Worthwhile Ivy League stealth s*c**l*st initiative

An amateur report card on the progress of the US health IT plan.

(Long wonkish post for a quiet weekend)
During the election, the handover runup and the negotiation of the stimulus package, I posted various grumbles that the Obama team´s enthusiasm for health IT was creating a risk of a rerun of the British NHS near-fiasco. I thought I´d better check to see how things are going.

As far as I can see. the short answer is: pretty well. Here´s the website of the Office of the National Health Coordinator for Health Information Technology You can judge for yourself probably as well as I can. But since I went on record with criticisms, it´s only fair to retract them.

Longer report below the jump.

Flashback. What went wrong with the £6bn £12bn health IT plan for NHS England was basically that consultants sold a visionary top-down plan (1) to Tony Blair, which was rushed into implementation (2) managed by more IT types (3), with privacy issues treated as marginal (4), and before the medical professions or patients were on board (5). Result: long delays, huge cost overruns, and many unhappy users. It´s not a total disaster: the budget has stabilised, the applications are being rolled out one by one, and it will no doubt eventually deliver a very homogeneous working national health IT network.

The US scheme is a big contrast.

Bottom-up

Making a virtue of necessity, the US plan aims at a quilt of different health IT systems and providers, coordinated through common technical standards and institutional agreements to solidify trust. It will inevitably be uneven in coverage. Dogpatch will get good health IT long after Llareggub. Even throwing resources at the interfaces and conversions, there are bound to be holes and snags. I would guess that the network won´t be completely transparent until version 2.0, a long way ahead. But the price is IMHO well worth paying to get a scheme that´s more acceptable to professionals, has early payoffs, and is more open to innovations.

Run by medics

The US health IT satrap is a physician: Dr David Blumenthal of the Harvard School of Public Health, and unquestionably an expert on the subject. He appears to have so little vanity that his bio page is quite hard to find. Of the 13 senior staff named on the website, four have MDs and one an MPH. The qualifications of the 24 members of the main policy advisory committee aren´t listed but half have institutional affiliations clearly in health provision, and that´s probably an understatement. There´s only one from an insurance company.

Ideally, one would like to see as strong representation of patients as of medics. But the whole structural problem of the market for health care is the extreme asymmetry of expertise and anxiety between providers and consumers, for which there is no quick fix. The choice in practice is between a project run by medics who have learnt a fair amount about IT, and one run, as in Britain, by IT experts who have learnt a a fair amount about medicine. It´s absolutely no contest. Doctors are half the users and represent themselves accurately; they are imperfect representatives of patients, but have at least listened to and cared for lots of them; and they are powerful and have a de facto veto. The Obama Administration has made the right call here.

The fast start of the ONC has been helped by the good luck (or good deed of some guardian angel) that the Coordinator didn´t need Senate confirmation. This may be down to the historical accident that the office was set up in 2004 by GW Bush by executive order. It was an unimportant piece of stage scenery then with a trivial budget. In ten years, it will be a bigger but routine operation, keeping a large show running. Just now Blumenthal´s team are making strategic decisions for the next few decades of US health care. The job is far more important today than that of Surgeon-General (delayed) or ambassador to Kyrgyzstan (still held up), which both require confirmation.

Coddling physicians

Blumenthal &c clearly give top priority to securing acceptance by the reluctant population of American physicians. They have already set up 68 regional extension centres to hold their hands and jolly them into taking the plunge. At a time when there are precisely three software suites with even provisional certification!

Doctors are not stupid in resisting IT. It´s a classic issue of externalities; the physician faces large costs – mainly in changing daily working methods – while the benefits will mainly accrue elsewhere, when the patient moves town or goes to hospital, or to the doctor´s assistants. This is particularly true when doctors or hospitals are already using an incompatible IT system and have to change it, for no direct benefit to themselves.

The ONC is therefore also emphasising the blue-sky potential for making doctors´ lives easier through ¨clinical decision support¨. They carefully avoid using the words artificial intelligence, let alone golem, Igor, and autodoc, which suggest the awful prospect of robots taking over much of the job. (They eventually will. Doctors can´t have it both ways. If medicine is a science, it can be specified as algorithms, and algorithms are best implemented in reliable and memorious silicon. The non-algorithmic component that must remain in human hands is the craft, non-scientific aspect of healing, epitomised by Ambroise Paré: guérir parfois, soulager souvent, consoler toujours.)

