In addition to its vulnerability to spoofing, for example, there is its gross inefficiency. “For a child to learn to recognize a cow,” says Hinton, “it’s not like their mother needs to say ‘cow’ 10,000 times”—a number that’s often required for deep-learning systems. Humans generally learn new concepts from just one or two examples.
There is a nice review on Deep Learning in PNAS. The spoofing referred to, is an ‘adversarial patch’ — a patch comprising an image of something else. In the example here, a mini-image of a toaster confuses the AI such that a very large banana is seen as a toaster (the paper is here on arXiv — an image is worth more than a thousand of my words).
Hinton, one of the giants of this field, is of course referring to Plato’s problem: how can we know so much given so little (input). From the dermatology perspective, the humans may still be smarter than the current machines in the real world, but pace Hinton our training sets need not be so large. But they do need to be a lot larger than n=2. The great achievement of the 19th century clinician masters was to be able to create concepts that gathered together disparate appearances, under one ‘concept’. Remember the mantra: there is no one-to-one correspondence between diagnosis and appearance. The second problem with humans is that they need continued (and structured) practice: the natural state of clinical skills is to get worse in the absence of continued reinforcement. Entropy rules.
Will things change? Yes, but radiology will fall first, then ‘lesions’ (tumours), and then rashes — the latter I suspect after entropy has had its way with me.
General practice has been undergoing a quiet revolution in recent years that has had little fanfare: it is now an overwhelmingly part-time profession.
Official figures suggest almost 70% of the workforce work less than full time in general practice – the highest proportion ever.
This is from an interview with Geoffrey Hinton who — to paraphrase Peter Medawar’s comments about Jim Watson — has something to be clever about. The article is worth reading in full, but here are a few snippets.
Now if you send in a paper that has a radically new idea, there’s no chance in hell it will get accepted, because it’s going to get some junior reviewer who doesn’t understand it. Or it’s going to get a senior reviewer who’s trying to review too many papers and doesn’t understand it first time round and assumes it must be nonsense. Anything that makes the brain hurt is not going to get accepted. And I think that’s really bad…
What we should be going for, particularly in the basic science conferences, is radically new ideas. Because we know a radically new idea in the long run is going to be much more influential than a tiny improvement. That’s I think the main downside of the fact that we’ve got this inversion now, where you’ve got a few senior guys and a gazillion young guys.
I would make a few comments:
All has been said before, I know, but no apology will be forthcoming.
This is from an article by Stephen Senn in Nature. He keeps making this point — for the very good reason that people want to pretend there is no problem. But there is.
Personalized medicine aims to match individuals with the therapy that is best suited to them and their condition. Advocates proclaim the potential of this approach to improve treatment outcomes by pointing to statistics about how most drugs — for conditions ranging from arthritis to heartburn — do not work for most people. That might or might not be true, but the statistics are being misinterpreted. There is no reason to think that a drug that shows itself to be marginally effective in a general population is simply in want of an appropriate subpopulation in which it will perform spectacularly.
When you treat patients with chronic diseases such as psoriasis, it quickly becomes clear that there is considerable within person variation is response to treatments. We do not understand what this variation is due to. What we do know however, is that assuming variation in response between people at single time points may be misleading in that we have no measure of within person variance. This is only one of the problems. But hey, precision, personalised.. whatever: it shifts units (as Frank Zappa once said of Michael Jackson).
This is from a book review in the FT of American Overdose. by Chris McGreal — prescription for carnage.
McGreal has written an interview-based book, with especially vivid reporting from West Virginia, the state hit hardest by the epidemic. In the little town of Williamson, or Pilliamson as people came to call it, pharmacies were dispensing opioids at a staggering rate both to locals and to out-of-state visitors, who clogged its streets with their cars but boosted some local businesses as well as city tax revenues.
