In this post, I’d like to take a look at a key point made in Howard Brody & Donald Light’s 2011 paper published in American Journal of Public Health entitled The Inverse Benefit Law: How Drug Marketing Undermines Patient Safety and Public Health.

(In the interests of full disclosure, I will say that Howard Brody is the Director of the Institute for the Medical Humanities, which is where I trained as a doctoral student.  Dr. Brody was a member of my dissertation committee, and I count him as a mentor and a colleague).

Although I commend to all readers the entire article, in this post I am just going to quick look at one specific graph the authors utilize, because I think it illuminates some of the key points in the article nicely.

Brody Light Inverse BenefitSo, I know there’s lots of people, especially health wonks, that really love charts and graphs.  I will be the first to admit that I am actually much more of a word guy than a visual media guy, but this is a great graph which I use all the time in teaching about the population health implications of conflicts of interest, drug marketing, disease inflation, and the pharmaceuticalization of health.

It’s quite a simple graph.  The y-axis is the number of patients in a given population.  The best way to think of the X-axis is “symptom severity,” or, generally, how sick the patients are in the given population.

So what is going on here? I usually frame the analysis through an example of a typical drug class such as atypical antipsychotics.  In terms of the history of this drug class, atypical antipsychotics have for much of their life been used only in a relatively small percentage of the overall population — often severely mentally ill patients.  We can imagine this pool of patients represented by the vertical bar labeled X in the population curve.  Why were atypical antipsychotics used in such a small proportion of patients? I am neither a geriatrician nor a psychiatrist, but the primary answer, I believe, is that atypical antipsychotics as a class of drugs have featured an historically significant side effect profile, including well-known risks of tardive dyskinesia, among many other drawbacks such as weight gain and diabetes.  So we might want to be extremely judicious in our utilization of such drugs, ensuring that they only go to the most severely ill patients (in whom it is of course easier to show efficacy in clinical trials as well).

All well and good.  (Not really, but bear with me).  Now, we would need to ask, who is unhappy with this graph at the vertical X-bar? The answer, of course, is generally the manufacturers of the atypical antipsychotics, at least from a business perspective.  If the class of drugs is only used in the most severely ill patients, that is a relatively small proportion of the overall population, and exerts significant downward pressure on sales of the drug.  The obvious solution — or at least ONE obvious solution — is to expand the patient population for whom the drug is being prescribed.  This pushes the vertical X-bar in the graph to the left, now marked as the vertical Y-bar.

Brody and Light emphasize that an enormous shift leftward is not required to dramatically expand the patient pool:

The shape of the bell curve dictates that a relatively small left shift in the threshold will demarcate a disproportionately larger area under the curve (under the curve to the right of Y). Thus, a small change in the cutoff point for drug prescribing can lead to major increases in company revenues.

Please note, as Brody and Light point out, this process is not simply speculation or conjecture.  This has happened multiple times, as they go on to discuss.  But from the leftward shift in the user pool we derive two troubling implications for population health.  First, any time we expand the pool of people taking a drug, we are guaranteed to have more adverse events.  More adverse events in general is undesirable unless the increase is somehow offset by greater health benefits that also correspond with the increase, but this is extremely unlikely to be the case.  Why? It is epidemiologically axiomatic that it is easier to show proof of efficacy in a more severely ill pool of patients/participants.  So as we expand the user pool leftward, the added users are simply less sick and as a group are less likely to derive benefits from the intervention.  This leads to the second important implication for population health, which, as the analysis suggests, is really just derived from the first point: the NNT goes up.

“NNT” stands for “number-needed-to-treat,” and it is IMO one of the most important statistics we can use in thinking about priority-setting in public health policy.  Given that virtually all interventions pose some risk (even if they are low), it follows that, other things being equal, we want NNTs to be as low as possible.  The less people we need administer an intervention in order for one person to benefit, the better.  But when we shift the user pool in terms of drugs lower down the symptom severity scale, we necessarily increase the NNT, which is generally not something we want to be doing in terms of population health unless we have a very good justification for doing so (increasing drug sales doesn’t qualify as a good population health justification).

As Brody and Light put it:

. . . when less severely afflicted or lower-risk patients are given the drug, many more must be treated for 1 patient to benefit (high NNT).  A high NNT has important public health implications if the drug is expensive and competes for scarce resources.

An additional problem is that expanding the user pool may often require off-label use, which means the added prescriptions may be being taken and used in a population for whom we have virtually no good safety & efficacy data.

The example of atypical antipsychotics again shows this entire process unfolding, as this 2012 paper documents:

It is important to note that although the FDA has approved second-generation antipsychotic medications for these conditions, most pediatric use is off label, that is, prescribed for conditions not approved by the FDA (Crystal et al., 2009; Zito, Derivan, et al., 2008).  In addition, a twofold to fivefold increase in the use of antipsychotic medications in children younger than 6 years has occurred, despite little information on their long-term effects on child health and the developing brain.

I like the Brody and Light paper and the chart for a variety of reasons, not least because it shows how clinical decisions aggregated together can and do have population health implications, even if reducing population health outcomes to a mere aggregate of clinical decisions commits the fallacy of division.