Assessing the Performance of Chat Generative Pretrained Transformer (ChatGPT) in Answering Andrology-Related Questions.


Journal

Urology research & practice
ISSN: 2980-1478
Titre abrégé: Urol Res Pract
Pays: Turkey
ID NLM: 9918608789006676

Informations de publication

Date de publication:
Nov 2023
Historique:
medline: 7 11 2023
pubmed: 7 11 2023
entrez: 7 11 2023
Statut: ppublish

Résumé

The internet and social media have become primary sources of health information, with men frequently turning to these platforms before seeking professional help. Chat generative pretrained transformer (ChatGPT), an artificial intelligence model developed by OpenAI, has gained popularity as a natural language processing program. The present study evaluated the accuracy and reproducibility of ChatGPT's responses to andrology-related questions. The study analyzed frequently asked andrology questions from health forums, hospital websites, and social media platforms like YouTube and Instagram. Questions were categorized into topics like male hypogonadism, erectile dysfunction, etc. The European Association of Urology (EAU) guideline recommendations were also included. These questions were input into ChatGPT, and responses were evaluated by 3 experienced urologists who scored them on a scale of 1 to 4. Out of 136 evaluated questions, 108 met the criteria. Of these, 87.9% received correct and adequate answers, 9.3% were correct but insufficient, and 3 responses contained both correct and incorrect information. No question was answered completely wrong. The highest correct answer rates were for disorders of ejaculation, penile curvature, and male hypogonadism. The EAU guideline-based questions achieved a correctness rate of 86.3%. The reproducibility of the answers was over 90%. The study found that ChatGPT provided accurate and reliable answers to over 80% of andrology-related questions. While limitations exist, such as potential outdated data and inability to understand emotional aspects, ChatGPT's potential in the health-care sector is promising. Collaborating with health-care professionals during artificial intelligence model development could enhance its reliability.

Identifiants

pubmed: 37933835
doi: 10.5152/tud.2023.23171
doi:

Types de publication

Journal Article

Langues

eng

Pagination

365-369

Auteurs

Ufuk Caglar (U)

Department of Urology, Haseki Training and Research Hospital, Istanbul, Turkey.

Oguzhan Yildiz (O)

Department of Urology, Haseki Training and Research Hospital, Istanbul, Turkey.

M Fırat Ozervarli (MF)

Department of Urology, Istanbul University, Istanbul School of Medicine, Istanbul, Turkey.

Resat Aydin (R)

Department of Urology, Istanbul University, Istanbul School of Medicine, Istanbul, Turkey.

Omer Sarilar (O)

Department of Urology, Haseki Training and Research Hospital, Istanbul, Turkey.

Faruk Ozgor (F)

Department of Urology, Haseki Training and Research Hospital, Istanbul, Turkey.

Mazhar Ortac (M)

Department of Urology, Istanbul University, Istanbul School of Medicine, Istanbul, Turkey.

Classifications MeSH