Validation of a model-based virtual trials method for tight glycemic control in intensive care

Area of Science:

• Computational Biology
• Clinical Informatics
• Diabetes Management

Background:

• In-silico virtual patients and trials offer potential cost, time, and safety benefits for developing tight glycemic control (TGC) protocols.
• Previous methods lacked validation for the independence of virtual patient predictions from the data used for their creation.

Purpose of the Study:

• To validate in-silico virtual patients and virtual trial methods using matched cohorts from a TGC clinical trial.
• To assess the independence and predictive accuracy of virtual patient models.

Main Methods:

• Utilized a 211-patient subset from the Glucontrol trial, with cohorts targeted for different glycemic ranges (Glucontrol-A: 4.4-6.1 mmol/L, Glucontrol-B: 7.8-10.0 mmol/L).
• Created virtual patients by fitting a validated model to clinical data, generating time-varying insulin sensitivity profiles (SI(t)).
• Validated individual virtual patients using model fit and prediction errors; assessed trial prediction accuracy via self-validation and cross-validation against clinical data.

Main Results:

• Individual virtual patients demonstrated high accuracy, with median forward prediction errors of 4.3% (Group-A), 2.8% (Group-B), and 3.5% (Overall).
• Both self-validation and cross-validation results closely matched clinical data (within 1-10%).
• Cross-validation confirmed that virtual patients with patient-specific SI(t) profiles accurately predict the performance of independent TGC protocols.

Conclusions:

• This study provides the first rigorous validation of in-silico virtual patients and virtual trial methodologies.
• Validated virtual models can accurately simulate TGC protocol clinical results in advance.
• These methods enable rapid in-silico design and optimization of TGC protocols.