Development and Validation of a Clinical Prediction Model for Insulin Requirement in Diabetes Mellitus Patients: The Insulin Requirement Probability Score (IRPS)

Abstract

Background: Determining optimal pharmacological intervention in diabetes mellitus remains challenging, particularly in identifying patients requiring insulin therapy versus those manageable with oral agents. Current practice lacks validated quantitative tools integrating beta-cell function and glycemic parameters to predict insulin requirement.

Objective: To develop and validate a clinical prediction model estimating insulin requirement probability using readily available clinical and laboratory parameters.

Methods: This retrospective study included 300 adult patients with diabetes (85% Type 2, 15% Type 1). Using logistic regression, they developed the Insulin Requirement Probability Score (IRPS), incorporating serum C-peptide, glycated hemoglobin (HbA1c), diabetes duration, age, fasting blood glucose, and diabetes type. The dataset was split into training (n=240) and test (n=60) sets. Performance was assessed using the area under the receiver operating characteristic curve (AUC-ROC), sensitivity, specificity, and calibration statistics.

Results: The IRPS demonstrated excellent discrimination with an AUC-ROC of 0.91 (95% CI: 0.85-0.97). At the optimal threshold (0.50), the model achieved 86.7% accuracy, 89.3% sensitivity, and 84.4% specificity. C-peptide emerged as the strongest predictor (β = -2.0, p <0.001), followed by Type 1 diabetes classification (β = +3.5, p <0.001). The model explained 74% of variance (Nagelkerke R² = 0.74) with good calibration (Hosmer-Lemeshow p = 0.60). Risk stratification categorized patients into four groups, enabling individualized treatment recommendations.

Conclusion: The IRPS provides a validated, quantitative tool for predicting insulin requirement, integrating beta-cell function with clinical parameters to support evidence-based treatment decisions in diabetes management.