Contextualized Policy Recovery: Modeling and Interpreting Medical Decisions with Adaptive Imitation Learning

Authors: Jannik Deuschel, Caleb Ellington, Yingtao Luo, Ben Lengerich, Pascal Friederich, Eric P. Xing

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We assess CPR through studies on simulated and real data, achieving state-of-the-art performance on predicting antibiotic prescription in intensive care units (+22% AUROC vs. previous SOTA) and predicting MRI prescription for Alzheimer s patients (+7.7% AUROC vs. previous SOTA).
Researcher Affiliation Collaboration 1Carnegie Mellon University 2Karlsruhe Institute of Technology 3Broad Institute of MIT and Harvard 4MIT 5MBZUAI 6Petuum, Inc.
Pseudocode No The paper describes the methods conceptually and mathematically but does not include any pseudocode or algorithm blocks.
Open Source Code Yes Code for data processing, model training, and figure generation is available on Git Hub 1. 1https://github.com/JADEUSC/ contextualized_policy_recovery
Open Datasets Yes We look at 4195 patients in the intensive care unit over up to 6 timesteps extracted from the Medical Information Mart for Intensive Care III (Johnson et al., 2016) dataset and predict antibiotic prescription based on 7 observations temperature, hematocrit, potassium, white blood cell count (WBC), blood pressure, heart rate, and creatinine.
Dataset Splits Yes Each dataset is split up into a training set (70% of patients), validation set (15% of patients) for hyperparameter tuning, and test set (15% of patients) to report model performance.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments, such as GPU or CPU models.
Software Dependencies No The paper mentions using “the Adam optimizer (Kingma & Ba, 2014)” but does not specify version numbers for any software dependencies or libraries used in the implementation.
Experiment Setup Yes The initial learning rate chosen for CPR is 5e-4 and 1e-4 for the baseline RNNs. We select the dimensions of the hidden state for both CPR and the baseline RNNs from [16,32,64]. For CPR, λ is chosen from [0.0001,0.001,0.01,0.1]. The batch size is selected as 64 for all models. Table 4 shows the optimal hyperparameters chosen based on the validation set performance.