Medical Moneyball!!

The story below changes or omits all the proper names, because that's the decent thing to do, but it would be even more decent not to write this at all, as it wouldn't be that difficult to figure out who these people are, and they don't come off well. Balanced against my need to get all this off my chest, I'm afraid I'm going to err on the side of selfishness.

Background, some of it anyway. I spent the last two decades at a variety of financial institutions. There have been interesting technical challenges and smart people to work with (especially early on), but over the years an unfortunate pattern emerged. After a brief period of intellectual excitement, one is drawn - ratcheted with audible viscous smacks, like the beginnings of a death by quicksand - into managerial and procedural purgatory. Despite the fact that the end result is stultifying boredom, the mechanism by which this happens is not uninteresting... but it's also another story. The summary version is that you agree incrementally to new responsibilities and end up never getting to code or, it seems, to think.

Determined to end this sad pattern, I managed to engineer extrication from my last engagement. If, hypothetically, contract law were relevant to that extrication, then I would be contractually bound not to go into the details, which might also be interesting, again despite the the topic itself being lack of interest. I'm under no constraints whatsoever in discussing my current activities, and anyway it's unlikely that Hacker School would object to being described as unbelievably marvelous. Essentially, I get to program and study study programming all day, surrounded by people who validate the importance of this self-indulgence. I'm happy.

The motto of Hacker School - never graduate - notwithstanding, most good things come to an end, and I think this one does in around a month (I've refused to look up the exact details). I will have to find gainful, respectable employment.

With this in mind, I was not immediately dismissive when a former colleague, who is such an unambiguously nice guy, so I won't even give him an incriminating pseudonym, and his potential business partner, whom we'll call Danny, approached me for a CTO role (whatever that means) in their incipient venture. Danny has been associated with a venture capital company associated with a few well known internet start-ups. You've heard of both the company and the start-ups; you've probably been a "customer" of at least one of the latter, and an unwitting user of another. A shallow internet search confirms Danny's direct responsibility for these humdingers, though remarkable textual similarities in the results do suggest the hand of a busy PR firm. More complex searches are ambiguous. For example, Danny's real name and the name of the well-known CEOs of the these companies never show up together in one article. Every time Danny's name is mentioned, it's part of a friendly interview about his new VC/incubator project.

Danny describes his latest idea as "moneyball for medicine." This is not an original formulation, and his use of it is inaccurate and silly enough that I expect it to be dropped. If he doesn't have the sense to drop it, then I won't feel bad about your ability to discover his identity through web searches. (For the moment, the google won't help you.) The idea begins reasonably enough, with an actual (but apparently not yet published) paper by real people at Johns Hopkins that seems to demonstrate that, when doctors are told the cost of the procedures they are about to order, the cost of treatment declines in a statistically significant way, without a corresponding decline in the quality of treatment. This is a powerful result, because it is unlikely to inspire whining about "death panels" and because it could be implemented really cheaply. While it's notoriously complicated to calculate the web of negotiated cash flows that result from any medical event, it's relatively easy to estimate the economic cost, e.g. whether or not insurance pays for the CT scan, we can still approximate its by amortizing the device over its lifetime, factoring in salaries of the technician, etc. If these actual costs are what we want to reduce, and they're easy to calculate, then some large electronic health record provider like Epic could just add an extra column to their existing systems and be done with it. In other words, this result is compelling precisely because it is not a good business model.

True says Danny, but these systems were all built decades ago, and they're absolutely loathed by the physicians who use them because of their clunky Windows-95y interfaces. In fact, says Danny, patients "literally bleed to death" while doctors and nurses are fumbling with this ancient software. How awful! "Is it fair to say our first deliverable would be a more ergonomic GUI? Because GUIs are hard, and competing on the basis of your ergonomics is an expensive game, requiring both programmers and a deep understanding of medical workflow, and..." Danny interrupts me with a look that says, "too many notes." These are hospitals, he says, so the doctors will really have to use whatever system the administrators choose. OK... so it's putting the cost column on somebody else's GUI? No, says Danny, going on to repeat the "literally bleeding" bit and to say the worst thing he can think of about the existing systems, which turns out to be that "it's like they hadn't heard of AJAX. That's what it's called, right?" I'm confused, the skype connection is not great, and I might have misunderstood something. "So," I ask, "is it fair to say that our first deliverable would be a more ergonomic GUI? Because GUIs are hard, and..." No, that's not the point at all. (I learn later that this sort of warble is called pivoting.)

