An approach to make general practitioner referrals suitable for artificial intelligence deployment

Abstract

Outpatient referrals for hospital specialist assessment are an increasing workload that carry significant risk if not attended to in a timely manner. This viewpoint discusses how decision support (including artificial intelligence and machine learning) may address this problem. Of the many possible approaches, we choose a combination of two that illustrate the breadth of available tools and how they combine to complement each other. To understand the issues and inform this discussion, a survey of general practitioners' views was conducted (Appendix 2), an audit of declined referrals was undertaken (Appendix 3) and draft decision trees were constructed (Appendix 4). To have data suitable for automated decision support, the current referral needs to change from free text to a structured format that ensures every patient has a complete minimum dataset. Regarding triaging decisions, at present there is human variability, but the decision support tools will need to be trained on a set of referrals that have an agreed gold-standard decision. In order to maintain patient safety throughout, the process needs to be incremental. We suggest that one way to assure patient safety is to combine simple decision trees with sophisticated contemporary machine learning.

Ähnliche Arbeiten