Published in:
Open Access
01-07-2017 | COMMENTARY
Epidemiology and ‘big data’
Author:
Bruno H. Stricker
Published in:
European Journal of Epidemiology
|
Issue 7/2017
Login to get access
Excerpt
In the past 70 years, epidemiology has gone through a substantial development of both methodology and more sophisticated analyses with determinant- and disease-specific elements. Despite the discussion earlier in this journal as to whether the duality of epidemiology’s embrace of both research methods and content-based topics is a positive development [
1,
2], it shows at least that epidemiology as a discipline is able to re-invent itself and to adapt to the rapid scientific evolution which currently takes place. Whereas the methodological ‘toolkit’ keeps its solid basis in cohort- and case-control designs, determinant-oriented genetic epidemiology and pharmacoepidemiology, and disease-oriented cardiovascular and infectious diseases epidemiology are examples of branching within one discipline. While the start of epidemiology had its main focus on hypothesis-testing concerning the association between determinants and disease as well as risk estimation, the importance of hypothesis-generating studies is increasing. A nice example is the technique of genome-wide association studies which led to an explosion of discoveries in genetic epidemiology, also by a new model of collaborative science [
3]. However, genes remain while drugs come and disappear again from human society. This makes pharmacoepidemiology an important hypothesis-generating and -testing vacuum cleaner of the pharmaceutical market. Elsewhere in this journal, Chia-Cheng Lai et al. [
4] reviewed ‘sequence symmetry analysis’ as a technique for detecting adverse drug events by utilizing computerized claims data. Data mining techniques such as this one are considered of increasing importance. There are two reasons for that development. First, the enormous expansion of information technology in the past 30 years facilitated the easy access to huge amounts of healthcare information. In the stone age of pharmacovigilance and pharmacoepidemiology, a new adverse reaction not discovered in clinical trials was usually reported in the first years of marketing by intelligent medical observers with an open eye for new and unexpected events in their patients [
5]. The more relevant of these reports were actively published in medical journals as ‘short reports’ or ‘letters to the editor’. The most serious ones were occasionally investigated in relatively small hospital-based cohort studies or population-based case-control studies with the aim to confirm signals from adverse event reports and calculate the magnitude of a risk increase. Such studies were expensive and time-consuming because data had to be collected and collated
de novo. Nowadays, expensive multi-center studies on adverse reactions to drugs such as the International Agranulocytosis and Aplastic Anemia Study [
6] and the International Primary Pulmonary Hypertension Study Group [
7] would probably no longer be performed in that way. The more and more easy access to healthcare information in databases in most developed countries in the world is too efficient to omit. Second, there is the growing risk aversion in Western countries and fear for litigation. Consequently, increasingly strict legislation for obtaining a marketing authorization by regulatory authorities such as the FDA and the European Medicines Agency EMA have forced the pharmaceutical industry to demonstrate their active surveillance of literature, and signal generating activities. How far these requirements go, can be read in EMA guidelines for signal management [
8]. Also, jurisdiction regarding adverse reactions in the past, especially in the U.S.A., demonstrates the important consequences of inadequate adverse reaction management [
9]. …