Over the past few years Azara has come across a large variety of clinical quality measures (CQMs). As with many things in the complicated health IT landscape, the devil is in the details. Even the simplest CQMs have gone through years of debate on how their different components should actually work. The point of this post is to provide a rough overview of the "anatomy" of a CQM – a foundation for future posts in which we'll take a deeper dive into some of these measure concepts and hopefully reduce some of the confusion around CQMs.
The first –and most obvious– component of a measure is the measure logic. This is the "Boolean logic" that determines the measure numerator and denominator. Most specifications will define the measure logic as looking something like this:
Denominator: All patients who meet the following criteria:
o Face to face visit during the measurement period
o Age at the end of the measurement period >= 19 and <= 75
o Active diabetes diagnosis during the measurement period
Numerator: All patients in the denominator who meet the following criteria:
o Most recent A1c lab result in the measurement period < 8%
It’s vitally important that measure logic has as little ambiguity as possible. Simply saying "percentage of diabetic patients age 18-75 with A1c < 8" raises several questions, such as: What if they haven't been seen in 3 years? What is the lookback period for the A1c test? At what time should we assess the patient's age? A good spec leaves very little up to interpretation.
Even with an extremely detailed measure logic breakdown, we still run into issues with how to define specific data elements, including: What do you actually mean by "diabetes"? Whose definition is it? Fortunately, the healthcare industry has widespread adoption of a number of standard terminologies, such as ICD-9-CM/ICD-10-CM for diagnoses, CPT for various services, etc. This means that we can define data elements by using what we call a value set – a list of standard codes used to define a clinical concept. For example, a diabetes value set may include ICD-9-CM codes 250.00, 250.01 (and so on), as well as a number of ICD-10-CM and SNOMED-CT codes. A good spec has a discrete list of codes for all data elements. This leaves very little up to interpretation and allows us to calculate measures on anyone's data if they’re using the standard codes. This is where "structured data" becomes really important: if your data is stored using the same codes as the CQM value sets, the data mapping process is much easier.
The last – and least obvious – component of a CQM is called attribution. CQMs really shine for quality improvement when specific providers, locations, etc., can run them. This is also the area where specifications provide the least clarification, which means we at Azara are often tasked with determining which logic should be used when placing patients into various categories or buckets. This is where concepts such as rendering provider vs. usual provider, service line, etc., come into play. For example, what if you provide certain specialty services for patients who are not also seeing you for primary care - should they be included in your diabetes measures? Attribution facilitates the interesting "slicing and dicing" analysis of a standardized measure.
In future posts we will consider clinical quality measures in the context of measure logic, value sets, and attribution. As I mentioned, the devil is in the details, and there is no shortage of edge cases and gotchas for us to explore. If you have a specific question that you would like us to tackle, please email us at Info@azarahealthcare.com.
Eric Gunther is an engineer at Azara Healthcare.