Quantum Quandaries, Diagnostic Decoherence and Probabilisticians

So I imagine you have just read the title of this post and thought – “Casey has finally lost it!” – or something to that effect….

Diagnostic Decoherence – it sounds like the title of an obscure SMACC lecture.

So what am I rambling on about – what is this concept? Where does it come from? And why do you care?

Let me explain.

At SMACC 2013 I had the pleasure of meeting Prof. Simon Carley – a man with a brain the size of a small planet.  One evening over a beer he stumped me with a statistical problem – the Monty Hall paradox

It is a simple problem that demonstrates our often-flawed intuition when it comes to probability and decision-making.  And even smart, experienced clinicians get it wrong a lot of the time.   I recall walking away from that conversation feeling like a small part of the rug had been pulled from under my clinical gestalt.

The next morning Simon gave a great lecture on the real world application of statistics at the bedside – this talk was called “Wrestling with Risk” – you can see it on the 2013  SMACC podcast.  Free, awesomeness.

Simon’s talk was great – a beautiful illustration of  balancing risk of missing vs. over investigation and diagnosis that leads to unnecessary intervention.  It is all about understanding the imperfection of our science and being able to think past the “results” of tests. We need to be thinking about what the results really mean and how they apply to the individual in your care.  IN recent months Simon has teamed up with Dr Ian Beardsell [@docib] to produce the St. Emlyn’s podcast – the first few episodes go right to the heart of probability and testing in ED.

Simon used the PE workup as an example – it is one of those diagnoses where we have good data – well-defined pretest risk assessment tools and calculated likelihood ratios for our “diagnostic tests”.  The problems we see with PE workup come from the common misunderstanding many clinicians have about the use of these tests.  For example, performing a CTPA on a Well’s Score low-risk patient is more  likely to yield a false positive over a true positive if reported as “PE”.

OK, great.  We need to teach our students and trainees all these stats – educate them about test characteristics and application to various populations.  If they knew all the LRs, scores and risks then they could make statistically-sensible decisions.  Less harms, less false diagnoses and probably a swag of money saved.

But…. here’s the problem.  We are still practicing in the Wild west of science. There are not a lot of rules, plenty of unknown unknowns and yet we need to get the job done – for most of what we do there is just not a great evidence base.  Even when we do have evidence – it is often of questionable quality or cannot be applied to our patients’ particular context.  Would lead many of us to EBM-nihlism or drink!

Ever since that talk I have been thinking – contemplating and trying to come up with a counterpoint.  Examining the way I use probability in daily practice and searching for a good analogy in another field of science.  And I think I have found it!  So here it goes…. but first a quick disclaimer.

I, Dr Casey Parker, am a total physics nerd and make no apology for this.  So the following rant makes complete sense to me and maybe a few other people I know!  This next bit is very heavy on non-medical, theoretical ideas which I believe are a nice analogy to the conundrum posed by Prof. Carley.  If this type of thing bores you to tears – stop reading now! OK, still there? Begin…

Einstein is famously quoted as stating that “God does not play dice with the Universe.”  Possibly the most famous quote from the great man – and possibly the most flawed!  Einstein was expressing his incredulity at the emerging quantum physics of the early 20th Century – he felt fundamentally uneasy with the idea that probabilistic mechanisms determined the nature of the universe.  Specifically the Copenhagen interpretation stated that: “A system is completely described by a [probability] wave function , representing the state of the system, which evolves smoothly in time, except when a measurement is made, at which point it instantaneously collapses to an eigenstate of the observable that is measured.”

Here is where the infamous thought experiment of Schrodinger’s cat comes in (or doesn’t… if it is dead ;-).  The cat exists in a state where it is both alive and dead (probabilistically) up until the point that the box is opened and an observation is made.

So when you are working up a lowish-risk PE patient – you might have a Well’s score, a modified Geneva or your trusty clinical gestalt.  You plug it into MDCalc and you get a probability of a PE being present.  The D-dimer is done… DOH!… it is “not negative” and you are unable to say “go home!”   So you decide to make an ‘observation’ ie. a CTPA – which really just changes the probability.  There is no waveform collapse, no eigenstate, no.   Medicine is not like physics – there is no absolute.

But, here is the but…  at some point you have to make a call.  We work in the real world – you cannot go to the patient and tell them that they have a 73.2% PE.  You cannot give them a 73 % dose of heparin? We need to be able to to decide who gets treated and who does not.  This is analogous to the classical-quantum divide:  we know that particles are composed of probability waves, yet when you look around you – you see objects, collisions, hard surfaces all of a predictable nature.  When your patient falls – their radius breaks, when the clot sticks, they get ST elevation.  So how can we reconcile this in our practice.