When the federal authorities belatedly raided one Williamson clinic in late 2009, they found that an individual doctor had written 355,132 opioid prescriptions over the previous seven years — about 1,000 for every inhabitant of the town. Another wrote 118,443 scrips over the same period. Most were handed out for cash fees without the doctors bothering to see their patients. The investigators estimated that the clinic took in $4.6m cash during 2009 and they found banknotes stuffed into safes and cupboards in the doctors’ homes and offices.
At my old university, we were encouraged to explore our subjects and to love what we were studying. Now, at medical school, the emphasis seems to be don’t burnout, focus on not making mistakes, and understand that life is going to be hard, so develop the resilience to cope.
The above is from a letter to this month’s Academic Medicine [83(12) 1745-1884, 2018] written by a graduate student at Warwick medical school (TC Shortland). The title is what caught my eye: “Enjoying, and Not Just Surviving, Medical School”
He goes on:
At Warwick Medical School, staff and students are trying to build a more positive environment. Staff and students have organized art classes, interstaff/ student sports events, and several baking competitions; the last winner featured cupcakes that could be injected with either a salted caramel or raspberry filling. As positive health care workplaces and positive cultures are associated with better patient outcomes,why shouldn’t medical schools try and create such environments for future medical professionals?
I am not against the various suggestions (…well, I am actually), but what I and others are in despair about is how much (?most) medical education has become so dull, tedious, and brutal, rather than humane. When I have spoken to others, some hold similar views: the students put up with it, because they want to be doctors, but they no not enjoy most of it. If they are obliged to attend, they do; but out of choice, many would skip much of what we offer.
Now this is not a new thought or phenomenon. I didn’t enjoy — in fact I actively hated — the preclinical years (aka: the prescientific years) — but I did get a big kick out of the clinical years, and loved my intercalated degree. What made the clinical years work, was that the opportunity for some kind of personal bond with some teaching staff made up for all the despots and dull souls who should have been destined to be gravediggers. And unless somebody has recently discovered something I have missed, scale and intimacy rarely go together.
Of course, what makes matters worse, is that the ennui and anomie will get worse: for many junior doctors, after the initial high of being qualified, their working jobs are miserable. If they get to higher training, things may improve, but not for all.
George Steiner’s comments in a slightly different context are apposite:
“Bad teaching is, almost literally, murderous and, metaphorically, a sin. It diminishes the student, it reduces to gray inanity the subject being presented. It drips into the child’s or the adult’s sensibility that most corrosive of acids, boredom, the marsh gas of ennui.”
The NHS (for this is the fault of the NHS rather thant the universities) is accumulating a massive moral debt, borrowing on the very market it has rigged (because it can!), forgetting that this is like PFI on steroids. It assumes it is too big to fail: I think otherwise.
How did you go bankrupt: slowly and then suddenly.
This is from an editorial in the NEJM, discussing the results of a trial of a synthetic peanut antigen to facilitate tolerance. Prevously the ‘raw’ stuff had been shown to be useful. The synethic version will of course cost a lot, and might be considered IPR created through regulatory arbitrage.
AR101 and other, similar products such as CA002, which is being developed by the Cambridge group, would therefore appear to have a role in initial dose escalation. The potential market for these products is believed to be billions of dollars. It is perhaps salutary to consider that in the study conducted by the Cambridge group, children underwent desensitization with a bag of peanut flour costing peanuts.
Costing penauts: I wish I had said that
Over 40 years ago, I remember lectures on basic demography, and what changes to the population mix of the UK were likely. The late John Grimley Evans was one lecturer, but others went over cognate themes including Klaus Bergmann who pointed out the relevance of increased mobility when considering family and / or social support. ‘Were we prepared to adopt each other people’s grannies?’, stuck in my mind. If you look at the family structures in some working class areas (think parts of Newcastle), and how they have changed, it was obvious what might happen. The sociologist Peter Townsend published studies based in London, highlighting the depth of community support that once existed.
Which is why I get so cross when you come across this as a new problem, as though all these baby boomers (like me) were born yesterday. As though we didn’t know; as though nothing could have been done to prepare for such a change. Remember too, when to study dementia or stroke, or seek funds for such topics was……well futile.
