(1236) Pharmacists' Impact on Quality Measure Achievement in a Multidisciplinary Primary Care Model

Objectives: To assess the impact of integrating pharmacists in a community health system-based multidisciplinary primary care model to achieve HEDIS and CMS quality measures in patients with diabetes. Additionally, to employ machine learning algorithms for predictive risk modeling, an underutilized approach to assess characteristics associated with quality measure achievement. In randomized clinical trials, pharmacist interventions have improved hemoglobin A1c (HbA1c) and blood pressure (BP) control. However, limited research has been done in formally evaluating pharmacists’ influence on the improvement of primary care quality measures in comparison to usual care.

Methods: This retrospective, observational cohort study used electronic health record data from a community health system in Virginia. Patients with a diagnosis of diabetes and an ambulatory visit in 2019 were included. The study had two conditions: an intervention group (IG) that received care from clinical pharmacists (e.g., medication management, access, and education), and a usual care comparator group (CG). Propensity score analysis was employed to match patients on demographic, insurance, and baseline HbA1c values to control for selection bias. HEDIS and CMS quality measures assessed at the end of the calendar year included: HbA1c control ( < 9%, < 8%, < 7%), HbA1c poor control (>9%), and BP control ( < 140/90 mmHg). Differences in quality measure achievement between the IG and CG were assessed using a generalized linear mixed model with a clinic-specific random effect. Advanced machine learning algorithms, specifically Random Forest and XGBoost, predicted the patient characteristics associated with meeting the quality measures. Statistical analyses were performed in SAS and the machine learning algorithms were implemented in R.

Results: Following 1:2 propensity score matching, a total of 1,263 patients with diabetes were included in the study (n=431 [IG]; n=832 [CG]). The mean age was 57.5 (SE = 0.65) and 57.0 (SE = 0.47) in the IG and CG, respectively. Both groups were predominantly female (IG: 55.0%; CG: 56.2%) and featured balanced racial distributions, with approximately 45% White and 46% Black patients. Baseline HbA1c was similar between groups (IG: 9.9%, SE = 0.09; CG=9.7%, SE = 0.8). Patients in the IG had higher odds of achieving HbA1c control relative to patients in the CG. Specifically, the odds ratios (ORs) were 4.2 (95% CI: 2.46, 7.06), 3.8 (95% CI: 2.15, 6.67), and 4.9 (95% CI: 2.54, 9.29) for HbA1c targets of < 9%, < 8%, and < 7%, respectively. Additionally, patients in the IG were 0.22 times less likely to have poor HbA1c control (95% CI: 0.13, 0.37) and were 1.5 times more likely to achieve BP control (95% CI: 1.03, 2.27). However, late-year initiation of care (OR = 0.17, 95% CI: 0.10, 0.30), higher baseline HbA1c levels (OR = 0.54, 95% CI: 0.49, 0.60), and the total number of chronic medications (OR = 0.95, 95% CI: 0.92, 0.98), were negative predictors of quality measure achievement for HbA1c. Utilizing machine learning to predict quality measure achievement, XGBoost displayed a superior predictive power (AUC = 0.85) compared to logistic regression (AUC = 0.84) and Random Forest (AUC = 0.78). Key predictive factors included: baseline HbA1c values, study group allocation, and late-year initiation of care.

Conclusions: The incorporation of pharmacists into primary care resulted in higher attainment of quality measures for HbA1c and BP in patients with diabetes. Machine learning algorithms, particularly XGBoost, corroborated these findings. The models emphasized the relevance of baseline HbA1c levels and the timing of initial patient care as substantial influencers of HbA1c control. These results underscore the importance of an early, proactive, and integrated approach to diabetes and blood pressure management. Including pharmacists as part of the multidisciplinary team is a promising strategy that demonstrates benefits in improving population health outcomes.