Potential source of bias in AI models: Lactate measurement in the ICU as a template

Potential source of bias in AI models: Lactate measurement in the ICU as a template | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Potential source of bias in AI models: Lactate measurement in the ICU as a template Nebal S. Abu Hussein, Pratiksha Pradhan, Fredrik Willumsen Haug, and 11 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5836145/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Objective: Health inequities may be driven by demographics such as sex, language proficiency, and race-ethnicity. These disparities may manifest through likelihood of testing, which in turn can bias artificial intelligence models. The goal of this study is to evaluate variation in serum lactate measurements in the Intensive Care Unit (ICU). Methods: Utilizing MIMIC-IV (2008-2019), we identified adults fulfilling sepsis-3 criteria. Exclusion criteria were ICU stay <1-day, unknown race-ethnicity, <18 years of age, and recurrent stays. Employing targeted maximum likelihood estimation analysis, we assessed the likelihood of a lactate measurement on day 1. For patients with a measurement on day 1, we evaluated the predictors of subsequent readings. Results: We studied 15,601 patients (19.5% racial-ethnic minority, 42.4% female, and 10.0% limited English proficiency). After adjusting for confounders, Black patients had a slightly higher likelihood of receiving a lactate measurement on day 1 (odds ratio 1.19, 95% confidence interval (CI) 1.06-1.34), but not the other minority groups. Subsequent frequency was similar across race-ethnicities, but women had a lower incidence rate ratio (IRR) 0.94 (95% CI 0.90-0.98). Interestingly, patients with elective admission and private insurance also had a higher frequency of repeated serum lactate measurements (IRR 1.70, 95% CI 1.61-1.81, and 1.07, 95% CI, 1.02-1.12, respectively). Conclusion: We found no disparities in the likelihood of a lactate measurement among patients with sepsis across demographics, except for a small increase for Black patients, and a reduced frequency for women. Variation in biomarker monitoring can be a source of data bias when modeling patient outcomes, and thus should be accounted for in every analysis. Sepsis lactate MIMIC-IV Critical Care Health Equity Figures Figure 1 Figure 2 INTRODUCTION Disparities in healthcare are widely recognized, especially regarding discrimination based on race and ethnicity ( 1 , 2 ). Such disparities can unveil themselves as differences in quality of care, unequal medical device performance, or access to services reflecting structural inequities ( 3 ). These biases are not only harmful for patient care, but can also impact the development of machine learning-based clinical algorithms that train on electronic health records ( 4 ). Ensuring the development of fair AI models is crucial, and addressing missing information is a key initial step in achieving this objective, especially when such information is not missing at random ( 5 , 6 ). Unfortunately, this variation in the level of monitoring is often not taken into consideration in the development of machine learning-based clinical algorithms. In a 2017 study that evaluated 107 electronic health record (EHR)-based risk prediction tools, 49 did not account for missing data ( 7 ). A common approach to imputation is the use of normal values based on the assumption that laboratory tests that are not ordered are presumed to be within normal range, a practice that likely introduces bias ( 8 ). The probability of detecting an abnormal finding is contingent on the frequency of testing. Consequently, non-randomly missing data can lead to spurious correlations—non-causal relationships between features and outcome—that are learned and then incorporated into clinical algorithms ( 9 ). When the etiology of missing data stems from social determinants of care, these biases can become ingrained in subsequent AI models, perpetuating and even scaling existing disparities ( 10 , 11 ). This is even more important in a high-stake environment such as in patients with sepsis admitted to the Intensive Care Unit (ICU). Sepsis is a severe life-threatening systemic infection and effective management of this condition requires prompt diagnosis, aggressive treatment and continuous monitoring. Despite current advances, one key challenge remains the timely delivery of care. Herein, serum lactate level is one of the two key diagnostic tools of septic shock according to the guidelines ( 12 , 13 ). Disparities in sepsis outcomes are known to exist ( 14 ). However, the drivers of sepsis disparities are unknown and the question of whether disparities extend to serum lactate monitoring remains underexplored. This paper seeks to examine whether race and ethnicity, sex, and language differences influence the frequency of serum lactate determination conducted during the management of sepsis in the ICU. By shedding light on this dimension of care, we aim to contribute to a more comprehensive understanding of the social patterning of the data generation process in healthcare. Methods This observational retrospective study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement ( 15 ). The health equity language, narrative and concepts of this paper follows the American Medical Association’s recommendations ( 16 ). Data Extraction Data was extracted from the publicly available MIMIC-IV database ( 17 ). The MIMIC database is maintained by the Laboratory for Computational Physiology at the Massachusetts Institute of Technology and shared via the PhysioNet platform ( 18 ). The dataset has been de-identified, and the institutional review boards of the Massachusetts Institute of Technology (No. 0403000206) and Beth Israel Deaconess Medical Center (2001-P-001699/14) both approved the use of the database for research. The MIMIC-IV database includes physiologic data collected from bedside monitors, laboratory test results, medications, medical images and clinical progress notes captured in the electronic health record from patients admitted to the ICU between 2008–2019. Hypothesis We hypothesized that both the likelihood for a patient to have a serum lactate measurement and the frequency of subsequent measurements are not the same across race-ethnicity, sex, and English proficiency (as recorded by providers). Cohort Selection The following exclusion criteria were applied to create a study cohort: those without sepsis as defined by the sepsis-3 criteria ( 12 ), patients under 18 years of age, and those with length of ICU stay less than 1 day. Patients with recurrent hospital stays in the database, and those with a racial description other than White, Asian, Black, or Hispanic, especially excluding those of the heterogenous group “other”. For the negative binomial regression, we further excluded patients with absent serum lactate values on day one. Covariates We drew directed acyclic graphs (DAG) to understand which variables to extract ( Supplementary Fig. 1, Supplementary Table 1 ). Twelve confounders were extracted, including non-time-varying variables such as demographics, comorbidities, admission information, and source of infection and time-varying variables including Sequential Organ Failure Assessment (SOFA) score ( 19 ), and fluids normalized by length of stay. Time-varying variables were modeled as follows: SOFA score was calculated for the day of ICU admission; serum lactate measurements were used as a binary variable for whether or not it was measured on day one, in addition to taking the overall number of measurements for the whole ICU stay normalized by length of ICU stay. Outcomes We had two primary outcomes: the first was a binary variable predicting whether a patient received serum lactate measurement on day 1, and the second was a prediction of how many lactate measurements a patient would receive per day throughout the length of their ICU stay. Statistical Analysis Statistical analysis was performed using Python 3.10.9 ( 20 ) and R 4.2.1 ( 21 ). For the outcome of whether or not a patient had a serum lactate measurement on day 1, we fitted a Targeted Maximum Likelihood Estimation (TMLE) model ( 22 ). From the TMLE model, we extracted and utilized the odds ratio (OR) to estimate the likelihood of receiving a serum lactate measurement. For the outcome of the number of serum lactate measurements during an ICU stay, we fitted a non-penalized, negative binomial regression ( statsmodel package ( 23 )) adjusted for confounders to estimate the number of serum lactate measurements for each patient each day in the ICU. We report our findings as incident rate ratios (IRR). All findings are reported with 95% CI and with White patients as the reference group. RESULTS Baseline Study Cohort The MIMIC-IV database has 73,140 ICU stays, of which 15,601 were included in our final cohort following application of the inclusion and exclusion criteria (Fig. 1 ). The race-ethnicity distribution was 10.8% Black, 3.8% Hispanic, 2.9% Asian, 68.8% White and 14.6% others (without specified race). The demographic distribution did not change after applying exclusion criteria. SOFA score had a median of 6.00 (interquartile range (IQR) 4.00, 8.00), regardless of the race-ethnicity reported at baseline, with the Charlson comorbidity index at 6.00 (IQR 4.00, 8.00). Serum lactate on day 1 was slightly higher in the Non-White group at 2.50mmol/l (IQR 1.60, 4.00), compared to the White group at 2.20 (1.50, 3.50). In addition, Non-White patients received more fluids on the first day in the ICU than White patients (2,060 ml (IQR 640, 5,000) versus 1,690 (461, 4,540)),respectively. Of note, the volume of fluids received prior to admission to the ICU is not available in the dataset (Table 1 ). Table 1 Baseline information on the study cohort, derived from MIMIC-IV Race and Ethnicity Missing Overall Non-White White N (%) 15,601 (100) 2,801 (17.9) 12,800 (82.1) Age, median [Q1,Q3] 0 68.0 (57.0, 78.0) 64.0 (52.0, 76.0) 68.0 (59.0, 79.0) Sex, n (%) Female 0 