An Algorithmic Approach for Causal Health Equity: A Look at Race Differentials in Intensive Care Unit (ICU) Outcomes
Drago Plečko, Paul Secombe, Andrea Clarke, Amelia Fiske, Samarra Toby, Donisha Duff, David Pilcher, Leo Anthony Celi, Rinaldo Bellomo, Elias Bareinboim · 2025
Read An Algorithmic Approach for Causal Health Equity: A Look at Race Differentials in Intensive Care Unit (ICU) Outcomes by Drago Plečko, Paul Secombe, Andrea Clarke, Amelia Fiske, Samarra Toby, Donisha Duff, David Pilcher, Leo Anthony Celi, Rinaldo Bellomo, Elias Bareinboim (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-05197.txt.
Introduction
Drago Plečko and a team of co-authors build a causal frame for health equity in twenty twenty-five. Big data now lets researchers trace why health outcomes differ between groups. Their framework analyzes disparities with causal inference rather than plain statistics. They apply it to intensive care outcomes by race and ethnicity.
Cases come from Australia, comparing Indigenous and non-Indigenous patients, and from the United States. There, African-American and White patients are compared. The authors show common statistical measures of inequity are not enough. Instead, a disparity should be attributed to the mechanisms that generate it. Among their findings, minority patients tended to be younger at admission with worse chronic health. Readers get a method for moving from gap-watching to cause-finding.
Catalog record
| CREATOR | Drago Plečko, Paul Secombe, Andrea Clarke, Amelia Fiske, Samarra Toby, Donisha Duff, David Pilcher, Leo Anthony Celi, Rinaldo Bellomo, Elias Bareinboim |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 64 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-05197
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
An Algorithmic Approach for Causal Health Equity: A Look at Race Differentials in Intensive Care Unit (ICU) Outcomes Drago Plečko, Paul Secombe, Andrea Clarke, Amelia Fiske, Samarra Toby, Donisha Duff, David Pilcher, Leo Anthony Celi, Rinaldo Bellomo, Elias Bareinboim Abstract Health equity is defined as the state in which everyone has a fair and just opportunity to attain their highest level of health. Achieving health equity is believed to improve the well-being of communities, reduce healthcare costs, and increase productivity and longevity. However, disparities in health are still significant. In this context, the new era of large-scale data collection and analysis presents an opportunity for diagnosing and understanding the causes of health inequities. In this study, we describe a framework for systematically analyzing health disparities using tools of causal inference. We illustrate the framework by investigating racial and ethnic disparities in intensive care unit (ICU) outcome between majority and minority groups in Australia (Indigenous vs. Non-Indigenous) and the United States (African-American vs. White). We demonstrate that commonly used statistical measures for quantifying inequity are insufficient, and focus on attributing the observed disparity to the causal mechanisms that generate it. We find that minority patients are younger at admission, have worse chronic health, are more likely to be admitted for urgent and non-elective reasons, and have higher illness severity. At the same time, however, we also find a protective direct effect of belonging to a minority group, with minority patients showing improved survival compared to their majority counterparts, with all other variables being equal. We then demonstrate that this protective effect is related to the increased probability of being admitted to ICU, with minority patients having an increased risk of ICU admission. Additionally, we also find that minority patients, while showing improved survival, are in fact more likely to be readmitted to ICU. These findings support the hypothesis that, due to worse access to primary health care, minority patients are more likely to end up in ICU for preventable conditions, causing a reduction in the mortality rates and creating an effect that appears to be protective. Since the baseline risk of ICU admission may serve as proxy for lack of access to primary care, we developed the Indigenous Intensive Care Equity (IICE) Radar, a monitoring system for tracking the over-utilization of ICU resources by the Indigenous population of Australia across geographical areas. keywords health equity, causal inference, algorithmic fairness, intensive care medicine 1 Introduction Health equity is defined as the state in which everyone has a fair and just opportunity to attain their highest level of health ( Braveman et al 2017 ) . One of the key goals in pursuing health equity is the elimination of economic, social, and