Risk Adjusting Episodes of Care to Account for Illness Burden in Payment and Evaluation
- Presenter:
Chair: Timothy McBride; Discussant: TBA Tue June 6, 2006 10:45-12:15 Room 309
There are many instances when it is convenient to use an ‘episode of care’ (e.g., a bout of pneumonia) as the unit of analysis rather than a patient-level measure (e.g., annual health care cost). The episode approach groups health care claims data into clinically homogenous periods of care typically based on diagnoses, procedures and dates. Episodes are widely used by health plans in profiling and in payment schemes to evaluate healthcare providers and are seen as constituting the critical analytic link between process and outcome measures, potentially providing an efficient tool for measuring quality of care. One criticism of the episode approach is that the classification system is too coarse and that some diagnosis data at the patient level are often ignored. Providers and analysts alike are concerned that episodes alone may not sufficiently account for the diversity of comorbidity among patients that may influence both total cost and therapeutic treatment choices within an episode. This paper explores this issue by evaluating the effect of risk-adjusting episodes of care with two years of data from MedStat’s MarketScan® database of large commercial employers. Episodes were created using Episode Treatment Group (ETG) software from Symmetry Health Data Systems Inc., and patient-level disease burden is computed using DxCG’s Hierarchical Condition Categories (HCCs) to predict concurrent year risk.
ETGs account for severity of illness and comorbid conditions by taking note of the presence of relevant surgical procedures and comorbid complicating conditions. For tractability reasons, included comorbidities are specific to an ETG. For instance, osteoporosis is a comorbidity when the diagnosis is arthritis but not a comorbidity for congestive heart failure. This approach is intuitively appealing and helps distinguish between more and less costly patients with a particular condition. However, ETGs stop short of fully risk adjusting for illness burden of the patient. This is illustrated by comparing predicted cost without risk-adjustment for paired ETGs, one with comorbidity and the other without. As expected, predicted cost for the ETG with comorbidity is higher that for the ETG without comorbidity. However, for many ETGs, the remaining variation in actual cost was found to be high and is further explained by the overall disease burden of the patient.
We compare the predictive power of unadjusted ETGs versus risk-adjusted ETGs by regressing actual episode payment separately as a function of: (1) unadjusted episode dummies (R2=0.29); and (2) risk-adjusted episode dummies (R2=0.34). However, part of the predictive power of the ETGs comes from grouping conditions with surgery separately from conditions without surgery, thereby classifying patients based on observed treatment rather than the disease condition alone. When ETGs are pooled so that only diagnoses are used and the regressions re-run, the gap widens (R2=0.31 for risk-adjusted ETGs vs. R2=0.24 for unadjusted ETGs). These findings support the case for risk adjusting episodes of care to promote fairness in comparing provider performance at the episode level and encourage efficient allocation of resources.
