Risk adjustment promotes fair and accurate comparison of health care outcomes across measured entities. Risk stratification allows for comparison of health care performance within peer groups of measured entities rather than across all measured entities. This content offers insight into risk adjustment and risk stratification models and supplements the information found on other Measure Specification sections.

The purpose of risk adjustment and risk stratification is to deconstruct the measured entity-level variation into factors that are and are not correlated with (meaning, are independent of) the quality construct. Risk adjustment refers to the inclusion of risk factors associated with a measure score in a statistical model of measured entity performance captured at the person, facility, community, or other levels. Risk stratification groups patients or resource services with similar characteristics and then calculates multiple performance scores for the measured entity. Measure developers often risk adjust and/or stratify outcome measures and cost and resource use measures, however not all outcome measures need risk adjustment or risk stratification. 

Risk adjustment at the person level, also referred to as case-mix adjustment, aims to increase the likelihood of fair comparison of measured entity performance, which is to compare apples with apples.  [cite] It involves controlling for confounding factors -- meaning systematic differences within the population of interest -- in the modeling of measured entity performance. Confounding factors may be clinical (e.g., types, number, or severity of conditions), demographic (e.g., age, sex), functional (e.g., dementia) and/or social (e.g., income, education, geography) in nature.

Taking confounding factors into account could prevent the model from incorrect specification or the estimates of performance scores from bias. The variation in measured entity-level (e.g., clinician or facility) performance may be due to variation in quality or variation in factors independent of quality (e.g., factors like the age or severity of illness of patients). Independent of quality means the clinician treats the patients the same, but patients who have the factor (older or sicker) have worse outcomes than patients who do not (younger or less sick). In such a circumstance, selecting one clinician over another based on a quality measure not accounting for these independent factors would not result in improved outcomes for the population. Risk adjustment attempts to solve that problem and increase the likelihood of selecting a clinician or facility based on performance results in improved outcomes for the population.

Considering confounding factors becomes even more important with use of performance scores as a basis for calculating the amount of incentives or penalties for value-based purchasing and many Alternative Payment Models (APMs).