Analysis of Discrepancies Between Pulse Oximetry and Arterial Oxygen Saturation Measurements by Race and Ethnicity and Association With Organ Dysfunction and Mortality | Emergency Medicine | JAMA Network Open | JAMA Network
[Skip to Navigation]
Sign In
Figure 1.  Flow Diagram
Flow Diagram

Aggregate flow diagram for all data sets. ABG indicates arterial blood gas; Sao2, arterial oxygen saturation; Spo2, oxygen saturation by pulse oximetry.

Figure 2.  Modified Bland-Altman Plots by Race and Ethnicity
Modified Bland-Altman Plots by Race and Ethnicity

On each plot, the bold horizontal lines represent the mean bias (defined as the difference between the oxygen saturation measured by pulse oximetry [Spo2] and arterial blood gas value [Sao2]) for each of the 2 Spo2 groups. The blue lines, with dashed blue lines indicating 95% CIs, are for the group with Spo2 of 88% to 92%, and the orange lines, with dashed orange lines indicating 95% CIs, for the group with Spo2 of 93% to 96%. The solid black line indicates the absence of bias (ie, Spo2 − Sao2 = 0). When the mean bias is above the black line, there is positive bias in the pulse oximetry measurement (ie, the observed Spo2 is greater than the true Sao2). In contrast, when the mean bias is below the black line, there is negative bias in the pulse oximetry measurement (ie, the observed Spo2 is below the true Sao2).

Table 1.  Characteristics of Patients with Spo2 of at Least 88% 5 Minutes Before First ABG Measurementa
Characteristics of Patients with Spo2 of at Least 88% 5 Minutes Before First ABG Measurementa
Table 2.  Descriptive Statistics for Patients With Hidden Hypoxemia vs Patients Without Hypoxemiaa
Descriptive Statistics for Patients With Hidden Hypoxemia vs Patients Without Hypoxemiaa
Table 3.  Risk of Hidden Hypoxemia by Race and Ethnicitya
Risk of Hidden Hypoxemia by Race and Ethnicitya
Supplement.

eAppendix 1. Supplementary Methods

eAppendix 2. Supplementary Results

eFigure 1. Spo2-Sao2 by Race and Ethnicity, for All ABG Measurements Taken Throughout 268 904 Hospital Encounters

eFigure 2. Spo2-Sao2 by Race and Ethnicity, for the First ABG Measurement per 87 971 Hospital Encounters

eFigure 3. Spo2-Sao2 by Race and Ethnicity, for the First ABG Measurement per 87 971 Hospital Encounters, Stratified by Age Group

eFigure 4. Spo2-Sao2 by Race and Ethnicity, for the First ABG Measurement per 87 971 Hospital Encounters, Stratified by Sex

eFigure 5. Spo2-Sao2 by Race and Ethnicity, for the First ABG Measurement per 87 971 Hospital Encounter, Stratified by Cardiovascular SOFA Score

eFigure 6. Directed Acyclic Graph for Race and Ethnicity, Hidden Hypoxemia, Organ Dysfunction, and Mortality

eTable 1. Literature Review

eTable 2. SOFA Score Components, as per Vincent et al, 1996

eTable 3. Other Patient Characteristics

eTable 4. Characterization of Variable Missingness Across the 5 EHR Data Sets

eTable 5. Spo2 Variability for All Spo2 Values Within the 5 Minutes Preceding the ABG Measurement

eTable 6. Percentage of Encounters With Arterial Blood Gases, by Race and Ethnicity

eTable 7. Rate of ABG Measurements Obtained Throughout a Hospital Encounter by Race and Ethnicity, Stratified by Cardiovascular SOFA Score

eTable 8. Descriptive Statistics for All Sao2-Spo2 Pairs With Spo2 of at least 88%

eTable 9. Total Number of Patients per Spo2 Level and Characterization of Hidden Hypoxemia Incidence, Stratified by Race and Ethnicity

eTable 10. Characterization of the Distribution of Hidden Hypoxemia by Respiratory SOFA Score

eTable 11. Distribution of In-Hospital Mortality, Stratified by Respiratory SOFA Score and Presence of Hidden Hypoxemia

eTable 12. Descriptive Statistics for Patients With Hidden Hypoxemia vs Patients Without Hypoxemia, Carboxyhemoglobin Less than 2 and Methemoglobin Less Than 2

