“Electronic Health Data for Postmarket Surveillance: A Vision Not Realized”


One primary justification for building large electronic health records systems has been that they will make it easier to detect rare but dangerous side effects of drugs, or to identify previously undetected risks by studying trends in large user populations. In Drug Safety, Thomas Moore of the Institute for Safe Medication Practices and Wake Forest University School of Medicine’s Curt Furberg ask, after years of efforts, “What has been learned about electronic health data as a primary data source for regulatory decisions regarding the harms of drugs?”

“Electronic health data for postmarket surveillance became a key element in the new paradigm for drug regulation, which involved fewer and smaller clinical trials prior to marketing approval,” the authors state in their summary points. They then review what was awry in a slew of attempts to use massive databases of electronic health records to better understand drug risks, including inconsistent practice tendencies and record-keeping among different physicians, difficulty in validating findings, and problems with identifying and adjusting for confounders.

The authors conclude, “There is no credible evidence that electronic health data today has the capacity to provide robust, reliable ‚Äėactive surveillance‚Äô, meaning identifying new drug risks not previously identified through other means. The results thus far dramatize the difficulties in confirming known adverse effects found using other methods. The high levels of variability in almost every parameter render findings difficult to replicate and vulnerable to substantial bias, either as an accident of data and method selection or through intentional manipulation of study criteria.”

Moore, Thomas J., and Curt D. Furberg. ‚ÄúElectronic Health Data for Postmarket Surveillance: A Vision Not Realized.‚ÄĚ Drug Safety, May 30, 2015, 1‚Äď10. doi:10.1007/s40264-015-0305-9. (Abstract) (Full text)


  1. Computer programming wasn’t very well taught yet when my terms became those of nontraditional survivor retaking English and opting out of clueless pscyh some more times. Tellingly, I see potential error in the implementation for the IT ontology, such that it the procedure doesn’t let the machine decide every detail of input and all input is variably defined for mthe outset no matter what entity it intends to “collect” in to the form of one term. No matter what happens about that and any speculation on it, though, no hurry should seem needed for giving Their troops a big survivor reform movement shout out about it from here. They are, after all, the medico-statists authorities’ own. The model for prescribing isn’t market-driven enough according to drug availability for the consumer, and that matters most for the data and outcomes measured–alike. And for health and safety and information sharing rituals and lookup codes at point of sale.

    Report comment