Use of principal component analysis in the evaluation of adherence to statin treatment: a method to determine a potential target population for public health intervention

The prevalence of statin use is high but adherence low. For public health intervention to be rational, subpopulations of nonadherent subjects must be defined. To categorise statin users with respect to patterns of reimbursement, this study was performed using the main French health reimbursement database for the Aquitaine region of south‐western France. The cohort included subjects who submitted a reimbursement for at least one delivery of a statin (index) during the inclusion period (1st of September 2004–31st of December 2004). Indicators of adherence from reimbursement data were considered for principal component analysis. The 119 570 subjects included and analysed had a sex ratio of 1.1, mean (SD) age of 65.9 (11.9), and 13% were considered incident statin users. Principal component analysis found three dimensions that explained 67% of the variance. Using a K‐means classification combined with a hierarchical ascendant classification, six groups were characterised. One group was considered nonadherent (10% of study population) and one group least adherent (1%). This novel application of principal component analysis identified groups that may be potential targets for intervention. The least adherent group appears to be one of the most appropriate because of both its relatively small size for case review with prescribing physicians and its very poor adherence.