1.
Sjoding  MW, Dickson  RP, Iwashyna  TJ, Gay  SE, Valley  TS.  Racial bias in pulse oximetry measurement.   N Engl J Med. 2020;383(25):2477-2478. doi:10.1056/NEJMc2029240PubMedGoogle ScholarCrossref
2.
US Food and Drug Administration. Pulse oximeter accuracy and limitations: FDA safety communication. February 19, 2021. Accessed March 22, 2021. https://www.fda.gov/medical-devices/safety-communications/pulse-oximeter-accuracy-and-limitations-fda-safety-communication
3.
Seguin  P, Le Rouzo  A, Tanguy  M, Guillou  YM, Feuillu  A, Mallédant  Y.  Evidence for the need of bedside accuracy of pulse oximetry in an intensive care unit.   Crit Care Med. 2000;28(3):703-706. doi:10.1097/00003246-200003000-00017PubMedGoogle ScholarCrossref
4.
Van de Louw  A, Cracco  C, Cerf  C,  et al.  Accuracy of pulse oximetry in the intensive care unit.   Intensive Care Med. 2001;27(10):1606-1613. doi:10.1007/s001340101064PubMedGoogle ScholarCrossref
5.
Perkins  GD, McAuley  DF, Giles  S, Routledge  H, Gao  F.  Do changes in pulse oximeter oxygen saturation predict equivalent changes in arterial oxygen saturation?   Crit Care. 2003;7(4):R67. doi:10.1186/cc2339PubMedGoogle ScholarCrossref
6.
Wilson  BJ, Cowan  HJ, Lord  JA, Zuege  DJ, Zygun  DA.  The accuracy of pulse oximetry in emergency department patients with severe sepsis and septic shock: a retrospective cohort study.   BMC Emerg Med. 2010;10:9. doi:10.1186/1471-227X-10-9PubMedGoogle ScholarCrossref
7.
Singh  AK, Sahi  MS, Mahawar  B, Rajpurohit  S.  Comparative evaluation of accuracy of pulse oximeters and factors affecting their performance in a tertiary intensive care unit.   J Clin Diagn Res. 2017;11(6):OC05-OC08. doi:10.1183/1393003.congress-2017.PA2127PubMedGoogle Scholar
8.
Vandenbroucke  JP, Von Elm  E, Altman  DG,  et al.  Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): explanation and elaboration; translation to Russian.   Digital Diagnostics. 2021;2(2):1-10. doi:10.17816/dd70821Google ScholarCrossref
9.
Pollard  T, Johnson  A, Raffa  J, Celi  LA, Badawi  O, Mark  R. eICU Collaborative Research Database. PhysioNet. April 15, 2019. Accessed February 15, 2020. https://physionet.org/content/eicu-crd/2.0/
10.
Pollard  TJ, Johnson  AEW, Raffa  JD, Celi  LA, Mark  RG, Badawi  O.  The eICU Collaborative Research Database, a freely available multi-center database for critical care research.   Sci Data. 2018;5:180178. doi:10.1038/sdata.2018.178PubMedGoogle Scholar
11.
Goldberger  AL, Amaral  LAN, Glass  L,  et al.  PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals.   Circulation. 2000;101(23):E215-E220. doi:10.1161/01.cir.101.23.e215PubMedGoogle ScholarCrossref
12.
Johnson  AEW, Pollard  TJ, Shen  L,  et al.  MIMIC-III, a freely accessible critical care database.   Sci Data. 2016;3:160035. doi:10.1038/sdata.2016.35PubMedGoogle Scholar
13.
Vincent  J-L, Moreno  R, Takala  J,  et al; the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine.  The SOFA (Sepsis-Related Organ Failure Assessment) score to describe organ dysfunction/failure.   Intensive Care Med. 1996;22(7):707-710. doi:10.1007/s001340050156PubMedGoogle ScholarCrossref
14.
Coustet  B, Lhuissier  FJ, Vincent  R, Richalet  JP.  Electrocardiographic changes during exercise in acute hypoxia and susceptibility to severe high-altitude illnesses.   Circulation. 2015;131(9):786-794. doi:10.1161/CIRCULATIONAHA.114.013144PubMedGoogle ScholarCrossref
