C-peptide Index under Non-fasting Conditions Can Predict the Likelihood of Future Insulin Therapy in Outpatients with Type 2 Diabetes Mellitus Not Receiving Insulin at Baseline.

Objective This study aimed to develop a clinical model in which the C-peptide index (CPI) under non-fasting conditions can predict future insulin therapy in patients with type 2 diabetes mellitus (T2DM). Methods This was a single-center retrospective study. We analyzed the correlation between non-fasting CPI and future insulin therapy in 464 patients who T2DM not receive insulin therapy who attended our clinic and were evaluated for non-fasting CPI on an outpatient basis. We used machine learning as an adjunct method to create a clinical model to predict future insulin therapy. Results At the end of the observation period, 22 participants remained on insulin therapy (continuous insulin therapy group: CIG), and 442 were not on insulin therapy (without insulin therapy group: WIG). HbA1c and serum creatinine (Cre) were significantly higher in the CIG than in the WIG (p<0.001 and p=0.045, respectively). The non-fasting CPI was lower in the CIG than in the WIG (p<0.001). The cutoff value of non-fasting CPI to predict the need of future insulin therapy was 1.62 (sensitivity 40.9%, specificity 90.7%), and the cutoff value to indicate the low possibility of future insulin therapy was 3.11 (sensitivity 90.1%, specificity 48.9%). A clinical model was created to predict future insulin therapy by machine learning using the Easy Ensemble Classifier method with non-fasting CPI, HbA1c, and Cre as features (accuracy, 78.5%; area under the ROC curve 0.86, false positive rate, 18.2%; false negative rate, 22.2%). Conclusion CPI under non-fasting conditions can predict the likelihood of future insulin therapy in outpatients with T2DM.