Predictive and Interpretable Machine Learning of Economic Burden: The Role of Chronic Conditions Among Elderly Patients with Incident Primary Merkel Cell Carcinoma (MCC)

• Published in: ClinicoEconomics and outcomes research 2024, Vol.16, p.847-868
• Main Authors: Mbous, Yves Paul Vincent, Siddiqui, Zasim Azhar, Bharmal, Murtuza, LeMasters, Traci, Kolodney, Joanna, Kelley, George A, Kamal, Khalid M, Sambamoorthi, Usha
• Format: Article
• Language: English
• Subjects: Accountable care organizations; Beneficiaries; cancer care; Cardiovascular diseases; Chemotherapy; chronic conditions; Chronic illnesses; Codes; Cost control; economic burden; Economics; Epidemiology; Health care expenditures; Hypertension; Medicaid; Medical diagnosis; Medical prognosis; Medicare; merkel cell carcinoma; Older people; Original Research; out-of-pocket expenditures; Patients; Prediction models; Radiation therapy; Skin cancer; Surveillance
• ISSN: 1178-6981; 1178-6981
• DOI: 10.2147/CEOR.S456968

To evaluate chronic conditions as leading predictors of economic burden over time among older adults with incident primary Merkel Cell Carcinoma (MCC) using machine learning methods. We used a retrospective cohort of older adults (age ≥ 67 years) diagnosed with MCC between 2009 and 2019. For these elderly MCC patients, we derived three phases (pre-diagnosis, during-treatment, and post-treatment) anchored around cancer diagnosis date. All three phases had 12 months baseline and 12-months follow-up periods. Chronic conditions were identified in baseline and follow-up periods, whereas annual total and out-of-pocket (OOP) healthcare expenditures were measured during the 12-month follow-up. XGBoost regression models and SHapley Additive exPlanations (SHAP) methods were used to identify leading predictors and their associations with economic burden. Congestive heart failure (CHF), chronic kidney disease (CKD) and depression had the highest average incremental total expenditures during pre-diagnosis, treatment, and post-treatment phases, respectively ($25,004, $24,221, and $16,277 (CHF); $22,524, $19,350, $20,556 (CKD); and $21,645, $22,055, $18,350 (depression)), whereas the average incremental OOP expenditures during the same periods were $3703, $3,013, $2,442 (CHF); $2,457, $2,518, $2,914 (CKD); and $3,278, $2,322, $2,783 (depression). Except for hypertension and HIV, all chronic conditions had higher expenditures compared to those without the chronic conditions. Predictive models across each of phases of care indicated that CHF, CKD, and heart diseases were among the top 10 leading predictors; however, their feature importance ranking declined over time. Although depression was one of the leading drivers of expenditures in unadjusted descriptive models, it was not among the top 10 predictors. Among older adults with MCC, cardiac and renal conditions were the leading drivers of total expenditures and OOP expenditures. Our findings suggest that managing cardiac and renal conditions may be important for cost containment efforts.