15.
Bickler  PE, Feiner  JR, Lipnick  MS, Batchelder  P, MacLeod  DB, Severinghaus  JW.  Effects of acute, profound hypoxia on healthy humans: implications for safety of tests evaluating pulse oximetry or tissue oximetry performance.   Anesth Analg. 2017;124(1):146-153. doi:10.1213/ANE.0000000000001421PubMedGoogle ScholarCrossref
16.
Dewan  NA, Nieto  FJ, Somers  VK.  Intermittent hypoxemia and OSA: implications for comorbidities.   Chest. 2015;147(1):266-274. doi:10.1378/chest.14-0500PubMedGoogle ScholarCrossref
17.
MacIntyre  NR.  Tissue hypoxia: implications for the respiratory clinician.   Respir Care. 2014;59(10):1590-1596. doi:10.4187/respcare.03357PubMedGoogle ScholarCrossref
18.
Colgan  SP, Campbell  EL, Kominsky  DJ.  Hypoxia and mucosal inflammation.   Annu Rev Pathol. 2016;11:77-100. doi:10.1146/annurev-pathol-012615-044231PubMedGoogle ScholarCrossref
19.
Jansen  TC, van Bommel  J, Schoonderbeek  FJ,  et al; LACTATE study group.  Early lactate-guided therapy in intensive care unit patients: a multicenter, open-label, randomized controlled trial.   Am J Respir Crit Care Med. 2010;182(6):752-761. doi:10.1164/rccm.200912-1918OCPubMedGoogle ScholarCrossref
20.
Kruse  O, Grunnet  N, Barfod  C.  Blood lactate as a predictor for in-hospital mortality in patients admitted acutely to hospital: a systematic review.   Scand J Trauma Resusc Emerg Med. 2011;19:74. doi:10.1186/1757-7241-19-74PubMedGoogle ScholarCrossref
21.
Scholz  H, Boivin  FJ, Schmidt-Ott  KM,  et al.  Kidney physiology and susceptibility to acute kidney injury: implications for renoprotection.   Nat Rev Nephrol. 2021;17(5):335-349. doi:10.1038/s41581-021-00394-7PubMedGoogle ScholarCrossref
22.
US Food and Drug Administration. Pulse oximeters—premarket notification submissions [510(k)s]: guidance for industry and Food and Drug Administration staff. March 2013. Accessed April 2, 2021. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/pulse-oximeters-premarket-notification-submissions-510ks-guidance-industry-and-food-and-drug
23.
US Census Bureau. 2017 National population projections tables: main series. Accessed April 2, 2021. https://www.census.gov/data/tables/2017/demo/popproj/2017-summary-tables.html
24.
US Census Bureau. QuickFacts: United States. Accessed April 4, 2021. https://www.census.gov/quickfacts/fact/table/US/PST045219
25.
Ghosh  I. Visualizing the US population by race. December 28, 2020. Accessed April 2, 2021. https://www.visualcapitalist.com/visualizing-u-s-population-by-race/
26.
Blum  FE, Lund  ET, Hall  HA, Tachauer  AD, Chedrawy  EG, Zilberstein  J.  Reevaluation of the utilization of arterial blood gas analysis in the intensive care unit: effects on patient safety and patient outcome.   J Crit Care. 2015;30(2):438.e1-438.e5. doi:10.1016/j.jcrc.2014.10.025PubMedGoogle ScholarCrossref
27.
Feiner  JR, Severinghaus  JW, Bickler  PE.  Dark skin decreases the accuracy of pulse oximeters at low oxygen saturation: the effects of oximeter probe type and gender.   Anesth Analg. 2007;105(6)(suppl):S18-S23. doi:10.1213/01.ane.0000285988.35174.d9PubMedGoogle ScholarCrossref
28.
Bickler  PE, Feiner  JR, Severinghaus  JW.  Effects of skin pigmentation on pulse oximeter accuracy at low saturation.   Anesthesiology. 2005;102(4):715-719. doi:10.1097/00000542-200504000-00004PubMedGoogle ScholarCrossref
Limit 200 characters
Limit 25 characters
Conflicts of Interest Disclosure

