Background: The case-population approach or population-based case-cohort approach is derived from the case-control design and consists of comparing past exposure to a given risk factor in subjects presenting a given disease or symptom (cases) with the exposure rate to this factor in the whole cohort or in the source population of cases. In the same way as the case-control approach, the case-population approach measures the disproportionality of exposure between cases of a given disease and their source population expressed in the form of an odds ratio approximating the ratio of the risks in exposed and not-exposed populations (relative risk).
Objective: The aim of this study was to (i) present the case-population principle design in a way understandable for non-statisticians; (ii) propose the easiest way of using it for pharmacovigilance purposes (mainly alerting and hypothesis testing); (iii) propose simple formulae for computing an odds ratio and its confidence interval; (iv) apply the approach to several practical and published examples; and (v) discuss its pros and cons in the context of real life.
Methods: The approach used is derived from that comparing two rates expressed as person-time denominators. It allows easy computation of an odds ratio and its confidence interval under several hypotheses. Results obtained with the case-population approach were compared with those of case-control studies published in the literature.
Results: Relevance and limits of the proposed approach are illustrated by examples taken from published pharmacoepidemiological studies. The odds ratio (OR) reported in a European case-control study on centrally acting appetite suppressants and primary pulmonary hypertension was 23.1 (95% CI 6.9, 77.7) versus 31 (95% CI 16.2, 59.2) using the case-population approach. In the European case-control studies SCAR (Severe Cutaneous Adverse Reactions) and EuroSCAR on the risk of toxic epidermal necrolysis associated with the use of medicines, the OR for cotrimoxazole was 160 and 102, respectively, versus 44.4 using the case-population approach. Similarly, these two case-control studies found ORs of 12 and 72 for carbamazepine versus 24.4 using the case-population approach, 8.7 and 16 for phenobarbital versus 21.9, 12 for piroxicam (analysed in the SCAR study only) versus 14.5, and 5.5 and 18 for allopurinol versus 3.4 using the case-population approach.
Conclusions: Being based on the estimate derived from sales statistics of the total exposure time in the source population of cases, the method can be used even when there is no information about the actual number of exposed subjects in this population. Although the case-population approach suffers from limitations stemming from its main advantage, i.e. impossibility to control possible confounders and to quantify the strength of associations due to the absence of an ad hoc control group, it is particularly useful to use in routine practice, mainly for purposes of signal generation and hypothesis testing in drug surveillance.