Clinical decision support, with epidemiology, is also one of the few parts of the scheme that could one day bend the health cost curve, the doomsday machine of the federal budget, downwards to affordability. Fewer medical errors and duplicate tests and smoother administration are more certain benefits, and obviously worth having, but they will be one-off reductions in costs. A longish shot, but this really is a huge problem, and you have to try everything.

Priority for physicians is also good politics. So far, the scheme has managed to stay below the Tea Party radar. A national health information network is a very large, collective, government-led piece of social infrastructure, like the Interstate highways. It does raise very real privacy issues as well, so it is vulnerable to both informed criticism and yahoo attack. Securing the backing of the Tea Partiers´ doctors, of whom they stand in awe as the Best in the World, is vital insurance against paranoia. In the US, headlines like this recent British one could kill the plan:

The British Medical Association has said the government should suspend its new £600 million “Spine” project because of inaccurate data.

Strong network standards

The incremental approach is carried through to the standards for the national health information network, inevitably the most centralised part of the plan. The IT people have been given their head and are churning out technical documents that are quite beyond me. But the institutional design is comprehensible and makes sense.

  • The plan assumes different levels and phases of participation. VA and Kaiser go first, Dogpatch Community Physicians much later.
  • Information exchange depends on trust, and a lot of effort goes into securing this through formal institutional agreements over privacy, security, etcetera.
  • The pilot group for the network includes the DOD and VA, who already have a standardised individual health record, as well as Kaiser.

The last point is worth reflecting on. The Pentagon has access to an ocean of knowledge about network security and how to break it. If you are paranoid (and who isn´t these days?) you can suspect that the health IT network is being designed to be just insecure enough to allow NSA snooping. That will still be much, much more secure than a network a teenager can hack into with an off-the-Web toolkit. The English NHS national spine records are protected by access through smart cards, of which they have already lost 4,000.

Phase-in

During the stimulus debate I argued that it would be a mistake to put too much money into this as a stimulus measure because you must have the plan and standards in place first (Exhibit A: England NHS project). In this context, Watson-Watt´s usually excellent motto ¨second best tomorrow¨ has to be severely qualified.. I wrongly thought it would be straightforward to boost the spending in later budgets, not realising that the window for rational policymaking would close so quickly. Anyway, I´m sure nobody was listening to me.

What is the ARRA budget for health IT? It´s given by Wikipedia as $25.8bn, by the HHS as $19.8 bn; but I can only find $2bn in the ARRA text in up-front money in Title VIII, section 2.6.1. The rest looks to be just a reversible statement of intention about future budgets (Title XIII, section 3018). Still, $2bn isn´t chump change. I´ve no real evidence whether the funding, whatever it is, hit the sweet spot or not. The ONC`s website does not give the impression of an underfunded or overfunded organisation for now. Things may change when it has to fight for more money to subsidise IT adoption on a large scale.

Frankly. the only niggles I have left are that the ONC doesn´t recognize the problem of international coordination created by the growing number of people like me who move across borders, and I can´t see any mention of regional data silos for scanned images, a British idea that looks good and has already been implemented in England.

Hard to draw many general conclusions from this. But:

  • It´s just one example of the difference it makes to elect a President and a party that believes in the competent management of public business. They make policy mistakes, sure, but the execution is chalk and cheese. Do you recall who the US ambassador to the United Nations is, or the head of FEMA? That´s the way it ought to be. But I bet you remember John Bolton and Donald Michael Brown.
  • Second, the project luckily isn´t (touch wood) on the Republican ACA hate-list. Isn´t it curious that for all the talk of states´ rights, Republicans in fact always seem to target the most decentralised features of the reform? They target the public option and the health exchanges, and leave alone centralised Medicare on the large scale and the national health IT network in the small. They even want to gut states´ regulation of health insurers and allow cross-state marketing of inferior products – a national broken system. Every Republican win on this takes the US further away from Otto von Bismarck and closer to Nye Bevan.