Identify all potential conflicts of interest that might be relevant to your comment.

Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.

Err on the side of full disclosure.

If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.

Not all submitted comments are published. Please see our commenting policy for details.

Limit 140 characters
Limit 3600 characters or approximately 600 words
    2 Comments for this article
    EXPAND ALL
    Discrepancies Between Results, and MetHb & COHb findings
    Albert Donnay, MS, MHS | Donnay Detoxicology LLC
    I appreciate the authors publishing so much data on this critical issue. In trying to make sense of them, I noticed the numbers of Asian, Black, Hispanic and White patients with SpO2 levels from 88% to 100% are given twice, in supplemental Table e8b and e9.

    But only 6 of the 52 results match, and only at SpO2=100% do they match for all four patient groups. All the other results are higher in Table e8b (total 87 971 vs 85 458), ranging from 1.3% more Black patients to 6.1% more Hispanic patients.
    /> While the summary results given in the abstract, figure 1, and table 2 match those in Table e8b, other sections of the text do not match any of the tables or figures. The results on Hidden Hypoxemia by Race and Ethnicity, for example, include 15 more Hispanic patients than in Tables 2 and e8b, who are offset by 2 fewer Asians, 4 fewer Black patients, and 9 fewer White patients.

    Can the authors explain these discrepancies, and if necessary, correct them?

    I am also curious about the carboxyhemoglobin (COHb) and methemoglobin (MetHb) findings. The authors found  "hidden hypoxemia associated increased hospital mortality... [in range of 20.6% for Black patients to 26.1% for White patients] and that "these [results] continue to hold true when restricted to ABGs with carboxyhemoglobin and methemoglobin levels less than 2%," citing table e12.

    It looks like the subgroups with hidden hypoxemia (HH) and COHb and MetHb over 2% were much larger and had higher mortality, ranging from 21.9% among Black patients to 29.6% among Hispanic patients. Mortality among the smaller HH subgroups with COHb and/or MetHb below 2% ranged from 18.2% among Asian patients to 21.5% among White patients.

    Most surprising is that 69.8% of all White patients had COHb and/or MetHb over 2% on their first ABG, including 76.2% of all the White patients who later died and 87.9% of those who died with HH. These rates of CO and Met were lowest among Black patients but still very high: 35.9% had their first COHb and/or MetHb over 2%, including 37.6% of all Black patients who later died, and 57.7% of those who died with HH. While some of these high COHb may be due to smoking, only about 15% of Americans smoke.

    In contrast, the authors reported rates of hidden hypoxemia ranging from just 4.9% of White patients to 6.9% of Black patients. This suggests that abnormally high levels of COHb and/or MetHb are a much more common source of error among pulse oximeters used in hospitals, and that White hospital patients, not Asian, Black, or Hispanic patients, are at highest risk.

    Can the authors tell us the mean and range of the first COHb and MetHb measured in each group among those who were later discharged compared to those who died? It would be very helpful to know if in-patient mortality is more closely correlated with COHb or MetHb.
    CONFLICT OF INTEREST: The author holds a US patent for a method of measuring the levels of free CO diffusing from lungs, arteries, veins and the average of all tissues, but it is not licensed and he gets no income from it.
    READ MORE
    Revised discrepancies, MetHb + COHb
    An-Kwok Wong, MD, PhD | Duke University
    Thank you for thoroughly reading our work.