other obstacles to health and health care, together with eliminating existing, preventable differences in health ( Braveman 2014 ) . The goal is to reduce preventable illnesses and deaths, improve the overall well-being of communities, and create a more inclusive healthcare system that serves everyone effectively ( Braveman et al 2011 ) . Despite a commitment of public health organizations and government agencies to address issues of health equity ( Centers for Disease Control and Prevention (2024) CDC ; U.S. Department of Health & Human Services (2024b) HHS ; U.S. Department of Health & Human Services (2024a) HHS ; World Health Organization (2024) WHO ; European Union (2024) EU ; Healthy People 2030 2023 ) , health disparities are still large ( Commonwealth Fund 2023 ; Dwyer-Lindgren et al 2022 ; Health 2023 ) , and many consider the solutions to health inequities to be in their early stages. In this context, the rise of the new generation of computational tools and the widespread adoption of electronic health records (EHRs) ( Evans 2016 ) offer a major opportunity to quantify and understand health disparities. This task is even more important due to a broad transition to using artificial intelligence (AI) tools in healthcare, which may perpetuate or amplify existing biases. In fact, recent works demonstrate formally that understanding disparities in human decision-making is an essential step for understanding disparities in automated AI systems ( Plečko and Bareinboim 2024 ) . Therefore, a successful transition to healthcare systems that benefit from the numerous advantages of AI requires careful scrutiny of the data that is given to such AI systems for training, which includes understanding and quantifying the biases in the data itself. In this paper, we demonstrate that significant disparities in health exist in the data even before applying AI tools. Thus, training AI systems using such biased data may lead to further biases, if not handled appropriately ( Plecko and Bareinboim 2024 ) . At the same time, while great care needs to be taken when training AI systems, a transition to such systems also presents the opportunity to possibly correct for existing human biases. So far, a systematic framework for analyzing health equity that is compatible with the new needs of the data-driven era has not been proposed. In this paper, we discuss such a framework that may appeal to a broad range of practitioners, data scientists, and AI engineers. The main aim of the framework is to use a causal lens to obtain actionable insights that can improve patient engagement and outcomes. To illustrate this, we perform a case study and focus on the question of racial and ethnic disparities in outcome following admission to an intensive care unit (ICU) ( McGowan et al 2022 ) , in particular focusing on disparities in mortality. We analyze data from Australia ( Secombe et al 2023 ) for disparities between the Indigenous (Aboriginal and Torres Strait Islander) and Non-Indigenous (majority group) population of Australia. In parallel, we also analyze a database from a large tertiary hospital in Boston, Massachusetts ( Johnson et al 2020 ) , focusing on the racial disparity between the African-American and White population. We illustrate how a systematic approach for analyzing equity can illuminate different causal mechanisms that generate disparities observed in the data, and demonstrate that racial/ethnic disparities in ICU outcome are markedly similar between the United States and Australia. 2 Results The research in this paper was approved by the ethics committee of the Alfred Hospital, Melbourne (Project #661/24), and was done in collaboration with the Indigenous Data Network (IDN) of Australia ( Indigenous Data Network 2024b ) . 181 centers from Australia ANZICS APD MIMIC-IV Large tertiary center from United States Synchronized Data \mathbf{X} \mathbf{Z} \mathbf{W} \mathbf{Y} x^{(1)} z^{(1)} w^{(1)} y^{(1)} \vdots \vdots \vdots \vdots x^{(n)} z^{(n)} w^{(n)} y^{(n)} Standard Fairness Model Domain Knowledge Causal Fairness Analysis TV Decomposition Interaction Testing Effect Heterogeneity Baseline Risks of ICU Admission IICE Radar Figure 1 : Diagram of the performed analysis steps. 2.1 Data The first dataset we analyzed is the Australian and New Zealand (ANZ) Intensive Care Society (ANZICS) Adult Patient Database (APD) ( Secombe et al 2023 ) . The ANZICS APD receives submissions from 98% of ICUs in Australia. We analyzed all admissions from 181 hospitals across Australia between 2018 and 2024, yielding a cohort of 1,035,890 patients, of whom 38,736 ( 3.7\% ) were Indigenous. The overall mortality rate was 7.5\% . The second dataset we analyzed is the Medical Information Mart for