6,520 (41.8%) 1,341 (47.9%) 5,179 (40.5%) English proficient, n (%) Yes 0 14,113 (90.5%) 1,894 (67.6%) 12,219 (95.5%) Insurance, n (%) Medicaid 0 1,042 (6.7%) 398 (14.2%) 644 (5.0%) Medicare 0 7,476 (47.9%) 1,064 (38.0%) 6,412 (50.1%) Other 0 7,083 (45.4%) 1,339 (47.8%) 5,744 (44.9%) Charlson comorbidity index, mean [Q1,Q3] 0 6.00 (4.00, 8.00) 6.00 (4.00, 8.00) 6.00 (4.00, 8.00) SOFA, median [Q1,Q3] 0 6.00 (4.00, 8.00) 6.00 (4.00, 9.00) 6.00 (4.00, 8.00) Elective admission, n (%) 2,876 (18.4%) 312 (11.1%) 2,564 (20.0%) Length of stay, median [Q1,Q3] days 0 3.13 (1.83, 6.25) 3.21 (1.88, 6.83) 3.13 (1.83, 6.17) Lactate day 1 (mmol/L) 0 2.20 (1.50, 3.50) 2.50 (1.60, 4.00) 2.20 (1.50, 3.40) Number of lactate measurements day 1, median [Q1,Q3] 3.00 (2.00, 5.00) 3.00 (2.00, 5.00) 3.00 (2.00, 5.00) Lactate day 2 (mmol/L) 9,397 (60.2%) 1.70 (1.20, 2.60) 1.80 (1.30, 2.90) 1.70 (1.20, 2.60) Number of lactate measurements day 2, median [Q1,Q3] 9,397 (60.2%) 2.00 (1.00, 3.00) 2.00 (1.00, 3.00) 2.00 (1.00, 3.00) Mechanical Ventilation, n (%) 0 8,841 (56.7%) 1,566 (55.9%) 7,275 (56.8%) Renal Replacement Therapy, n (%) 0 1,550 (9.9%) 397 (14.2%) 1,153 (9.0%) Vasopressor(s), n (%) 0 9,243 (59.2%) 1,455 (51.9%) 7,788 (60.8%) Fluids received day 1 (mL), median [Q1,Q3] 446 (2.9%) 1,750 (498, 4,620) 2,060 (640, 5,000) 1,690 (461, 4,540) Abbreviations: Q1, lower quartile range; Q3, upper quartile range; SOFA, sequential organ failure assessment Model Results We adjusted our models for confounders according to a DAG ( Supplementary Fig. 1, Supplementary Table 1 ). Using the TMLE model with being White, male and English proficient as a reference, Black patients were more likely to have a serum lactate measurement on day 1 with OR 1.19 (95% CI 1.06, 1.34). Asian and Hispanic patients had a similar likelihood compared to White patients, with an OR of 1.08 (95% CI 0.93, 1.24), and an of OR 0.98 (95% CI 0.89, 1.08), respectively (Table 2 , Fig. 2 a). We validated these findings with a cross-validated logistic regression model ( Supplementary Table 2 ). Table 2 Likelihood of receiving a lactate measurement on day 1 fitted by a Targeted Maximum Likelihood Estimation model Demographic OR 2.50% CI 97.5% CI White Reference Black 1.19 1.06 1.34 Asian 1.08 0.93 1.24 Hispanic 0.98 0.89 1.08 Male Reference Female 1.02 0.96 1.09 English Proficient Reference English Non-Proficient 0.96 0.86 1.07 Abbreviations: OR, odds ratio; CI, confidence intervall The negative binomial model was fitted to predict the total frequency of serum lactate measurements during a patient’s ICU stay (Table 3 , Fig. 2 b). We found no significant difference in the frequency of measurements across race-ethnicities compared to Whites as reference. Hispanic (IRR 1.12, 95% CI 0.99, 1.26), Black (IRR 1.01, 95% CI 0.94, 1.09), and Asian (IRR 1.08, 95% CI 0.95, 1.23) patients had a non-significant difference in their frequency of serum lactate measurements. In addition, English proficiency had no significant impact on measurement frequency (IRR 1.06, 95% CI 0.97, 1.16). On the other hand, female sex (IRR 0.94, 95% CI 0.90, 0.98) and having a urinary tract infection (IRR 0.68, 95% CI 0.50, 0.93) were associated with a decreased serum lactate measurement frequency, while having private insurance (IRR 1.07, 95% CI 1.02, 1.12) and being admitted electively (IRR 1.7, 95% CI 1.61, 1.81) significantly increased the frequency of receiving a measurement. Table 3 Results of the negative binomial regression for outcome of lactate measurement frequency on day 1 Variable IRR 2.5% CI 97.5% CI Intercept 0.72 0.62 0.85 Age 1.00 1.00 1.00 Charlson comorbidity index 1.01 1.00 1.02 SOFA 1.10 1.09 1.10 Race : White Reference Asian 1.08 0.95 1.23 Black 1.01 0.94 1.09 Hispanic 1.12 0.99 1.26 Binary variables : Female sex 0.94 0.90 0.98 English proficient 1.06 0.97 1.16 Private insurance 1.07 1.02 1.12 Elective admission 1.70 1.61 1.81 Volume of fluids normalized by LOS 1.00 1.00 1.00 Pneumonia 1.01 0.90 1.13 Urinary tract infection 0.68 0.50 0.93 Biliary infection 1.22 0.81 1.84 Skin infection 1.03 0.61 1.72 Abbreviations: IRR, incidence rate ratio; CI, confidence Interval; SOFA, sequential organ failure assessment; LOS, length of stay; DISCUSSION In this retrospective cohort study in patients with sepsis, we observed no discernible disparities between sexes and non-native English speakers in receiving a serum lactate measurement on day one, although Black patients had a slightly increased likelihood. Furthermore, no apparent racial or language disparities were evident when examining the frequency of subsequent measurements, although a lower frequency was observed for women, those with private insurance, and those admitted electively. As Non-white patients were more likely to have Medicaid, there might still be disparities in care not captured in our data. Health equity has become a priority in clinical research and among policymakers not only in the US but globally ( 24 – 26 ). In recent years, significant legislative changes around AI and health equity outcomes have been proposed and implemented. The European Parliamentary Research Service conducted a study on AI in healthcare in 2022 and recommended the implementation of specific coordination and support programs to address issues pertaining to AI and bias ( 27 ). In December 2023, the European Union approved the world’s first legislation to regulate AI ( 28 ). Beyond the obvious risks associated with feeding non-representative data to a model, variation in the clinical monitoring of patients presents a problem in the development of prediction, classification and optimization models using real-world data. The non-random sparsity of data from minoritized groups, even when represented in the dataset, has implications in the development of machine learning-based decision support tools that are seldomly being investigated. Providers often intentionally refrain from measuring a variable especially in the ICU because of increasing recognition of the harm from over-testing ( 29 ). But the rationale behind such decisions is typically more complex, and confounded by both clinical and non-clinical (i.e., social determinants of care) features. In result, AI models learn wrong associations between clinical features and outcomes of interest. The problem becomes more pronounced in the advent of multi-modal modeling that requires black box deep learning representations ( 9 ). Models built on real-world data are thus subject to the human biases of the people who collected the primary data. For instance, a recent study found that large language models recommended low paying jobs more frequently to Mexicans, or implied that administrative work is solely a female job ( 30 ). In an effort to mitigate biases, some studies have suggested the use of causal inference frameworks for machine learning( 30 – 32 ), which should help understand and avoid embedding biases into AI algorithms. Evaluating data inputs used in AI models for biases and disparities as done in our work is a prerequisite even before employing causal inference frameworks and should become standard practice as the understanding gained aids in building better, more equitable, and trustworthy AI models. This study provides a framework and approach for future work, as health care professionals, engineers, and developers have the moral accountability to ensure safe deployment of AI models ( 33 , 34 ). The optimal frequency of monitoring of serum lactate measurement is unknown. Two studies in tertiary centers demonstrated that serial lactate measurements were independently associated with 28-day mortality ( 33 , 35 ) and highlight gain for patient care as the information provided by serum lactate seems not to be captured by other biomarkers or clinical scores. Many EHRs have already incorporated automated sepsis alerts to clinicians which rely on data such as the lactate to be present; disparities in collecting the data leads to disparities in usage of such alerts ( 36 , 37 ). As such, the inputted data must be evaluated for bias. Other studies have shown that racially diverse Non-White ICU patients have nearly double the incidence of sepsis and higher rates of sepsis-related mortality compared to White patients ( 36 , 38 , 39 ). Furthermore, some studies in pediatric patients have reported higher mortality rates for those of lower socioeconomic status in the ICU ( 40 ). As such, all possible efforts need to be undertaken to close this disparity in patient care. LIMITATIONS While our research provides valuable insights into the discourse on disparities and biases within critical care, it is essential to acknowledge the limitations of our study. Firstly, selection bias could be a potential concern, as our data only encompassed patients admitted to the ICU in an academic tertiary care center in the USA whose patients are predominantly White. In general, race-ethnicity is self-reported in MIMIC-IV or provided by relatives, however in instances where this was not possible, data is recorded by the providers themselves. Additionally, our study design precludes us from testing for unmeasured confounding variables. Future research endeavors should make concerted efforts to address these limitations, such as including Social Determinants of Health and fostering a more comprehensive understanding of the topic by employing causal inference frameworks as the next prerequisite step before validating AI models. Moreover, future studies should extend their scope to cover other facets of care, including emergency departments, regular wards, or ambulatory care, to provide a more holistic perspective. CONCLUSION The implications of our study extend beyond the realm of lactate monitoring during sepsis management. In addition to the ongoing challenge of achieving healthcare equity within a system marked by systemic biases, clinicians and researchers must remain cognizant of these disparities before endeavoring to enhance patient care at their local institution or constructing any AI model. These biases not only have the potential