Automatic Detection of Vaccine Adverse Reactions by Incorporating Historical Medical Conditions

Zhonghua Jiang and George Karypis
ACM Conference on Bioinformatics, Computational Biology and Biomedicine. Chicago, 2011
Download Paper
This paper extends the state of art by bringing the historical medical conditions into the vaccine adverse reaction discovery process. The goal is to identify evidences which suggest that given adverse reaction is likely to be developed for individuals with certain medical conditions when they are vaccinated with certain vaccines. We propose a novel measure called dual-lift for this task. It is shown that the dual-lift measure can not only identify medical conditions associated with known vaccine adverse reactions, but also have the potential of detecting new adverse reactions that are otherwise hidden. We formulate this problem in the framework of constraint pattern mining. Three constraints are developed. The rst is based on the dual-lift measure which aims to discover meaningful patterns, the second is used to remove redundancy, and the third is used to ensure prevalence of generated patterns. We present a pattern mining algorithm DLiftMiner which utilizes a novel approach to upper bound the dual-lift measure for reducing the search space. Experimental results show that our pruning methods lead to dramatic performance improvement. It is also shown that DLiftMiner scales linearly with the size of input database. Some interesting vaccine adverse reactions discovered from VAERS database are presented.
Research topics: Adverse reactions | Data mining | Health care | Information retrieval