Coding-Driven Changes In Measured Risk In Accountable Care Organizations.

Claims data, which form the foundation of risk adjustment in payment for health care services, may reflect efforts to capture more-or more severe-clinical conditions rather than true changes in health status. This can distort payments. We quantify this in the context of Medicare's accountable care organization (ACO) program by comparing risk scores derived from two different measurement approaches. One approach uses diagnoses coded on claims based on Centers for Medicare and Medicaid Services Hierarchical Condition Categories (HCC), and the other uses self-reported, survey-based health data from the Consumer Assessment of Healthcare Providers and Systems (CAHPS). During 2013-16 HCC-based risk scores grew faster than CAHPS-based risk scores (2.1 percent versus 0.3 percent annually), and the gap in HCC- and CAHPS-based risk score growth varied widely across ACOs. The average gap in risk score growth appears to be the result primarily of HCC coding practices rather than poor performance of the CAHPS model, suggesting that coding practices (not necessarily driven by ACO contracts) may account for most of the observed risk score growth for ACO beneficiaries.

Evaluating Hospital Readmission Rates After Discharge From Inpatient Rehabilitation.

OBJECTIVE: To examine facility-level rates of all-cause, unplanned hospital readmissions for 30 days after discharge from inpatient rehabilitation facilities (IRFs). DESIGN: Observational design. SETTING: Inpatient rehabilitation facilities. PARTICIPANTS: Medicare fee-for-service beneficiaries (N=567,850 patient-stays). INTERVENTIONS: Not applicable. MAIN OUTCOME MEASURES: The outcome is all-cause, unplanned hospital readmission rates for IRFs. We adapted previous risk-adjustment and statistical approaches used for acute care hospitals to develop a hierarchical logistic regression model that estimates a risk-standardized readmission rate for each IRF. The IRF risk-adjustment model takes into account patient demographic characteristics, hospital diagnoses and procedure codes, function at IRF admission, comorbidities, and prior hospital utilization. We presented national distributions of observed and risk-standardized readmission rates and estimated confidence intervals to make statistical comparisons relative to the national mean. We also analyzed the number of days from IRF discharge until hospital readmission. RESULTS: The national observed hospital readmission rate by 30 days postdischarge from IRFs was 13.1%. The mean unadjusted readmission rate for IRFs was 12.4%±3.5%, and the mean risk-standardized readmission rate was 13.1%±0.8%. The C-statistic for our risk-adjustment model was .70. Nearly three-quarters of IRFs (73.4%) had readmission rates that were significantly different from the mean. The mean number of days to readmission was 13.0±8.6 days and varied by rehabilitation diagnosis. CONCLUSIONS: Our results demonstrate the ability to assess 30-day, all-cause hospital readmission rates postdischarge from IRFs and the ability to discriminate between IRFs with higher- and lower-than-average hospital readmission rates.