DIMACS Working Group on Adverse Event/Disease Reporting, Surveillance, and Analysis

October 16 - 18, 2002
DIMACS Center, CoRE Building, Rutgers University

Donald Hoover, Rutgers, Statistics, drhoover@stat.rutgers.edu
David Madigan, Rutgers, Statistics, madigan@stat.rutgers.edu
Henry Rolka, (CDC), hrr2@cdc.gov
Presented under the auspices of the of the Special Focus on Computational and Mathematical Epidemiology. Co-sponsored by the American Statistical Association, Section on Statistics in Epidemiology.

DIMACS Subgroup on Adverse Event/Disease Reporting, Surveillance, and Analysis

DIMACS Working Group on Adverse Event/Disease Reporting, Surveillance, and Analysis II

Disease or event reporting and surveillance systems represent a primary epidemiological data source for the study of/alert to adverse reactions to medication, emerging diseases, or bioterrorist attacks. These systems synthesize data from millions of reports. The working group will bring together pharmacoepidemiologists, statisticians and computer scientists to investigate current major issues confronting adverse event/disease reporting, surveillance, and analysis. We describe the issues for drug reaction reports; those for disease or symptom reports are similar though they raise their own set of issues as well. In the US, the two major reporting/surveillance systems are AERS and VAERS. AERS, the Adverse Event Reporting System (http://www.fda.gov/cder/aers/) is a data base of drug adverse reactions reported by health professionals and others. AERS is administered by the Food and Drug Administration (FDA). The system contains adverse reactions detected and reported after marketing of a drug for a specified time period. AERS contains over two million cases. VAERS, the Vaccine Adverse Event Reporting System (http://www.vaers.org), is a Cooperative Program for Vaccine Safety of the FDA and the Centers for Disease Control and Prevention (CDC). VAERS is a post-marketing safety surveillance program, collecting information about adverse events that occur after the administration of US licensed vaccines. Reports are provided by all concerned individuals: patients, parents, health care providers, pharmacists, and vaccine manufacturers. The VAERS database is publicly available and contains over 100,000 reports. Analyses of AERS and VAERS data must confront several difficulties including adverse event recognition, underreporting, biases, estimation of population exposure, report quality, and, most importantly, no denominator or control group of persons not taking the medication. In many cases it is difficult to discern whether or not a reported adverse reaction was from the medication or instead was a consequence of the underlying conditions that necessitated the medication (see, for example, [Koch-Weser, Sellers, and Zacest (1977), Rawlins (1994), Strom and Tugwell (1990)]). Several methodological issues relating to these reporting mechanisms and others emphasizing disease/symptom reporting will form the agenda for this working group. These include application of computational and statistical methods for early detection of emerging trends; modification of algorithms of streaming data analysis designed to set off early warning alarms; application of data mining methods; development of causal inferential methods in the absence of controls; study of ways to eliminate bias; design of verification methodology. This last issue is especially pressing since large-scale medical record databases that now exist in certain sub-populations (e.g., HMOs, military) can provide a basis both for assessing the quality of AERS and VAERS data and for validating analyses. Still another set of research issues for the working group arises from the use of natural language in reporting systems: Devise effective methods for translating natural language input into formats suitable for statistical analysis (prior work on machine natural language processing and information retrieval is relevant); develop computationally efficient methods to provide automated responses consisting of follow-up questions; develop semi-automatic systems to generate queries based on dynamically changing data, indicating developing epidemiological trends. Relevant to these questions is work in [VanLehn and Niu (to appear)] on interpreting natural language reports based on probabilistic models of context, work in [Langlotz, Shortliffe and Fagan (1986), McConachy, Korb and Zuckerman (1998), Zuckerman, McConachy and Korb (1998)] on communicating uncertain information and summarizing rough trends, and work in [Horvitz and Paek (1999), Horvitz and Paek (2001), Walker (2000)] on decision-theoretic methods for asking followup questions in natural language processing. Earlier work on electronic surveillance reporting from public health reference laboratories is very relevant here (e.g., [Bean and Martin (2001), Bean, Martin and Bradford (1992), Hutwagner, Maloney, Bean, Slutsker and Martin (1997), Mahon, Rohn, Pack and Tauxe (1995)]). Subgroups might be formed to concentrate on drug reactions, emerging diseases, or bioterrorism.


Bean, N.H., and Martin, M. (2001), "Implementing a network for electronic surveillance reporting from public health reference laboratories: An international perspective," Emerging Infectious Diseases, 7, http://www.cdc.gov/ncidod/EID/vol7no5/bean.htm

Bean, N.H., Martin, M., and Bradford, H. (1992), "PHLIS: An electronic system for reporting public health data from remote sites," Am. J. Public Health, 82, 1273-1276.

Horvitz, E. and Paek, T. (1999), "A computational architecture for conversation," User Modeling Conference, 201-210.

Horvitz, E. and Paek, T. (2001), "Harnessing models of users' goals to mediate clarification dialog in spoken language systems," User Modeling Conference.

Hutwagner, L.C., Maloney, E.K., Bean, N.H., Slutsker, L., and Martin, S.M. (1997), "Using laboratory-based surveillance data for prevention: An algorithm for detecting salmonella outbreaks," Emerging Infectious Diseases, 3, 395-400.

Koch-Weser, J., Sellers, E.M., and Zacest, R. (1977), "The ambiguity of adverse drug reactions," Eur. J. Clin. Pharmacol., 11, 75-78.

Langlotz, C.P., Shortliffe, E.H., and Fagan, L.M. (1986), "A methodology for computer-based explanation of decision analysis," Stanford University, KSL-86-57.

Mahon, B.E., Rohn, D.D., Pack, S.R., and Tauxe, R.V. (1995), "Electronic communication facilitates investigation of a highly dispersed foodborne outbreak: Salmonella on the superhighway," Emerging Infectious Diseases, 1, 94-95.

McConachy, R., Korb, K.B., and Zuckerman, I. (1998), "Deciding what not to say: An attentional probabilistic approach to argument presentation," Proceedings of the Twentieth Annual Conference of the Cognitive Science Society (CogSci 98).

Rawlins, M.D. (1994), "Pharmacovigilance: Paradise lost, regained or postponed?" The William Withering Lecture. Appeared in J. R. Coll. Physicians London, 29, (1995), 41-49.

Strom, B.L., and Tugwell, P. (1990), "Pharmacoepidemiology: Current status, prospects, and problems," Ann. Intern. Med., 113, 179-181.

VanLehn, K., and Niu, Z. (to appear), "Bayesian student modeling, user interfaces and feedback: A sensitivity analysis," International Journal of Artificial Intelligence in Education, 12.

Walker, M.A. (2000), "An application of reinforcement learning to dialogue strategy selection in a spoken dialogue system for email," Journal of Artificial Intelligence Research, 12, 387-416.

Zuckerman, I., McConachy, R., and Korb, K.B. (1998), "Bayesian reasoning in an abductive mechanism for argument generation and analysis," Proceedings of the Fifteenth National Conference on Artificial Intelligence (AAAI 98).

Next: List of Participants
Working Group Index
DIMACS Homepage
Contacting the Center
Document last modified on December 10, 2001.