Data Mining Methods for Signal Detection
Ram Chandra Tiwari, Jyoti N. Zalkikar, Lan Huang · 2021
Frequentist and Bayesian methods for safety surveillance can be used for analyzing large post-market safety data. The Frequentist methods include relative odds ratio (ROR), proportional reporting ratio (PRR), Chi-squared tests, and a recently developed likelihood ratio test (LRT)-based method that controls the type-I error for multiple comparisons. Commonly used Bayesian methods for signal detection from a large database discussed are: Bayesian Confidence Propagation Neural Network (BCPNN), Multi-item Gamma Poisson Shrinker (MGPS, also known as Gamma Poissor Shrinker (GPS)), New IC Method, and Simplified Bayes (sB) Method. Bayesian methods assign a common prior distribution to the reporting rates in order to smooth the reporting rates estimated from the data. The presence or absence of signals may be affected by stratification. A score exceeding the threshold indicates a strong association between the specific drug and AE.