The following is from an article in the FT (How Britain can heal its ailing social care system | Financial Times). Once again the dismal leadership of the UK compares unfavourably with that of some other European countries
The pressure on staff is becoming unbearable. Good social care is about creating relationships. Staff who work for domiciliary care operations like those run by Allied Healthcare must walk into the home of a stranger, reassure them, figure out what is needed, and build trust. That takes maturity, emotional resilience and time. Yet, all too often, the reality is rushed visits from a plethora of different faces.
A few years ago, I met an 89-year-old man who had made a note of every carer who had crossed his threshold in the past year — Meals on Wheels, district nurse, domiciliary care staff. He showed me the list. There were 102 names on it. Some had only come once, then vanished — probably into better paying jobs at a supermarket. That is the stark reality of how little we value our elderly.
I do not have a coherent overview of many of the traditional professions, but I wonder if people will soon say similar things about doctors.[Link]
“The big issue that concerns me at the moment in the English education system is the supply of high-quality teachers. We’ve seen quality issues in recruitment to teaching and our schools are getting increasingly desperate to find decent teachers. The whole workload issue has come to a big head again in England with teachers having very big workloads and their conditions of service is deteriorating a lot recently. We’re seeing a big exodus in teaching and so of course, we need a bigger inflow to maintain the balance.”
Yet despite these innovations and those to come, quantitative risk prediction in medicine has been available for several decades, based on more classical statistical learning from more structured data sources. Despite reports that risk models outperform physicians in prognostic accuracy, application in actual clinical practice remains limited.
It seems unlikely that incremental improvements in discriminative performance of the kind typically demonstrated in machine learning research will ultimately drive a major shift in clinical care. In this Viewpoint, we describe 4 major barriers to useful risk prediction that may not be easily overcome by new methods in machine learning and, in some instances, may be more difficult to overcome in the era of big data.
The hype cycle marches on.
When working in Africa in the 1980s with my good friend Victor Pretorius, I heard a legend about an important tribe in Central Africa, the Masai. The legend claimed that a genius member of the tribe in the nineteenth century or earlier had the idea that cow’s urine was the safest fluid for washing cooking utensils. Compared with the previous practice of using far from clean river water, it avoided the dangers of dysentery and probably saved many lives. This simple and effective public heath practice was cast out by medical missionaries who had quite different ideas, more religious than medical, about what was clean and what was dirty. Neither the original genius, nor the missionaries, knew anything about the epidemiology of water-borne disease. Whether or not there is any substance to this legend, it has stayed in my mind as a metaphor appropriate for many of our problems today. Inventions such as Newcomen’s steam engine, Faraday’s electrical machines, and the idea that fresh urine is a sterile fluid, all came long before their scientific understanding.
James Lovelock, A Rough Ride to the Future. This is like so much of real discovery in clinical medicine, although the academy gets to write the history of how it is supposed to work.
If a doctor expressed concern about a patient showing signs of addiction, Ms Panara was trained to counter those fears by educating them on so-called pseudo-addiction, she says. For example, an addict might turn up at the surgery requesting a fresh batch of pills before their 30-day supply should have run out, claiming they had lost the tablets or accidentally dropped them down the toilet. The advice that she was told to give the doctor was that the patient’s dosing was too low and should be increased, she says.
“The theory of pseudo-addiction was that a patient might exhibit these drug-seeking behaviours, but if their pain were adequately managed by giving a higher dose, then that drug-seeking behaviour would cease,” she says. “Thereby we were building their tolerance, building their physical dependence, and making them an addict.”
I still fail to see why we need drug representatives, nor why they are allowed.
For a baseline life expectancy of 80 years:
Well these are all taken from John Ioannadis’ article in JAMA. He asks : “Could these results possibly be true?”
The great financial crash led to some (but not enough) soul-searching about the state of academic economics and, in turn, the academy. Whole swathes of the modern research university are geared to the production of unreliable knowledge. There is money in it. Without wishing to understate in any way Ioannadis’ major contributions, we have known that there are fundamental methodological flaws in much of observational epidemiology for a long time (for instance see the late Alvan Feinstein’s article in Science). A must read.
