Module 3 - Strategic case studies in practice

The suitability of the study population should be considered carefully. Study designs should focus on patient groups that differ – as much as possible - only in the contamination status of the examined medicinal product. As cancer occurrence is rare among younger adults, inclusion of mainly young adults may lead to studies that require an unrealistic sample size, especially in the case of uncommon cancers. For example, a cohort study would have to observe about 250,000 patient years per group to exclude a 2-fold increased risk of liver cancer in adult patients (20 to 85+ years of age) with a power of 80% (significance level 5%) if the two groups to be compared are equally represented. However, raising the lower age limit to 40 leads to a reduction in the number of patient years required by about 70,000 per group. The total sample size would even increase if the relation between both group sizes is unbalanced. This example is based on incidence data from the German Centre for Cancer Registry Data (ZfKD) for the calendar year 2014 11 . In the case of rare events, case control studies might be a suitable alternative to cohort studies. Apart from the patient population to be included, a study should cover a sufficiently long observation time to enable the assessment of both, early and late cancer risks. Different lag-times should be considered in the analyses, as it is unlikely that very recent N- nitrosamines exposure affects an individual’s risk of receiving a cancer diagnosis. Statements about required study periods are difficult, but a study period of at least 10 years is recommended in order to adequately address the risk of cancer. As long-term follow-up may increase imbalances between treated and untreated patients in cohort studies, risk factors for cancer that change over time – such as age – should be considered as time-dependent variables in the analyses to reduce time-varying confounding. The observation of large populations over a long time period is challenging and may lead to biased results in case of excessive loss to follow-up. Therefore, studies using routinely collected data over time including larger sample sizes are expected to be more promising than studies with primary data collection. Appropriate data sources may be, for example, nationwide registries or large healthcare databases. Nevertheless, it should be considered that routinely collected data were not designed to answer the study question at hand. Depending on the initial purpose of the data sources, information on relevant variables may be missing. For example, administrative claims data are routinely collected for billing purposes and drugs that are not reimbursed by insurance companies or purchased without a prescription (over the counter, OTC) are rarely covered in administrative databases. To assess the risk of cancer associated with the use of OTC drugs is therefore challenging or even impossible. Furthermore, information on cancer risk factors may be limited or lacking in healthcare databases, such as nutrition, smoking, radiation exposure, alcohol consumption, obesity, the socioeconomic status or family history of cancer. One of the most serious events experienced by cancer survivors is cancer recurrence or the diagnosis of a second cancer. If possible, patients with records of previous cancer should be excluded or stratified analyses should be considered. However, information on previous cancer may be missing in several data sources. Especially, if the primary cancer has occurred long ago, such as childhood cancer. Despite the lack of information on individual cancer risk factors in healthcare databases, it should be considered that most cancer risk factors are unlikely to be associated with the contamination status of the drug of interest. Therefore, the chance for confounding is considered to be low in studies observing a uniform patient population, such as valsartan users only. However, if exposure to different types of drugs is compared in a study, confounding is more likely to be present, as the respective target populations may differ in their cancer risk factors. The definition of exposure displays a major problem in the conduct of future studies, especially if the onset, the extent or the cause of the contamination is unclear or cannot be precisely determined. Non- exposed subjects may be categorized as exposed and/or patients treated with contaminated

11 taken from: https://www.krebsdaten.de/Krebs/EN/Database/databasequery_step1_node.html, accessed on 18/11/2019)

EMA/369136/2020

Page 57/90