Adherence Analysis Using Visual Analog Scale Versus Claims-Based Estimation

Abstract

Although visual analog scales (VAS) have been used frequently in outcomes research, there is little evidence regarding the validity of this scale for measuring medication adherence. To determine whether a VAS self-report measure of medication adherence is concordant with claims-based measurement of adherence. A mail survey was conducted in 2005 of persons with diabetes. Prescription claims were obtained for the 1985 survey respondents who used oral diabetes medications and lipid-modifying drugs. The self-reported measure of adherence was a VAS scored 0-100%, and the claims-based measure was the continuous measure of medication gaps (CMG), reverse-coded to yield a score of 0-100%. Dichotomous measures (highly adherent vs poorly adherent) were also created from the VAS and CMG using a cutoff value of 80%. For diabetes and lipid-modifying drugs, the scores on the VAS and CMG (continuous versions) were compared using a Pearson correlation coefficient, while the concordance of the dichotomous versions of the measures was compared using the kappa coefficient. The mean +/- SD for the VAS and CMG for oral diabetes drugs were 95.9 +/- 9.2 and 84.1 +/- 19.2, respectively, and for lipid-modifying drugs, 95.2 +/- 11.2 and 85.3 +/- 20.0, respectively. The VAS-diabetes and CMG-diabetes scales were moderately correlated (r = 0.22), as were the VAS-lipid and CMG-lipid (r = 0.26). The majority (69.0%) of subjects had consistent adherence classifications across the dichotomous versions of VAS-diabetes and CMG-diabetes (kappa = 0.13), while 73.1% of subjects had consistent classifications for the dichotomous VAS-lipid and CMG-lipid (kappa = 0.19). The VAS self-reports of adherence to medications had moderate concordance with estimates derived from drug benefit claims. Although the majority of subjects were consistently classified by the VAS and claims, the concordance may not be sufficient for direct comparisons of studies using VAS data with studies using claims-based estimates.