    Mismatching numbers for eTables 8b, 9

    Thank you for bringing this to our attention - this was an erroneous old version. The counts per level in eTable 8b are correct. What eTable 8b provides is an exact characterization of hidden hypoxaemia as seen in the data.

    Based upon the data used for eTable 8b, we wanted to further investigate our estimate of hidden hypoxemia per SpO2 level by bootstrapping, captured in eTable 9. As you can predict, the confidence interval for SpO2 = 88%, with much smaller n,
    is much wider than that for, say, SpO2 = 100%, with much larger n. The counts data was added afterwards and linked to the wrong data. We will have eTable 9’s n’s updated to match.



    Abstract Results & Hidden hypoxemia by Race and Ethnicity

    Thank you for also bringing this to our attention. Table 2 is correct; we will update the numbers in “Hidden hypoxemia by Race and Ethnicity” in the manuscript; this was also based off an erroneous old version.



    Carboxyhemoglobin / methemoglobin

    Thank you for asking this question. However, the presumption that all patients not listed in eTable 12 had MetHB > 2% or COHb > 2% is not correct. Many patients did not have co-oximetry, so their MetHB and COHb values are simply missing.

    What can be said is that of all patients with co-oximetry, 648 patients had MetHB and/or COHb > 3%, and of these, 114 had MetHB and/or COHb > 5% - much smaller than the 36,843 in eTable 12. This will have to be analyzed in a future study.
    CONFLICT OF INTEREST: Dr Wong reported holding equity and management roles in Ataia Medical outside the submitted work.
    READ MORE
    Original Investigation
    Pulmonary Medicine
    November 3, 2021

    Analysis of Discrepancies Between Pulse Oximetry and Arterial Oxygen Saturation Measurements by Race and Ethnicity and Association With Organ Dysfunction and Mortality

    Author Affiliations
    • 1Division of Pulmonary, Allergy, Critical Care, and Sleep Medicine, Emory University, Atlanta, Georgia
    • 2Division of Pulmonary, Allergy, and Critical Care Medicine, Duke University, Durham, North Carolina
    • 3MIT Institute for Data, Systems and Society, Cambridge, Massachusetts
    • 4Department of Biomedical Engineering, Johns Hopkins University, Baltimore, Maryland
    • 5Department of Surgery, Emory University, Atlanta, Georgia
    • 6Leiden University Medical Centre, Department of Biomedical Data Sciences, Leiden, the Netherlands
    • 7Leiden University Medical Centre, Department of Information Technology and Digital Innovation, Leiden, the Netherlands
    • 8Department of Neurology, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts
    • 9Department of Biomedical Informatics, Emory University, Atlanta, Georgia
    • 10School of Biological Science and Medical Engineering, Beihang University, Beijing, China
    • 11Respiratory Institute, Cleveland Clinic, Cleveland, Ohio
    • 12Sociedade Mineira de Terapia Intensiva, Belo Horizonte, Brazil
    • 13Department of Infectious Diseases, Boston Medical Center, Boston, Massachusetts
    • 14Connected Care and Personal Health, Philips Research North America, Cambridge, Massachusetts
    • 15Department of Surgery, Emory University, Atlanta, Georgia
    • 16Medicine–Primary Care and Population Health, Stanford University, California
    • 17Office of Research, Stanford Health Care, Stanford, California
    • 18Massachusetts Institute of Technology, Laboratory for Computational Physiology, Cambridge
    • 19Division of Pulmonary, Critical Care, and Sleep Medicine, Beth Israel Deaconess Medical Center, Boston, Massachusetts
    • 20Department of Biostatistics, Harvard T.H. Chan School of Public Health, Boston, Massachusetts
    JAMA Netw Open. 2021;4(11):e2131674. doi:10.1001/jamanetworkopen.2021.31674
    Key Points

    Question  Do pulse oximetry discrepancies, hidden hypoxemia, and clinical outcomes differ among racial and ethnic subgroups?

