Introduction and Objective: in-hospital glucose management of people living with type 2 diabetes (PLWT2D) has a direct impact on morbidity and mortality. Artificial intelligence (AI) tools could improve patient care. This study aimed to assess the correlation between in-patient glucose management suggested by an AI chatbot and hospital protocols and compare it with real life medical management.

Methods: we conducted a cross-sectional study of PLWT2D hospitalized in a tertiary care center in 2023 who required an Endocrinology consultation during their hospital stay. We excluded patients with corticosteroid treatment or need for intravenous insulin.  We provided Chat-GPT4 with the protocol for T2D management used in the hospital and asked for a suggested prescription based on individual de-identified variables. We assessed the accuracy of the output using Spearman (r) and interclass correlation (ICC) coefficients. We then analyzed the correlation between the treatment prescribed by physicians on the first day of hospital stay and the protocol.

Results: 85 patients met the inclusion criteria. There was a strong correlation between the dose of basal insulin suggested by Chat-GPT and the protocol (r=0.995), CCI=0.996 [IC 95% 0.994;0.998] as opposed to a weak correlation between the basal insulin dose prescribed by physicians and the protocol (r=0.223), ICC=0.134 [IC95% -0.072;0.331]. The same strong correlation was found between the dose of prandial insulin suggested by Chat-GPT and the protocol (r=0.978), CCI=0.984 [IC 95% 0.975;0.990], and a moderate correlation between physicians and the protocol (r=0.399), CCI=0.466 [IC95% 0.283;0.616]. The chatbot suggested the discontinuation of non-insulin therapies in all patients, as stipulated in the protocol.

Conclusions: the use of an AI chatbot resulted in a higher correlation with established protocols compared with medical prescription.

Disclosure

T. Rojas-López: None. D. Alvarez-Martin: None. A. Pujol-deCastro: None. O. Moreno-Dominguez: Speaker's Bureau; Novo Nordisk, Lilly Diabetes. Research Support; Sanofi. B. Barquiel: None. E. Garcia-Perez-de-Sevilla: None. P. Parra-Ramírez: None. N. Gonzalez-Perez-de-Villar: Speaker's Bureau; Abbott Diagnostics. Advisory Panel; Medtronic. Speaker's Bureau; Air Liquide.

© 2024 by the American Diabetes Association
2024
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at http://www.diabetesjournals.org/content/license.