to distort predictions, but may also endanger patient’s safety when the predictions are employed for treatment or management decisions. Abbreviations ICU intensive care unit LOS length of stay Proficient English proficient Limited Prof. limited English proficiency Declarations Conflicts of Interest None of the authors have any conflicts of interest relevant to this work. Author Contributions Dr Struja and Mr. Matos designed the study and had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the analysis. Concept and design: Struja, Celi, Matos. Acquisition, analysis, and interpretation of data: All authors. Drafting of the manuscript: Struja and Abu Hussein. Critical revision of the manuscript for important intellectual content: All authors. Statistical analysis: Pradhan and Haug. Administrative, technical, or material support: Celi. Supervision: Struja, Celi, Matos. Funding The funding organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. LAC is funded by the National Institute of Health through R01 EB017205, DS-I Africa U54 TW012043-01 and Bridge2AI OT2OD032701, and the National Science Foundation through ITEST #2148451. JM was supported by a Fulbright / FLAD Grant, Portugal, AY 2022/2023. TS is supported by the Swiss National Science Foundation (P400PM_194497 / 1). NAH is supported by the Swiss National Science Foundation (P500PM_210847). Ethics approval and consent to participate The dataset in this study was obtained from MIMIC-IV. We had completed the CITI Program course known as Human Research and Data or Specimens Only Research to apply for permission to access the database. The individual information of the patients included in this database was anonymous, and ethical review and informed consent were waived. All methods were performed in accordance with the relevant guidelines and regulations. Data Availability The data that support the findings of this study are available in MIMIC-IV with the identifier doi.org/10.1093/jamia/ocx084 publicly available on PhysioNet (https://physionet.org/). Code Availability The code that produces the results in this manuscript can be accessed at https://github.com/joamats/mit-lactate, which includes detailed instructions for running the code. References Magesh S, John D, Li WT, Li Y, Mattingly-app A, Jain S, et al. Disparities in COVID-19 Outcomes by Race, Ethnicity, and Socioeconomic Status: A Systematic Review and Meta-analysis. JAMA Netw Open. 2021 Nov 11;4(11):e2134147. Hall WJ, Chapman MV, Lee KM, Merino YM, Thomas TW, Payne BK, et al. Implicit Racial/Ethnic Bias Among Health Care Professionals and Its Influence on Health Care Outcomes: A Systematic Review. Am J Public Health. 2015 Dec;105(12):e60–76. Charpignon ML, Byers J, Cabral S, Celi LA, Fernandes C, Gallifant J, et al. Critical Bias in Critical Care Devices. Crit Care Clin. 2023 Oct;39(4):795–813. Ferryman K, Mackintosh M, Ghassemi M. Considering Biased Data as Informative Artifacts in AI-Assisted Health Care. Drazen JM, editor. N Engl J Med. 2023 Aug 31;389(9):833–8. Nazer LH, Zatarah R, Waldrip S, Ke JXC, Moukheiber M, Khanna AK, et al. Bias in artificial intelligence algorithms and recommendations for mitigation. Kalla M, editor. PLOS Digit Health. 2023 Jun 22;2(6):e0000278. Chen IY, Pierson E, Rose S, Joshi S, Ferryman K, Ghassemi M. Ethical Machine Learning in Healthcare. Annu Rev Biomed Data Sci. 2021 Jul 20;4(1):123–44. Gold R, Cottrell E, Bunce A, Middendorf M, Hollombe C, Cowburn S, et al. Developing Electronic Health Record (EHR) Strategies Related to Health Center Patients’ Social Determinants of Health. J Am Board Fam Med. 2017 Jul;30(4):428–47. Wells BJ, Nowacki AS, Chagin K, Kattan MW. Strategies for Handling Missing Data in Electronic Health Record Derived Data. EGEMs Gener Evid Methods Improve Patient Outcomes. 2013 Dec 17;1(3):7. Yang Y, Zhang H, Katabi D, Ghassemi M. Change is Hard: A Closer Look at Subpopulation Shift [Internet]. arXiv; 2023 [cited 2024 Oct 15]. Available from: http://arxiv.org/abs/2302.12254 Panch T, Mattie H, Atun R. Artificial intelligence and algorithmic bias: implications for health systems. J Glob Health. 2019 Dec;9(2):010318. Martin K. Ethics of data and analytics: concepts and cases. First edition. Boca Raton, FL: CRC Press; 2022. 1 p. Singer M, Deutschman CS, Seymour CW, Shankar-Hari M, Annane D, Bauer M, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). JAMA. 2016 Feb 23;315(8):801. Surviving-Sepsis-Guidelines-2021#Recommendations [Internet]. 2024. Available from: https://www.sccm.org/Clinical-Resources/Guidelines/Guidelines/ Black LP, Hopson C, Puskarich MA, Modave F, Booker SQ, DeVos E, et al. Racial disparities in septic shock mortality: a retrospective cohort study. Lancet Reg Health - Am. 2024 Jan;29:100646. Von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. J Clin Epidemiol. 2008 Apr;61(4):344–9. Flanagin A, Frey T, Christiansen SL, AMA Manual of Style Committee. Updated Guidance on the Reporting of Race and Ethnicity in Medical and Science Journals. JAMA. 2021 Aug 17;326(7):621. Johnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, et al. MIMIC-IV, a freely accessible electronic health record dataset. Sci Data. 2023 Jan 3;10(1):1. Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov PCh, Mark RG, et al. PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals. Circulation [Internet]. 2000 Jun 13 [cited 2024 Oct 15];101(23). Available from: https://www.ahajournals.org/doi/10.1161/01.CIR.101.23.e215 Vincent JL, Moreno R, Takala J, Willatts S, De Mendonça A, Bruining H, et al. The SOFA (Sepsis-related Organ Failure Assessment) score to describe organ dysfunction/failure: On behalf of the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine (see contributors to the project in the appendix). Intensive Care Med. 1996 Jul;22(7):707–10. van Rossum G. Python reference manual. CWI.; 1995. R Core Team. R: A Language and Environment for Statistical Computing [Internet]. R Core Team, Vienna, Austria; 2022. Available from: https://www.R-project.org/ Gruber S, Laan MJVD. tmle : An R Package for Targeted Maximum Likelihood Estimation. J Stat Softw [Internet]. 2012 [cited 2024 Oct 15];51(13). Available from: http://www.jstatsoft.org/v51/i13/ Seabold S, Perktold J. Statsmodels: Econometric and Statistical Modeling with Python. In Austin, Texas; 2010 [cited 2024 Oct 15]. p. 92–6. Available from: https://doi.curvenote.com/10.25080/Majora-92bf1922-011 Centers for Medicare & Medicaid Services. CMS Proposes Policies to Improve Patient Safety and Promote Health Equity. 2023; Yao Q, Li X, Luo F, Yang L, Liu C, Sun J. The historical roots and seminal research on health equity: a referenced publication year spectroscopy (RPYS) analysis. Int J Equity Health. 2019 Dec;18(1):152. Coley RY, Duan KI, Hoopes AJ, Lapham GT, Liljenquist K, Marcotte LM, et al. A call to integrate health equity into learning health system research training. Learn Health Syst. 2022 Oct;6(4):e10330. Lekadir K, Quaglio G, Garmendia AT, Gallin C. Artificial Intelligence in Healthcare-Applications, Risks, and Ethical and Societal Impacts. [Internet]. Eur Parliam [Internet]; 2022. Available from: Available from: https://www.europarl.europa.eu/RegData/etudes/STUD/2022/729512/EPRS_STU(2022)729512_EN.pdf European Parliament [Internet]. EU AI Act: first regulation on artificial intelligence | News | . [Internet]. 2023 [cited 2023 Dec 9]. Available from: https://www.europarl.europa.eu/news/en/headlines/society/20230601STO93804/eu-ai-act-first-regulation-on-artificial-intelligence Kox M, Pickkers P. “Less Is More” in Critically Ill Patients: Not Too Intensive. JAMA Intern Med. 2013 Jul 22;173(14):1369. Scholkopf B, Locatello F, Bauer S, Ke NR, Kalchbrenner N, Goyal A, et al. Toward Causal Representation Learning. Proc IEEE. 2021 May;109(5):612–34. Struja T, Matos J, Lam B, Cao Y, Liu X, Jia Y, et al. Evaluating equitable care in the ICU: Creating a causal inference framework to assess the impact of life-sustaining interventions across racial and ethnic groups [Internet]. Health Informatics; 2023 [cited 2024 Oct 15]. Available from: http://medrxiv.org/lookup/doi/10.1101/2023.10.12.23296933 Plecko D, Bareinboim E. Causal Fairness Analysis [Internet]. arXiv; 2022 [cited 2024 Oct 15]. Available from: https://arxiv.org/abs/2207.11385 Xie Y, Zhuang D, Chen H, Zou S, Chen W, Chen Y. 28-day sepsis mortality prediction model from combined serial interleukin-6, lactate, and procalcitonin measurements: a retrospective cohort study. Eur J Clin Microbiol Infect Dis. 2023 Jan;42(1):77–85. Habli I, Lawton T, Porter Z. Artificial intelligence in health care: accountability and safety. Bull World Health Organ. 2020 Apr 1;98(4):251–6. Park H, Lee J, Oh DK, Park MH, Lim CM, Lee SM, et al. Serial evaluation of the serum lactate level with the SOFA score to predict mortality in patients with sepsis. Sci Rep. 2023 Apr 18;13(1):6351. Barnato AE, Alexander SL, Linde-Zwirble WT, Angus DC. Racial Variation in the Incidence, Care, and Outcomes of Severe Sepsis: Analysis of Population, Patient, and Hospital Characteristics. Am J Respir Crit Care Med. 2008 Feb 1;177(3):279–84. Raman J, Johnson TJ, Hayes K, Balamuth F. Racial Differences in Sepsis Recognition in the Emergency Department. Pediatrics. 