For the vast majority of diseases or problems that we see in the ED there are no numbers, no starting position – we are sailing in uncharted waters and would be lost.  I say “would”.  Because we do have our own personal onboard navigation systems – we all carry a rich library of experience and “gut instinct” around in our heads.  This is what we call “Gestalt” – that oft spoken German word which is rarely understood.  So how does this work?

Back to the quantum world… For most of the 20th century theoretical physics was stuck with a big headache.  On one hand there was a great theory that explained the big things – Einstein’s relativity.  And then there was the new kid on the block – quantum theory which used probability and wave equations to describe and predict the tiny world of the fundamental particles.  However, there was a demarcation dispute.  At some scale the particles had to start behaving like classical physical objects.  But there was no way to marry these two perfectly functional theories in such a way that made any sense.  This is where Decoherence comes in.

When we take a history and examine a patient – we are building up a set of probabilistic waves.  Each piece of data increases or decreases the amplitude of the wave.  Each wave carries the probability of a diagnosis – and to reach the threshold of action / reality / or the “treatment threshold” the wave must break over the wall.  Stick with me – it gets clearer soon.

Simon has beautifully described how various factors create waves of various amplitude in the work up of chest pain.  Check out his SMACC Gold lecture from 2014 – on risk factors.  When we take the history from a chest pain patient – we know that the typical atherosclerosis risk factors are poor predictors of acute coronary syndromes.  In other words – they produce only small waves, they are subtle ripples in the ACS pond.  In order to get over the threshold – we need a collection of waves which positively interfere with one another in order to make it to an “actionable” probability.  Chest pain is easy – we have good data to measure the waves and the amplitude of the threshold.  Alas, when you are floating on the “non-specific abdo pain pond” the waters are choppy.  Lots of waves interfering with one another – leaving us with murky water.  That is why these patients seem tougher to sort out.  And by the way – no matter what the Surg Reg tells you – the white cell count is merely a ripple on that pond!

Now of course, some of our “tests” create big waves – think about “thunderclap headaches”, “shingles rash” or a “3 mm of ST elevation pattern in II, III and aVF”.  These features create waves that easily carry us over the “action threshold”.  This is the stuff we all learn in Med School.  Sometimes it is wrong – for example – Framingham risk factors and ACS – no real wave in the ACS setting.  However, more often it is just plain messy.

When the quantum physicists look at probability waves – they isolate a couple of particles and do complex probability wave calculations  – very smart.  But if you try and do this with more than 3 particles (waves) at a time – the numbers get crazily complex – it is just too messy.  And yet we all do this intuitively on a daily basis.   One of my mentors believed in “Diagnostic Triads” – the idea that many diseases could be reduced to 3 symptoms – this is an example of three simple probability waves interfering to create an ‘eigendisease’ -e.g.  fever, jaundice plus (R)UQ pain = cholangitis.  Unfortunately during our daily grind in the ED the act of taking a clinical history and exam is an infinitely complex assessment in which we are constantly integrating new data and probability waves into the mix.  At some point the waves may coalesce into a meaningful reality – and this is what a quantum physicist might describe as Decoherence.  The waves become a real thing – something that we can name, refer, incise or thromobolyse.

So in summary – I agree with Simon. [ ‘twould be daft not to! ]  We are probabilisticians rather than diagnosticians.  However, the math is far more complex than our textbooks and teachers would have us believe.   Medical students have always struggled with the complexity of clinical diagnosis.  There are just too many variables for even the brightest minds to integrate into a cohesive model of what is going on in most patients.  Of course – spending more time, doing more tests can give us more data points and help us “realise” a diagnosis.  And as we gain experience we learn to recognise “disease scripts” – patterns in the waves that we recognise as “a thing” before they reach the threshold.

Now – one might read all of this and get a sense of nihilism.  It all just seems too complex and hard.  But I think there is a way forward.  Here is what I believe we should do:

1.  We need more research to define the probabilities – clinical medicine is full of dogmatic belief in our clinical skills.  E.g.  “the chest is clear – therefore no pneumonia….”  In reality the negative LR for auscultation in pneumonia is close to 1.0.  Utterly useless!  By knowing and teaching the reality of our skills we can do better, make more reasoned decisions.

2.  We need to teach probability to Medical Students – personally I am a fan of the Bayesian approach and teach likelihood ratios as the most easily applied stats to keep in the back of one’s head.  These are more intuitive than sensitivity or specificity.  They tell you more about the “test” and can be applied directly tot he individual patient in front of you.

