Improving the measurement of self-reported medication nonadherence

Abstract

Objective: Medication nonadherence is a significant clinical problem in chronic disease management. Self-report measures have inadequate reliability and poor distributional properties. We demonstrate how two fundamental measurement issues have limited the usefulness of self-reported medication nonadherence measures and offer recommendations for improving measurement.

Study design and methods: We reviewed existing self-report measures of medication nonadherence in the context of hypertension, one of the most common chronic conditions in which medication nonadherence is a paramount concern. We evaluated these measures with regard to two issues: (1) conflation of causal indicators (which give rise to a latent construct) and effect indicators (which are determined by a latent construct), and (2) a lack of evidence regarding the stability of nonadherence over time.

Results: Nonadherence measurement could be improved by using effect indicators to assess the extent of nonadherence and causal indicators to assess reasons for nonadherence. Moreover, nonadherence should be assessed longitudinally, so that recent developments in statistical modeling can illuminate the extent to which medication nonadherence is transient vs. stable.

Conclusion: Attention to these measurement issues can improve the assessment of self-reported nonadherence, thereby allowing more accurate conclusions to be drawn about medication-taking behavior and informing the development of improved interventions that target medication nonadherence.

Fig. 1

Decision tree: measuring a construct.

Fig. 2

Path diagrams of effect indicator (A) and causal indicator (B) measurement models. e, measurement error; d, disturbance or measurement error associated with prediction.