[A new artificial intelligence tool for assessing symptoms in patients seeking emergency department care: the Mediktor application].

OBJECTIVES: To analyze agreement between diagnoses issued by the Mediktor application and those of an attending physician, and to evaluate the usefulness of this application in patients who seek emergency care.

MATERIAL AND METHODS: Prospective observational study in a tertiary care university hospital emergency department. Patients with medical problems and surgical conditions (surgery and injuries) who did not require immediate emergency care responded to the Mediktor questions on a portable computer tablet. The software analyzed the answers and provided a list of 10 possible preliminary diagnoses in order of likelihood. The patient and the attending physician were blinded to the list to so that the usual care process would not be altered. The level of agreement between the physician's diagnosis and the Mediktor diagnosis was analyzed.

RESULTS: A total of 1015 patients were included; 622 cases were considered valid for study. Cases were excluded if the patients did not meet the inclusion criteria, they did not have a discharge diagnosis, they had a final diagnosis expressed as a symptom or their final diagnosis was not included in the Mediktor database. The physician's diagnosis (the gold standard) coincided with one of the 10 MEDIKTOR diagnoses in 91.3% of the cases, with one of the first 3 diagnoses in 75.4%, and with the first diagnosis in 42.9%. Sensitivity was over 92% and specificity over 91% in the majority of common diagnostic groups; the κ statistic ranged from 0.24 to 0.98.

CONCLUSION: The Mediktor application is a reliable diagnostic aid for the most prevalent problems treated in a hospital emergency department. The general public finds it easy to use.