Friday, September 09, 2005

More on drug safety surveillance from the Medical Informatics perspective: the drug 'risk portfolio'

As a result of some recent discussions and after reading my essay "Personal reflections on the future of drug safety surveillance from the Medical Informatics perspective", Mark D. Uehling, Senior Science Editor of Bio-IT World Magazine, who has published a number of my Op-Eds, has recently asked some good questions.

He asks: "It’s not clear EXACTLY what information in the EMR will distinguish the MI caused from Vioxx from the one caused by pre-existing factors. Is EMR adoption really going to help pharma? ... What, aside from EMR and pharmacogenomics, might allow sponsors to make product-withdrawal decisions quickly when necessary and stick to their guns when the drug is not causing problems? Will this always take 5-10 years to figure out?"

My response was:

Patients know there is inherent risk to medical treatment. Rather than trying to minimize notification and availability of data about risk, I think pharma needs to create up-to-date 'risk portfolios' of their drugs that clinicians and consumers could access via the web. it's a constrained version of the idea of putting RCT results on the web (see below).

Think about it. Consumers can access gambling data and odds online e.g., Off Track Betting at sites such as But they cannot access up-to-date risk information categorized by adverse event type on drugs. Even clinicians must rely on package inserts, PDR, the detail people with their likely dated material, etc. and the occasional call to a pharma company's clinical information people.

The risk information needs to come from numerous sources. MedWatch, clinical trials, existing national registries (e.g., Genentech National Registry of MI) etc.

The question about the use of EMR in adverse event detection is a good one. It is here that statistical techniques and data mining could provide the associations. Retrospective cohort studies and other methodologies are useful. This is what Graham at FDA and others do with retrospective data such as claims data; such work would be more credible if the data utilized was decent clinical data from EMR. Further, the adverse events (AE's) could be partitioned into directly observed (e.g., Medwatch) vs. derived (e.g., from EMR data analysis) categories.

The results are not RCT-perfect, but clearly the public is unhappy with the current minimizing of facile access to concurrent information about drug AE's, as represented by those fast-pitched warnings at the end of the ad with the pretty girl and the farm that says how the medicine will turn an arthritic cripple into a champion figure skater. (/sarcasm off) ...

The problem right now is one of pharma intransigence towards non-RCT data. Graham hit the nail on the head here. As I mentioned, there is a marked asymmetry between testing drugs for effectiveness vs. for low-level AE's via clinical trials using the p<.05 constraint. It would be a wonderful world if we could do RCT's on tens of thousand or millions before drugs hit the shelves, but we cannot. RCT's will always take time to show low-incidence AE's and need supplementation if we are going to avoid the "5-10 years to figure out" problem.

If such concurrent drug 'risk portfolios' data were readily available via the Web, clinicians would have an easier time explaining the best-of-our-knowledge risks more granularly to patients. And putting full clinical trials results online is not the answer, either (and is probably not necessary), because the AE data of use at the business end of clinical encounters might get lost in the details. I envision a virtual "package-insert AE table" that can be used by clinician and patient to decide if they want to take the risk of using a particular drug.

If such a facility were available, VIOXX might still be on the market. I believe many would be willing to take it even knowing the risks. What caused the problem was the sudden turnabout and the perception of concealment of risk information.

It would, of course, be assumed that companies would do their utmost to keep the AE clinician/consumer tables updated, and when new low-incidence risks are discovered, the openness might prevent the need to have to suddenly withdraw a drug from the market. Of course, it would also minimize the risks of lawsuit for AE's that were reported via the site.

Transparency via 'risk portfolio' information (i.e., up to date, "virtual package insert", web-available AE tables) is the key concept to avoiding spectacular pharmaceutical company "ELE's" a.k.a. extinction level events, as in those asteroid-hits-earth movies...

In my opinion, faith in RCT's has become somewhat of a religion. By this, I mean that it is sometimes forgotten that accepting results as reflecting truth when "p is less than .05" is a convention for a commonly-accepted level of uncertainty regarding the meaning of experimental results. The p value of .05 is not a universal constant such as c (the speed of light, 299,792,458 m/s), or a proof of reality outside the scope and limitations of a particular trial. Marketing people would perhaps have us believe otherwise...

It must also be asked regarding clinical trials: "what is the p[robability] that due to trial scope and limitations, findings of importance were missed?"

The chronology of Vioxx clinical trials and the latest industry statements that Vioxx risks only exist after 18 months of consumption is likely a sign that this question is not being asked, especially since retrospective data hint otherwise. To trust the latest APPROVe trial regarding the 18-month period, when earlier trials (e.g., "Study 090") were conflicting regarding whether Vioxx was associated with cardiac adverse events at all, are evidence that faith in individual clinical trials may have become dogmatic and short-sighted. What will the next, even larger trial show?

Albert Einstein had a saying: Insanity is doing the same thing over and over again and expecting different results -- which in this situation means depending only on progressively-less limited (but limited nonetheless) RCT's, in concert with a very broken postmarketing surveillance "system", over and over again for drugs sold to millions.

In any case, regarding the Web-based 'risk portfolio' mentioned earlier based on a number of sources of data, if pharmas won't do it, then someone else will. That "someone" will be government or private entrepreneurs, or a combination. Perhaps an independent authority would be best for such an endeavor, anyway.

-- SS