Assessing polypharmacy in the general older population : comparison of findings from a health survey and health insurance data

Introduction Polypharmacy (i.e. the use of multiple drugs administered to the same patient) is common in the older population and an important public health problem. Usually polypharmacy measures are based on billing data, and only reimbursed medicines are considered. In Belgium, however, polypharmacy is also assessed through a systematic recording of all medicines used in the past 24 hours among respondents of the national Health Interview Survey (HIS). This method has two major advantages. First, it is based on actual use of medicines, not on the prescription of medicines. Moreover, it also takes into account non-reimbursed medicines, such as e.g. benzodiazepines. The method also has disadvantages. It is based on a population sample, hence a selection bias is possible. Furthermore, the completeness of the information depends on the willingness of respondents to show to the interviewer all medicines that were used in the past 24 hours. Finally, the method does not allow taking into account the chronicity and the volume of the use of medicines. In this study a HIS based polypharmacy indicator was compared with a similar indicator based on health insurance data. Methods Data from the Belgian HIS 2008 were linked with data from the Belgian mandatory health insurance. Only the population of 65 years and older was considered ( n  = 2722). From both data sources an indicator was created with 3 categories: no polypharmacy (0–4 medicines), polypharmacy (5–8 medicines) and excessive polypharmacy (9+ medicines). Information on the date of dispensary of the medicine, the number of packages supplied, the quantity per package and the daily defined dose were used to define an active medicine in the health insurance database on the date of the interview. To assess the selection bias an indicator was created from a completely random sample of the health insurance ( n  = 46,376). For each individual in this sample, polypharmacy was calculated at a random date in 2008 in the same way as in the linked HIS sample. Determinants of polypharmacy were explored through a generalized ordered logit model. Agreement between the HIS and the insurance-based indicator was assessed with kappa statistics and a conditional logistic regression model. Results The insurance-based estimates for polypharmacy and excessive polypharmacy were, respectively, 25% and 4.4% in the HIS sample, and 23% and 4.2% in the random health insurance sample. The HIS based method yielded estimates of respectively 25% and 8.4%. Age, suffering from a chronic disease, region, number of contacts with a general practitioner and a hospitalisation in the past year were significantly associated with polypharmacy, regardless of the measurement method. Educational attainment was only a significant determinant of the HIS based polypharmacy indicator. The weighted kappa, assessing the agreement between the indicators calculated by both methods, was 0.369 (95% confidence interval [95% CI] 0.363–0.397). The odds of classifying excessive polypharmacy and polypharmacy versus no polypharmacy was 1.38 (95% CI: 1.08–1.76) times higher for the HIS than for the insurance-based indicator. For excessive polypharmacy versus polypharmacy and no polypharmacy this was 2.56 (95% CI: 1.55–4.23) times. Conclusions The HIS based method yielded higher rates of polypharmacy in the general older population than an indicator based on billing data. This higher rate was not the result of a selection bias but could have been due to the fact that the HIS based method also takes into account non-reimbursed medicines. As a HIS based indicator considers medicines that are actually consumed, rather than those that are prescribed, it is a more relevant indicator from a public health perspective. Although the systematic collection of information on medicines that are taken during a national health survey increases the burden of the fieldwork, it is an added value for the assessment of polypharmacy in the general population.