A sparse-data-driven personalization framework for glucose-insulin modeling of parenteral nutrition administration

Authors: Jung, Sooyeon; Lee, Yein; Yoo, Junyoung; Rachim, Vega Pradana; Park, Sung-Min

Source journal: Computer methods and programs in biomedicine (2026-11-13)

DOI: 10.1016/j.cmpb.2026.109719

Abstract

BACKGROUND AND OBJECTIVE: Delivering carbohydrates intravenously during parenteral nutrition (PN) can produce rapid glycemic excursions. The management of patients receiving PN who are insulin deficient or need insulin therapy requires interpreting glucose exposure and exogenous insulin administration during the procedure. However, protocol-based care does not quantify subject-specific glucose-insulin responses, and routine inpatient measurements are often sparse. Hence, we developed a sparse-data-driven framework for subject-specific glucose-insulin modeling of patients receiving PN. METHODS: An established type 1 diabetes mellitus (T1DM) glucose-insulin model was adapted by replacing meal-related glucose appearance with known intravenous-glucose input while retaining the insulin kinetics, delayed insulin action, glucose distribution, utilization, and endogenous regulation. Personalization was defined as estimating subject-specific parameter groups from sparse glucose observations, insulin information, and known inputs using maximum a posteriori inference. The framework was evaluated first in silico using virtual T1DM subjects (n = 30) and then in vivo using a porcine T1DM PN model (n = 9). RESULTS: The subject-specific parameter setting in vivo reconstructed glucose trajectories from six observations over 2 h with a root mean square error (RMSE) of 9.15 +/- 1.18 mg/dL and a mean absolute relative difference (MARD) of 3.11 +/- 0.61%. Forward-prediction RMSE/MARD ratios were 8.20 +/- 5.38 mg/dL/5.99 +/- 6.14% at 30 min and 13.86 +/- 10.65 mg/dL/9.59+/-9.19% at 60 min. Under the same structure and inputs, subject-specific parameters reduced prediction errors relative to the mean-parameter vector. Sequential updating improved short-horizon prediction, while the in silico scenarios showed coherent responses to glucose and insulin input changes. CONCLUSIONS: Sparse PN-relevant data can support subject-specific glucose-insulin response modeling, thereby providing a preclinical computational foundation for individualized insulin-relevant modeling during PN.

Read the full text

Related topics

Related open access articles

Abstract and bibliographic data from PubMed (U.S. National Library of Medicine). Copyright of the abstract belongs to the original publisher or authors.