Early detection of adverse drug events in the post-market phase is essential for protecting the public from significant morbidity and mortality. The broad, long-term objectives of this project are to develop tools and techniques that enable scientists to discover adverse drug events earlier and more reliably. Current drug safety approaches rely on analyses of either spontaneous reports or healthcare claims, and scientists are over-whelmed by the large amounts of disparate drug safety information. Integration is urgently needed to combine these complementary perspectives to improve adverse event discovery. The goal of this project is to develop integrated pharmacovigilance methods that combine information across multiple drugs and data sources to provide a more comprehensive view of drug safety. Distributed integration methods will allow organizations to collaborate on pharmacovigilance without exchanging private health data. Methods will be evaluated using fifty drug use cases, US and Canadian spontaneous report data, and claims data from the US's largest insurer: I. Develop multivariate network approaches to improve adverse event discovery using claims data. Current claims-based methods rely on a single pharmacoepidemiological comparison between two drugs. A pharmacoepidemiological network approach will be developed that combines multiple drug-drug comparisons to produce a unified picture of the drug safety environment, employing a sequential analysis approach to address multiple-testing over time. Detection performance will be evaluated, and will be compared to the standard single-reference-drug approach. The effects of network size and composition will also be studied. II. Integrate multiple data sources to improve adverse event discovery using spontaneous reports. Traditional disproportionality-based signal detection methods, including PRR and RRR, will be applied to the US AERS and Canada Vigilance databases. The effects of reporting volume on signal detectability will be studied using sub-sampling. Aggregative and Bayesian multi-univariate approaches will be developed to integrate the US and Canadian data, and their performance will be compared to single-data-source approaches. Spontaneous report-based methods will be compared to claims-based methods in order to investigate their relative strengths and weaknesses and characterize their temporal interrelationships. III. Develop distributed discovery methods that integrate spontaneous reports with claims data. Three distributed approaches for integrating spontaneous reports and claims data will be developed to allow scientists to collaborate on pharmacovigilance across organizations without exchanging private health data: A) Extending spontaneous-report-based signal detection methods to incorporate the findings from claims data. B) Extending claims-based signal detection methods to incorporate the findings from spontaneous reports. C) Developing dynamic Bayesian network models that exploit the temporal relationships between sources. The performance of these integration approaches will be compared to single-data-source approaches.

Public Health Relevance

Some drugs that are approved for sale to the public may have dangerous unknown side-effects. It is important to detect these unknown side effects as soon as possible in order to prevent serious illness or death. This project will help protect the public health by improving the ability to detect unknown dangerous drug side effects earlier and more reliably.

National Institute of Health (NIH)
National Institute of General Medical Sciences (NIGMS)
Research Project (R01)
Project #
Application #
Study Section
Biomedical Computing and Health Informatics Study Section (BCHI)
Program Officer
Lyster, Peter
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
Children's Hospital Boston
United States
Zip Code
Cami, Aurel; Reis, Ben Y (2014) Concordance and predictive value of two adverse drug event data sets. BMC Med Inform Decis Mak 14:74
Cami, Aurel; Manzi, Shannon; Arnold, Alana et al. (2013) Pharmacointeraction network models predict unknown drug-drug interactions. PLoS One 8:e61468
Reis, Ben Y; Olson, Karen L; Tian, Lu et al. (2012) A pharmacoepidemiological network model for drug safety surveillance: statins and rhabdomyolysis. Drug Saf 35:395-406
Cami, Aurel; Arnold, Alana; Manzi, Shannon et al. (2011) Predicting adverse drug events using pharmacological network models. Sci Transl Med 3:114ra127