Related Experiment Video
Updated: May 3, 2026
Electrochemiluminescence Assays for Human Islet Autoantibodies
Published on: March 23, 2018
Assay-specific machine-learning models converting insulin to C-peptide or C-peptide index and C-peptide to HOMA2-IR:
Yuichiro Iwamoto1, Tomohiko Kimura1, Toshitomo Sugisaki1
1Department of Diabetes, Endocrinology and Metabolism, Kawasaki Medical School, Kurashiki, Japan.
Background:
In this study, we employed machine learning to develop a conversion method for comparing immunoreactive insulin (IRI) and C-peptide immunoreactivity (CPR), which are indicators of endogenous insulin secretion capacity, using a standardized approach.
Methods:
This is a single-center retrospective observational study of 449 patients with type 2 diabetes (T2D) who were hospitalized at our hospital and 63 patients with T2D who were treated as outpatients, focusing on patients in whom IRI and CPR were measured simultaneously.
Results:
The gradient boosting decision tree (GBDT) model constructed for hospitalized patients used seven features, including IRI, and showed an accuracy of R2 = 0.641 and MSE = 0.247 ng/mL after applying a nonlinear transformation to the CPR index (CPI). The correlation coefficient between actual CPI and predicted CPI was r = 0.943. The accuracy of the GBDT model, which nonlinearly transforms HOMA2-IR using seven features, including CPR, was R2 = 0.615 and MSE = 0.268. The correlation coefficient between the actual HOMA2-IR and the predicted HOMA2-IR was r = 0.943. When the model was applied to outpatients, CPI and HOMA2-IR were significantly correlated with actual values (r = 0.820 and r = 0.812, respectively).
Conclusions:
If either IRI or CPR is measured, it will be possible to evaluate endogenous insulin secretion capacity and insulin resistance using the same standards, and it is expected to be used as an auxiliary indicator in future clinical research and when integrating data from multiple institutions.
