The measure developer begins construction of measure specifications by outlining the target/initial population, numerator, denominator, denominator exclusions, numerator exclusions, denominator exceptions, and measure logic. Then, the measure developer gives the measure concept increasing amounts of detail, including precisely defined data elements and the appropriate values or value sets. Every part of the measure specification requires explicitly defined elements with accompanying analysis to identify constraints and criteria of the specification. Additional considerations for both numerator and denominator include alignment with other measures conceptually and technically.
Use Positive Evidence
Inquiries for measure specifications should be based on the principle of positive evidence, defined as data used to confirm a given criterion was met. This principle is particularly relevant when there are no data or there are conflicting data. Where, for instance, a numerator criterion is “low density lipoprotein (LDL) cholesterol is less than 100” and there is no LDL cholesterol result in the patient record, then there is no positive evidence, and the criterion is not met. When, for instance, a denominator criterion is “ejection fraction is less than 40%” and there is both an ejection fraction of less than 40% and an ejection fraction of greater than 40% in the same patient’s record, then because there is positive evidence of an ejection fraction less than 40%, the criterion is met.
The measure developer should consider the attribution model early in development. The attribution model can affect which patients to include in the population addressed by a value-based purchasing program or included in the denominator of a quality measure. The CMS consensus-based entity offers these considerations for attribution approaches:
- Is the attribution model for the new measure evidence-based?
- To what degree can the new accountable unit influence the outcomes?
- Are there multiple units to which the attribution model will be applied?
- What are the potential consequences?
- What are the qualifying events for attribution, and do those qualifying events accurately assign care to the right accountable unit?
- What are the details of the algorithm used to assign responsibility?
- Did the measure developer consider multiple methodologies for reliability?