Machine learning is coming to medicine. Even though I am a fan of computing even I have been shocked at the speed of the advancement of machine learning in medicine. We can discuss the ethics, we can proclaim that it’s not fair, however, this will not change the fact that machine learning is coming. A prominent example was the prediction of heart attacks. Not only was the machine learning algorithm for accurate that consultants it even considered parameters like if the patient was taking steroids, and the patient’s mental health. Bizarrely, it was more comprehensive than a human [link]. Does this mean that you’re going to be replaced? Of course not. Here we can discuss the areas to focus on to minimise the effect of machine learning.
To be effective, we must discuss what machine learning is. In brief, it’s an equation with several variables. Let’s say that we want to predict a patient’s blood pressure by their age. We all know this would be a terrible algorithm, it’s more for demonstration. We would have the following prediction model:
BP = (weighting_A x age) + (weighting_B x pulse)
A load of training data is pushed through this equation and the results are compared with real results and the error is calculated. There’s a bunch of steps and this is where the science of machine learning gets complicated but the result is that the weighting for each variable is slightly altered to achieve less of an error. This cycle is repeated. Step size, conditioning before calculating the rate of change to avoid local minimas, logistic functions to account for yes/no answers and neural networks are just some of the things to consider that cannot be covered in a single post. However, the take home message here is that the machine learning algorithm tweaks the weighting of each variable until it comes up with an optimum prediction model. The advantage of computing is that you can do this with thousands of variables. It can consider a variable even if it only contributes 0.001% to the outcome. Trying to compete with this is not an effective use of your time. There is so much more to you than pattern recognition and memorisation. Here are some areas that you have time focus on now Google and machine learning are focusing on pattern recognition and memorisation.
When Google made the information available to anyone with an internet connection. Ok it took the conversation points away from that particular type of guy in the pub who memorised facts from the encyclopedia. Now people are no longer interested in the fact that Dave knows the economic statistics of a country. They are interested in what Dave thinks about those stats. In a social context, you meet a friend and after a few beers, he confesses his burning desire to move and live it up in Detroit. Even if he doesn’t know much about Detroit and you’ve never heard of the place, within minutes you can be looking at the pictures, discussing the house prices and digging deeper with your friend to see if he’s truly thought this through. Dave would shed some light on the situation before Google. But now we have Google at the tips of our fingers, Dave is going to have to work on his social skills. Machine learning gives us the freedom to advance personalised medicine. Those who use the findings of machine learning algorithms to zoom in and personalise treatment plans will enhance their worth. Don’t be a Dave, see machine learning for what it is, a tool that you can use to make more decisions. Traditionally medics read up on trials, assessed guidelines, and considered the personal situations of the case that they were dealing with. Let’s say we’re managing blood pressure. Machine learning can assess 20 drugs and find the weighting of hundreds of variables for a favourable outcome. Depending on the patient’s parameters we will be able to pick the most effective antihypertensive for that patient. The doctors that find ways to bring this into practice will be the leaders of tomorrow. In one last context let’s look at when Excel was created and what impact it had on accounting. One accountant looking at Excel and thinking, no, I’m going to try and trash it as it does part of my job better than I do. Whilst the other accountant is looking at it and thinking, great! I can do my calculations very fast. I can then look at my Client’s personal situation and see which is the best tax bracket I can get him in. Once I have done this I only have to alter a few things in my excel sheet and see if he’s better off financially. The second accountant is using Excel to provide a more personalised service.
Whilst machine learning has made its way into writing and can most probably write better than I can this field isn’t easy to conquer by the machines. Clinicians are always coming up with practical solutions to problems. You most probably do it more than you realise. Again, you can use computing as a tool here. Is there a problem that keeps popping up that can be streamlined or automated? You can come up with the solution. Furthermore, you can spread it to other departments if it’s effective. Whilst we’re all aware of the fact that computing power is getting stronger this isn’t the only way you can be creative. Ways of solving the problem is a big thing. Even if you don’t want to learn how to code at the very least revisit the math you learned at school and how to map out a problem logically. There are mathematical proofs on which way is quicker to sort a range of books on a bookshelf. The way you solve a problem has a knock-on effect. If you combine your medical knowledge with logical mapping and a bit of math, you will be able to shave time off solving medical problems. These considerations will be unique to the field of medicine, the department and how the hospital runs as well as the demographics of the patients and staff. Shaving time greatly reduces the computational load as these computations will be repeated. I know a UCL computing professor who did math as an undergrad at Cambridge. He always said, before you go running for faster hardware, you should try and solve the problem in a more efficient way. For a basic example, lets say we want to work out the average of the following set: [1, 2, 3, 4, 5]. We could divide every number by 5 and then add them together. This would result in 10 steps. Or we could get the same answer by adding them all together and dividing the sum of that addition by 5. It would get the same result but in 6 steps. Medical computing is just begging for clinicians who understand the clinical context and variables and know enough math to quantify their logic.
Learn how to code
If you can’t beat them join them. This isn’t for everybody but there is a growing number of doctors learning how to code. Ok, you won’t be leading the field of coding. There are computer science grads and math geniuses who are already at the forefront of this field, however, with the high-level programming languages that are out there it won’t take too long before you’re coding basic scripts that will be of use. Basic machine learning can also be used by the non-math graduate. There are prebuilt modules like SciKit learn where you can plug in and play with machine learning algorithms that have been developed for you. The trick for you here is developing a platform that will get the data, and cleaning the data so it can be processed by the machine learning algorithm. Be careful with this field though. It can reveal high rewards but there are some traps. I advocate that every clinician should learn code but this is just because I love coding. I appreciate that it’s irrational. Whilst I’ve met my fair share of clinicians who are great at coding I have taught enough to realise that the average clinician starts to get confused when it gets to loops within loops. There’s no correlation to medical academic performance here. I’ve met Ph.D. medics who have graduated from top universities struggle to logically map out basic problems, whilst I’m been impressed by the coding skills of an F2 from a med school with a low ranking. Skills that get rewarded in medical academia don’t translate to coding.
Whatever you choose to do I hope you do it well. There is a clinical developers club that I help with that tries to get clinicians into coding. I also teach at events the basics of coding and I hope doctors and nurses continue to take it up. For small minds tech will be scary, but for the problem solvers, tech gives them the opportunity to put their problem-solving skills on steroids.