Composite Measure Development

Composite Measures in CQL

Composite measure make use of multiple component measures to produce a combined score. In the most general case, a composite measure is akin to a continuous variable measure, where the measure observation for each population member is some combination of their measure component measure scores. However, the calculation logic involved is tedious, and a higher-level representation of the most common composite measure calculation approaches would be much simpler to work with and understand.

Based on an analysis of the example composite measures so far, the following calculation methods have been identified:

Opportunity Scoring

Opportunity scoring simply combines the scores from composite measures by combining the numerators and denominators for each component. In this approach, the measure score uses the same calculation logic as any proportion measure would, but the individual populations are the union of the component measure populations. Combining scores from component measures

Two-Step Calculation Methods (Continuous Variable Composites)

The other calculation approaches all involved a two step process where the first step involved computing the score for each patient based on some calculation method involving component measure populations, and the second step involved aggregating those scores in the same way that a continuous variable measure is defined.

Note that these methods may be applied at any cross-section of the population (i.e. by Provider, by Department, by Institution, etc.)

All-or-nothing Scoring

• Denominator of composite is the union of component denominators
• Numerator of composite is 1 if all numerators are 1, 0 otherwise

Linear Scoring

• Denominator is sum of component denominators
• Numerator is sum of component numerators

Weighted Scoring

• Denominator of composite is the union of component denominators
• Each component is assigned some percentage weight

Proposed Approach

Rather than require measure authors to write the tedium, only require the unique inputs necessary for each calculation type.

All composite measure types would need to specify the component measures. For opportunity scoring, this is all that is required, the implicit logic appropriate for the specified measure type is applied.

For methods other than opportunity scoring, an aggregation method is required, which would be specified in the same way as continuous variable aggregation methods are specified.

For weighted scoring, the percentage weight of each measure is required.

These could all be represented as Continuous Variable Measures, with tooling to automatically produce the required Measure Observation function based on the Component Measures and the selected calculation method.

In this way, composite measures could be handled by implementers with the same machinery they would already be providing for consumption of CQL-based measures.

And note also that nothing prevents the use of libraries and direct CQL to write more complex composite measures. The patterns identified here would enable simple expression of what are so far the most common patterns for composites.