The elderly, the young, and the ill: How different databases shape epidemiological research questions

Pharmacoepidemiology can be defined as the science of the use and the effect of drugs in large human populations. Although originally limited to post-marketing drug surveillance of rare or long-latency adverse drug events, the relatively young science emerged to be an invaluable tool across different stages of drug development, including the assessment of drug utilisation and cost-effectiveness, the characterisation of target populations of drugs in development, the evaluation of undiscovered beneficial or detrimental drug effects, or the provision of evidence of effectiveness when randomised controlled trials are not feasible due to ethical or practical barriers. The possibilities of conducting pharmacoepidemiological studies have substantially increased with the unprecedented emergence of large health care databases over the past decades. Available databases can vastly be divided into administrative databases (data are collected primarily for administrative purposes, such as reimbursement of health care costs) and medical record databases (data are collected by physicians, usually primary care providers, who enter information on patients while providing medical care). A peculiarity of most epidemiological database research is that the number of accessible databases within an organisation, a research group, or a company is usually restricted, i.e., a study’s underlying database is predetermined and precedes the research question. As the available databases vary in the type of questions they can optimally answer, their strengths, but especially their inherent limitations inevitably need to be considered when establishing a fit-for-purpose research question.
This thesis aimed to contribute to the general understanding of how different databases shape epidemiological research questions. To do so, three exemplary studies encountering different research questions, study populations (which are reflected in the title of each study), and study designs were conducted within the administrative Helsana claims database ('the elderly' and 'the young') and the Clinical Practice Research Datalink (CPRD; 'the ill').
In 'the elderly', we comprehensively assessed drug utilisation in Swiss nursing homes in 2016 using the Helsana claims database. In particular, the aim of this study was to examine the prevalence of polypharmacy and use of potentially inappropriate medication (PIM) in individuals aged ≥65 years, who were either nursing home residents (NHR) or living in the community (reference group, RG). Compared with the RG, polypharmacy was more prevalent in NHR (86%; RG, 50%). On top, more than half of all NHR (56%) were long-term users of at least one PIM. Although the number of different drugs taken concurrently was high in NHR (9.4; RG, 5.5), it decreased with increasing age, suggesting a de-prescribing of drugs with an unfavourable benefit-risk profile. Drug costs seemed to constitute a minor part of total health care costs of patients in nursing homes, as per capita drug costs per day for NHR were relatively low with CHF 8.55. The high prevalence of polypharmacy and of long-term PIM use may indicate the need for effective interventions to optimise drug prescribing in Swiss nursing homes.
In 'the young', we shed light on coverage rates and timeliness of nationally recommended vaccinations in Swiss preschool children by investigating cohorts of children born between January 2010 and December 2016 using the Helsana claims database. Time trend analysis revealed a continuously increasing proportion of children with up-to-date vaccinations for all vaccinations under study. Nevertheless, national vaccination goals have not been reached. On average, only seven out of ten children had an up-to-date status for completed basic vaccinations; even less (six out of ten) were up-to-date for recommended supplementary vaccinations at 37 months of age. Additionally, delay of vaccine administration was common, particularly for measles, mumps, and rubella combination vaccines, where 23% of children vaccinated with the first dose experienced delays relative to age-appropriate standards. Every twenty-fifth child did not receive any vaccination under study up to 37 months of age (substantially varying according to canton of residence), leaving this vulnerable population unprotected against vaccine-preventable diseases, and calling for effective public health measures to optimise vaccination coverage in Switzerland.
'The ill' aimed at assessing the risk of melanoma and non-melanoma skin cancer in new users of thiazides and thiazide-like diuretics (TZs) such as hydrochlorothiazide (HCTZ), indapamide, and bendroflumethiazide (BFT) compared to two other first-line antihypertensive drug groups. Applying a cohort study design that adhered to key principles of pharmacoepidemiological methods, we were able to quantify absolute and relative risks of cutaneous malignant melanoma (CMM), basal cell carcinoma (BCC), and squamous cell carcinoma (SCC) using the CPRD (1998-2017). Long-term use (≥20 prescriptions) of HCTZ was associated with a 1.95-fold increased risk of SCC (number needed to harm, 1 144 patients per year) when compared with long-term use of calcium channel blockers. While use of BFT appears to be safer than other TZs with regard to skin cancer risk, long-term use of indapamide was associated with a 1.43-fold increased risk of CMM. Future studies are warranted to corroborate this finding, especially given the severe clinical implications and high mortality of CMM when diagnosed at advanced stages.
In summary, the studies presented in this thesis contribute to the understanding of when and how different databases can be used to answer particular types of pharmacoepidemiological research questions. While administrative databases such as the Helsana claims database are considered the gold standard regarding drug exposure information and are thus ideally suited for descriptive research questions concerning drug utilisation, medical record databases such as the CPRD contain more detailed information on diagnoses and potential confounders, and are therefore better equipped to address analytical research questions seeking causal associations. Finally, and perhaps most crucially, the database only provides the raw materials to address a research question, but does nothing to ensure a study is appropriately designed and conducted. It is the interplay of the accessible database, a suitable study question, and a high-quality study design and analysis plan that determines whether a study can successfully reproduce the truth. This thesis is an important reminder that rigorous observational research is not only possible, it is essential to further improve clinical practice and health policy.