    Findings  In this cross-sectional study of 5 databases with 87 971 patients, significant disparities in pulse oximetry accuracy across racial and ethnic subgroups (ie, Asian, Black, Hispanic, and White individuals) were found, with higher rates of hidden hypoxemia associated with mortality, future organ dysfunction, and abnormal laboratory test results.

    Meaning  In this study, discrepancies in pulse oximetry accuracy among racial and ethnic subgroups were associated with higher rates of hidden hypoxemia, mortality, and organ dysfunction.

    Abstract

    Importance  Discrepancies in oxygen saturation measured by pulse oximetry (Spo2), when compared with arterial oxygen saturation (Sao2) measured by arterial blood gas (ABG), may differentially affect patients according to race and ethnicity. However, the association of these disparities with health outcomes is unknown.

    Objective  To examine racial and ethnic discrepancies between Sao2 and Spo2 measures and their associations with clinical outcomes.

    Design, Setting, and Participants  This multicenter, retrospective, cross-sectional study included 3 publicly available electronic health record (EHR) databases (ie, the Electronic Intensive Care Unit–Clinical Research Database and Medical Information Mart for Intensive Care III and IV) as well as Emory Healthcare (2014-2021) and Grady Memorial (2014-2020) databases, spanning 215 hospitals and 382 ICUs. From 141 600 hospital encounters with recorded ABG measurements, 87 971 participants with first ABG measurements and an Spo2 of at least 88% within 5 minutes before the ABG test were included.

    Exposures  Patients with hidden hypoxemia (ie, Spo2 ≥88% but Sao2 <88%).

    Main Outcomes and Measures  Outcomes, stratified by race and ethnicity, were Sao2 for each Spo2, hidden hypoxemia prevalence, initial demographic characteristics (age, sex), clinical outcomes (in-hospital mortality, length of stay), organ dysfunction by scores (Sequential Organ Failure Assessment [SOFA]), and laboratory values (lactate and creatinine levels) before and 24 hours after the ABG measurement.

    Results  The first Spo2-Sao2 pairs from 87 971 patient encounters (27 713 [42.9%] women; mean [SE] age, 62.2 [17.0] years; 1919 [2.3%] Asian patients; 26 032 [29.6%] Black patients; 2397 [2.7%] Hispanic patients, and 57 632 [65.5%] White patients) were analyzed, with 4859 (5.5%) having hidden hypoxemia. Hidden hypoxemia was observed in all subgroups with varying incidence (Black: 1789 [6.9%]; Hispanic: 145 [6.0%]; Asian: 94 [4.9%]; White: 2831 [4.9%]) and was associated with greater organ dysfunction 24 hours after the ABG measurement, as evidenced by higher mean (SE) SOFA scores (7.2 [0.1] vs 6.29 [0.02]) and higher in-hospital mortality (eg, among Black patients: 369 [21.1%] vs 3557 [15.0%]; P < .001). Furthermore, patients with hidden hypoxemia had higher mean (SE) lactate levels before (3.15 [0.09] mg/dL vs 2.66 [0.02] mg/dL) and 24 hours after (2.83 [0.14] mg/dL vs 2.27 [0.02] mg/dL) the ABG test, with less lactate clearance (−0.54 [0.12] mg/dL vs −0.79 [0.03] mg/dL).

    Conclusions and Relevance  In this study, there was greater variability in oxygen saturation levels for a given Spo2 level in patients who self-identified as Black, followed by Hispanic, Asian, and White. Patients with and without hidden hypoxemia were demographically and clinically similar at baseline ABG measurement by SOFA scores, but those with hidden hypoxemia subsequently experienced higher organ dysfunction scores and higher in-hospital mortality.

