Functional Limitations are Key Predictors of Avoidable Utilization and Readmission Nicholas K. Schiltz, PhD Case Western Reserve University AcademyHealth Annual Research Meeting June 27, 2017
Background Preventing readmissions and avoidable care are key policy goals to improve quality of care and reduce cost Multimorbidity -> associated with higher rates
Background However, the specific combinations of MM that have the greatest effect on utilization outcomes are not well understood
Multimorbidity “…[ Multimorbidity] population is categorized by tremendous clinical heterogeneity” – “Developing means for determining homogenous subgroups…is an important step to improve health status of this population.” 2 million unique chronic disease combinations in Medicare alone (Sorace, 2011) Sorace J, Wong H-H, Worrall C, Kelman J, Saneinejad S, MaCurdy T. The complexity of disease combinations in the Medicare population. Popul. Health Manag. 2011;14(4):161-166
Study Aim Use a data-driven approach to investigate the specific combinations of conditions and disabilities that are the most pivotal predictors of avoidable utilization and readmission
Data Sources • Health and Retirement Study – U.S. representative survey of age 50 and older population – Cohort study – interviewed every two years • Linked to Medicare claims • 31,487 person-waves (2002 – 2012) – Age 65 with 24 months Part A & B.
Independent Variables • Chronic conditions – (heart disease, lung disease, cancer, psychiatric disorders, diabetes, stroke, arthritis, hypertension) • Functional limitations – (strength, upper & lower body mobility, limitations in Activities of Daily Living, and Instrumental Activities of Daily Living (IADL)) • Geriatric syndromes – (visual / hearing impairment, incontinence, depressive symptoms, cognitive impairment) • Demographics, economic, behavioral
Dependent Variable • Potentially avoidable ED Visits – NYU avoidable ED visit algorithm • Preventable Hospitalizations – AHRQ Prevention Quality Indicator #90 • All-cause 30 day readmission
Analytic Approach • Classification and Regression Tree – Tree represents a complete model – GLMM: to account for repeated measures – 10-fold cross-validation • Bootstrap Aggregation: Random Forest • R packages: partykit, caret, randomForest
Study Population Characteristic Total No. of person-waves 31,487 Age categories 65 – 69 26% 70 – 74 25% 75 – 79 20% 80+ 29% Female 58% Race/Ethnicity Black non-Hispanic 12% Hispanic 6% Other 1% White non-Hispanic 81%
Multimorbidity Functional Limitations % Chronic Conditions % Upper mobility 41% Heart Disease 37% Lower mobility 71% COPD 12% Strength 65% Stroke 21% Activities of Daily Living (ADL) 6% Cancer 16% Instrumental ADLs 19% Diabetes 23% Geriatric Syndromes Arthritis 68% Impaired vision 24% Hypertension 67% Impaired hearing 27% Psychiatric Disorders 15% Depressive symptoms 13% Incontinence 26% Severe Pain 6% Poor Cognitive Functioning 8%
Avoidable Utilization / Readmission • Avoidable ED Use: 29.0% • Preventable hospitalizations: 8.6% • 30-day readmissions: 8.9%
Avoidable ED Visits IADL Limitations No Yes Heart ADL Disease Limitations None Mild or Severe No Yes Upper Lung Heart Mobility Lim. Disease Disease None No Yes Mild or Severe None Mild or Severe Lung HH Income Upper Age Age Disease (% of FPL) Mobility Lim. ≤79 >80 >85 >200% Mild Severe ≤85 ≤200% No Yes 100% 50% 0%
Avoidable ED Visits IADL Limitations No Yes Heart ADL Disease Limitations None Mild or Severe No Yes Upper Lung Heart Mobility Lim. Disease Disease None No Yes Mild or Severe None Mild or Severe Lung HH Income Upper Age Age Disease (% of FPL) Mobility Lim. ≤79 >80 >85 >200% Mild Severe ≤85 ≤200% No Yes 100% 50% 0%
Avoidable ED Visits IADL Limitations No Yes Heart ADL Disease Limitations None Mild or Severe No Yes Upper Lung Heart Mobility Lim. Disease Disease None No Yes Mild or Severe None Mild or Severe Lung HH Income Upper Age Age Disease (% of FPL) Mobility Lim. ≤79 >80 >85 >200% Mild Severe ≤85 ≤200% No Yes 100% 50% 0%