(The Challenge of Reforming Nutritional Epidemiologic Research John P. A. Ioannidis, AMA. Published online August 23, 2018. doi:10.1001/jama.2018.11025)
It is not only taxi drivers that are being “uberised” but radiologists, lawyers, contractors and accountants. All these services can now be accessed at cut rates via platforms.
The NHS became such a platform, for good and bad. That is the real lesson here. The tech is an amplifier, but the fundamentals were always about power.
Those who rent seek on biomedical knowledge wish to seek to define the norms of what is foundational. What is foundational for the practice of medicine should be contested more. Anatomy for surgeons is an easy case to make. But for most non-surgeons, the case for much anatomy is far from simple.
In any historical account of the ascent of modern medicine, Versalius looms large. But this Nature article (Sex, religion and a towering treatise on anatomy) intrigues me for a not so obvious reason: the counterpoint between how such knowledge was represented and understood.
Even Vesalius realized that his images could be confusing, and devised an ingenious method to explain them. A letter or number was printed onto the image of each body part, with a separate key. Unfortunately, the characters were often too small to pick out against the swirling background….
Faced by such challenges, many medics might have given up on the images. Indeed, when we reconstructed what early modern readers and scholars found fascinating about the Fabrica, it was evidently the text. The clear majority of sixteenth- and seventeenth-century readers who annotated the book focused on that and left no traces of having engaged with the illustrations. Sixteenth-century reviews of the Fabrica confirm this impression, because they tended to discuss only the text.
This is no surprise. The Fabrica’s scholarly readership was trained in the traditions of Renaissance humanism, which put a strong emphasis on textual analysis. Even if they found it difficult to interpret visual information, medical practitioners were expert at making sense of long Latin texts.
Davidson had had enough. “I wasn’t even making 6 figures, and I was killing myself,” she recalled.
“Frustrated, she googled “ideal practice” one sleepless night and came across Atlas MD in Wichita, Kansas. That practice does not accept insurance, although patients still need to have insurance to cover health care beyond the scope of primary care. Instead of co-payments and deductibles, Atlas MD patients pay a monthly “membership fee” that covers all of the primary care their physician provides. But more importantly, this retainer guarantees unhurried, same-day appointments and round-the-clock accessibility to their physician, who would get to know their story “inside and out,” thanks to having to care for only around 500 patients.”
The most interesting thing in this article is the lengths the opponents go to to oppose such a change:
“For now, said Weisbart, chair of the Missouri chapter of Physicians for a National Health Program, there’s no evidence to support the argument that DPC [the model describe here], by allowing physicians to spend more time with patients, can prevent expensive downstream medical problems. “If they could prove it, I’d be one of their advocates,” he said, adding that he understands the attraction of DPC for physicians. “They can see a third or a quarter of the number of patients (as fee-for-service practices) and preserve their income.”
“Weisbart remains skeptical, though. Direct primary care practices might attract a different population of patients. The only way to compare how well the 2 models improve health and cut costs would be to conduct a trial that randomly assigned patients to DPC or fee-for-service practices. But, Weisbart added, such a trial would be difficult if not impossible to conduct. For one, it’s unlikely that a representative sample of US patients would agree to enroll in a study in which they were randomly assigned to a primary care physician. “And,” he added, “it would have to be large to show meaningful impact, which means the study would be expensive.”
Polling undertaken for the NHS recruitment campaign found that many people had an outdated view of nursing. It suggested most saw nurses primarily as “caring”, with far fewer regarding them as “leaders” or “innovators”.
Apparently this is not approved of. My mother would be turning over in her grave.
Did the NHS save your life, or did Doctors and Nurses save your life?
It’s an earnest question. A comment on an excellent FT piece: “Is Britain loving the NHS to death?”
