From the Editor
As artificial intelligence advances, what role will computers play in mental health care?
Today, computers touch practically every aspect of our lives – from suggesting books that may be of interest to us on Amazon to helping fly our planes to tropical destinations. But will computers soon help us with diagnosing and treating our patients? Will some parts of clinical medicine be replaced or assisted by computers?
This week, we look at a new paper from Acta Psychiatrica Scandinavica considering AI and care. University of New South Wales’ Professor Gordon Parker sees a role for computers to help humans with diagnosis – but not more. “[R]ather than seeking to develop a computer program that will have diagnostic superiority to an ace clinical psychiatrist, it may be more important to develop programs that complement the psychiatrist’s judgement.”
AI: The next great doctor – or just a pretty face?
And in the second selection, we look back, not forward, and consider the career and contributions of psychiatrist Ronald R. Fieve, who recently passed. Dr. Fieve’s work helped bring lithium to North America.
AI and Psychiatry
“Thinking fast and slow – in clinical psychiatry”
Acta Psychiatrica Scandinavica, 23 November 2017 Online First (Open Access)
He walked into my rooms slowly, barely swinging his arms and showing no light in his eyes in response to my welcome. Parkinson’s Disease? Melancholia?
He provided a succinct reason for the appointment. “I developed depression 3 months ago. Just after my 63rd birthday. Never had it before. My mother developed depression at exactly the same age. She needed ECT to get better”.
I undertook a review of his depressive symptoms before taking a general history covering other possible diagnostic issues and his personal narrative. He was clearly depressed, but my primary objective was to ‘subtype’ his depression. His response to a set of questions that I have progressively weighted over the years (e.g. anhedonic and non-reactive mood, marked anergia, impaired ‘foggy’ concentration, mood and energy worse in mornings) argued for a melancholic depression as a ‘no brainer’ diagnosis and as a stretched metaphor. I commenced him on an antidepressant and, at reviews, his depression had significantly improved by 2 weeks and fully remitted at 6 weeks. I recommended he remains on the same dose of the medication in the light of the gravity of his initial presentation and as he was not experiencing distinct side-effects.
A year later he returned to say that ‘it’ had returned in the last fortnight. He again showed distinct psychomotor retardation and affirmed the same set of depressive features. Asked about his medication, he stated that he had stopped it 6 weeks previously. My ‘fast brain’ encouraged me to state the obvious – melancholia is highly likely to return if maintenance medication is ceased, and that he should immediately restart it. However, I held such censuring back as my ‘slow brain’ kicked in and I broadened my questioning. Were there any new symptoms? Well, he had had a disastrous golf round that week – his swing just didn’t work at all. Anything else? He had been embarrassed to drop a grocery bag at the supermarket that morning. Anything else? Well, yes. He had noticed some numbness in two fingers in his left hand. Anything else? Now that he thought about it, there was also some left-sided facial numbness.
The prescription pad was retracted and instead a neurology referral form completed. An MRI that afternoon showed a subdural haemorrhage which was successfully drained that evening. At review a fortnight later, he appeared completely normal and no symptoms of ‘melancholia’ were present. We both judged that it might be best not to reintroduce the antidepressant at that time.
This vignette serves as an introduction to clinical reasoning and one of its key components: pattern analysis.
So begins an Editorial written by Dr. Gordon Parker. This piece is short but nuanced.
Parker considers diagnostic skills. Noting Malcolm Gladwell’s observation that it takes 10,000 hours of practice to develop a skill set, he then draws on the work of Daniel Kahneman:
In his monograph, ‘Thinking, Fast and Slow’, Daniel Kahneman noted how chess masters are able to move down an assembly line of chess boards slaved over for hours by amateur chess tragics to make rapid and brilliant chess moves seemingly intuitively. But Kahneman dismissed intuition, arguing more for the masters seeing each chess board differently, reflecting the 10 000 h of dedicated performance to become familiar with thousands of configurations and with each involving arrangements of related pieces that could defend or threaten each other. In essence, pattern recognition acquired by repeated experience.
He then considers diagnosis and thinking:
- “Fast thinking operates automatically and without voluntary control, seeks categories or prototypes and looks to construct the best possible narrative or formulation. Such thinking is, however, gullible, possesses a bias to believe, neglects ambiguity….”
- “‘Slow’ thinking [is] intrinsically lazy but that it can be called on… to consider alternate explanations.”
- Neither style is ideal for psychiatric diagnosis. Slow thinking is for the early years of training with its reliance on formalized diagnostic criteria; fast thinking comes with years of experience.
- He argues that the “optimal endpoint” is fast thinking towards a diagnosis, and then slow thinking for second guessing it. This is a particularly important point in a field like psychiatry, since it lacks laboratory tests, and thus places emphasis on clinical reasoning and pattern analysis.
Noting that the best chess playing involves man and machine, he argues that computers should be designed to complement clinical reasoning of humans. How might this work?
