Big Picture Medicine

Isn't it strange when everyone is raving about something — but you just — don't get it? That's the position Dr David Watkins (@DrMurphy11) found himself in in 2017. Babylon had launched in the UK and promised to provide the powers of AI and telemedicine to the NHS. Babylon lets you video consult with a doctor, often on the same day. This was slightly controversial, but not compared to its other head. Babylon had developed a sophisticated chatbot AI. After a game of 20 questions, the chatbot would suggest a diagnosis — although crucially (from a regulatory point of view) it wouldn't diagnose you. You would then be directed (triaged) to appropriate care. Mild headache and flu-like symptoms? No problem! Stay at home and take paracetamol. One-sided weakness and slurring your words? Call 999 🚑 Dr Watkins installed the app and began testing it. The details of this testing are disputed — Babylon who refer to him as a 'troll' claim that he ran 2400 tests, Dr Watkins claims just hundreds. More insidiously, Dr Watkins began to notice mistakes the chatbot was making — mistakes that no doctor would ever make. In one test, a 59-year old male smoker presents with central chest pain and nausea. Cookie-cutter symptoms of a heart attack, and he is duly warned. The exact same situation but this time, the patient is a woman— She is warned of a panic attack. This week's podcast is the story of Dr Watkin's battle against Babylon Health.* *Views expressed are those of Dr Watkins'* This week's episode presents a single side of the story. For context, Babylon have done a lot of good in the world. They've brought affordable healthcare to Rwanda, created a clinical AI fellowship for doctors and generally offer lucrative remote working opportunities.  —————— You can find Dr Watkins on Twitter @DrMurphy11. The title of this podcast episode is adapted from Dr Watkins' talk at the Royal Society of Medicine: "Cowboys and unicorns in #DigitalHealth — why validation is more important than valuation". You can find me on Twitter @MustafaSultan and subscribe to my newsletter on www.musty.io

How do you get into MedTech and collaborate with DeepMind/Google Health to lead world-renowned Deep Learning research? All to help stop people from going blind?  This is 30 minutes of pure gold — Dr Pearse Keane's advice on how to get into exciting Deep Learning projects, approaches that worked for him and books he recommends. Dr Pearse Keane is a consultant ophthalmologist and NIHR Clinician Scientist at Moorfields Eye Hospital in London. Pearse was responsible for starting the collaboration between Moorfields and DeepMind. Some of his most famous research uses deep learning to identify retinal disease from OCT scans. Clinically applicable deep learning for diagnosis and referral in retinal disease: https://www.nature.com/articles/s41591-018-0107-6)

Ever wondered if astronauts are honest with their doctors? Or what it feels like to look out of the window at Earth? Helen Sharman was the first British astronaut. Project Juno, was a joint Soviet-British programme which took Helen to the Mir Space Station. Helen was working as a research technologist for Mars (the confectionery company) before hearing about the opportunity on the radio. We talked about the selection and training, what it’s like being in Space and the peculiar doctor-astronaut relationship.

What’s AI? What’s machine learning? What’s deep learning? This week’s episode is a crash course on machine learning for medics. Dr Chris Lovejoy is a doctor, data scientist and content creator. He runs some exceptional ‘machine learning for medics’ day courses in London which is where I met him — and he also puts out loads of great content on his YouTube channel and newsletter. He also recently published a systematic review in the BMJ titled: AI Versus Clinicians: a Systematic Review. If you’re someone who has heard of concepts in AI and machine learning, but don’t necessarily understand them — Chris is the very best person I could think of to help explain them. I sat down with Chris to go from zero -> hero in machine learning as it relates to healthcare. Hopefully you'll get a better idea of what some of the buzzwords and key ideas are, how you can interpret a medical machine learning paper — and where you can go to keep your finger on the pulse and to learn more. Find Chris Website: https://chrislovejoy.me Newsletter: https://chrislovejoy.substack.com/p/artificial-intelligence-in-drug-discovery YouTube: https://www.youtube.com/channel/UCDoIFHMTKsgIweO-OZt1ecw Resources and Links Mentioned Luke Oakden-Rayner's post explaining metrics used to measure machine learning algorithm performance: https://lukeoakdenrayner.wordpress.com/2017/12/06/do-machines-actually-beat-doctors-roc-curves-and-performance-metrics/ Eric Topol: Twitter and book Chris's BMJ Paper: https://www.bmj.com/content/368/bmj.m689 Doctor Penguin Newsletter: http://doctorpenguin.com/about Andrew Ng's Machine Learning Course: https://www.coursera.org/learn/machine-learning AI for Medicine Coursera: https://www.coursera.org/specializations/ai-for-medicine Find Me www.bigpicturemedicine.co.uk Twitter @MustafaSultan

