This release of ChemMeta introduces data and analytics derived from the Food and Drug Administration’s Adverse Event Reporting System for ChemMeta’s inventory of active pharmaceutical ingredients. This release also adds 36 certified bodily injury hypotheses involving 25 chemicals and one pharmaceutical and updates general acceptance scores with the most recent scientific publications investigating ChemMeta's inventory of 3,583 hypotheses of bodily injury attributable to chemical exposures and 532 hypotheses of bodily injury attributable to pharmaceutical exposures.
The FDA Adverse Event Reporting System
The Food and Drug Administration’s Adverse Event Reporting System (FAERS) is a database of pharmaceutical adverse event reports submitted to the FDA by healthcare professionals, pharmaceutical manufacturers, consumers, and other interested parties. A single adverse event report can list multiple pharmaceuticals and multiple adverse health effects (or harms as we refer to them in ChemMeta). More than two million adverse event reports are filed with the FDA each year.
Critically, the existence of an adverse event report, by itself, does not demonstrate a causal link between a pharmaceutical and a harm. Although the reporter may infer a causal link between a pharmaceutical and a harm, a report should be treated as merely an observation that an individual was taking one or more pharmaceuticals and experienced one or more harms. And, while the FDA encourages and, in some cases, requires healthcare professionals, manufacturers, and consumers to report adverse events, it should be assumed that the reporting is substantially incomplete. Even with these key limitations in mind, however, the FAERS data constitute a critical component of the FDA’s post-market surveillance program and are mined routinely by scientists throughout the world in an effort to identify serious pharmaceutical side effects as early as possible. We explain our approach to using FAERS for this purpose in the section to follow.
The FAERS data require considerable processing in order to make them useful in ChemMeta. Data processing includes the following activities:
- Determining the date of each adverse event. Drug adverse events can be reported multiple times and are commonly updated after the initial report was filed. Where available, we use the date the adverse event was first observed by the reporter. Otherwise we use the date FDA first received the adverse event report.
- Matching adverse event reports to active pharmaceutical ingredients (APIs). Reporters are not required to employ a controlled vocabulary when reporting which APIs the patient was taking at the time of the adverse event. Thus, it is possible that some adverse event reports will be falsely linked to a given API. And some adverse event reports cannot be linked to any API.
- Matching adverse event reports to harms. All adverse events are coded using terms in the Medical Dictionary for Regulatory Activities (MedDRA). We translate these terms to the Unified Medical Language System (UMLS) and then to the much broader harm categories employed by ChemMeta.
- Limiting adverse event reports. We limit reports to those containing at least one ChemMeta API and one ChemMeta harm.
- Tabulating adverse events over time. The number of adverse event reports filed each year has grown substantially over time as a result of more pharmaceutical use and increases in the reporting of adverse events. We begin our tabulations of adverse event data in 1999.
- Updating adverse event data. FDA releases FAERS data to the public on a quarterly basis. The last available quarter of FAERS data as of this ChemMeta release is for the first quarter of 2019. We will add subsequent quarters of FAERS data to ChemMeta as soon as they become available. Each quarterly release of FAERS not only contains adverse events not previously reported (and some where the underlying event may have occurred in prior quarters), but also revisions to adverse event reports already in the system. Thus, each quarterly release will result in revisions to the full time series of adverse event data.
As of this release, the universe of adverse event reports in ChemMeta contains 18,186,904 distinct pharmaceutical events. Adverse event data reported on the ChemMeta dashboard are for both certified and non-certified APIs. Adverse event data reported on individual pharmaceutical profile pages are currently available for certified pharmaceuticals only and will be available for all pharmaceuticals in a future release.
The adverse event risk score
ChemMeta now reports an adverse event risk score for every certified pharmaceutical-harm hypothesis. The adverse event risk score data are visible on the ChemMeta dashboard (1-2), underlying pharmaceutical, harm hypothesis, and company tables (3), and on the individual pharmaceutical profile page (4).
The adverse event risk score is a modification of the reporting odds ratio commonly employed in the pharmacovigilance literature. The reporting odds ratio provides an indication of the relative frequency of adverse events involving a given harm and pharmaceutical. The reporting odds ratio is high when individuals report a given harm more frequently when taking a given pharmaceutical than when taking other pharmaceuticals. We compute reporting frequencies cumulatively starting in the first month an adverse event is reported for a given pharmaceutical or in January 1999, whichever is later. As a result, the adverse event risk score is typically more volatile early in a pharmaceutical’s life-cycle when the cumulative number of adverse event reports is small.
We use a hyperbolic tangent function to translate the reporting odds ratio into the [-1,1] interval so that we can present it on the same scale as the existing general acceptance risk score. An adverse event risk score of zero indicates adverse events involving the harm and pharmaceutical are no more likely to be observed than adverse events involving the harm and all other pharmaceuticals; positive scores indicate relatively high frequency of adverse event reporting and negative scores indicate relatively low frequency. It bears repeating that a high adverse event risk score does not necessarily mean that the pharmaceutical is causing the harm, just that the level of adverse event reporting for that pharmaceutical-harm pair is unusually high. By contrast, a high general acceptance risk score means that the scientific community broadly accepts the hypothesis that the pharmaceutical causes the harm.
Adverse event report count data
ChemMeta also now provides interactive tabulations of adverse event reports by pharmaceutical, harm, and other event characteristics. On the ChemMeta dashboard, users will see tabulations of adverse events by year and the “seriousness” of the patient’s health condition: death, serious, and non-serious. “Serious” includes the categories of life-threatening, hospitalization, disability, congenital anomaly, and required intervention.
On the individual pharmaceutical profile page, users can tabulate adverse event reports by year, harm, UMLS harm, seriousness, suspect level, and type of reporter (5). Suspect level is either primary, secondary, or non-suspect. “Primary suspect” means that the reporter believes the pharmaceutical is the primary factor causing the patient’s harm. “Secondary suspect” means that the reporter believes the pharmaceutical may have contributed to the patient’s harm along with other factors. And “non-suspect” means that the reporter believes the pharmaceutical was not a factor in the patient’s harm.
Additional enhancements
This release includes three other enhancements to the ChemMeta user interface. The dashboard has been reorganized to feature the highest risks scores and greatest change in risk score cards at the top of the page. On the individual chemical or pharmaceutical profile page, users will see that the risk score plot displays by default the harm hypothesis with the highest current general acceptance risk score. The current general acceptance risk score is also now displayed in the plot legend.