The point, it seems, is... MONEYBALL!! At this point we're sitting at a monumental, heavily polished dining room table in a Long Island mansionette. I call it that instead of just mansion, because I know that there are larger single-family residences in existence, but in truth I've only seen such places in movies and on guided tours. This one is owned by a cardio-thoracic surgeon who not only has a lot of money but is quite sure that he deserves it - unless the objectivist shrine of ornamentally bound Ayn Rand volumes in the entrance foyer is meant ironically, which seems unlikely, or at any rate would be very lonely for it. His wife clops in and out of Carmela Soprano's kitchen on skinny heels, to ask which machine we would most like coffee from. Cardio-thoracic man, whom we'll call Vince, seems an odd choice if medical cost savings are the goal, though it later occurred to me that we could make a lot of headway simply by taking back all of his money and using it on, say, measles vaccines. Vince is more taken with the MONEYBALL!! angle. MONEYBALL means that we'll provide "real-time, actionable information" and "quality of care metrics" to code patients by cost-effectiveness. Danny illustrates this by taking from his briefcase a pair of Google Glasses, which we spend about 15 minutes trying to get to perform.

Despite the distraction (the glasses are bright red), it occurs to me to ask for an example of a "metric," because I'm still having a hard time with this. A few hours later, we get an answer - from a different doctor actually, who was smarter and generally more sympathetic and so, following precedent, won't get a nickname. It turns out that negative outcomes in heart surgery tend to occur immediately. 1% of the patients die in the hospital, while the other 99 both survive and experience the benefits they were supposed to derive from the procedure. This differs from other sorts of medical intervention, where various factors make the outcome a little murkier. Dead/not-dead is about as clean as it gets, metric-wise. Isn't, I wonder aloud, standard error going to be a problem? Danny looks perplexed, but quickly rallies, by recapping the plot of Moneyball (the movie). Pointdexter (me) rudely interrupts with something about $\sqrt(N)$ and the Poisson distribution. Sympathetic doctor begins to understand. Danny rolls his eyes. Vince looks serious and fondles his phone. After 100 surgeries, the expected mortality is one person... plus or minus one person. You'd have to accumulate 10000 surgeries to start distinguishing between surgeons who differed in skill by 10%, or to pass judgment on the spending of surgeons whose patients ended up differing in cost by 10%. 10000 operations is about the number performed over the career of a cardio-thoracic surgeon in a major hospital. So "real time" might mean telling a surgeon how he did at his retirement party.

I never saw or read Moneyball, but I don't think the point was that, if one wants hard enough to become fabulously successful by using statistics, one inexorably will. (That was a different baseball movie, other than the part about statistics.) I am, to be clear, pro-statistics. We will get an amazing data set from mandatory electronic health records, and, given the vastness of the data and the enormous coefficient by which you get to multiply even the smallest improvements in outcome or cost-effectiveness, this is an area to which considerable resources should be directed. That's what NIH grants are for. It's not a startup.

Danny's take on the days proceedings was an email consisting primarily of this sentence: "I would hope that you both will be able to get the ball rolling and have an evolving conversation as we move forward."

My friend reminds me that people like Danny, who can tell a good story, fervently and in the right plutocratic echelons, can be very useful when starting a business. That's probably true. It may even be true, along the lines suggested by F. Scott Fitzgerald, that the abiding of nonsense is a form of genius. I don't need Danny to be a fraud to reach the conclusion that I must stay far away from him. Evasion drives me crazy. Glib misuse of science enrages me. And, I've discovered, I get physically ill if I go the work day without coding.