For example, the psychiatrist might diagnose ‘depression’ and then turn to the computer for subtyping the specific depressive condition. Thus, the computer would assist by determining whether the depressed patient was most likely to have a unipolar depression (and, if so, a biological melancholic or an environmental non-melancholic disorder) or a bipolar disorder based on the presence (and absence) of multiple symptoms and their weightings.
A few thoughts:
- This is a provocative and well-argued piece.
- The fast thinking-slow thinking construct is interesting.
- And it seems about right, at least for what we know now.
- That said, things may be very different in the future.
He draws on chess. Let’s look that example: I would point out that a variant of his core argument – that there is an “art” to chess, not just positional analysis, and so grandmasters will always have an edge over computers – was made by chess players for years… that is, until Kasparov was beaten by Deep Blue. In the end, Deep Blue showed that incredible positional analysis (or “brute force,” to use the computer science term) trumps all else.
Is this relevant to psychiatry? Today, no. But it is possible that one day machines could tap large data sets that we can’t. Think of the GPS on the phone of a person with bipolar that could be used by a computer to “see” subtle changes in geographic patterns, indicating the start of a manic episode long before his psychiatrist notes the pressured speech and flight of ideas.
Chess is, of course, relatively simple. The data set is small (pieces and their positions on a board of 64 squares) and uncomplicated by “noise.” Humans are complex and we generate incredible data sets – and incredible noise. (To pick up on the last example: a subtle change in geographic pattern could be caused by an obscure childhood memory being triggered by a random TV ad.)
So… I’m not rushing back to school to study law just yet. But I also think that the potential of computers as diagnosticians might be underappreciated. And, of course, the neural networks – pushing computers well past brute force – could be a game-changer.
- AI has been much in the news – and in the journals. In an Editorial, The Lancet notes that 2017 was a bad year for AI. “2017 saw noticeable setbacks for two of the largest commercial companies operating in this space, as IBM Watson’s project with the MD Anderson Cancer Centre was halted after 4 years of development and Google DeepMind’s partnership with Royal Free London NHS Foundation trust came under fire for inappropriate sharing of confidential patient data.”
Still, they go on to argue:
There is no doubt that AI in health care remains overhyped and at risk of commercial exploitation… but, like other disruptive technologies in the past, the potential for impact should not be underestimated.
You can find the piece here: http://www.thelancet.com/journals/lancet/article/PIIS0140-6736(17)31540-4/fulltext.
- Many thanks to Drs. David Goldbloom and Ivan Silver for these suggested pieces, as well as Dr. Albert Wong for the lively conversation (and the chess counter-example).
“Dr. Ronald Fieve, 87, Dies; Pioneered Lithium to Treat Mood Swings”
The New York Times, 12 January 2018
Dr. Ronald R. Fieve, who was a pioneer in the prescription of lithium to treat mania and other mood disorders — while avowing that some gifted individuals, like Abraham Lincoln, Theodore Roosevelt and Winston Churchill, might have benefited from being bipolar — died on Jan. 2 at his home in Palm Beach, Fla. He was 87.
The cause was congestive heart failure, his daughter Vanessa Fieve Willett said.
Alerted by his adviser, Dr. Lawrence Kolb, to groundbreaking research in Australia, Dr. Fieve (pronounced FEE-vee) began experimenting with lithium to mitigate depression in the 1950s, when he was a resident at the New York State Psychiatric Institute.
He and a colleague, Dr. Ralph N. Wharton, went on to identify lithium as the first naturally occurring medication to prevent and control a specific psychiatric disorder and reduce the risk of relapses.
In 1966 Dr. Fieve established the first lithium clinic in North America.
Ronald R. Fieve
So begins an obituary for Dr. Fieve. Born in Stevens Point, Wisconsin, he originally studied to become an engineer but switched to medicine after being diagnosed with diabetes.
The article notes a few things about his life:
- “Dr. Fieve and several other researchers persuaded the Food and Drug Administration to approve the prescription of lithium salts for acute mania.”
- “Before it was approved to treat depression, lithium was found in the late 1940s to be potentially unsafe as a salt substitute. But Dr. Fieve pointed out that lithium had been found in natural mineral waters prescribed by Greek and Roman physicians 1,500 years earlier to treat what were then called manic insanity and melancholia.”
- “He promoted the use of lithium in the 1970s on radio and television talk shows, where he often appeared with the theatrical and film director Joshua Logan, a former patient.”
The article notes that he wrote widely and had influence on the definitions of Bipolar I and II found in DSM-III and DSM-IV.
A few thoughts:
- This essay provides a nice overview of an important psychiatrist.
- Modern psychiatry is very modern – Dr. Fieve opened the first lithium clinic in North America just five decades ago.
- Lithium remains an important mood stabilizer for those with bipolar – arguably the best mood stabilizer. But it has grown less used. Do we need a Dr. Fieve for the 21st century to re-popularize medication?
Reading of the Week. Every week I pick articles and papers from the world of Psychiatry.