2019 Oct 1;144(4):e20190348. Mayr FB. Infection Rate and Acute Organ Dysfunction Risk as Explanations for Racial Differences in Severe Sepsis. JAMA. 2010 Jun 23;303(24):2495. Martin GS, Mannino DM, Eaton S, Moss M. The Epidemiology of Sepsis in the United States from 1979 through 2000. N Engl J Med. 2003 Apr 17;348(16):1546–54. Reddy AR, Badolato GM, Chamberlain JM, Goyal MK. Disparities Associated with Sepsis Mortality in Critically Ill Children. J Pediatr Intensive Care. 2022 Jun;11(02):147–52. Additional Declarations No competing interests reported. Supplementary Files Supplementarymaterial.docx GraphicalAbstractLactate.jpg Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5836145","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":409043825,"identity":"f21a7531-8eb4-47c9-ae95-db56242e937a","order_by":0,"name":"Nebal S. Abu Hussein","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABE0lEQVRIie3QwUrDMBjA8a/s4CU6jxlj7/BBoE6QPUtCobuUIhSkoOhOeql6rW9RGXhuCbhL9RxQsA+wQ3vrQNRMPAwxG94E8z/lC/klEACb7U9GACoADvRzotAFcHK9cibrCF8lPX30VwQA8w1kb+uyqDmMQnxKWNPGw5Apr8prOBhk+c9kP3n0KAcvwufS7ZOSRq7ysUjBZyaCKkBNOiJTgdtxzqm4UxwlASmM5GXOWg5nS8KaxRsV03Rcy1d4NxNFXP2KXBKk2xMqMhqg1D9gJmXoDznOop7yj/rknoq0nB8WCXrsxkRmD1LV8XG4o7xp056ciuuL8W3VxqPBlYF8wdVhl3/f2Vh37e02m832D/sAS3ppw3NO0UEAAAAASUVORK5CYII=","orcid":"","institution":"Yale University","correspondingAuthor":true,"prefix":"","firstName":"Nebal","middleName":"S. Abu","lastName":"Hussein","suffix":""},{"id":409043826,"identity":"f0839f5a-4804-47d3-b765-3a64212a7ed2","order_by":1,"name":"Pratiksha Pradhan","email":"","orcid":"","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Pratiksha","middleName":"","lastName":"Pradhan","suffix":""},{"id":409043827,"identity":"9793204d-fc7a-4601-8dca-72480c5abaf5","order_by":2,"name":"Fredrik Willumsen Haug","email":"","orcid":"","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Fredrik","middleName":"Willumsen","lastName":"Haug","suffix":""},{"id":409043828,"identity":"45d0bea9-3450-4b14-b620-c656efaeef0a","order_by":3,"name":"Dana Moukheiber","email":"","orcid":"","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Dana","middleName":"","lastName":"Moukheiber","suffix":""},{"id":409043829,"identity":"99fde82d-1f98-4474-a206-fb39bf387c15","order_by":4,"name":"Lama Moukheiber","email":"","orcid":"","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Lama","middleName":"","lastName":"Moukheiber","suffix":""},{"id":409043830,"identity":"5dd6eeb6-f203-433a-9502-a57627f2fcde","order_by":5,"name":"Mira Moukheiber","email":"","orcid":"","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Mira","middleName":"","lastName":"Moukheiber","suffix":""},{"id":409043831,"identity":"369444f7-a343-4771-af64-91fb666f795d","order_by":6,"name":"Sulaiman Moukheiber","email":"","orcid":"","institution":"Worcester Polytechnic Institute","correspondingAuthor":false,"prefix":"","firstName":"Sulaiman","middleName":"","lastName":"Moukheiber","suffix":""},{"id":409043832,"identity":"a8d052e2-a3fa-4713-9bf2-b5585523a906","order_by":7,"name":"Luca Leon Weishaupt","email":"","orcid":"","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Luca","middleName":"Leon","lastName":"Weishaupt","suffix":""},{"id":409043833,"identity":"7443d305-d54b-4a7a-9799-ee342b22ca14","order_by":8,"name":"Jacob G. Ellen","email":"","orcid":"","institution":"Harvard University","correspondingAuthor":false,"prefix":"","firstName":"Jacob","middleName":"G.","lastName":"Ellen","suffix":""},{"id":409043834,"identity":"eb2d31a7-8787-4429-9d07-78abac2c3aff","order_by":9,"name":"Helen D’Couto","email":"","orcid":"","institution":"Georgetown University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Helen","middleName":"","lastName":"D’Couto","suffix":""},{"id":409043835,"identity":"dc8ee730-d176-44a8-b423-65357a4625b5","order_by":10,"name":"Ishan C. Williams","email":"","orcid":"","institution":"University of Virginia","correspondingAuthor":false,"prefix":"","firstName":"Ishan","middleName":"C.","lastName":"Williams","suffix":""},{"id":409043836,"identity":"e438c214-a7f2-4bea-9055-a9eeaac7ee37","order_by":11,"name":"Leo Anthony Celi","email":"","orcid":"","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Leo","middleName":"Anthony","lastName":"Celi","suffix":""},{"id":409043837,"identity":"939d9d9a-3c2e-4fbb-b492-37782d8d09e3","order_by":12,"name":"João Matos","email":"","orcid":"","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"João","middleName":"","lastName":"Matos","suffix":""},{"id":409043838,"identity":"241d8f87-61aa-49e4-991f-2fc796578cf3","order_by":13,"name":"Tristan Struja","email":"","orcid":"","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Tristan","middleName":"","lastName":"Struja","suffix":""}],"badges":[],"createdAt":"2025-01-15 16:38:05","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5836145/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5836145/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":75591658,"identity":"e59afdde-694a-4934-84d6-a7539541faed","added_by":"auto","created_at":"2025-02-06 07:10:40","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":300283,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eStudy cohort selection flow chart, MIMIC-IV\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLegend:\u003c/strong\u003e Right panels depicts the change of key demographic factors through application of the exclusion criteria\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAbbreviations:\u003c/strong\u003e ICU, intensive care unit; LOS, length of stay; Proficient, English proficient; Limited Prof., limited English proficiency\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5836145/v1/12707325e7c3380e691b49a3.jpg"},{"id":75591646,"identity":"c1389a4a-6be8-4966-8786-1ceeabb1f169","added_by":"auto","created_at":"2025-02-06 07:10:39","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":48138,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSummary of main results from TMLE and negative binomial regression models.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAbbreviations:\u003c/strong\u003e ATE, average treatment effect; IRR, incidence rate ratio; CI, confidence interval;\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5836145/v1/d02c36a03b3811bf3e8e8a4f.jpg"},{"id":75593668,"identity":"ffe66973-8ca2-475b-950b-5e2847758bca","added_by":"auto","created_at":"2025-02-06 07:26:39","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1407980,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5836145/v1/76708269-7c1e-4248-959b-a837811d3119.pdf"},{"id":75591647,"identity":"3742f9a2-a5d7-4898-974d-8f11f6b7875e","added_by":"auto","created_at":"2025-02-06 07:10:39","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":157100,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-5836145/v1/b07e9935cdc9e0d04a78891b.docx"},{"id":75592022,"identity":"764f9bd5-3baa-43d1-bf1b-58d5339f4048","added_by":"auto","created_at":"2025-02-06 07:18:39","extension":"jpg","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":789585,"visible":true,"origin":"","legend":"","description":"","filename":"GraphicalAbstractLactate.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5836145/v1/9ce8de56e96c6bfeeb300ecc.jpg"}],"financialInterests":"No competing interests reported.","formattedTitle":"Potential source of bias in AI models: Lactate measurement in the ICU as a template","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eDisparities in healthcare are widely recognized, especially regarding discrimination based on race and ethnicity (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e). Such disparities can unveil themselves as differences in quality of care, unequal medical device performance, or access to services reflecting structural inequities (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). These biases are not only harmful for patient care, but can also impact the development of machine learning-based clinical algorithms that train on electronic health records (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eEnsuring the development of fair AI models is crucial, and addressing missing information is a key initial step in achieving this objective, especially when such information is not missing at random (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). Unfortunately, this variation in the level of monitoring is often not taken into consideration in the development of machine learning-based clinical algorithms. In a 2017 study that evaluated 107 electronic health record (EHR)-based risk prediction tools, 49 did not account for missing data (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). A common approach to imputation is the use of normal values based on the assumption that laboratory tests that are not ordered are presumed to be within normal range, a practice that likely introduces bias (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe probability of detecting an abnormal finding is contingent on the frequency of testing. Consequently, non-randomly missing data can lead to spurious correlations\u0026mdash;non-causal relationships between features and outcome\u0026mdash;that are learned and then incorporated into clinical algorithms (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e). When the etiology of missing data stems from social determinants of care, these biases can become ingrained in subsequent AI models, perpetuating and even scaling existing disparities (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e). This is even more important in a high-stake environment such as in patients with sepsis admitted to the Intensive Care Unit (ICU).\u003c/p\u003e \u003cp\u003eSepsis is a severe life-threatening systemic infection and effective management of this condition requires prompt diagnosis, aggressive treatment and continuous monitoring. Despite current advances, one key challenge remains the timely delivery of care. Herein, serum lactate level is one of the two key diagnostic tools of septic shock according to the guidelines (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). Disparities in sepsis outcomes are known to exist (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). However, the drivers of sepsis disparities are unknown and the question of whether disparities extend to serum lactate monitoring remains underexplored.\u003c/p\u003e \u003cp\u003eThis paper seeks to examine whether race and ethnicity, sex, and language differences influence the frequency of serum lactate determination conducted during the management of sepsis in the ICU. By shedding light on this dimension of care, we aim to contribute to a more comprehensive understanding of the social patterning of the data generation process in healthcare.