Artificial intelligence and diagnosis in general practice

INTRODUCTION

Diagnosis in general practice is difficult. Within the populations that GPs care for, serious diseases and disorders are less prevalent and, generally, at an earlier stage in the development of symptoms than in other healthcare settings.¹ However, the public, the media, and our specialist colleagues want primary care clinicians to become much better diagnosticians. GPs are now expected to spot individuals with cancer earlier while also avoiding excessive testing or inappropriate onward referrals.

Currently, there is growing interest in using artificial intelligence (AI) to supplement, enhance, or even replace GP diagnostic intelligence. Advocates of AI have suggested that such technologies could improve diagnostic accuracy (with fewer underdiagnoses and overdiagnoses) in addition to diagnostic efficiency. But others have argued that it will add yet another information burden during an already busy surgery. It might do very little to improve patient outcomes, GP stress levels, or healthcare finances.²

AI refers to computing techniques that resemble human cognitive functions, such as learning and problem solving. Rather than following pre-programmed instructions, AI approaches draw inferences from, or discover, patterns in data. Expert systems emulate human decision making by using an inference engine to apply rules to facts held in a knowledge base. Machine learning — including neural networks or deep learning — develops models to make decisions, or to classify examples. This is achieved by using a training algorithm — supplied with data of suitable quantity and quality — to adjust the weights within a model to optimise the goodness of fit between the input and the output.

All AI techniques depend on high-quality information on which to learn and classify clinical data in relation to outcomes. To date, most AI …