OBJECTIVES: Artificial intelligence (AI) used in medical diagnostics is growing rapidly. Diagnostic AI can facilitate improved triage and diagnosis, prioritise caseload, monitor physiology and chronic disease, and maximise diagnostic efficiency. We reviewed how AI is currently used in the NHS and summarised the uncertainties and challenges its implementation faces.

METHODS: A targeted review of the NICE website was conducted to identify MedTech innovation briefings (MIBs) reviewing AI technologies. Data extracted included disease area, clinical outcomes, costs, and limitations.

RESULTS: Six MIBs featuring AI were identified, covering 14 technologies. All were used in diagnostics or monitoring, with the most common application being computerised tomography. Disease areas covered included respiratory medicine, cardiovascular medicine, neurology, and oncology. Evidence used in the MIBs reported that all the technologies were effective, including improvements in performance and reduced time for interpretation of diagnostic information compared with human readers. However, considerable uncertainties regarding the informing studies were identified. These included use of small sample sizes and participants not representative of their intended population, poor methodological quality featuring retrospective analysis, increasing the risk of bias and confounding, and a lack of reference standards, meaning diagnostic accuracy data could not be ascertained. Additionally, many of the supporting studies were not published or peer-reviewed, casting further doubt on the robustness of their findings. Capital costs ranged from £1,699 to £80,000, and annual costs were as high as £60,000. Where costs were higher than standard care, AI had the potential to reduce resource use through decreased staff workload or improve clinical management through faster diagnosis.

CONCLUSIONS: AI has the potential to improve diagnostic efficiency leading to benefits in the patient pathway, which could release healthcare resources. Going forward, a key challenge for diagnostic AI will be to improve the quality of the evidence base used by decision-makers such as NICE.