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eThis observational retrospective study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e). The health equity language, narrative and concepts of this paper follows the American Medical Association\u0026rsquo;s recommendations (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e).\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData Extraction\u003c/h2\u003e \u003cp\u003eData was extracted from the publicly available MIMIC-IV database (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e). The MIMIC database is maintained by the Laboratory for Computational Physiology at the Massachusetts Institute of Technology and shared via the PhysioNet platform (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e). The dataset has been de-identified, and the institutional review boards of the Massachusetts Institute of Technology (No. 0403000206) and Beth Israel Deaconess Medical Center (2001-P-001699/14) both approved the use of the database for research. The MIMIC-IV database includes physiologic data collected from bedside monitors, laboratory test results, medications, medical images and clinical progress notes captured in the electronic health record from patients admitted to the ICU between 2008\u0026ndash;2019.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eHypothesis\u003c/h3\u003e\n\u003cp\u003eWe hypothesized that both the likelihood for a patient to have a serum lactate measurement and the frequency of subsequent measurements are not the same across race-ethnicity, sex, and English proficiency (as recorded by providers).\u003c/p\u003e\n\u003ch3\u003eCohort Selection\u003c/h3\u003e\n\u003cp\u003eThe following exclusion criteria were applied to create a study cohort: those without sepsis as defined by the sepsis-3 criteria (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e), patients under 18 years of age, and those with length of ICU stay less than 1 day. Patients with recurrent hospital stays in the database, and those with a racial description other than White, Asian, Black, or Hispanic, especially excluding those of the heterogenous group \u0026ldquo;other\u0026rdquo;. For the negative binomial regression, we further excluded patients with absent serum lactate values on day one.\u003c/p\u003e\n\u003ch3\u003eCovariates\u003c/h3\u003e\n\u003cp\u003eWe drew directed acyclic graphs (DAG) to understand which variables to extract (\u003cb\u003eSupplementary Fig.\u0026nbsp;1, Supplementary Table\u0026nbsp;1\u003c/b\u003e). Twelve confounders were extracted, including non-time-varying variables such as demographics, comorbidities, admission information, and source of infection and time-varying variables including Sequential Organ Failure Assessment (SOFA) score (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e), and fluids normalized by length of stay. Time-varying variables were modeled as follows: SOFA score was calculated for the day of ICU admission; serum lactate measurements were used as a binary variable for whether or not it was measured on day one, in addition to taking the overall number of measurements for the whole ICU stay normalized by length of ICU stay.\u003c/p\u003e\n\u003ch3\u003eOutcomes\u003c/h3\u003e\n\u003cp\u003eWe had two primary outcomes: the first was a binary variable predicting whether a patient received serum lactate measurement on day 1, and the second was a prediction of how many lactate measurements a patient would receive per day throughout the length of their ICU stay.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eStatistical analysis was performed using Python 3.10.9 (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e) and R 4.2.1 (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). For the outcome of whether or not a patient had a serum lactate measurement on day 1, we fitted a Targeted Maximum Likelihood Estimation (TMLE) model (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e). From the TMLE model, we extracted and utilized the odds ratio (OR) to estimate the likelihood of receiving a serum lactate measurement. For the outcome of the number of serum lactate measurements during an ICU stay, we fitted a non-penalized, negative binomial regression (\u003cem\u003estatsmodel\u003c/em\u003e package (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e)) adjusted for confounders to estimate the number of serum lactate measurements for each patient each day in the ICU. We report our findings as incident rate ratios (IRR). All findings are reported with 95% CI and with White patients as the reference group.\u003c/p\u003e \u003c/div\u003e"},{"header":"RESULTS","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\n \u003ch2\u003eBaseline Study Cohort\u003c/h2\u003e\n \u003cp\u003eThe MIMIC-IV database has 73,140 ICU stays, of which 15,601 were included in our final cohort following application of the inclusion and exclusion criteria (Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e). The race-ethnicity distribution was 10.8% Black, 3.8% Hispanic, 2.9% Asian, 68.8% White and 14.6% others (without specified race). The demographic distribution did not change after applying exclusion criteria.\u003c/p\u003e\n \u003cp\u003eSOFA score had a median of 6.00 (interquartile range (IQR) 4.00, 8.00), regardless of the race-ethnicity reported at baseline, with the Charlson comorbidity index at 6.00 (IQR 4.00, 8.00). Serum lactate on day 1 was slightly higher in the Non-White group at 2.50mmol/l (IQR 1.60, 4.00), compared to the White group at 2.20 (1.50, 3.50). In addition, Non-White patients received more fluids on the first day in the ICU than White patients (2,060 ml (IQR 640, 5,000) versus 1,690 (461, 4,540)),respectively. Of note, the volume of fluids received prior to admission to the ICU is not available in the dataset (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eBaseline information on the study cohort, derived from MIMIC-IV\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"6\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\" colspan=\"5\"\u003e\n \u003cp\u003eRace and Ethnicity\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMissing\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eOverall\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNon-White\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eWhite\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eN (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15,601 (100)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2,801 (17.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12,800 (82.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eAge, median [Q1,Q3]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e68.0 (57.0, 78.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e64.0 (52.0, 76.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e68.0 (59.0, 79.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eSex, n (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6,520 (41.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1,341 (47.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5,179 (40.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eEnglish proficient, n (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14,113 (90.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1,894 (67.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12,219 (95.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"3\"\u003e\n \u003cp\u003e\u003cstrong\u003eInsurance, n (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMedicaid\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1,042 (6.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e398 (14.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e644 (5.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMedicare\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7,476 (47.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1,064 (38.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6,412 (50.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOther\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7,083 (45.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1,339 (47.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5,744 (44.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eCharlson comorbidity index, mean [Q1,Q3]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.00 (4.00, 8.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.00 (4.00, 8.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.00 (4.00, 8.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eSOFA, median [Q1,Q3]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.00 (4.00, 8.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.00 (4.00, 9.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.00 (4.00, 8.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eElective admission, n (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2,876 (18.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e312 (11.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2,564 (20.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eLength of stay, median [Q1,Q3]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003edays\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.13 (1.83, 6.25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.21 (1.88, 6.83)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.13 (1.83, 6.17)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eLactate day 1 (mmol/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.20 (1.50, 3.50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.50 (1.60, 4.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.20 (1.50, 3.40)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eNumber of lactate measurements day 1, median [Q1,Q3]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.00 (2.00, 5.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.00 (2.00, 5.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.00 (2.00, 5.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eLactate day 2 (mmol/L)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9,397 (60.