Why should doctors understand machine learning?

We are seeing increasing use of, and discussion about, artificial intelligence (AI) and machine learning (ML) across all sectors; healthcare included. Greater volumes of data, improved computing power and break-throughs in machine learning techniques have provided fertile soils for innovation. In healthcare, this is reflected by an exponential increase in research involving AI, and a corresponding surge in publications and academic funding.

But do doctors really need to understand machine learning? We think so, and here are our top reasons why.

Communication with patients

We are beginning to see machine learning tools used in clinical settings and with our current wave of new research, it’s only a matter of time before we see more widespread use. This means that your ‘average’ doctor will be using machine learning-based decision support tools and recommending machine learning-based treatments as part of their daily practice in the near future.

When interpreting tests, it’s important for us to understand terms like specificity, sensitivity, positive and negative predictive values, amongst others. We know a patient with raised PSA doesn’t necessarily have prostate cancer, or that a raised D-dimer doesn’t definitely mean a DVT/PE, because the tests have high sensitivity but low specificity. This often forms a part of our discussions with patients.

However, we will now need to expand this to include discuss of AUCs and F1 scores, and other nuances of machine learning-based predictions.

Additionally, many patients will want to understand how the new interpretations and predictions work. They will find the explanation “then a very sophisticated algorithm spits out a prediction” much less satisfying than “the algorithm is trained on 1000s of patients to learn to spot patterns that can predict X vs Y, so that we can take into account your age, genetic profile, comorbidities, etc to give a personalised prediction”.

Contribute to research and innovation

Artificial intelligence in healthcare is one of the most dynamic and exciting research fields, with huge potential and so much left to be explored. There are many ways that, as a doctor who understands ML, you can contribute to this field.

Many hospitals store large quantities of data, which are fertile soil for exciting healthcare AI research. However, without people in the department who understand the ways in which it could be used, the data often sits unused. As someone who understands ML, you could help to establish a department working on cutting-edge AI research. (We’re going to upload a guide for doing so in the future: watch this space.)

There’s also huge potential for collaboration with the many companies working in the healthcare AI space. Their teams often involve a combination of machine learning engineers, researchers and, of course, healthcare professionals. While it is ML engineers who will build the tools, they will need input from healthcare professionals to guide their efforts. There is a lot of demand for medics who can “speak their language”; who can translate their medical insights into terms of ‘data’ and ‘variables’. You can also provide insight into what could feasibly be incorporated into the medical workflow and what would be of little use. Should we train an algorithm to recognise absolute observations/vitals signs, or fluctuations from the individual’s baseline? Would an algorithm with high sensitivity for AKI, that provides pop-up notifications, be helpful or a hindrance? There are many insights that can only be gained from working within healthcare systems, so we shouldn’t leave the development of healthcare AI to those who haven’t.

Be a part of the discussion

Understanding ML will also enable you to cut through the hype and help shift the discussion towards a more measured, accurate appraisal of the current state of affairs. When a leading machine learning researcher reports that AI can now diagnosed pneumonia better than radiologists, you will have the insight to explain why this is not the case.

The implementation of ML across society and in healthcare will represent one of the key transformations of our generation. As doctors, we want to have a say about how this takes place. There are many ethical issues to be considered and these do not always align with the financial incentives of companies working in the area. Understanding ML can enable us to be informed and responsible members of our medical community, who can contribute to discussions and policy decisions, thereby facilitating the introduction of AI in a safe, effective and patient-centred way.

How?

The next step is how to develop understanding of machine learning. Check out our recommended resources for building the foundation of understanding and then keeping up-to-date.