APPROACHES TO STRUCTURE-RELATED PREDICTION OF DRUG TOXICITY BY ANALYSIS OF HUMAN TOXICITY DATA (NSAIDS, SIDE-EFFECTS, ADR REPORTING)

There is a need for organizing and analyzing existing human toxicity data bases and linking them with the more structured chemical information data bases for the purposes of correlation and development of coherent concepts of chemical toxicity. This work explores the use of selected chemical and human toxicity data bases in relating generic chemical structural moieties to the occurrence of toxic events in exposed human populations. Data from the Adverse Drug Reaction System (ADR) data base, Division of Drug Experience, Food and Drug Administration and from the Michigan-Minnesota Medicaid data base, Health Information Designs, Inc. (HID), Washington, D.C., were analyzed in a pilot project using the non-steroidal antiinflammatory drugs (NSAIDS) as a test group of pharmacologically similar drugs with diverse chemical structures. Estimates of comparative toxicities were determined for the chemically different groups in the ADR data base using hepatic, renal, hematologic, serious upper gastro-intestinal, serious skin, and anaphylactoid side effects as the main categories. In analysis of spontaneous ADR data, it is an apparent given condition that reporting of ADRs varies with time of marketing, being proportionately higher relative to the market and eventually decreasing and leveling off at some point in time. However, when considered over a period of time, the ADR data base appears to contain information that may be representative of the experience with a drug of the U.S. population. Supporting this it was found that the demographic profiles of ADR reports correlate with population exposure and are similar for all drugs in the class. The ADR data were correlated with the relative rates (events per 1,000 patients) of hematologic and anaphylactic reactions in the NSAID exposed patients in the Michigan Medicaid data base of 1.1 million patients which links drug and diagnoses by data of service, thereby providing an unbiased estimate of clinical events after drug exposure. Although structural hypotheses were developed, logistic factors prevented full access and use of chemical structure data bases. These illustrate the need for additional steps to facilitate the linkage between these diverse data bases. The establishment of models for analyzing and correlating these toxicological data bases represents an important step towards the integration of chemical and drug toxicity information.