Geography and patient history in long-term lipid lowering medication adherence for primary prevention of cardiovascular disease. (June 2019)

• Record Type: Journal Article
• Authors: Hu, Feiyu; Warren, Jim; Exeter, Daniel J.
• Is Part Of: Spatial and spatio-temporal epidemiology. Volume 29(2019)
• Page Start: 13
• Page End: 29
• Publication Date: 2019-06
• Subjects: Medication adherence -- Spatial analysis -- Refill history -- Prediction models; Epidemiology -- Statistical methods -- Periodicals; Epidemiology -- Periodicals; 614.4072
• DOI: 10.1016/j.sste.2018.12.001
• Languages: English
• ISSNs: 1877-5845

Abstract

Highlights: Significant variation by age, ethnicity and being a new therapy user were present in the unadjusted and adjusted models. A large, contiguous 'cold spot' with low medication adherence was detected in South Auckland. Prediction was improved by past refill history but not by geographical variables for continuing lipid-lowering therapy users. BRT provided more accurate adherence prediction than logistic regression. Abstract: In order to determine the role of geographical and patient history factors in long-term medication adherence in cardiovascular disease (CVD), we analysed adherence to lipid-lowering therapy in a primary care cohort based on CVD decision support and linked health systems and census data from Auckland, New Zealand. Two-year adherence was examined for 10, 410 patients aged between 30 and 74 with neither diabetes nor a history of CVD. Using logistic regression we found significant variation in adherence by age, ethnicity and being a new therapy user, and in 9 of 86 geographic zones. A large low-adherence 'cold-spot' of 13 contiguous geographic zones was detected through local Getis-Ord Gi* analysis. A set of 42 models to predict adherence was formulated on sets of demographic, geographic and refill history factors. We observed prediction ability to be improved by addition of refill history but not geographical variables, and boosted regression tree (BRT) models outperformed logistic regression.