Leveraging Cluster Causal Diagrams for Determining Causal Effects in Medicine

Presentation Time: 08:30 AM - 08:45 AM

Abstract Keywords: Causal Inference, Knowledge Representation and Information Modeling, Rule-based artificial intelligence
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Causal inference, or the task of estimating the causal effect of an exposure or interventional variable on an outcome from an observational dataset, requires precise and rigorous methods, based on assumptions about the system under study. Such assumptions can be articulated as a causal diagram, however use of this technique in medicine is uncommon due to challenges in causal diagram construction in high-dimensional settings. Recent introduction of cluster causal diagrams or C-DAGs promise to ease the task of diagram construction by allowing for the representation of some unknown or partially defined relationships. We evaluate the practical application of C-DAGs in simulated medical contexts. We estimate causal effects under varying sets of assumptions, determined by both causal diagrams and C-DAGs and compare our results. Our findings show empirically similar results, with little discrepancy between causal effect sizes or variance across experimental runs, although estimation and efficiency challenges remain to be explored.

Speaker(s):
Tara Anand, MA
Columbia University

Author(s):
George Hripcsak, MD - Columbia University Irving Medical Center;

Deep Reinforcement Learning for Efficient and Fair Allocation of Healthcare Resources

Presentation Time: 08:45 AM - 09:00 AM

Abstract Keywords: Deep Learning, Fairness and elimination of bias, Data Mining
Primary Track: Applications
Programmatic Theme: Clinical Informatics

During health crises like the COVID-19 pandemic, scarce resources like ventilators necessitate rationing. Current allocation protocols vary widely, lacking a standardized approach. We explore reinforcement learning for critical care resource allocation optimization. Our transformer-based deep Q-network integrates patient disease progression and interactions for fairer, more effective resource distribution. Experiments show reduced excess deaths and improved equity compared to existing methods across different levels of ventilator shortage.

Speaker(s):
Yikuan Li, M.Sci
Northwestern University

Author(s):
Chengsheng Mao, Ph.D - Northwestern University - Feinberg School of Medicine; Hanyin Wang, PhD - Northwestern University; Kaixuan Huang, BS - Princeton University; Zheng Yu, PhD - Princeton University; Mengdi Wang, PhD - Princeton University; Yuan Luo, PhD - Northwestern University;

Pruning the Path to Optimal Care: Identifying Systematically Suboptimal Medical Decision-Making with Inverse Reinforcement Learning

Presentation Time: 09:00 AM - 09:15 AM

Abstract Keywords: Machine Learning, Fairness and elimination of bias, Clinical Decision Support, Healthcare Quality, Diversity, Equity, Inclusion, Accessibility, and Health Equity, Informatics Implementation
Primary Track: Applications

In aims to uncover insights into medical decision-making embedded within observational data from clinical settings, we present a novel application of Inverse Reinforcement Learning (IRL) that identifies suboptimal clinician actions based on the actions of their peers. This approach centers two stages of IRL with an intermediate step to prune trajectories displaying behavior that deviates significantly from the consensus. This enables us to effectively identify clinical priorities and values from ICU data containing a mix of optimal and suboptimal clinician decisions. We observe that the benefits of removing suboptimal actions vary by disease and impact certain demographic groups more than others.

Speaker(s):
Fiona McBride, MS
Harvard University

Author(s):
Adi Carmel, BS - Harvard University; Rebecca Hurwitz, BA - Harvard University; Finale Doshi-Velez, PhD - Harvard University; Leo Benac, BS - Harvard University; José Roberto Tello Ayala, BS - Harvard University; Inko Bovenzi, BA Candidate - Harvard University; Michael Hu, BS Candidate - Harvard University;

Where do doctors disagree? Characterizing Decision Points for Safe Reinforcement Learning in Choosing Vasopressor Treatment

Presentation Time: 09:15 AM - 09:30 AM

Abstract Keywords: Clinical Decision Support, Machine Learning, Information Visualization
Working Group: Student Working Group
Primary Track: Applications
Programmatic Theme: Clinical Informatics

In clinical settings, domain experts sometimes disagree on optimal treatment actions. These ``decision points" must be comprehensively characterized, as they offer opportunities for Artificial Intelligence (AI) to provide statistically informed recommendations. To address this, we introduce a pipeline to investigate ``decision regions", clusters of decision points, by training classifiers for prediction and applying clustering techniques to the classifier's embedding space. Our methodology includes: a robustness analysis confirming the topological stability of decision regions across diverse design parameters; an empirical study using the MIMIC-III database, focusing on the binary decision to administer vasopressors to hypotensive patients in the ICU; and an expert-validated summary of the decision regions' statistical attributes with novel clinical interpretations. We demonstrate that the topology of these decision regions remains stable across various design choices, reinforcing the reliability of our findings and generalizability of our approach. We encourage future work to extend this approach to other medical datasets.