I posted this awhile back, but it still makes me smile. I wrote:
Well my knowledge of Neanderthals is rather limited to the work showing that some of them would likely had red hair. But now a reviewer (Clive Gamble) in Nature of a book on Neanderthals states that
Wynn and Coolidge conclude that today, Neanderthals would be commercial fishermen or mechanics, based on their enormous strength and ability to learn the motor procedures needed. Their capacity for empathy might even have made them competent physicians, the authors say, although a lack of mathematical ability means that they would never have been able to graduate from medical school. Neanderthals would also make excellent army grunts, with their high levels of pain tolerance, and would be good tacticians in small combat units. They would never rewrite the tactical manual — although tearing it up, however thick, would not be a problem.
“It’s quite obvious that we should stop training radiologists,” said Geoffrey Hinton, an AI luminary, in 2016. In November Andrew Ng, another superstar researcher, when discussing AI’s ability to diagnose pneumonia from chest X-rays, wondered whether “radiologists should be worried about their jobs”. Given how widely applicable machine learning seems to be, such pronouncements are bound to alarm white-collar workers, from engineers to lawyers.
The Economist’s view is (rightly) more nuanced than Hinton’s statement on this topic might suggest, but this is real. For my own branch of clinical medicine, too. The interesting thing for those concerned with medical education is whether we will see the equivalent of the Osborne effect (and I don’t mean that Osborne effect).
In Britain, an Epipen — a simple device that saves lives in the case of severe allergic reactions — costs $70. In France and Germany, roughly the same. In America, it costs $600. But in 2007, it cost in America what it did in Britain, France, and Germany. What happened? A drug company called Mylan bought the rights to it — and then it didn’t just send prices soaring, it uses all kinds of shady tactics to maximize profits from insurance companies and healthcare systems both. How?
Well, what does it cost to “make” an Epipen? Not a whole lot. It’s just a device for delivering a dose of epinephrine. The dose used in it “costs” maybe $1. I put “cost” in quotes because even those numbers are mostly fictional — the marginal cost of producing a basic chemical like this is pennies. In fact, the real problem is that epinpehrine became too cheap to manufacture — so cheap that many producers stopped making it altogether. And so a company like Mylan swooped in, put two and two together: corner the supply, gain a monopoly on the demand side, and hey presto — mega profits.
That’s predatory capitalism — a drug that should cost pennies, if the economy were run a little more sanely, costs hundreds, without any regard for the human possibility that is destroyed. Mylan didn’t create any real value whatsoever, only extracted it, siphoned it off.
Yep. Wealth creators, and wealth aggregators. Profit harvesting mode.
Well, this was the modest description of a ‘new’ way to test blood. Except it wasn’t. The reality distortion field in hyperspace. If you don’t know the Theranos story — or doubt the importance of real journalism — have a look.
Science 21 June 2013: 1394-1399.
For most alumni, university fundraising may seem to be uncoordinated and lacking in focus—an assortment of phone calls, solicitous letters, and invitations to a class reunion. But for Steven Rum, it’s a science. And the goal is to carry out more research.
Rum is senior vice president for development and chief fundraiser for Johns Hopkins Medicine in Baltimore, Maryland. Last year, his team had a banner year, raising $318 million. Their approach places the physician scientists at Hopkins on the donor front lines. The goal is to turn the positive feelings of “grateful patients” into support for new research, faculty chairs, academic scholarships, bricks and mortar, or simply defraying the cost of running a multibillion-dollar medical center.
Rum has 65 full-time fundraisers on a staff of 165. Each one is responsible for meeting weekly with physicians—their “caseloads” range from a dozen to more than 30 docs—to discuss which of their patients might be potential donors. The conversation is designed to help them identify what Rum calls a donor’s “qualifying interest” and connect it to their “capacity,” that is, the ability to make a donation.
More often than not, Rum’s team finds that sweet spot…..
”Ideally, I’d like to have one gift officer manage no more than six doctors,” he says.
