Duration-general fasting and insulin-dosing strategy in type 1 diabetes: code and data for a preregistered in-silico factorial study

Complete computational record for a preregistered within-subject factorial in-silico study of daily fasting duration and insulin-dosing strategy in type 1 diabetes, run on the UVA/Padova type 1 diabetes model via the open-source simglucose implementation (S2008 lineage).

Design. Twenty virtual patients (10 adults, 10 adolescents; children excluded a priori) crossed daily fasting duration (0, 12, 14, 16, 18 h) with four dosing strategies (standard; basal reduction 25%; 30-minute glucose-trend-aware correction with predictive low-glucose suspend; combined), in two populations (default and an illustrative Gulf/Asian parameter shift). Total daily carbohydrate was held constant and, for each virtual patient, the CGM sensor-noise and carbohydrate-counting-error sequences were shared across all arms, so arms differ only in the manipulated factors. Each arm was simulated for 14 days. Co-primary endpoints, computed over the fasting window, were time in range 70–180 mg/dL and time below 70 mg/dL; time below 54 mg/dL was a key secondary endpoint.

Contents. Simulation, metrics, analysis and figure-generation scripts; the exploratory dataset (seed 42, 800 simulations), the preregistered confirmatory dataset (seed 2025, 800 simulations), an exploratory sensitivity dataset repeating the basal-reduction arm at a 40% cut (60 simulations), the derived 25%-versus-40% contrast table, the published figures and their exploratory counterparts, a pinned dependency lockfile, and SHA-256 checksums for every file.

Reproducibility. Simulations are deterministic given their seeds; re-running with the archived seeds reproduces the deposited CSV files exactly, and parallel execution was verified to give output identical to sequential execution. Note the setuptools<81 pin: later versions remove pkg_resources, which the gym dependency of simglucose still imports.

Two runs, two roles. The seed-42 dataset is exploratory and generated the directional hypotheses; the seed-2025 dataset is the preregistered confirmatory replication and is the primary dataset for the accompanying manuscript. This sequence is declared in the OSF preregistration.

Licensing. Code is released under Apache-2.0; data files, derived tables and figures under CC-BY-4.0. See LICENSE and LICENSE-DATA in the archive.

Not a medical device. Research and educational use only; not for clinical insulin dosing. simglucose is an open reimplementation of the UVA/Padova equations and is not the licensed regulatory simulator. The regional arm is an illustrative sensitivity manipulation, not a validated Gulf/Asian cohort, and supports no ethnicity-specific claim. All findings are hypothesis-generating and require prospective clinical validation.