Recent Publication: On shapes of ADR report accumulation data

A recent article published by Cytel authors Samadhan Ghubade, Sharayu Paranjpe, Kushagra Gupta, Anil Gore and colleague Krishna Asvalayan in the journal Current Science, tackles the topic of adverse drug reactions (ADRs) – a matter of great concern in drug research. The authors focused their research on drugs which had been either banned or withdrawn due to a serious problem of ADRs and applied quantitative modeling techniques to see if a systematic pattern of safety signals could be detected within the ADR count data. In this blog, the publication’s authors share their thoughts on the goals, takeaways and next steps for the research and we also link to the full article.

What were the goals of the publication?
 The team was already involved in modeling Adverse Event (AE) data in clinical trials, and we wanted to explore the possibilities of conducting similar modeling with pharmacovigilance data, while of course acknowledging the fact that pharmacovigilance ADR reports are distinct by being made voluntarily. Within our team, we were very keen to develop interdisciplinary initiatives between pharmacovigilance, statistics, and programming and decided to use ADR counts that were readily available via the World Health Organization's www.vigiaccess.org  as the basis of our research. We began by selecting a subgroup of drugs that had been withdrawn from the market, with the hope that a systematic pattern of safety-related signals would be detectable in the ADR count data.

The team worked with cumulative totals of ADRs over years, and the exploratory part of the work involved fitting different mathematical models to these cumulative counts as a function of time. The outcomes were more interesting than expected and so the scope of the project expanded from the originally planned conference presentation to a full paper.

New Thinking and Takeaways
Our research showed that the plot of cumulative counts over years assumes different shapes in case of different drugs. We believe that only a few models can handle most of the data: saturating growth; sigmoidal growth; linear growth and exponential growth. In the exponential growth model, the coefficient in the exponent is independent of the absolute counts and represents the fast-rising shape. The coefficient can be used to compare any group of drugs for the same indication and rank them in terms of safety. The issue of how to compare the safety of drugs if the shape is not exponential but, for example, linear remains an open question. This may be resolved by making the data comparable by correcting for usage of the drug.

Next Steps
Ultimately, for the future development of the technique, we would like to develop a ranking method that works even when the cumulative ADR count curve does not fit an exponential curve.

To apply the technique further in practice, through literature review, we could take one group of drugs at a time (for example opioids, contraceptives etc.) and check how researchers view them in terms of safety. If there is a lack of consensus among experts about the drugs' safety profiles then we can conduct quantitative analysis and apply our approach to come up with rankings that may help researchers and practitioners.

Another important next step for the work is to effectively communicate the potential of the approach to clinicians. Understandably, there may be concerns that the story cannot be summarized by a single number. In addition, we would need to take steps to separate the total ADR count into subgroups – for example, deaths, serious AEs and other AEs.

With thanks to Dr. Anil Gore,  Samadhan Ghubade, Sharayu Paranjpe, Kushagra Gupta, and Krishna Asvalayan

To read the publication in full, click the button below.