3.  We need to change the language that we use when we talk to our patients.  A while back I defined the Zeroth Law of Diagnostics – “there is no such thing as Zero risk”.  And I believe that we ought to fess up to our patients – certainty is a rare beast in Medicine.  We should tell our patients this.  We can have a reasonable belief that they do not have a PE, or that they do have ‘smouldering diverticulitis’ – however, there is a real chance that we are wrong.  If things change, or they get new symptoms [waves – new waves ] then they need a review and rethink.  The catch phrase for this is = “shared decision-making”… or as we in GP-land call it : consultation.  The patient is the one with the disease [or not] and they take the risk.

4.  We need a bit of Dogmalysis – there are a lot of pseudo axioms and long-held truths in our game that need re-examination.  David Newman has produced a nice series of podcasts on this topic – check it out at SMART EM.

5.  We need to develop a healthy insight into the way we think – I believe the buzz word is metacognition.  Understanding the sources of error and recognising the false waves will help us to make better, more accurate decisions – and to do better by our patients.

Ok.  That was quite a ramble.  Back to something more concrete soon!   Or is that concrete really just the summation of a pile of waves of probable concrete particles ?

Let me know what you think.

Andif you want to read more about Decoherence in quantum physics – I recommend reading this quick review paper.  For more reading check out the Decoherence website here.  Or just go outside and get some sun 😉



  1. I have long thought that we should teach more history & philosophy of science / medicine in medical schools, but given that less than 1 in 3 interns I have surveyed this year know who Sir William Osler was, I don’t think this is going to be feasible.

    [I started writing a rant about Popperian diagnosis here, but I think I should finish it and put it on my own blog.]

  2. Pik Mukherji says

    Nice- some of the physics analogy seems stretched, but I can see it. Probability waves “interfering” to create a diagnosis is a bit much though. We’d consider this a constellation of symptoms that is recognizable as a syndrome, or perhaps a differential diagnosis of 3 separate symptoms that each included a common diagnosis that then is elevated to the top of the queue.

    I’d rather talk in terms of signal, noise and “weighting.” We interpret lots of symptoms, physical findings and tests without much attention to whether they are signal or noise. WBC? Noise. Or ripple, or LR near 1, whatever you like.
    When it’s more than noise (LR=1) and less than signal (LR>10), then weighting comes into play where various data points (waves, if you like) carry differing heft. Probabilistic interpretation of these weightings is paramount to deciding risk of disease or likelihood of benefit to pursuing a diagnosis: achieving the “threshold.” For CP evaluation, weighting would look something like: EKG > Cardiac marker > HPI > PMH(classic risks)

    So I quite agree Bayesian analysis and realistic risk assessment is key. “Recognizing false waves” = eliminate noise. In the metacognition model that should help eliminate cognitive load due to fewer variables being considered, reducing sources of error.

    • Hi Pik
      I agree it is a well stretched metaphor. I accept that the physics isn’t exactly as described – but in my head it is a useful analogy. Of course, my head may not be a normal place!!

      Just another way of conceptualising the fluid nature of clinical practice – a call to move away from binary / concrete thinking. This is the source of a lot of error & bias in modern medicine IMHO

      Agree with your conclusions about noise : signal in practice. Hard to teach to interns & students. A graphic illustration using the wave metaphor may help some get there?

  3. Spot on mate

    I as inspired by Simon last year in Sydney, when I was pestering him about likelihood ratios and ‘why dont we teach this stuff in medical school’

    S in past 18/12 I’ve been trying to incorporate this into teaching medical students – sadly it seems too mindblowingly hard for them – actually, no, that’s unfair – they get the concept and are keen to apply…but sadly it seems a luxury when they face a barrage of OSCEs, minCEX, MCQs and SAQs where there has to be a ‘right answer’

    Seems that our training sets us up to accept as gospel ‘undeniable truths’ – then our later postgrad years helps to unwind this, often borne through the harsh tutor of experience.

    I;m gonna keep banging on about LRs and their ilk though, and accepting that the more I know, the less certain I am.