Author: James Wimberley

James Wimberley (b. 1946, an Englishman raised in the Channel Islands. three adult children) is a former career international bureaucrat with the Council of Europe in Strasbourg. His main achievements there were the Lisbon Convention on recognition of qualifications and the Kosovo law on school education. He retired in 2006 to a little white house in Andalucia, His first wife Patricia Morris died in 2009 after a long illness. He remarried in 2011. to the former Brazilian TV actress Lu Mendonça. The cat overlords are now three. I suppose I've been invited to join real scholars on the list because my skills, acquired in a decade of technical assistance work in eastern Europe, include being able to ask faux-naïf questions like the exotic Persians and Chinese of eighteenth-century philosophical fiction. So I'm quite comfortable in the role of country-cousin blogger with a European perspective. The other specialised skill I learnt was making toasts with a moral in the course of drunken Caucasian banquets. I'm open to expenses-paid offers to retell Noah the great Armenian and Columbus, the orange, and university reform in Georgia. James Wimberley's occasional publications on the web

9 thoughts on “Worthwhile Ivy League stealth s*c**l*st initiative”

  1. "…which suggest the awful prospect of robots taking over much of the job. (They eventually will. Doctors can´t have it both ways. If medicine is a science, it can be specified as algorithms, and algorithms are best implemented in reliable and memorious silicon. … )

    You are right in the specifics but not in the general point. It is absolutely true that computer systems are becoming incrementally "smarter" (in some sense) all the time; and that even work that we normally think of as pretty high-powered and professional will be increasingly automated. Moreover some aspects of medicine (diagnostics) are likely to prove particularly amenable to data-centric, statistical methods. However this doesn't really have anything to do with medicine as a science or whether science — or more accurately scientific practice — can be specified as algorithms. If anything could be specified as algorithms it would be mathematics — but theorem-proving technology hasn't put mathematicians out of business yet and it isn't likely to for a long time.

  2. Larry,

    Theorem proving machinery will never replace mathematicians. Goedel's Theorem shows that there are true propositions that cannot be proven within the system. An automated proof machine may prove some theorems, but proof of other theorems seems to require insight into the the system.

    Similarly, diagnostic decision support will provide fast updating via Bayes' Theorem. But it will still require a human to assess and make sense of those posterior probabilities.

  3. I'm sure there's actually a much *stronger* representation of patients than of medics, if you stop to think about it.

  4. Dennis, please note that I was not disputing James's conclusion (that powerful decision aids are here or soon will be) but his argument (medicine is science, and science is algorithmicizable, therefore medicine is algorithmicizable). It seems to me that your comment in fact makes exactly the same point, that this isn't a very good argument.

    I should add that the argument against AI via Godel's theorem has never seemed very convincing to me. What needs mechanizing isn't mathematics but the processes of a mathematician (a distinction I alluded to in my initial comment). But of course human mathematicians aren't infallible. Computerized mathematicians are also unlikely to be so. In other words, forget completeness, they won't even be sound.

  5. The argument about Gödel and mathematics is interesting but is it relevant? Mathematics isn´t a science in the Baconian sense of a system of structured and verifiable knowledge about nature. (I´ll grant you that Platonist mathematicians think of their art as discovering things about an invisible objective reality, but constructivism works just as well as a justification for the activity.) The question is whether you can reduce medical diagnosis and treatment to an algorithm: if symptoms a1, a2, then possible causes b1, b2; eliminate causes by tests c1, c2..; then unique diagnosis d. If d, and patient´s other characteristics e1, e2…, then try in order treatments f1, f2.

    Am I wrong in thinking the agenda of scientific medicine is to codify and update such algorithms? And if you have such algorithms, why can´t they be automated? I´m sure that there are limits to this agenda. But they are limits to the whole reductionist approach – for instance, around the patient´s faith in the healer – not the need to insert an expensively trained gatekeeper within it. We have driverless trains already (eg in the Lille metro). There comes a point in the development of automation when including a fallible human in the loop adds risk rather than reducing it.

  6. James, I apologize if I wasn't clear in my initial comment. I believe you are entirely correct as far as where technology is going. It won't be too long before people (not to mention insurance companies) insist that a computerized system be in the loop. And beyond trains, within a couple of decades it will seem incredible that people took the risk of driving in, or even walking near roads travelled by, cars controlled by human beings. However I just don't see any of this depending on the "grand" question of whether science (or any other human endeavor) can be, will be, or should be entirely "reduced" to algorithms. Like a lot of other grand questions I think that's just going to turn out to have been hopelessly vague or wrongly posed.

  7. Simply getting a computer in the loop as a memory prosthetic will be an enormous improvement. Doctors can't know the current state of the art for every ailment they see. So a simple "That test will have diagnostic value only if you suspect XYZ" or "That medication is better than placebo only in the following subsets of the population; are you sure you want to prescribe it?" could have a substantial impact on treatment and cost.

    Meanwhile, I'm not surprised at the omission of silos from early stages of the process. First, data warehousing works best when you already have a good process for feeding your repository. Second, a big part of the initial buy-in from doctors will be allaying fears of losing control of patient information (more than already happens).

Comments are closed.