    Introduction

    Recently, reports of systemic racial bias in which oxygen saturation measured by pulse oximeter (Spo2) overestimates the true arterial oxygen saturation (Sao2) in patients with darkly pigmented skin has raised concerns about the clinical accuracy of pulse oximetry.1 Sjoding et al1 used race listed in the EHR as a proxy for skin color to retrospectively analyze the accuracy of pulse oximetry during hypoxemia in 2 high-acuity adult cohorts. They described occult hypoxemia as an Sao2 of less than 88% when the Spo2 was between 92% and 96%. In both cohorts, the incidence of hidden hypoxemia was almost 3 times higher among patients self-reported as Black vs White.1

    Pulse oximetry is a useful tool to monitor blood oxygen saturation without obtaining an invasive arterial blood gas (ABG) measurement. The US Food and Drug Administration (FDA) requires root mean square accuracy within 2% for values between 70% and 100%, implying that an adequate pulse oximeter returns an Spo2 value within 2% to 3% of the Sao2 value (ie, a range of 4%-6%) only two-thirds of the time.2

    Studies have highlighted the inaccuracy of pulse oximetry in critically ill patients; however, smaller sample sizes hindered in-depth analysis of race, ethnicity, and outcomes (eTable 1 in the Supplement).3-7 This study used 5 large EHR data sets of critically ill patients to further evaluate the incidence and clinical outcomes of hidden hypoxemia across racial and ethnic groups.

    Methods

    This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.8 Data in Medical Information Mart for Intensive Care III (MIMIC-III), MIMIC-IV, and Electronic Intensive Care Unit–Clinical Research Database (eICU-CRD) had been previously deidentified and did not require a waiver for informed consent. The Medical Information Mart for Intensive Care (MIMIC) database is a collaboration between the Beth Israel Deaconess Medical Center and the Laboratory for Computational Physiology at the Massachusetts Institute of Technology. The database contains granular, deidentified ICU data from the Beth Israel Deaconess Medical Center. The data have been generated from more than 70 intensive care unit beds with medical, surgical, cardiac, and neurological patients. We used the latest data version, MIMIC-III (version 1.4), which contains deidentified data associated with 53 423 ICU admissions.9 Physionet approved the use of MIMIC for this study. Emory University approved the use of the Emory and Grady databases for research, with a complete HIPAA and informed consent waiver.

    Data Sources
    Open Source PhysioNet Databases

    The eICU-CRD (comprising 335 ICUs across 208 hospitals) and MIMIC databases (comprising 6 ICUs in 1 hospital) were used. The Sequential Organ Failure Assessment (SOFA) scores (eTable 2 in the Supplement), including the cardiovascular SOFA (CVSOFA) and respiratory SOFA (RSOFA) scores, were provided by MIT Laboratory of Computational Physiology.9-13

    Emory Healthcare and Grady Memorial

    Data were collected from all units in Emory Healthcare (277 units, including 26 ICUs, across 4 hospitals) and Grady Memorial Hospital (73 units, including 9 ICUs, in 1 hospital). Emory Healthcare and Grady patient data spanned 2014 to 2021 and 2014 to 2020, respectively. SOFA scores and its components were not available for the Emory and Grady databases.

    Data Extraction

    Data analysis was conducted with R version 3.6.3 (R Project for Statistical Computing) and Python version 3.6 (Python Software Foundation). All ABG and Spo2 values were extracted from the EHR.

    Inclusion and Exclusion Criteria

    An Spo2 range of 88% to 100% was selected as an interval in which patients may have hypoxemia but falsely be considered as having arterial blood oxygenation in the reference range according to Sao2. Each ABG-measured Sao2 was matched with the closest Spo2 value recorded within the previous 5 minutes. To eliminate repeated measurements and limit confounding, only the first ABG measurement from each hospital encounter was used. Spo2 measurements of less than 88% were not examined because of low prevalence in the EHR.

    Race and Ethnicity

    In each EHR data set, race and ethnicity were defined using self-reported demographic data, including administrative entries with additional identifiers. All patients with race and ethnicity information who could not be classified as Asian, Black, Hispanic, or White were excluded. Patients were stratified by age, sex, race and ethnicity, and CVSOFA score. If any of these characteristics were missing, a patient was excluded from the corresponding subgroup analysis but included in the overall analysis.

