EUVIMEDEUVIMED entdecken
Crossref

1749-P: Predicting Insulin in Type 2 Diabetes Using Gradient Boosting Machine Learning

INES KHOCHTALI, EKRAM HAJJI, MOHAMED KENANI

Diabetes · 05.06.2026 · 75 (Supplement_1)

DOI: 10.2337/db26-1749-p ↗

Kurz gefasst

Fachartikel in der Zeitschrift Diabetes, erschienen am 05.06.2026. Verfasst von INES KHOCHTALI und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Introduction and Objective: Initiating insulin therapy in type 2 diabetes (T2D) is often delayed despite clinical indications, contributing to increased complication risk. Early identification of patients who will require insulin initiation may enable proactive therapeutic intervention. This study aimed to predict progression toward insulin therapy in diabetic adults using baseline demographic, clinical, lifestyle, and treatment-related variables, leveraging machine learning.

Methods

Analysis was conducted on 496 adults followed in our Endocrinology Department. Clinical variables included age at diabetes onset, disease duration, initial diagnostic context, initial treatment, comorbidities, and metabolic/lifestyle parameters. Categorical variables were binarized. Univariate analysis (p<0.05) was used to preselect features prior to Gradient Boosting model development. Model optimization was achieved using GridSearchCV and evaluated using a stratified train/test split (75/25%), confusion matrix metrics, ROC-AUC, precision-recall indices, and learning-curve stability.

Results

The predictive model demonstrated excellent discrimination for insulin initiation (ROC-AUC: 0.999) and high accuracy (0.984). Sensitivity was 0.974 and specificity 1.000, with PPV = 1.000 and NPV = 0.958. Confusion matrix analysis showed 76 true positives, 46 true negatives, 0 false positives, and 2 false negatives. Learning curves indicated rapid convergence and model stability, supporting strong generalization capacity. The most influential predictors were age at diagnosis, ketoacidosis at presentation, sedentary lifestyle, and type of initial antidiabetic treatment.

Conclusion

A Gradient Boosting-based model can accurately predict future insulin use using routinely collected clinical and lifestyle data in patients with T2D. Its excellent performance suggests potential utility for early therapeutic stratification and anticipatory diabetes care planning in clinical practice. Disclosure I. Khochtali: None. E. Hajji: None. M. Kenani: None.

Zitieren

APA

KHOCHTALI, I., HAJJI, E., & KENANI, M. (2026). 1749-P: Predicting Insulin in Type 2 Diabetes Using Gradient Boosting Machine Learning. Diabetes, 75(Supplement_1). https://doi.org/10.2337/db26-1749-p

Vancouver

KHOCHTALI I, HAJJI E, KENANI M. 1749-P: Predicting Insulin in Type 2 Diabetes Using Gradient Boosting Machine Learning. Diabetes. 2026;75(Supplement_1). doi:10.2337/db26-1749-p

BibTeX
@article{KHOCHTALI2026,
  author = {KHOCHTALI, INES and HAJJI, EKRAM and KENANI, MOHAMED},
  title = {1749-P: Predicting Insulin in Type 2 Diabetes Using Gradient Boosting Machine Learning},
  journal = {Diabetes},
  year = {2026},
  volume = {75},
  number = {Supplement_1},
  doi = {10.2337/db26-1749-p}
}
Kostenlos starten →