Drug regulations require that adverse event reports received by pharmaceutical companies must be entered into databases and analyzed for the detection of safety signals. In a couple of scenarios, these strict regulations are leading us to add what has been called “incidental data” to our safety databases, which can actually make the identification of new safety issues more difficult, rather than easier. For instance:
- A company runs a program where they send reminder postcards to patients. When these postcards are returned to the company with the notation “Addressee deceased”, the death may need to be recorded in the safety database and communicated to health authorities.
- A drug manufacturer (or an outsourced partner) may obtain a patient’s medical record, in paper or electronic form, while assisting a patient in obtaining insurance reimbursement. Anything unfavorable that has happened to the patient since they started taking the drug may be considered as an “adverse event”. The rules for storing and reporting such events are complicated and ambiguous.
There are three problems with the growth of incidental data in our safety databases:
- A great deal of effort goes into entering these adverse event reports, trying (often in vain) to obtain follow-up information, and analyzing the resulting data. In many cases, the scientific value of this exercise is unclear.
- Neither fish nor fowl. Incidental data do not come from a systematic data collection activity, where the goal is to collect safety data from a patient population. On the other hand, they are not truly spontaneous reports. Therefore they can confuse many of the statistical algorithms commonly used for safety signal detection, resulting in false signals. In the postcard scenario above, the number of reported deaths for a product could depend as much on the number of postcards sent as on the actual incidence of patient death while on the product.
- Thinking back to the original Monty Python sketch that gave rise to the term, incidental data can distract both safety scientists and statistical algorithms from the effective detection of new safety signals, by hiding them in noise.
The topic of incidental data came up in several pharmacovigilance sessions at the 2014 DIA Annual Meeting as a problem affecting both industry and regulators. Expect to hear more about this important topic in future DIA meetings and elsewhere.
Alan Hochberg is a Process Development Leader at F. Hoffmann-La Roche Ltd., Basel, Switzerland, where he is involved in drug safety signal detection. He was formerly Vice President of Research at ProSanos, where he developed signal detection algorithms. He has worked in medical informatics and biomedical engineering for 35 years, including roles at DuPont, Hologic, Inc., and Ortho Diagnostics (J&J). Mr. Hochberg holds several patents on medical devices and data analysis methods. He received his B.S. degree in Electrical Engineering from Princeton University.
The opinions in this posting are his own and not those of his employer.