Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unmasking Societal Biases in Respiratory Support for ICU Patients through Social Determinants of Health

Mira Moukheiber, Lama Moukheiber, Dana Moukheiber, Hyung-Chul Lee

TL;DR

This study addresses biases in ICU respiratory care by evaluating how social determinants of health shape predictions for two clinically important tasks: prolonged mechanical ventilation and successful weaning. It introduces a temporal benchmark derived from MIMIC-IV, housing 50,920 ICU stays with hourly ventilation and covariate data, and pairs this with fairness audits across demographic and SDOH attributes. The authors compare multiple sequence-based and MLP architectures, finding GRU-based hybrids to perform best (AUROC ~0.78 for MV and ~0.78 for weaning) while revealing systematic disparities across groups. By releasing the hourly dataset and conducting fairness analyses, the work provides a framework for equitable respiratory intervention decision support in critical care and encourages bias-aware deployment of predictive models.

Abstract

In critical care settings, where precise and timely interventions are crucial for health outcomes, evaluating disparities in patient outcomes is essential. Current approaches often fail to fully capture the impact of respiratory support interventions on individuals affected by social determinants of health. While attributes such as gender, race, and age are commonly assessed and provide valuable insights, they offer only a partial view of the complexities faced by diverse populations. In this study, we focus on two clinically motivated tasks: prolonged mechanical ventilation and successful weaning. Additionally, we conduct fairness audits on the models' predictions across demographic groups and social determinants of health to better understand health inequities in respiratory interventions within the intensive care unit. Furthermore, we release a temporal benchmark dataset, verified by clinical experts, to facilitate benchmarking of clinical respiratory intervention tasks.

Unmasking Societal Biases in Respiratory Support for ICU Patients through Social Determinants of Health

TL;DR

This study addresses biases in ICU respiratory care by evaluating how social determinants of health shape predictions for two clinically important tasks: prolonged mechanical ventilation and successful weaning. It introduces a temporal benchmark derived from MIMIC-IV, housing 50,920 ICU stays with hourly ventilation and covariate data, and pairs this with fairness audits across demographic and SDOH attributes. The authors compare multiple sequence-based and MLP architectures, finding GRU-based hybrids to perform best (AUROC ~0.78 for MV and ~0.78 for weaning) while revealing systematic disparities across groups. By releasing the hourly dataset and conducting fairness analyses, the work provides a framework for equitable respiratory intervention decision support in critical care and encourages bias-aware deployment of predictive models.

Abstract

In critical care settings, where precise and timely interventions are crucial for health outcomes, evaluating disparities in patient outcomes is essential. Current approaches often fail to fully capture the impact of respiratory support interventions on individuals affected by social determinants of health. While attributes such as gender, race, and age are commonly assessed and provide valuable insights, they offer only a partial view of the complexities faced by diverse populations. In this study, we focus on two clinically motivated tasks: prolonged mechanical ventilation and successful weaning. Additionally, we conduct fairness audits on the models' predictions across demographic groups and social determinants of health to better understand health inequities in respiratory interventions within the intensive care unit. Furthermore, we release a temporal benchmark dataset, verified by clinical experts, to facilitate benchmarking of clinical respiratory intervention tasks.

Paper Structure

This paper contains 18 sections, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. DataSet
  4. Dataset Overview
  5. Cohort Selection
  6. Data Extraction and Preprocessing
  7. Patient-level Static Variables
  8. Measurement Observations
  9. Treatment Interventions
  10. Outcome Variables
  11. Benchmark Tasks
  12. Model Architecture
  13. Model Evaluation
  14. Fairness Audits Along SDOH & Demographic Attributes
  15. Results & Discussion
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Visualization of time-varying covariates for a patient's stay, 30 days after ICU admission. The plots, listed from top to bottom, include ventilation parameters, laboratory results, and vital signs, respectively.
  • Figure 2: Model architecture of the sequence models. The architecture uses a joint-fusion strategy that concatenates hourly time-dependent features with static features and includes recurrent layers such as an LSTM, BiLSTM, or GRU, or neural networks like TCN.
  • Figure 3: Performance gap measures for the prolonged mechanical ventilation task under the best model (GRU). A positive bar indicates the model favors one group over the other group. Error bars denote a 95% confidence interval obtained through 1000 bootstrap samples. a) Performance gap evaluation for SDOH attributes. b) Performance gap evaluation for demographic attributes.
  • Figure 4: Performance gap measures for the successful prolonged weaning task under the best model (GRU). A positive bar indicates the model favors one group over the other group. Error bars denote a 95% confidence interval obtained through 1000 bootstrap samples. a) Performance gap evaluation for SDOH attributes. b) Performance gap evaluation for demographic attributes.