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.70 (1.20, 2.60)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.80 (1.30, 2.90)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.70 (1.20, 2.60)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eNumber of lactate measurements day 2, median [Q1,Q3]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9,397 (60.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.00 (1.00, 3.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.00 (1.00, 3.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.00 (1.00, 3.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eMechanical Ventilation, n (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8,841 (56.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1,566 (55.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7,275 (56.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eRenal Replacement Therapy, n (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1,550 (9.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e397 (14.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1,153 (9.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eVasopressor(s), n (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9,243 (59.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1,455 (51.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7,788 (60.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eFluids received day 1 (mL), median [Q1,Q3]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e446 (2.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1,750 (498, 4,620)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2,060 (640, 5,000)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1,690 (461, 4,540)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003cstrong\u003eAbbreviations:\u003c/strong\u003e Q1, lower quartile range; Q3, upper quartile range; SOFA, sequential organ failure assessment\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003eModel Results\u003c/h2\u003e\n \u003cp\u003eWe adjusted our models for confounders according to a DAG (\u003cstrong\u003eSupplementary Fig.\u0026nbsp;1, Supplementary Table\u0026nbsp;1\u003c/strong\u003e). Using the TMLE model with being White, male and English proficient as a reference, Black patients were more likely to have a serum lactate measurement on day 1 with OR 1.19 (95% CI 1.06, 1.34). Asian and Hispanic patients had a similar likelihood compared to White patients, with an OR of 1.08 (95% CI 0.93, 1.24), and an of OR 0.98 (95% CI 0.89, 1.08), respectively (Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e, Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003ea). We validated these findings with a cross-validated logistic regression model (\u003cstrong\u003eSupplementary Table\u0026nbsp;2\u003c/strong\u003e).\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eLikelihood of receiving a lactate measurement on day 1 fitted by a Targeted Maximum Likelihood Estimation model\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"4\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDemographic\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eOR\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e2.50% CI\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e97.5% CI\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eWhite\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003e\u003cem\u003eReference\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBlack\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.34\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAsian\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.24\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHispanic\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.08\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eMale\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003e\u003cem\u003eReference\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.09\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eEnglish Proficient\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003e\u003cem\u003eReference\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEnglish Non-Proficient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.07\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003cstrong\u003eAbbreviations:\u003c/strong\u003e OR, odds ratio; CI, confidence intervall\u003c/p\u003e\n \u003cp\u003eThe negative binomial model was fitted to predict the total frequency of serum lactate measurements during a patient\u0026rsquo;s ICU stay (Table \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e, Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003eb). We found no significant difference in the frequency of measurements across race-ethnicities compared to Whites as reference. Hispanic (IRR 1.12, 95% CI 0.99, 1.26), Black (IRR 1.01, 95% CI 0.94, 1.09), and Asian (IRR 1.08, 95% CI 0.95, 1.23) patients had a non-significant difference in their frequency of serum lactate measurements. In addition, English proficiency had no significant impact on measurement frequency (IRR 1.06, 95% CI 0.97, 1.16). On the other hand, female sex (IRR 0.94, 95% CI 0.90, 0.98) and having a urinary tract infection (IRR 0.68, 95% CI 0.50, 0.93) were associated with a decreased serum lactate measurement frequency, while having private insurance (IRR 1.07, 95% CI 1.02, 1.12) and being admitted electively (IRR 1.7, 95% CI 1.61, 1.81) significantly increased the frequency of receiving a measurement.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eResults of the negative binomial regression for outcome of lactate measurement frequency on day 1\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"4\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eVariable\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eIRR\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e2.5% CI\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e97.5% CI\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIntercept\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCharlson comorbidity index\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.02\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSOFA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eRace\u003c/strong\u003e:\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eWhite\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003e\u003cem\u003eReference\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAsian\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.23\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBlack\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.09\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHispanic\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.26\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eBinary variables\u003c/strong\u003e:\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFemale sex\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEnglish proficient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.16\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePrivate insurance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.12\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eElective admission\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.81\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVolume of fluids normalized by LOS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePneumonia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.13\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eUrinary tract infection\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBiliary infection\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.84\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSkin infection\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.72\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003cstrong\u003eAbbreviations:\u0026nbsp;\u003c/strong\u003eIRR, incidence rate ratio; CI, confidence Interval; SOFA, sequential organ failure assessment; LOS, length of stay;\u0026nbsp;\u003c/p\u003e\n\u003c/div\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eIn this retrospective cohort study in patients with sepsis, we observed no discernible disparities between sexes and non-native English speakers in receiving a serum lactate measurement on day one, although Black patients had a slightly increased likelihood. Furthermore, no apparent racial or language disparities were evident when examining the frequency of subsequent measurements, although a lower frequency was observed for women, those with private insurance, and those admitted electively. As Non-white patients were more likely to have Medicaid, there might still be disparities in care not captured in our data.