Abstract
This systematic review aimed to explore the current applications, potential benefits, and issues of artificial intelligence (AI) in medical questionnaires, focusing on its role in 3 main functions: assessment, development, and prediction. The global mental health burden remains severe. The World Health Organization reports that >1 billion people worldwide experience mental disorders, with the prevalence of depression and anxiety among children and adolescents at 2.6% and 6.5%, respectively. However, commonly used clinical questionnaires such as the Hamilton Depression Rating Scale and the Beck Depression Inventory suffer from several problems, including the high degree of overlap of symptoms of depression with those of other psychiatric disorders and a lack of professional supervision during administration of the questionnaires, which often lead to inaccurate diagnoses. In the wake of the COVID-19 pandemic, the health care system is facing the dual challenges of a surge in patient numbers and the complexity of mental health issues. AI technology has now been shown to have great promise in improving diagnostic accuracy, assisting clinical decision-making, and simplifying questionnaire development and data analysis. To systematically assess the value of AI in medical questionnaires, this study searched 5 databases (PubMed, Embase, Cochrane Library, Web of Science, and China National Knowledge Infrastructure) for the period from database inception to September 2024. Of 49,091 publications, a total of 14 (0.03%) studies met the inclusion criteria. AI technologies showed significant advantages in assessment, such as distinguishing myalgic encephalomyelitis or chronic fatigue syndrome from long COVID-19 with 92.18% accuracy. In questionnaire development, natural language processing using generative models such as ChatGPT was used to construct culturally competent scales. In terms of disease prediction, one study had an area under the curve of 0.790 for cataract surgery risk prediction. Overall, 24 AI technologies were identified, covering traditional algorithms such as random forest, support vector machine, and k-nearest neighbor, as well as deep learning models such as convolutional neural networks, Bidirectional Encoder Representations From Transformers, and ChatGPT. Despite the positive findings, only 21% (3/14) of the studies had entered the clinical validation phase, whereas the remaining 79% (11/14) were still in the exploratory phase of research. Most of the studies (10/14, 71%) were rated as being of moderate methodological quality, with major limitations including lack of a control group, incomplete follow-up data, and inadequate validation systems. In summary, the integrated application of AI in medical questionnaires has significant potential to improve diagnostic efficiency, accelerate scale development, and promote early intervention. Future research should pay more attention to model interpretability, system compatibility, validation standardization, and ethical governance to effectively address key challenges such as data privacy, clinical integration, and transparency.