ESCADA: Efficient Safety and Context Aware Dose Allocation for Precision Medicine

Authors: Ilker Demirel, Ahmet Alparslan Celik, Cem Tekin

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we make in silico experiments on the bolus-insulin dose allocation problem in type-1 diabetes mellitus disease and compare our algorithms against the famous GP-UCB algorithm, the rule-based dose calculators, and a clinician.
Researcher Affiliation Academia Ilker Demirel Bilkent University ilkerd@ee.bilkent.edu.trA. Alparslan Celik Bilkent University acelik@ee.bilkent.edu.trCem Tekin Bilkent University cemtekin@ee.bilkent.edu.tr
Pseudocode Yes Algorithm 1 ESCADA algorithm
Open Source Code Yes Our code is available at https://github.com/Bilkent-CYBORG/ESCADA.
Open Datasets Yes We make in silico experiments using the open-source implementation [57] of the U.S. FDA approved University of Virginia (UVA)/PADOVA T1DM simulator [23]
Dataset Splits No The paper describes how data is generated through simulation (e.g., 'We make 15 consecutive dose recommendations for a meal event in a single run. We repeat this experiment with 30 different meal events for all 30 patients.'), but it does not specify explicit training, validation, or test dataset splits in terms of percentages or counts for a pre-existing dataset.
Hardware Specification Yes Appendix E: We use a workstation with Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz with 8GB of RAM, and NVIDIA GeForce GTX 1080 GPU.
Software Dependencies Yes Appendix E: We use Python 3.7.4. The following libraries with their versions are used: numpy==1.18.5, scipy==1.5.0, scikit-learn==0.23.1, GPy==1.9.9, matplotlib==3.2.2. Simglucose v0.2.1.
Experiment Setup Yes We set the target blood glucose (BG) level to 112.5 mg/dl [22]. We sample carbohydrate intake for each meal event from [20, 80] g, and fasting blood glucose from [100, 150] mg/dl. We make 15 consecutive dose recommendations for a meal event in a single run. We repeat this experiment with 30 different meal events for all 30 patients.