Drug exposure assessment in pharmacoepidemiological database studies: Reporting and impact of exposure misclassification

Abstract

Health care databases are often used in pharmacoepidemiological studies to determine patient drug use. These data can include drug dispensings from the pharmacy or drug prescriptions by doctors. These data do not always reflect the actual use of a medicine by patients, as patients can decide to not use a drug or use the drug differently than what is registered in these databases. When these two do not align, it is called exposure misclassification. In this thesis, we evaluated how researchers report on assessing drug use in databases, the extent of exposure misclassification and the impact thereof on when studying the association between a drug and (un)intended effects. First, we found that reporting of how researchers assess drug use was insufficient in most studies. We also saw that exposure misclassification and other suboptimal methodological choices when designing observational studies may have an impact on the study conclusions. This was for example shown for studies on benefits of hydroxychloroquine for COVID19 patients. In addition, we looked at how using different databases might lead to misclassification of drug exposure assessment. We found, for example, that about 10% of the prescriptions prescribed by GPs are not collected from the pharmacy. Patients’ drug use is thus overestimated when prescription databases are being used. We also showed that misclassification can occur when a database contains prescriptions from a single type of doctor (e.g., general practitioners), while drugs can also be prescribed by different doctors (e.g., hospital specialists). And lastly, we show that it is important to take into account how long drugs remain in the body when measuring their effects. For amiodarone, the impact of not accounting for this on the effect estimates was however limited. To conclude, measuring use of drugs, or assessment of drug exposure, in pharmacoepidemiological studies and reporting thereof could be improved, thereby contributing to more reliable information on effective and safe use of drugs.