\u003c/p\u003e \u003cp\u003eHealth equity has become a priority in clinical research and among policymakers not only in the US but globally (\u003cspan additionalcitationids=\"CR25\" citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e). In recent years, significant legislative changes around AI and health equity outcomes have been proposed and implemented. The European Parliamentary Research Service conducted a study on AI in healthcare in 2022 and recommended the implementation of specific coordination and support programs to address issues pertaining to AI and bias (\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). In December 2023, the European Union approved the world\u0026rsquo;s first legislation to regulate AI (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eBeyond the obvious risks associated with feeding non-representative data to a model, variation in the clinical monitoring of patients presents a problem in the development of prediction, classification and optimization models using real-world data. The non-random sparsity of data from minoritized groups, even when represented in the dataset, has implications in the development of machine learning-based decision support tools that are seldomly being investigated. Providers often intentionally refrain from measuring a variable especially in the ICU because of increasing recognition of the harm from over-testing (\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e). But the rationale behind such decisions is typically more complex, and confounded by both clinical and non-clinical (i.e., social determinants of care) features. In result, AI models learn wrong associations between clinical features and outcomes of interest. The problem becomes more pronounced in the advent of multi-modal modeling that requires black box deep learning representations (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e). Models built on real-world data are thus subject to the human biases of the people who collected the primary data. For instance, a recent study found that large language models recommended low paying jobs more frequently to Mexicans, or implied that administrative work is solely a female job (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn an effort to mitigate biases, some studies have suggested the use of causal inference frameworks for machine learning(\u003cspan additionalcitationids=\"CR31\" citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e), which should help understand and avoid embedding biases into AI algorithms. Evaluating data inputs used in AI models for biases and disparities as done in our work is a prerequisite even before employing causal inference frameworks and should become standard practice as the understanding gained aids in building better, more equitable, and trustworthy AI models. This study provides a framework and approach for future work, as health care professionals, engineers, and developers have the moral accountability to ensure safe deployment of AI models (\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe optimal frequency of monitoring of serum lactate measurement is unknown. Two studies in tertiary centers demonstrated that serial lactate measurements were independently associated with 28-day mortality (\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e) and highlight gain for patient care as the information provided by serum lactate seems not to be captured by other biomarkers or clinical scores. Many EHRs have already incorporated automated sepsis alerts to clinicians which rely on data such as the lactate to be present; disparities in collecting the data leads to disparities in usage of such alerts (\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e). As such, the inputted data must be evaluated for bias. Other studies have shown that racially diverse Non-White ICU patients have nearly double the incidence of sepsis and higher rates of sepsis-related mortality compared to White patients (\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e). Furthermore, some studies in pediatric patients have reported higher mortality rates for those of lower socioeconomic status in the ICU (\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e). As such, all possible efforts need to be undertaken to close this disparity in patient care.\u003c/p\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eLIMITATIONS\u003c/h2\u003e \u003cp\u003eWhile our research provides valuable insights into the discourse on disparities and biases within critical care, it is essential to acknowledge the limitations of our study. Firstly, selection bias could be a potential concern, as our data only encompassed patients admitted to the ICU in an academic tertiary care center in the USA whose patients are predominantly White. In general, race-ethnicity is self-reported in MIMIC-IV or provided by relatives, however in instances where this was not possible, data is recorded by the providers themselves. Additionally, our study design precludes us from testing for unmeasured confounding variables. Future research endeavors should make concerted efforts to address these limitations, such as including Social Determinants of Health and fostering a more comprehensive understanding of the topic by employing causal inference frameworks as the next prerequisite step before validating AI models. Moreover, future studies should extend their scope to cover other facets of care, including emergency departments, regular wards, or ambulatory care, to provide a more holistic perspective.\u003c/p\u003e \u003c/div\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eThe implications of our study extend beyond the realm of lactate monitoring during sepsis management. In addition to the ongoing challenge of achieving healthcare equity within a system marked by systemic biases, clinicians and researchers must remain cognizant of these disparities before endeavoring to enhance patient care at their local institution or constructing any AI model. These biases not only have the potential to distort predictions, but may also endanger patient\u0026rsquo;s safety when the predictions are employed for treatment or management decisions.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eICU\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eintensive care unit\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLOS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003elength of stay\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eProficient\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eEnglish proficient\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLimited Prof.\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003elimited English proficiency\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eConflicts of Interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNone of the authors have any conflicts of interest relevant to this work.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDr Struja and Mr. Matos designed the study and had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the analysis.\u003c/p\u003e\n\u003cp\u003eConcept and design: Struja, Celi, Matos.\u003c/p\u003e\n\u003cp\u003eAcquisition, analysis, and interpretation of data: All authors.\u003c/p\u003e\n\u003cp\u003eDrafting of the manuscript: Struja and Abu Hussein.\u003c/p\u003e\n\u003cp\u003eCritical revision of the manuscript for important intellectual content: All authors.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eStatistical analysis: Pradhan and Haug.\u003c/p\u003e\n\u003cp\u003eAdministrative, technical, or material support: Celi.\u003c/p\u003e\n\u003cp\u003eSupervision: Struja, Celi, Matos.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe funding organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.\u003c/p\u003e\n\u003cp\u003eLAC is funded by the National Institute of Health through R01 EB017205, DS-I Africa U54 TW012043-01 and Bridge2AI OT2OD032701, and the National Science Foundation through ITEST #2148451.\u003c/p\u003e\n\u003cp\u003eJM was supported by a Fulbright / FLAD Grant, Portugal, AY 2022/2023.\u003c/p\u003e\n\u003cp\u003eTS is supported by the Swiss National Science Foundation (P400PM_194497 / 1).\u003c/p\u003e\n\u003cp\u003eNAH is supported by the Swiss National Science Foundation (P500PM_210847).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe dataset in this study was obtained from MIMIC-IV. We had completed the CITI Program course known as Human Research and Data or Specimens Only Research to apply for permission to access the database. The individual information of the patients included in this database was anonymous, and ethical review and informed consent were waived. All methods were performed in accordance with the relevant guidelines and regulations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data that support the findings of this study are available in MIMIC-IV with the identifier doi.org/10.1093/jamia/ocx084 publicly available on PhysioNet (https://physionet.org/).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe code that produces the results in this manuscript can be accessed at https://github.com/joamats/mit-lactate, which includes detailed instructions for running the code.\u003cstrong\u003e\u003c/strong\u003e\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eMagesh S, John D, Li WT, Li Y, Mattingly-app A, Jain S, et al. Disparities in COVID-19 Outcomes by Race, Ethnicity, and Socioeconomic Status: A Systematic Review and Meta-analysis. JAMA Netw Open. 2021 Nov 11;4(11):e2134147.\u003c/li\u003e\n \u003cli\u003eHall WJ, Chapman MV, Lee KM, Merino YM, Thomas TW, Payne BK, et al. Implicit Racial/Ethnic Bias Among Health Care Professionals and Its Influence on Health Care Outcomes: A Systematic Review. Am J Public Health. 2015 Dec;105(12):e60\u0026ndash;76.\u003c/li\u003e\n \u003cli\u003eCharpignon ML, Byers J, Cabral S, Celi LA, Fernandes C, Gallifant J, et al. Critical Bias in Critical Care Devices. Crit Care Clin. 2023 Oct;39(4):795\u0026ndash;813.\u003c/li\u003e\n \u003cli\u003eFerryman K, Mackintosh M, Ghassemi M. Considering Biased Data as Informative Artifacts in AI-Assisted Health Care. Drazen JM, editor. N Engl J Med. 2023 Aug 31;389(9):833\u0026ndash;8.\u003c/li\u003e\n \u003cli\u003eNazer LH, Zatarah R, Waldrip S, Ke JXC, Moukheiber M, Khanna AK, et al. Bias in artificial intelligence algorithms and recommendations for mitigation. Kalla M, editor. PLOS Digit Health. 2023 Jun 22;2(6):e0000278.\u003c/li\u003e\n \u003cli\u003eChen IY, Pierson E, Rose S, Joshi S, Ferryman K, Ghassemi M. Ethical Machine Learning in Healthcare. Annu Rev Biomed Data Sci. 2021 Jul 20;4(1):123\u0026ndash;44.\u003c/li\u003e\n \u003cli\u003eGold R, Cottrell E, Bunce A, Middendorf M, Hollombe C, Cowburn S, et al. Developing Electronic Health Record (EHR) Strategies Related to Health Center Patients\u0026rsquo; Social Determinants of Health. J Am Board Fam Med. 2017 Jul;30(4):428\u0026ndash;47.\u003c/li\u003e\n \u003cli\u003eWells BJ, Nowacki AS, Chagin K, Kattan MW. Strategies for Handling Missing Data in Electronic Health Record Derived Data. EGEMs Gener Evid Methods Improve Patient Outcomes. 2013 Dec 17;1(3):7.