Signal Detection from Spontaneous Reports
10.3820/jjpe1996.6.101
- VernacularTitle:自発報告からのシグナル検出
- Author:
Kiyoshi KUBOTA
- Publication Type:Journal Article
- Keywords:
adverse drug reaction reporting systems;
drug monitoring;
data mining;
pharmacovigilance
- From:Japanese Journal of Pharmacoepidemiology
2001;6(2):101-108
- CountryJapan
- Language:Japanese
-
Abstract:
Objective : To outline new methods developed in Medicines Control Agency (MCA) in the UK, Food and Drug Administration (FDA) in the USA and WHO Uppsala Monitoring Centre (UMC) to detect signals from spontaneous reports on suspected drug reactions.
Methods : Presentations in the Signal Generation Symposium (Southampton, UK, June 2001) and related articles identified by hand searching were examined.
Results : All of the 3 methods compare the number or probability of reports on a particular drug-event combination with the expected number or probability for the combination. For example, in the MCA's method, the expected number is estimated as (the total number of reports on a drug) × (the fraction of an event among all spontaneous reports). A signal is detected when Proportional Reporting Ratio (PRR) defined as the ratio of observed/expected numbers>2 and the corresponding chi-square value> 4. In the FDA's method, the observed number of a drug-event combination is supposed to have a Poisson distribution with a mean of μ and the signal score is defined as the expected value of a random variable λ=μ/E where E is the expected number of reports on that combination. A signal is detected when signal score>2. The “Information Component” (IC) in the UMC's methods is estimated from the ratio of posterior to prior probabilities for a particular drug-event combination. A signal is detected when the 95% confidence interval for the IC is positive and does not include 0.
Conclusion : New methods outlined in this article require further theoretical development and its application to the analysis of spontaneous reports.
