Isolating Effects of Medicare Code Slot Expansion on Longitudinal Risk Assessment

BackgroundThe evaluation of trends in patient outcomes requires adjustment for the changes in case-mix over time and, thus, could be influenced by the expansion of code slots on inpatient claims that occurred in January 2011. We tested the hypothesis that the changes in code slots caused an artifactual increase in the case mix over time compared with a strategy that restricted inpatient codes to the same number of slots over time, excluding consideration of codes beyond the first 9 after the expansion. MethodsIn Medicare claims over a 5-year period spanning the inpatient code slot expansion (2008-2012), we created cohorts of hospitalizations for heart failure (HF), acute myocardial infarction (AMI) and pneumonia, common hospitalization conditions included in federal policies. We obtained information on risk factors for 30-day post-discharge mortality or readmission for each condition from inpatient facility Medicare claims, outpatient facility claims and professional (or carrier) claims. We evaluated the effects of additional codes captured from the expanded slots in inpatient claims on the number of risk factors or model covariates, overall and based on their contribution to the risk of mortality or readmission. We modelled the effects of code expansion on risk-assessment using an interrupted time series framework. ResultsThere were 2,102,509 eligible hospitalizations for HF, 872,734 for AMI and 1,824,079 for pneumonia. The average number of risk factors increased across all covariate selection strategies. There was a larger increase in monthly average covariate count that included all codes at the time of the code slot increase relative to a strategy that consistently used only 9 inpatient codes (level change in interrupted time series model, 0.9% [95% CI 0.7% to 1.1%] in HF, 0.6% [0.5% to 0.7%] for AMI and 0.6% [0.4% to 0.8%] for pneumonia). Using both inpatient and outpatient/carrier codes for assessing risk factors, there was a smaller difference between strategies using 9 inpatient codes, compared with all inpatient claims (relative excess increase in covariates by 0.6% [0.4% to 0.8%] in HF, 0.4% [0.3% to 0.5%] for AMI, and 0.3% [0.1% to 0.6%] in pneumonia). However, the additional codes were limited to covariates with small contributions to the risk-adjustment models for mortality, without a significant inflection in measured risk of mortality across code expansion (P>0.05 in interrupted time-series models). Measured readmission risk increased with using only inpatient claims for risk assessment, but not if all outpatient and carrier claims were also used (P>0.05 in interrupted time-series models). ConclusionsThe expansion of inpatient code slots did not meaningfully affect the measurement of the risk of mortality or readmission, especially if comprehensive inpatient and outpatient claims are used, because the additional covariates only included conditions with a modest influence on risk adjusted models. The use of all versus limited codes after the code slot expansion has a minimal effect on evaluating trends in these conditions.