  4. Ron Cassano says

    Casey Casey!! It becomes clearer every day that you want to be a statistician or scientist or writer, not a doctor!! However an excellent piece of writing indeed.
    For the record, Einstein’s most famous quote was ” the definition of insanity is doing the same thing over and over again, the same way, and expecting a different result” ( i.e the way most Health Systems and Departments work). So certainly new teaching MAY produce better results and diagnoses. This is by no means certain however. In your rave, I feel you have forgotten the 2 most important things : the actual patient, and the actual doctor.
    It is easily observable in any busy ED that a majority of young doctors are much more interested in their social life, ( and maybe they should be), having ample breaks, getting out of the hospital exactly on time, than in learning even basic procedures, let alone probability theory and new formulas for treatment, ( except of course if they can use such a treatment formula to justify mistakes in a medico legal case)! So maybe the teaching and responsibility of young doctors should begin with what medicine is, and their very important role in it , and that they are dealing with another person’s life , and their families, not a set of symptoms or conditions or waves!! Otherwise it is highly likely that Gestalt will never develop in junior doctors going forward.
    Regarding “modern tertiary top of their field specialists” who work in the highest regarded hospitals in this country, and their highly trained senior registrars : in these famous institutions, probability algorithms are used and taught much more frequently, and in fact that is their claim to fame! It is with great disappointment that I have witnessed , over about 2 months of very frequent visits to such a hospital, that common sense and Gestalt are totally absent : the results are disastrous, leading to premature deaths in many patients; I could give examples, but I don’t see the point! So use scientific formulas by all means, to improve patient outcomes, but be very wary about thinking they hold the answer BEFORE listening to the patient ( history) and doing a focussed examination. My experience is that far more mistakes happen today, with new scientific formulas and uber scientific discussions, which are mainly ego driven, than happened 20 years ago!


    • Ron. Are you sure you don’t want a regular editorial job? Pay is terrible !
      I think you’ll like my SMACC lecture when it comes out: The history of Empathy in Medicine


    • Isn’t the whole point of internship to develop that gestalt? I think you’ll find that the best scoring and probability systems are derived from a detailed history and focussed examination, and that gestalt plays a much larger part in clinical synthesis than the actual score. As a case in point, when I was a (very) junior surgical registrar I used the Alvarado score religiously; now I probably subconsciously score it but rely on my impression to determine the management pathway that the patient goes down.

      Perhaps we can draw some inspiration from information theory – google Claude Shannon for a scientific giant – and start scoring tests by how many bits of information they give us – this is just another way of representing the change between pre and post test probabilities. It relates to likelihood ratios in the same way that ARR relates to RRR.

  5. David Berger says

    Commenting in the wake of the illustrious Ron Cassano is a daunting prospect, but here goes!

    There are a couple of confounding factors here, which aren’t mentioned.

    Firstly, risk sharing is difficult because most of the population is functionally innumerate when it comes to understanding and weighing numerical values of probability and risk. You can use aids such as smiley face nomograms — which are incredibly helpful, but woefully underemployed — but the problem remains. Combine this with an increasing aversion to acceptance of ANY risk at all within society and the thrust is towards ruling out diagnoses at all costs, even if it means unnecessary tests and a fog of unwanted, confusing data. People are simply not ready to accept the small risk of a bad event in the way they used to be. This influences the way we practise medicine profoundly.

    Secondly, as doctors we are all influenced hugely by recent or very significant bad events or unexpected outcomes or diagnoses. If I’ve recently made a ‘Golly, who would ever have thought that?!” diagnosis then I am much more likely to think of and try to rule out that diagnosis for the next few months, or maybe forever if it was surprising enough. That’s human nature, but makes for bad probabilistically-based decision-making and it’s hard to get past.

    In the end, if we deconstruct our decision-making process too much, we risk being very disappointed by how little of our reasoning seems to have a sound basis. So what are we left with? ‘Gestalt’, experience, gut feel, whatever you want to call it. That’s hard to define, but I can certainly identify pretty easily which of my fellow practitioners I would trust more and which less to look after me well. Now, I may be wrong, of course, but maybe that’s my gestalt which is lacking!

    (But seriously, perhaps an interesting way forward is to examine why doctors identify other doctors as good clinicians / decision-makers. It may uncover some of our subconscious decision-making processes and identify what we think is important.)

    • Ron Cassano says

      Well said Dave! We need to trust ourselves at least as much as any algorithm ! (And I’d be more than happy for you to look after my family!)

  6. Hi casey, just started listening to your pocast after working my way the Scott W and David N. Great topic this. Although Im an ED reg I am also GP Obs reg. The last year working in the community has really opened my eyes to medicine and what it really means to practice. I cant agree more with Ron above. In all the science is also the patient and the doctor, our limitations, expectations and autonomous thinking. The conveyer-belt approach does not fit into peoples lives and flexibility of thinking and acting is so individualized to both parties. Anyway mate, great stuff and thanks for frying my already fried ‘on call’ head this weekend!

Speak Your Mind