What are the realities of trying to procure PPE during a pandemic? As the government has given the NHS a 'blank cheque' to fight against COVID-19 — why can't we just throw money at the PPE shortage? To get answers, I spoke to Dr Alisa Pearlstone — the project director at NHS Hero Support — a pop up, emergency PPE delivery service. They’ve teamed up with the Doctors Association and Letsbeatcovid to help fight the PPE shortage. You can donate to NHS Hero Support at www.nhsherosupport.co.uk Links to socials and the podcast newsletter can be found at: www.bigpicturemedicine.co.uk

What do you when hospitals are running out of Personal Protective Equipment (PPE) and your colleagues are at risk? Ryan and his team at Oxford Inspired have created a PPE mask out of scuba masks. They plan to deliver 3000 of these to the frontlines — but how do they work? And why are they having to step in? You can donate to the fundraiser at: www.oxford-inspired.com and you can find Ryan's innovation journal of the future at www.weshare.healthcare More about reverse innovation: https://www.bmj.com/content/367/bmj.l6205

"I'm going to go out on a limb and say that AI isn't going to get us out of the covid-19 crisis." Natasha Loder, Health Policy Editor at the Economist. Except, some researchers are trying just that. They're using deep learning to read CT scans and try to diagnose COVID-19, as well as building other types of models to predict the prognosis (mortality, length of hospital stay etc.) of the disease. A number of these models have been published as preprints or in academic journals — but Laure's team found that they were poorly reported, at a high risk of bias and and highly optimistic about their results. You can find Laure's systematic review here (available open access): https://www.bmj.com/content/369/bmj.m1328

Curious about how the UK made the decision to the lock the country down? It was made using epidemiological models. One research team has been particularly influential in their response — the Imperial College team led by Professor Neil Ferguson. Their model predicted that an unchecked COVID-19 epidemic would overwhelm the NHS and result in 500,000 UK deaths. They suggested that we may need to have some form of social distancing for 12 out of the next 18 months. More recently, an Oxford University team led by Professor Sunetra Gupta published their own model. Any model on covid-19 has to make some assumptions — Imperial looked at the deaths we’ve had in the UK and assumed that COVID-19 hadn’t infected much of the UK, but had quite a high death rate. Oxford assumed the opposite — they constructed a model which assumed that COVID-19 had infected most of the population, but had a relatively low death rate. This was picked up in news outlets such as the Financial Times — “Coronavirus may have infected half of UK population“. To find out how these types of COVID-19 models work, what their limitations are and how we should interpret them — I called up Dr James Hay — who’s a computational epidemiologist at the Harvard’s School of Public Health. For the record, this conversation was recorded on the 31st March 2020. Imperial COVID-19 Model: Impact of non-pharmaceutical interventions (NPIs) to reduce COVID- 19 mortality and healthcare demand. https://www.imperial.ac.uk/media/imperial-college/medicine/sph/ide/gida-fellowships/Imperial-College-COVID19-NPI-modelling-16-03-2020.pdf Oxford COVID-19 Model: Fundamental principles of epidemic spread highlight the immediate need for large-scale serological surveys to assess the stage of the SARS-CoV-2 epidemic. https://www.medrxiv.org/content/10.1101/2020.03.24.20042291v1 Financial Times article mentioning Oxford model: https://www.ft.com/content/5ff6469a-6dd8-11ea-89df-41bea055720b

What can 2 million retinal images and a machine-learning algorithm achieve? Early-detection of Alzheimers—at least that's what Dr Siegfried Wagner along with a team led by Dr Pearse Keane at the Moorfields Eye Hospital are working towards.  We discuss the study at Moorfields, before going down a deep dive into how a relatively novel AI technique—Generative Adversarial Networks (GANs) can be used in Medicine. You may have seen a number of 'deep fakes' online, all made using GANs. But what legitimate uses do they have in Medicine and research?  Mentioned Links AlzEye Study: https://readingcentre.org/workstreams/artificial_intelligence_hub/alzeye/ Economist article on AlzEye: https://www.economist.com/science-and-technology/2019/12/18/a-system-based-on-ai-will-scan-the-retina-for-signs-of-alzheimers Rotterdam Study: https://jamanetwork.com/journals/jamaneurology/fullarticle/2685868 Biobank Study: https://jamanetwork.com/journals/jamaneurology/fullarticle/2685869 Predicting age and sex from retinal fundus images: https://www.nature.com/articles/s41551-018-0195-0

A professor of Pathology who has accepted his impending redundancy from AI image recognition—Professor Neil Sebire is also the Chief Research Information Officer at Great Ormond Street Hospital (GOSH); the country's leading children's research hospital. We talk about the future of pathology and paediatrics in the context of AI, how he partnered with Microsoft to create the GOSH in Minecraft and I finish off by asking him for his advice on academic success and getting published.