Speaker(s):
Esther Brown


Author(s):
Esther Brown; Shivam Raval, Student - Harvard; Weiwei Pan, PhD - Haravrd; Yao Jiayu, PhD - Gladstone Institutes; Finale Doshi-Velez, PhD - Harvard; Siddharth Swaroop, PhD - Harvard;

Comparative Ranking of Marginal Confounding Impact of Natural Language Processing-Derived Versus Structured Features in Pharmacoepidemiology

Presentation Time: 09:30 AM - 09:45 AM

Abstract Keywords: Causal Inference, Natural Language Processing, Real-World Evidence Generation
Primary Track: Foundations

Objective: To explore the ability of natural language processing (NLP) methods to identify confounder information beyond what can be identified using claims codes alone for pharmacoepidemiology. Methods: We developed a retrospective cohort for high vs low dose proton pump inhibitors from linked Medicare claims (2008-2017) and clinical data for patients with a history of peptic ulcer disease. Clinical notes authored one year prior to cohort entry were processed via three NLP tools: bag-of-n-grams, MTERMS, and clustered BERT sentence embeddings. Candidate features were ranked using Bross formula. Results: The top 100 consisted of structured (75%; 19 prespecified) versus NLP-derived (25% with all tools accounted for) features. Conclusions: Bross formula is a simple way to rank the marginal confounding impact of binary features on estimated causal effects. NLP (especially n-grams) contributed to finding large numbers of features that can supplement claims data and prespecified variables to help in providing additional confounder information.

Speaker(s):
Joseph Plasek, PhD
Mass General Brigham

Author(s):
Richard Wyss, PhD - Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital, Harvard Medical School; Janick Weberpals, PhD, RPh - Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital, Harvard Medical School; Jie Yang, Ph.D. - Brigham and Women's Hospital/Harvard Medical School; Thomas Deramus, PhD - Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital, Harvard Medical School; Theodore Tsacogianis, MPH - Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital, Harvard Medical School; Kerry Ngan, MS - Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital, Harvard Medical School; Lily Bessette, MS - Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital, Harvard Medical School; Kueiyu Joshua Lin, MD, ScD - Department of Medicine, Massachusetts General Hospital, Harvard Medical School; Li Zhou, MD, PhD, FACMI, FIAHSI, FAMIA - Brigham and Women's Hospital, Harvard Medical School;

Better Blood Pressure Control for Stroke Patients in the ICU: A Deep Reinforcement Learning with Supervised Guidance Approach for Adaptive Infusion Rate Tuning

Presentation Time: 09:45 AM - 10:00 AM

Abstract Keywords: Clinical Decision Support, Deep Learning, Critical Care
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Blood pressure variability (BPV) plays a critical role in vascular diseases, particularly in acute ischemic stroke patients in intensive care units (ICUs), where higher BPV correlates with increased mortality rates. Current interventions lack effective methods for controlling BPV across consecutive time windows. To addressing this gap, we propose an offline deep reinforcement learning approach with supervised guidance to regulate systolic BPV in the following consecutive time windows by optimizing intravenous nicardipine infusion rates for intracerebral hemorrhage patients. Using clinically inspired reward functions, our method aims to tailor antihypertensive medication management within the critical 24-hour recovery window. Compared to human performance, our best method showed 57.52% and 126.01% improvements over the human baseline for maintaining BP within the desired range for the next time window and across two consecutive time windows. This research promises streamlined antihypertensive medication dosing, offering potential just-in-time adaptive interventions through automated pumps during stroke patients' ICU stays.

Speaker(s):
Kun-Yi Chen, M.S.
University of Missouri

Author(s):
Chi-Ren Shyu, PhD, FACMI, FAMIA - University of Missouri-Columbia; William Baskett, MS - University of Missouri; Kun-Yi Chen, M.S. - University of Missouri; Adnan Qureshi, M.D. - University of Missouri;