\u003c/li\u003e\n \u003cli\u003eYang Y, Zhang H, Katabi D, Ghassemi M. Change is Hard: A Closer Look at Subpopulation Shift [Internet]. arXiv; 2023 [cited 2024 Oct 15]. Available from: http://arxiv.org/abs/2302.12254\u003c/li\u003e\n \u003cli\u003ePanch T, Mattie H, Atun R. Artificial intelligence and algorithmic bias: implications for health systems. J Glob Health. 2019 Dec;9(2):010318.\u003c/li\u003e\n \u003cli\u003eMartin K. Ethics of data and analytics: concepts and cases. First edition. Boca Raton, FL: CRC Press; 2022. 1 p.\u003c/li\u003e\n \u003cli\u003eSinger M, Deutschman CS, Seymour CW, Shankar-Hari M, Annane D, Bauer M, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). JAMA. 2016 Feb 23;315(8):801.\u003c/li\u003e\n \u003cli\u003eSurviving-Sepsis-Guidelines-2021#Recommendations [Internet]. 2024. Available from: https://www.sccm.org/Clinical-Resources/Guidelines/Guidelines/\u003c/li\u003e\n \u003cli\u003eBlack LP, Hopson C, Puskarich MA, Modave F, Booker SQ, DeVos E, et al. Racial disparities in septic shock mortality: a retrospective cohort study. Lancet Reg Health - Am. 2024 Jan;29:100646.\u003c/li\u003e\n \u003cli\u003eVon Elm E, Altman DG, Egger M, Pocock SJ, G\u0026oslash;tzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. J Clin Epidemiol. 2008 Apr;61(4):344\u0026ndash;9.\u003c/li\u003e\n \u003cli\u003eFlanagin A, Frey T, Christiansen SL, AMA Manual of Style Committee. Updated Guidance on the Reporting of Race and Ethnicity in Medical and Science Journals. JAMA. 2021 Aug 17;326(7):621.\u003c/li\u003e\n \u003cli\u003eJohnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, et al. MIMIC-IV, a freely accessible electronic health record dataset. Sci Data. 2023 Jan 3;10(1):1.\u003c/li\u003e\n \u003cli\u003eGoldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov PCh, Mark RG, et al. PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals. Circulation [Internet]. 2000 Jun 13 [cited 2024 Oct 15];101(23). Available from: https://www.ahajournals.org/doi/10.1161/01.CIR.101.23.e215\u003c/li\u003e\n \u003cli\u003eVincent JL, Moreno R, Takala J, Willatts S, De Mendon\u0026ccedil;a A, Bruining H, et al. The SOFA (Sepsis-related Organ Failure Assessment) score to describe organ dysfunction/failure: On behalf of the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine (see contributors to the project in the appendix). Intensive Care Med. 1996 Jul;22(7):707\u0026ndash;10.\u003c/li\u003e\n \u003cli\u003evan Rossum G. Python reference manual. CWI.; 1995.\u003c/li\u003e\n \u003cli\u003eR Core Team. R: A Language and Environment for Statistical Computing [Internet]. R Core Team, Vienna, Austria; 2022. Available from: https://www.R-project.org/\u003c/li\u003e\n \u003cli\u003eGruber S, Laan MJVD. \u003cstrong\u003etmle\u003c/strong\u003e : An \u003cem\u003eR\u003c/em\u003e Package for Targeted Maximum Likelihood Estimation. J Stat Softw [Internet]. 2012 [cited 2024 Oct 15];51(13). Available from: http://www.jstatsoft.org/v51/i13/\u003c/li\u003e\n \u003cli\u003eSeabold S, Perktold J. Statsmodels: Econometric and Statistical Modeling with Python. In Austin, Texas; 2010 [cited 2024 Oct 15]. p. 92\u0026ndash;6. Available from: https://doi.curvenote.com/10.25080/Majora-92bf1922-011\u003c/li\u003e\n \u003cli\u003eCenters for Medicare \u0026amp; Medicaid Services. CMS Proposes Policies to Improve Patient Safety and Promote Health Equity. 2023;\u003c/li\u003e\n \u003cli\u003eYao Q, Li X, Luo F, Yang L, Liu C, Sun J. The historical roots and seminal research on health equity: a referenced publication year spectroscopy (RPYS) analysis. Int J Equity Health. 2019 Dec;18(1):152.\u003c/li\u003e\n \u003cli\u003eColey RY, Duan KI, Hoopes AJ, Lapham GT, Liljenquist K, Marcotte LM, et al. A call to integrate health equity into learning health system research training. Learn Health Syst. 2022 Oct;6(4):e10330.\u003c/li\u003e\n \u003cli\u003eLekadir K, Quaglio G, Garmendia AT, Gallin C. Artificial Intelligence in Healthcare-Applications, Risks, and Ethical and Societal Impacts. [Internet]. Eur Parliam [Internet]; 2022. Available from: Available from: https://www.europarl.europa.eu/RegData/etudes/STUD/2022/729512/EPRS_STU(2022)729512_EN.pdf\u003c/li\u003e\n \u003cli\u003eEuropean Parliament [Internet]. EU AI Act: first regulation on artificial intelligence | News | . [Internet]. 2023 [cited 2023 Dec 9]. Available from: https://www.europarl.europa.eu/news/en/headlines/society/20230601STO93804/eu-ai-act-first-regulation-on-artificial-intelligence\u003c/li\u003e\n \u003cli\u003eKox M, Pickkers P. \u0026ldquo;Less Is More\u0026rdquo; in Critically Ill Patients: Not Too Intensive. JAMA Intern Med. 2013 Jul 22;173(14):1369.\u003c/li\u003e\n \u003cli\u003eScholkopf B, Locatello F, Bauer S, Ke NR, Kalchbrenner N, Goyal A, et al. Toward Causal Representation Learning. Proc IEEE. 2021 May;109(5):612\u0026ndash;34.\u003c/li\u003e\n \u003cli\u003eStruja T, Matos J, Lam B, Cao Y, Liu X, Jia Y, et al. Evaluating equitable care in the ICU: Creating a causal inference framework to assess the impact of life-sustaining interventions across racial and ethnic groups [Internet]. Health Informatics; 2023 [cited 2024 Oct 15]. Available from: http://medrxiv.org/lookup/doi/10.1101/2023.10.12.23296933\u003c/li\u003e\n \u003cli\u003ePlecko D, Bareinboim E. Causal Fairness Analysis [Internet]. arXiv; 2022 [cited 2024 Oct 15]. Available from: https://arxiv.org/abs/2207.11385\u003c/li\u003e\n \u003cli\u003eXie Y, Zhuang D, Chen H, Zou S, Chen W, Chen Y. 28-day sepsis mortality prediction model from combined serial interleukin-6, lactate, and procalcitonin measurements: a retrospective cohort study. Eur J Clin Microbiol Infect Dis. 2023 Jan;42(1):77\u0026ndash;85.\u003c/li\u003e\n \u003cli\u003eHabli I, Lawton T, Porter Z. Artificial intelligence in health care: accountability and safety. Bull World Health Organ. 2020 Apr 1;98(4):251\u0026ndash;6.\u003c/li\u003e\n \u003cli\u003ePark H, Lee J, Oh DK, Park MH, Lim CM, Lee SM, et al. Serial evaluation of the serum lactate level with the SOFA score to predict mortality in patients with sepsis. Sci Rep. 2023 Apr 18;13(1):6351.\u003c/li\u003e\n \u003cli\u003eBarnato AE, Alexander SL, Linde-Zwirble WT, Angus DC. Racial Variation in the Incidence, Care, and Outcomes of Severe Sepsis: Analysis of Population, Patient, and Hospital Characteristics. Am J Respir Crit Care Med. 2008 Feb 1;177(3):279\u0026ndash;84.\u003c/li\u003e\n \u003cli\u003eRaman J, Johnson TJ, Hayes K, Balamuth F. Racial Differences in Sepsis Recognition in the Emergency Department. Pediatrics. 2019 Oct 1;144(4):e20190348.\u003c/li\u003e\n \u003cli\u003eMayr FB. Infection Rate and Acute Organ Dysfunction Risk as Explanations for Racial Differences in Severe Sepsis. JAMA. 2010 Jun 23;303(24):2495.\u003c/li\u003e\n \u003cli\u003eMartin GS, Mannino DM, Eaton S, Moss M. The Epidemiology of Sepsis in the United States from 1979 through 2000. N Engl J Med. 2003 Apr 17;348(16):1546\u0026ndash;54.\u003c/li\u003e\n \u003cli\u003eReddy AR, Badolato GM, Chamberlain JM, Goyal MK. Disparities Associated with Sepsis Mortality in Critically Ill Children. J Pediatr Intensive Care. 2022 Jun;11(02):147\u0026ndash;52.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Sepsis, lactate, MIMIC-IV, Critical Care, Health Equity","lastPublishedDoi":"10.21203/rs.3.rs-5836145/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5836145/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eObjective:\u003c/strong\u003e Health inequities may be driven by demographics such as sex, language proficiency, and race-ethnicity. These disparities may manifest through likelihood of testing, which in turn can bias artificial intelligence models. The goal of this study is to evaluate variation in serum lactate measurements in the Intensive Care Unit (ICU).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods:\u003c/strong\u003e Utilizing MIMIC-IV (2008-2019), we identified adults fulfilling sepsis-3 criteria. Exclusion criteria were ICU stay \u0026lt;1-day, unknown race-ethnicity, \u0026lt;18 years of age, and recurrent stays. Employing targeted maximum likelihood estimation analysis, we assessed the likelihood of a lactate measurement on day 1. For patients with a measurement on day 1, we evaluated the predictors of subsequent readings.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults: \u003c/strong\u003eWe studied 15,601 patients (19.5% racial-ethnic minority, 42.4% female, and 10.0% limited English proficiency). After adjusting for confounders, Black patients had a slightly higher likelihood of receiving a lactate measurement on day 1 (odds ratio 1.19, 95% confidence interval (CI) 1.06-1.34), but not the other minority groups. Subsequent frequency was similar across race-ethnicities, but women had a lower incidence rate ratio (IRR) 0.94 (95% CI 0.90-0.98). Interestingly, patients with elective admission and private insurance also had a higher frequency of repeated serum lactate measurements (IRR 1.70, 95% CI 1.61-1.81, and 1.07, 95% CI, 1.02-1.12, respectively).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion: \u003c/strong\u003eWe found no disparities in the likelihood of a lactate measurement among patients with sepsis across demographics, except for a small increase for Black patients, and a reduced frequency for women. Variation in biomarker monitoring can be a source of data bias when modeling patient outcomes, and thus should be accounted for in every analysis.\u003c/p\u003e","manuscriptTitle":"Potential source of bias in AI models: Lactate measurement in the ICU as a template","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-02-06 07:10:34","doi":"10.21203/rs.3.rs-5836145/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c51ef1fd-2b6b-4a35-a59d-e5b90acc0615","owner":[],"postedDate":"February 6th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-02-06T07:10:34+00:00","versionOfRecord":[],"versionCreatedAt":"2025-02-06 07:10:34","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5836145","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5836145","identity":"rs-5836145","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}