CARNETS DESCARTES

Drug safety 2.0

(Version française)

How long will we live without using available early warning signals from the Internet, particularly when it involves health care quality and safety?

Last February the 13th, a controversy raged, well summarized in Nature by Declan Butler (free of access, in English), regarding influenza surveillance through Google Flu Trends.  This year indeed, Google Flu Trends seem to have led health authorities from New York City to overestimate the winter epidemic wave. However, Dr Finelli said: “I’m in charge of flu surveillance in the United States and I look at Google Flu Trends and Flu Near You all the time, in addition to looking at US-supported surveillance systems. I want to see what’s happening and if there is something that we are missing, or whether there is a signal represented somewhat differently in one of these other systems that I could learn from.”

It is no question to say that existing epidemiologic tools are obsolete today, nor to foresee that they will be soon substituted by the Internet. What seems to me important to note is to realize that people in charge of epidemiologic surveillance in her/his country need to be trained in using social networks (the above mentioned paper on Nature refered to a study using Twitter for flu surveillance, conducted by researchers from Johns Hopkins). Public health professionals now need to acquire competencies in analysing data from search engines on the Web (Both Google and Yahoo have proved effective in this domain). More generally, public health professionnals must have skills in Big Data, characterized by their volume, variety, velocity and value, which means it is not a trivial game to analyse and interpret them. This includes database on drug consumption (i.e. data brokers such as IMS-Health or Celtipharm in France), as well as huge databases produceds by HMOs, and health care for billing purposes, providing they are available for research activities.

In the last online published issue of JAMIA (restricted access but the abstract, in English), a paper extend Google Flu Trends philosophy to drug safety.This paper acts as a proof for concept. Authors extracted a test case from the US FDA adverse event system (AERS), using data-mining algorithm. That was an interaction between paroxetin (an anti-depressant) and pravastatin (a cholesterol-lowering drug) reported to create hyperglycemia in 2011. This finding was confirmed in a retrospective analysis from electronic health records and in a mouse model. Authors analyzed the search logs of 6 million of consenting web users who opted to share search activities via the installation of a browser add-on, spanning a 12 month period in 2010, used to track anonymously 82 million drug, symptom and condition queries that users perform over time. We can say authors simulated (in 2012) a prediction their method had produced in analysing 2010 data, without the knowledge of this drug interaction (only known in 2011). On these conditions, internet search logs they worked on were unbiased.

Striking results were found: internet logs from users who searched on both paroxetin and pravastatin where twice as frequently associated with search on symptoms of hyperglycemia, than those who searched on either paroxetin or pravastatin only.  Moreover, these results remained quite stable over time in 2010, with no seasonal pattern. Authors provide details on their statistical methods for datamining and analyses, which are obviously not conventional ones.

Existing departments in charge of drug safety, both in pharmaceutical industry and in public health agencies will no longer be in position to resist anymore to acquire resources and competencies allowing them to perform such analyses. These analyses will probably soon prove their effectiveness (and usefulness) in early warning. And therefore in targeting enhance drug surveillance. Short time response is probably more crucial for drug safety than it is for influenza outbreaks. And we know that acquiring evidence of causality in drug safety is a matter of time. Clinical trials usually performed in a few thousands of patients are more often not able to detect rare and serious adverse drug reactions which are therefore identified later, after their approval, when already on the market, when millions of users are exposed. The sooner serious adverse events are detected, the safer for the patient. If hyperglycemia associated with the interaction between these two largely prescribded drugs had been detected earlier, let's say in 2010 rather than 2011, that would have avoided number of illnesses, perhaps deaths, not talking about costs, just by avoiding prescribing such dangerous associations of drugs in patients. Early warning benefits to patients first, but also to their confidence in the health care system, and to the business of pharmaceutical industry which may early target its drugs to those who express the highest benefit-risk ratio. Best usage of drugs is indeed good for all actors involved here: patients, doctors, manufacturers and health authorities.

It will not be my last post on Big Data. We have - in the field of health - to become more familiar with Big Data. We need to invest much more in skills, competencies, knowledge in Big Data. We need to use them in all sectors of health and healthcare: hospitals, administration of health, public health agencies, health industry, research labs (genomic data...). We must not be afraid with the use of these data. Again, we will continue to need careful and expert validation from  doctors, pharmacists and epidemiologists. However, how long will it remain acceptable to leave out of our daily work this huge amount of available, inexpensive and valuable information which is just waiting for you to be processed and analysed?

Antoine Flahault's blog (in English)

Antoine Flahault's blog (in English)

Antoine Flahault's blog. He is Faculty member, in public health, from Descartes School of Medicine, Sorbonne Paris Cité

Articles relatifs

Transforming Epidemiology for 21st Century

Transforming Epidemiology for 21st Century

Volume, variety, velocity and value from...
"Big Data" en santé et confidentialité

"Big Data" en santé et confidentialité

Les sites Internet de santé pour le grand public...
Are patients'data patient's property on health on the net?

Are patients'data patient's property on health on the net?

Health sites are blooming worldwide, and recent...
Epidémiologie du XXIème siècle

Epidémiologie du XXIème siècle

Le volume, la diversité, la rapidité et l'utilité...
  • http://carnets.parisdescartes.fr/blog/view/133759/drug-safety-20