs 14), and higher probability of UHA (UHA*: 19.7% vs 18.00%) than a similar UHA model without the specialized HF features. Conclusions Overall, enhancing our existing PCMP with HF-specific information and new technologies allows us to utilize our existing workflows to address HF within our attributed population more accurately. Moreover, it has shown that we can apply these new technologies to proactively identify members with HF, opening the door to future HF-targeted PCMP that can have significant impacts on HF progression within our population.
cardiac & cardiovascular systems