A beautifully written vignette in the NEJM by Abigail Luger
Now I’m the one contemplating a permanent departure. My health is fine, but my stamina is pretty much gone. Our health care system is not kind to the chronically ill and marginally insured, and it is not particularly kind to their doctors, either. Our patients are condemned to an unending swim against a hostile tide. Doctors can head for shore.
My third issue is more nuanced. The biggest reason for cataract blindness is the dearth of ophthalmologists. Orthopaedic surgeons in Leicester faced with a backlog of carpal tunnel surgery decided to train intensively one theatre nurse. As a result, our carpal tunnel surgery service is probably the best and most cost effective in the country. Having a person who is not a “fully qualified doctor and surgeon” operating on cataracts could be the best solution.
Quote from, John Sandford-Smith, retired ophthalmologist, Leicester. BMJ 2018;360:k640
This sort of argument is old, and largely correct. But you can only scale such a process with the help of some form of certification, because without it, there is no durable career structure. And without this, there is no investment.
Article in Nature. I largely agree, although my views are as much based on the hype-upon-hype that characterises so much of medical research, especially cancer. I do not have a reference, but whatever one’s views about the late David Horrobin, his Lancet article about cancer trials — written when he was dying from lymphoma — is worth a read. What a mess!
Key quotes from this article:
In 2017, my colleagues and I completed a study of all 48 cancer drugs approved by the European Medicines Agency between 2009 and 2013 (C. Davis et al. Br. Med. J. 359, j4530; 2017). Of the 68 clinical indications for these drugs (reasons to use a particular drug on a patient), only 24 (35%) demonstrated evidence of a survival benefit at the time of approval. Even fewer provided evidence of an improved quality of life for symptoms such as pain, tiredness and loss of appetite (7 trials; 10%). Most indications (36 of 68) still lacked such evidence three or more years after approval. Other groups in other regions have observed similar trends. For example, a 2015 study demonstrated that only a small proportion of cancer drugs approved by the FDA improved survival or quality of life (C. Kim and V. Prasad JAMA Intern. Med. 175, 1992–1994; 2015).
But the key point he makes is:
I believe that the low bar also undermines innovation and wastes money.
When assessments — whether in medicine or education — are flawed the loss in value is not in short term financial costs, but in what might have happened 10 years down the road.
“Why do doctors feel the need to do this? A study in The BMJ in 2015 suggested that there is an association between increased defensive practice and a reduced likelihood of being involved in litigation.2 One might conclude that defensive practice is a logical behaviour in the face of a culture that leads to doctors being fearful of the consequences of making an error or even of a known adverse outcome.
“No doctor sets out to practise defensively, but a system has been created where this is inevitable. The GMC acknowledges that medicine has become more defensive.3 Doctors often lack confidence in the fairness and competence of investigations and continue to see the GMC as threatening.”
Tom Bourne quoted in the BMJ
A “world renowned” consultant radiologist has been suspended from the medical register for six months after being convicted of failing to pay more than £400,000 in tax on his private practice earnings. XXX a professor of cardiovascular imaging, was given a 15 month suspended prison sentence and fined £200,000 in 2015. The GMC argued that he should be erased from the register. But the medical practitioners tribunal concluded that he made a unique contribution to the care of critically ill patients and the development of cutting edge techniques.
Bmj September 30 , 464, 2017
Modern medicine is full of doctors who are already robotic. But good medicine is like any other subject that requires one to make judgements based on prior cases. You can follow rules, or you can think. The first kidney doctor (who I saw multiple times) never bothered to actually think.
This article is about truck drivers in the US, where getting a license requires a big investment. The article states that there is now a shortage of truck drivers and argues that this may be a result of the ‘inevitable’ rise of automated drivers.
Many young people are reluctant to pay $5,000-$10,000 to learn to drive an 18-wheeler at a time when experts are predicting that it is a dead end career.
This is akin to what I wrote about before. When people know that a line of work is ‘unlikely’ to continue, they are prone not to want to invest in it. It is not too fanciful to see this happening in medicine. Start with imaging specialties first; and then look at the use of paramedical staff in restricted clinical domains.