Development and Validation of Algorithms to Predict Activity, Mobility, and Memory Limitations Using Medicare Claims and Post-Acute Care Assessments.
Sara E HeinsDenis AgnielJacob MannMelony E SorberoPublished in: Journal of applied gerontology : the official journal of the Southern Gerontological Society (2023)
Functional impairment predicts mortality and health care utilization. However, validated measures of functional impairment are not routinely collected during clinical encounters and are impractical to use for large-scale risk-adjustment or targeting interventions. This study's purpose was to develop and validate claims-based algorithms to predict functional impairment using Medicare Fee-for-Service (FFS) 2014-2017 claims data linked with post-acute care (PAC) assessment data and weighted to better represent the overall Medicare FFS population. Using supervised machine learning, predictors were identified that best predicted two functional impairment outcomes measured in PAC data-any memory limitation and a count of 0-6 activity/mobility limitations. The memory limitation algorithm had moderately high sensitivity and specificity. The activity/mobility limitations algorithm performed well in identifying beneficiaries with five or more limitations, but overall accuracy was poor. This dataset shows promise for use in PAC populations, though generalizability to broader older adult populations remains a challenge.
Keyphrases
- machine learning
- acute care
- big data
- healthcare
- deep learning
- artificial intelligence
- health insurance
- electronic health record
- working memory
- affordable care act
- mental health
- magnetic resonance
- physical activity
- cardiovascular disease
- type diabetes
- coronary artery disease
- weight loss
- adipose tissue
- peripheral blood
- skeletal muscle
- metabolic syndrome
- middle aged
- insulin resistance
- genetic diversity
- structural basis