Is artificial intelligence able to discriminate emergencies?

Introduction: over-frequent attendance in pediatrics is defined as repeated attendance in an emergency service for reasons that do not require urgent attention or could be managed at another level of care. There may be several factors related such as socioeconomic, cultural and psychological factors. Its impact on the health system is significant. Artificial intelligence (AI) has the potential to reduce over-frequent attendance.

Methodology: the agreement between the information provided by Gemini AI, a free and open Access service, for 101 common diseases in childhood, was analyzed in comparison with the available evidence. The analysis is performed by means of the adjusted kappa coefficient.

Results: AI gave a response/replied to the 101 pathologies analyzed. The kappa coefficient for the pathologies’ correct identification was 0.857 +/- 0.002, the proper detection of warning signs showed a coefficient of 0.888 +/- 0.003, the necessity of attending the emergency room had an accuracy of 0.876 +/- 0.005 and 0.915 +/- 0.003 for the adequacy of the measures needed.

Conclusions: text-based artificial intelligence has a good agreement with protocols for the identification of pathologies based on symptoms, and very good for assessing the need to visit an emergency department, assessing warning signs and therapeutic recommendations. This agreement is greater in children over three months of age and for common pathologies.