I remember falling in love with the idea of big data and computational modeling 5 years ago. Back then I was isolated. I was starting my physics degree and working in A and E. Both were good in themselves but they never crossed. When I spoke to doctors about big data and number crunching they simply brushed me off. They didn’t see it working, and speculated that I would find spurious correlations with little use. I remember trying to help out an audit where doctors were manually going through notes and logging a few one-dimensional markers. They couldn’t see how binary outcomes could contribute to a model, I’m being serious, years of education and they couldn’t weight binary outcomes. It’s not essential for being a good doctor but you’d think they would brush up on some basic math if they were going to do some research. They then made massive assumptions on these markers. One off the top of my head was being assessed by a consultant before cardiac arrest. They didn’t introduce thresholds. For instance if a patient arrests at 2 am after being in the ED for 5 minutes, it’s not really a reflection of unacceptable practice if a consultant didn’t assess them before the arrest happened, however, it didn’t stop them from saying stuff like “this is unacceptable” when the one-dimensional stats were presented. When I emailed them pointing this out I received a waffle of an email not addressing a single point but stating that the lead clinician has been doing this for years.
My introduction into the clinical coding community came when I gave a talk at Imperial College London about the advantages of high-level programming languages speeding up and automating tasks. I’d given up on selling the big data dream to clinicians who would laugh, shake their head and simply conclude that I just didn’t get it. After the talk and demonstration, a medical education fellow approached me and told me that that they were starting a computational medicine degree for doctors and that they were also giving weekend teaching sessions for doctors in Python…… I was excited, there were others. Another path that connected with others was this blog. In some of the books I was reading at the time they suggested that I start a blog to connect with others. At first, I was skeptical, but after interviewing, and posting I was meeting people all over London who shared the same vision. The coding community grew. Other clinicians are also forging clinical coding communities where excited doctors would share coding tips and articles. When I woke up today I came across an article that a doctor shared in a closed clinical coders group [link]. It presented a study that showed that a neural network machine learning algorithm was better at predicting heart attacks than doctors. It also recognized other risk factors and took into account variables like mental illness, if the patient was taking corticosteroids, and many more factors. The key thing is that it taught itself, and identified the variables. Furthermore, machine learning could generate guidelines and compute them resulting in superior outcomes.
The interesting politics that are yet to pan out is the split in the clinical professions. Before I was aware of these communities, I and people who I spoke to saw it as them and us. Computer scientists who just didn’t get how complex medicine and were just clueless if they thought that machine learning was going to play a major role in clinical decision making. Now there is a growing community of doctors who understand medicine, machine learning, can code, and understand tech. They seem to be embracing the advances, machine learning provides. I keep in contact with some of the medics that dismissed this stuff 5 years ago. I get the impression that the majority of them still fairly unaware of the advances and capabilities. Traditional appeals of authority and brushing off people is becoming less of an option, traditional medics will have to acknowledge that this technology exists.