reports

Creating meaningful composite scores for primary care

2015 North American Primary Care Research Group Annual Meeting, October 2015, Cancun, Mexico.

Context: Composite scores (CSs) provide a global measure of quality. CS development must be grounded in theory to best ensure that the aggregated indicators achieve their intended use.

Objective: Develop CSs for use in a study assessing what practice organizational structures support quality chronic disease prevention and management (CDPM).

Design: Cross sectional analysis.

Setting/Participants: Health administrative data (HAD) of all family physicians providing general care in Ontario (Canada) and their patients.

Method: We followed the guidelines described in the OECD Handbook on Constructing Composite Indicators: A)Ground in theory B)Select indicators; C)Conduct multivariate analyses to establish which indicators underlay a common concept; D)aggregate the selected indicators; E)Validate/test. Outcome: Valid CSs derived from HAD.

Results: A)Underlying theory: quality scores of indicators that are influenced by a common organizational structure are correlated within the same organization/provider; B)Indicator selection:10 common CDPM indicators; C)Principal Component Analysis suggested that all “test” and all “medication” indicators clustered under separate components; D)Aggregation: two CSs (CStest and CSmedication) were derived as: #indicators done/#indicators eligible; E)Validation: we assessed the ability of CStest to predict hospitalizations (1.all except non-elective or obstetric, 2.ambulatory care sensitive conditions (ACSC), and 3.diabetes). After adjusting for patient co-variates, CStest showed a clear dose-response relationship with all three hospitalization types. E.g., compared to individuals in the CStest quintile 5 (scores: 81-100%), rate ratio (95% confidence intervals) for individuals in quintiles 1 through 4 were: 2.72(2.63-2.80), 2.76(2.66-2.87), 1.95(1.87-2.02), 1.57(1.52-1.63), respectively, for ACSC hospitalizations (n=7,580,232) and 2.25(1.96-2.6), 1.47(1.31-1.66), 1.17(1.00-1.39), 0.94 (0.83-1.06) for diabetes hospitalizations (n=963,191). CSmedication will be tested next.

Conclusions: Clinically meaningful CSs of CDPM processes built for the purpose of finding associations between organizational factors and quality can be derived through a theoretically grounded, methodologically guided approach. Next steps will be to assess whether these CSs successfully identify organizational factors that would drive the CDPM performance.