Empiric therapy - not a substitute for rapid testing

In the last post I wrote about the case for the value proposition of narrowly-targeted antibiotics as developed by Spellberg and Rex. Their model has an implicit assumption - that the hypothetical high-priced antibiotic will be prescribed in a timely manner to patients who will benefit from it, and only to those patients. My point in that post was that a companion rapid diagnostic would be needed in order to realize this scenario, that such a test is likely to have a significant cost as well, and that this cost needs to be factored into the analysis.
David Shlaes has taken issue with this view, explaining that

"Luckily, physicians don't wait for the diagnosis before treating. They treat empirically - and if they are in a hospital where MDR Acinetobacter is a risk - they will use a new expensive drug to cover for that possibility until they have confirmation or not. "

He is of course correct. No physician faced with a failing patient would withhold a drug that might have a significant therapeutic benefit. But this practice has all sorts of implications for the value of narrowly-targeted antibiotics, and none of them are good. I'm not sure that "luckily" is the adverb I would have chosen to describe the situation.
Let's be clear - empiric prescribing is a necessary response to imperfect information. But there should be no illusions that physicians have some magic powers that make empiric prescriptions much better than guessing. Indeed, a careful study of prescribing practices is entitled "Empirical antimicrobial therapy for bloodstream infections ...: No better than a coin toss". As the title suggests, when faced with a binary choice between choosing an antibiotic that is optimal for MRSA and one that is optimal for MSSA, doctors did no better than predicted by chance. There is no reason to think they would do any better in choosing optimal antibiotics for Gram-negative infections.
Combine this element of chance with the fact that the organism of interest in this scenario is not all that common, and the case for the value of an antibiotic narrowly targeted toward it can deteriorate pretty quickly. Acinetobacter infections are not currently likely to constitute more than 10% of Gram-negative infections in any hospital in North America. Although not dominant, this level is high enough to be a threat in any patient with a serious infection, and physicians will have to treat accordingly.
If they prescribe the antibiotic to cover possible Acinetobacter infections in half of these patients when only 10% are actually infected, then there are several unintended consequences, all of them bad (except for the 10% who actually have Acinetobacter infections).
First, the value proposition for the antibiotic becomes far weaker. In Spellberg and Rex's median case the antibiotic costs $10K per dose, and the net cost for an additional year of life saved is only $3K (anything less than $50K is considered to have a favorable cost-benefit ratio). But if the antibiotic has no benefit for the patients who turn out not to have an Acinetobacter infection (and this is stipulated in their scenario), then use of the drug is a dead loss economically. If 50% of patients receive the antibiotic when only 10% are infected with Acinetobacter, then the cost per year of life saved goes from $3K to $15K, a much less attractive proposition.
Second, the non-Acinetobacter patients are exposed to the risk of adverse events, with no corresponding clinical benefit to compensate. Since this is a hypothetical drug, there is no point in trying to quantify this risk, but it is surely not zero.
Third, this level of overuse will accelerate the development of resistance to the drug. Although narrowly targeted by definition, it is biologically implausible that it would have no negative effect on the fitness of any of the thousands of species of bacteria that comprise the human microbiome. And once resistance develops in one species, it is only a matter of time until it spreads to many species, including the targeted pathogen.
Now imagine that there are several narrowly targeted antibiotics available: one for Pseudomonas, another for carbapenem-resistant Enterobacteriaceae, etc. In the absence of a rapid test, how will a physician decide which to prescribe? Should she prescribe all of them? If she does, all of the negative effects of inappropriate therapy listed above would be compounded.
All of these problems go away if there is a rapid test. If I were a pharma exec, I would not begin considering a program to develop a narrowly-targeted antibiotic unless I felt pretty confident that a companion rapid test could also be developed. The path to clinical and economic value is just too full of land mines otherwise.

Posted with Blogsy