Measuring and prediction mediction adherence using dispensing data and patient beliefs

H.C.J. Geers

Research output: ThesisDoctoral thesis 1 (Research UU / Graduation UU)


Introduction Medication adherence can be subdivided into drug initiation, drug taking compliance and (non)persistence. Objective of this thesis was to (1) investigate whether measurement methods and reporting of adherence influenced outcomes, (2) predict poor drug taking compliance and nonpersistence from historical dispensing data and patient beliefs. Methods To answer objective (1) a systematic review and meta-(regression)-analysis was done to investigate whether reporting of adherence, measurement methods and different definitions and cutoff values influenced adherence. In a retrospective cohort study we investigated the influence of four exposure outcome combinations on persistence. To answer objective (2) we undertook two database studies investigating whether historical dispensing data and therapeutic complexity predicted poor drug taking compliance. Second, we performed a prospective cohort study, investigating patient beliefs, using the Beliefs about Medicines Questionnaire (BMQ) and Satisfaction on Information about Medicines Scale (SIMS). These questionnaires were sent at the start of therapy and after one month. We used the answers to individual questions to construct a predictive model for nonpersistence. Results Both drug taking compliance and persistence were better if studies reported both outcomes, compared with reporting only drug taking compliance or (non)persistence. Self reported adherence was better than assessment of adherence using databases. Different definitions and cutoff values exist which are very likely to influence adherence outcomes. Accounting for supplies from previous dispensings was associated with better persistence, but differences seized to exist at gap lengths (period without medication available) of 90 days or longer. Long-term poor drug taking compliance was predicted by short-term poor drug taking compliance using a predictive model. The model increased efficiency in screening patients for poor drug taking compliance (Number Needed to Screen: 5). Nonpersistence could not be predicted from therapeutic complexity, although different predictors in multivariable analysis were statistically significant. We found early changes in attitudes towards chronic medication based on the BMQ-specific questionnaire. Compared to accepting patients (high necessity and low concern scores), skeptical patients (low necessity and high concerns scores) showed an almost seven times increased risk for nonpersistence after one month. A predictive model for nonpersistence was developed, using individual questions from the BMQ and SIMS with a high negative predictive value. Conclusions Adherence should be reported as both poor drug taking compliance and nonpersistence to obtain more reliable and comparable data. It is recommended that definitions for drug taking compliance should be standardized and preferentially both MPR and PDC to increase comparability with other studies. Gaps should be sufficiently long and cutoff values to classify poor drug taking compliance should be based on forgiveness of the drug. If short gaps are of clinical importance, supplies from previous dispensings should be accounted for. Historical dispensing data are probably not useful for the prediction of poor drug taking compliance and nonpersistence. We recommend that patient beliefs should be used and discussed with the patient during the early course of therapy, because during this period, the patient does a cost-benefit analysis and an intervention to increase adherence is expected to be most beneficial during this period
Original languageEnglish
QualificationDoctor of Philosophy
Awarding Institution
  • Utrecht University
  • Bouvy, Marcel, Primary supervisor
  • Heerdink, Rob, Co-supervisor
Award date1 Feb 2012
Print ISBNs978-90-818459-0-8
Publication statusPublished - 1 Feb 2012


Dive into the research topics of 'Measuring and prediction mediction adherence using dispensing data and patient beliefs'. Together they form a unique fingerprint.

Cite this