Case-Mix Adjustment of Adherence Based Pharmacy Quality Indicator Scores

Medication adherence has been shown to be influenced by demographics, health status and socio-economic status of the patient. Thus, adherence-based measures of pharmacy quality may be influenced by patient-related risk factors outside of the healthcare provider's control. This study examines the performance of a classical logistic regression model containing only patient characteristics and a random-effect model including patient characteristics and a pharmacy-specific effect in predicting medication adherence. These models were used to compute three different risk-adjusted scores on adherence-based pharmacy quality indicators: based on the classical logistic regression model (Method 1), the random effects model (Method 2) and the shrinkage estimators of the random-effects model (Method 3). Finally, we compared the classification as low, medium or high quality pharmacies based on unadjusted and adjusted scores. This retrospective cohort study used the 2007 Mississippi Medicare administrative claims dataset. Patient medication adherence was measured using the proportion of days covered (PDC) measure for seven therapeutic classes of medications. Pharmacy Quality scores on adherence-based measures were computed for all pharmacies serving Medicare beneficiaries in the state. The logistic regression model and the random-effect model displayed good predictive ability (c-statistic>0.7) for all therapeutic classes. The residual intra class correlation coefficient ranged from 0.008 to 0.012 indicating that although pharmacy level factors may have a significant impact, they may not be as important as patient level factors in determining adherence. Higher levels of agreement was observed between pharmacy classification based on unadjusted scores and risk-adjusted scores obtained from Methods 1 and 2 (0.5<?<0.74) with the percentage change in classification ranging from 16.3%–28.4%. Scores based on Method 3 produced fewer outliers and showed minimal agreement with unadjusted scores (0.19<?<0.35). When compared to risk-adjusted scores, unadjusted scores classified 8–12% of the low performing pharmacies as high performing and classified 20–30% of the pharmacies in the top 20% as low performers. Risk-adjusted scores produced more robust indicators of pharmacy quality than unadjusted scores. Not adequately addressing the effects of patient case-mix while measuring quality could have severe implications if these measures are used for pay for performance programs or generating quality report cards.

Name thesis:degree_name

M.S. in Pharmaceutical Science

Level thesis:degree_level

Thesis

Year dc:date.available

2011

Author dc:creator

• Dharmarajan, Sai Hurrish

Contributors dc:contributor

• Benjamin F. Banahan
• Donna West-Strum
• John P. Bentley

• Case-Mix
• Hierarchical Logistic Regression
• Logistic Regression
• Pharmacy Quality
• Risk-Adjustment
• Pharmacy and Pharmaceutical Sciences

Repository record dc:identifier

https://egrove.olemiss.edu/etd/91

OAI identifier oai:identifier

oai:egrove.olemiss.edu:etd-1090