A Novel Machine Learning Algorithm for Creating Risk-Adjusted Payment Formulas.

Importance: Models predicting health care spending and other outcomes from administrative records are widely used to manage and pay for health care, despite well-documented deficiencies. New methods are needed that can incorporate more than 70 000 diagnoses without creating undesirable coding incentives. Objective: To develop a machine learning (ML) algorithm, building on Diagnostic Item (DXI) categories and Diagnostic Cost Group (DCG) methods, that automates development of clinically credible and transparent predictive models for policymakers and clinicians. Design, Setting, and Participants: DXIs were organized into disease hierarchies and assigned an Appropriateness to Include (ATI) score to reflect vagueness and gameability concerns. A novel automated DCG algorithm iteratively assigned DXIs in 1 or more disease hierarchies to DCGs, identifying sets of DXIs with the largest regression coefficient as dominant; presence of a previously identified dominating DXI removed lower-ranked ones before the next iteration. The Merative MarketScan Commercial Claims and Encounters Database for commercial health insurance enrollees 64 years and younger was used. Data from January 2016 through December 2018 were randomly split 90% to 10% for model development and validation, respectively. Deidentified claims and enrollment data were delivered by Merative the following November in each calendar year and analyzed from November 2020 to January 2024. Main Outcome and Measures: Concurrent top-coded total health care cost. Model performance was assessed using validation sample weighted least-squares regression, mean absolute errors, and mean errors for rare and common diagnoses. Results: This study included 35 245 586 commercial health insurance enrollees 64 years and younger (65 901 460 person-years) and relied on 19 clinicians who provided reviews in the base model. The algorithm implemented 218 clinician-specified hierarchies compared with the US Department of Health and Human Services (HHS) hierarchical condition category (HCC) model's 64 hierarchies. The base model that dropped vague and gameable DXIs reduced the number of parameters by 80% (1624 of 3150), achieved an R2 of 0.535, and kept mean predicted spending within 12% ($3843 of $31 313) of actual spending for the 3% of people with rare diseases. In contrast, the HHS HCC model had an R2 of 0.428 and underpaid this group by 33% ($10 354 of $31 313). Conclusions and Relevance: In this study, by automating DXI clustering within clinically specified hierarchies, this algorithm built clinically interpretable risk models in large datasets while addressing diagnostic vagueness and gameability concerns.

Development and Assessment of a New Framework for Disease Surveillance, Prediction, and Risk Adjustment: The Diagnostic Items Classification System.

Importance: Current disease risk-adjustment formulas in the US rely on diagnostic classification frameworks that predate the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM). Objective: To develop an ICD-10-CM-based classification framework for predicting diverse health care payment, quality, and performance outcomes. Design Setting and Participants: Physician teams mapped all ICD-10-CM diagnoses into 3 types of diagnostic items (DXIs): main effect DXIs that specify diseases; modifiers, such as laterality, timing, and acuity; and scaled variables, such as body mass index, gestational age, and birth weight. Every diagnosis was mapped to at least 1 DXI. Stepwise and weighted least-squares estimation predicted cost and utilization outcomes, and their performance was compared with models built on (1) the Agency for Healthcare Research and Quality Clinical Classifications Software Refined (CCSR) categories, and (2) the Health and Human Services Hierarchical Condition Categories (HHS-HCC) used in the Affordable Care Act Marketplace. Each model's performance was validated using R 2, mean absolute error, the Cumming prediction measure, and comparisons of actual to predicted outcomes by spending percentiles and by diagnostic frequency. The IBM MarketScan Commercial Claims and Encounters Database, 2016 to 2018, was used, which included privately insured, full- or partial-year eligible enrollees aged 0 to 64 years in plans with medical, drug, and mental health/substance use coverage. Main Outcomes and Measures: Fourteen concurrent outcomes were predicted: overall and plan-paid health care spending (top-coded and not top-coded); enrollee out-of-pocket spending; hospital days and admissions; emergency department visits; and spending for 6 types of services. The primary outcome was annual health care spending top-coded at $250 000. Results: A total of 65 901 460 person-years were split into 90% estimation/10% validation samples (n = 6 604 259). In all, 3223 DXIs were created: 2435 main effects, 772 modifiers, and 16 scaled items. Stepwise regressions predicting annual health care spending (mean [SD], $5821 [$17 653]) selected 76% of the main effect DXIs with no evidence of overfitting. Validated R 2 was 0.589 in the DXI model, 0.539 for CCSR, and 0.428 for HHS-HCC. Use of DXIs reduced underpayment for enrollees with rare (1-in-a-million) diagnoses by 83% relative to HHS-HCCs. Conclusions: In this diagnostic modeling study, the new DXI classification system showed improved predictions over existing diagnostic classification systems for all spending and utilization outcomes considered.