BACKGROUND
Mild Cognitive Impairment (MCI) serves as a precursor to dementia, posing a significant public health challenge given the increasing global aging population. The imperative for early detection, with over 55 million individuals affected by dementia worldwide, necessitates efficient screening methods due to the resource-intensive nature of common tools.
OBJECTIVE
This study aims to assess the effectiveness of GPT-4 in screening for MCI in the elderly, comparing its performance with that of junior neurologists.
METHODS
An exploratory design was employed, involving 174 participants. GPT-4, trained using a set of language analysis indicators, evaluated MCI severity in participants' test texts (GPT-4 does not support voice assessment). Three junior neurologists independently assessed both text and voice components of the test corpus.
RESULTS
GPT-4 demonstrated a higher accuracy of 0.81, surpassing the neurologists' range of 0.41 to 0.49. Statistical analysis revealed significant discrimination of MCI by GPT-4 (p < 0.001). The study derived a clinical risk assessment nomogram from GPT-4's top ten weighted features, enhancing MCI patient evaluation.
CONCLUSIONS
The GPT-4 model exhibits promise as a diagnostic tool for MCI, potentially enhancing patient outcomes and alleviating healthcare burdens. Nevertheless, further investigation and clinical validation are essential to ascertain its practical applicability in real-world scenarios.