Avoidable ED Visits IADL Limitations No Yes Heart ADL Disease Limitations None Mild or Severe No Yes Upper Lung Heart Mobility Lim. Disease Disease None No Yes Mild or Severe None Mild or Severe Lung HH Income Upper Age Age Disease (% of FPL) Mobility Lim. ≤79 >80 >85 >200% Mild Severe ≤85 ≤200% No Yes 100% 50% 0%
Preventable hospitalization Lung Disease Severe None or Mild IADL Heart Limitations Disease No Yes Heart Heart Disease Disease None Mild or Severe None None Mild or Severe Mild or Severe Lower ADL Age Age Mobility Lim. Limitations No Yes No Yes ≤78 >85 >79 ≤85 100% 50% 0%
Preventable hospitalization Lung Disease Severe None or Mild IADL Heart Limitations Disease No Yes Heart Heart Disease Disease None Mild or Severe None None Mild or Severe Mild or Severe Lower ADL Age Age Mobility Lim. Limitations No Yes No Yes ≤78 >85 >79 ≤85 100% 50% 0%
Preventable hospitalization Lung Disease Severe None or Mild IADL Heart Limitations Disease No Yes Heart Heart Disease Disease None Mild or Severe None None Mild or Severe Mild or Severe Lower ADL Age Age Mobility Lim. Limitations No Yes No Yes ≤78 >85 >79 ≤85 100% 50% 0%
Preventable hospitalization Lung Disease Severe None or Mild IADL Heart Limitations Disease No Yes Heart Heart Disease Disease None Mild or Severe None None Mild or Severe Mild or Severe Lower ADL Age Age Mobility Lim. Limitations No Yes No Yes ≤78 >85 >79 ≤85 100% 50% 0%
Preventable hospitalization Lung Disease Severe None or Mild IADL Heart Limitations Disease No Yes Heart Heart Disease Disease None Mild or Severe None None Mild or Severe Mild or Severe Lower ADL Age Age Mobility Lim. Limitations No Yes No Yes ≤78 >85 >79 ≤85 100% 50% 0%
30-day readmission IADL Limitations No Yes Heart Heart Disease Disease None Mild or Severe None Mild or Severe ADL ADL Lower Upper Limitations Limitations Mobility Lim. Mobility Lim. No Yes No Yes ADL Diabetes No Yes Stroke No Yes Limitations No Yes None Mild or Severe None or Mild Severe 100% 50% 0%
30-day readmission IADL Limitations No Yes Heart Heart Disease Disease None Mild or Severe None Mild or Severe ADL ADL Lower Upper Limitations Limitations Mobility Lim. Mobility Lim. No Yes No Yes ADL Diabetes No Yes Stroke No Yes Limitations No Yes None Mild or Severe None or Mild Severe 100% 50% 0%
30-day readmission IADL Limitations No Yes Heart Heart Disease Disease None Mild or Severe None Mild or Severe ADL ADL Lower Upper Limitations Limitations Mobility Lim. Mobility Lim. No Yes No Yes ADL Diabetes No Yes Stroke No Yes Limitations No Yes None Mild or Severe None or Mild Severe 100% 50% 0%
Which predictors are most important?
Which predictors are most important?
Strengths Limitations • Non-parametric approach • Produces a single tree – non-linear relationships – Ensemble methods out perform for prediction • Can empirically identify – Prediction not our goal combinations without: – RF showed consistency – a priori knowledge of what • May not best approach if the most salient combos are hypothesis testing – Constraint of addititive (linear) relationship – Algorithm guides the splitting • Tree models easy to • Misspecification of interpret “avoidable” utilization • Useful for large number of predictors
Conclusions • Functional limitations, particular those related to IADL and ADL, are among the most influential characteristics in predicting readmission and other avoidable utilization measures. • Major chronic conditions like heart and lung disease are secondary in importance, while demographic, economic, and behavioral factors have less impact.
Implications for Policy/Practice • Limitations in functional abilities may hamper self-management of chronic disease – increasing likelihood of readmission and other avoidable utilization – Discharge to home health care, assisted living, or SNF could help this vulnerable group • Routine assessment of functional abilities in well- care settings would help identify those most at risk, and to permit tailored care management
Acknowledgements • Co-authors: Warner DF, Bakaki PM, Smyth KA, Stange KC, Gravenstein S, and Koroukian SM • Funding: – AHRQ #1R21HS023113-01 (PI: Koroukian) – CDC #3U48DP005030-01S3 (PI: Koroukian & Schiltz) – NIH #KL2 TR000440 (PI: Konstan)
Thank you! Nicholas Schiltz: nks8@case.edu
Recommend
More recommend