    Statistical Analysis

    For each racial and ethnic group, the frequency of ABG measurement was characterized using 2 complementary analyses. First, encounters with ABG measurements were compared with encounters without ABGs measurements to determine the likelihood of receiving an ABG measurement during an encounter by race and ethnicity. Second, to characterize the rate of ABG collection across encounters with at least 1 ABG, the total number of ABGs normalized by the length of stay (in days) was calculated. Given possible confounding by illness severity, the second estimates were stratified by CVSOFA score at the time of ABG.

    Differences in Spo2-Sao2 pairs were characterized by modified Bland-Altman plots. We used χ2 tests to compare the distribution of categorical variables (eg, sex) between any 2 groups, while Mann-Whitney nonparametric tests were used for continuous and ordinal variables (eg, age, SOFA score). Notably, differences in the distribution of numeric clinical end points (eg, SOFA score) were evaluated via bootstrapping (100 iterations), followed by a Mann-Whitney nonparametric test. Differences between stratified odds differences (eg, risk of hidden hypoxemia by race and ethnicity) were tested with the Breslow-Day test.

    Hidden hypoxemia was defined as an Spo2 of 88% or greater despite an Sao2 of less than 88%. Patients with and without hidden hypoxemia were compared at the time of ABG measurement by baseline demographic characteristics (age, sex) and by organ failure scores (SOFA, RSOFA, CVSOFA). The long-term association of hidden hypoxemia with clinical outcomes was analyzed by estimating differences in length of stay and in-hospital mortality. The short-term association of hidden hypoxemia with organ dysfunction was examined using SOFA and CVSOFA scores measured 24 hours after the baseline ABG measurement. Associations between hidden hypoxemia, RSOFA, and in-hospital mortality were also examined. The consequences of hidden hypoxemia were also evaluated by comparing the last serum lactate and serum creatinine levels in a 7-day window before the ABG measurement with the first value in a 7-day window starting 24 hours after the ABG measurement. Values were compared at each time in addition to the difference between the before and after values.

    Multivariate logistic regression was used for assessing binary end points (eg, in-hospital mortality), multivariate ordinal regression for numerical end points (eg, CVSOFA and RSOFA scores), and multivariate linear models for continuous end points (eg, creatinine lactate levels), using analysis of variance to test for the impact of hidden hypoxemia while adjusting for other covariates (eg, age, sex, SOFA score at time of ABG measurement) (eAppendix 1 in the Supplement). Missing data for any covariates were flagged with a separate binary variable.

    Statistical analysis was conducted with R version 3.6.3 (R Project for Statistical Computing) and Python version 3.6 (Python Software Foundation). Statistical significance was set at P < .05, and all tests were 2-tailed.

    Results

    The first Spo2-Sao2 pairs from 87 971 patient encounters (27 713 [42.9%] women; mean [SE] age, 62.2 [17.0] years) were analyzed among 4 race/ethnicity subgroups (Asian, 1919 patients [2.3%]; Black, 26 032 [29.6%]; Hispanic, 2397 patients [2.7%]; White, 57 623 patients [65.5%]), with 4859 (5.5%) having hidden hypoxemia. In total, 141 600 patients, with 679 909 ABGs and 5 435 144 Spo2-Sao2 pairs within 30 minutes of each other were identified. Patient characteristics are presented in Table 1 and Figure 1. Restricting to Spo2 measurements to those within the 5 minutes preceding the ABG measurement resulted in 268 904 Spo2-Sao2 pairs; further selecting the first ABG in an encounter led to 87 971 Spo2-Sao2 pairs (eAppendix 2 in the Supplement). Sample sizes were considerably smaller for Asian and Hispanic subgroups.

    Missingness

    Missing data on length of stay and in-hospital mortality was 0.1% (96 encounters) and 0.4% (369 encounters), respectively. Given that there were no SOFA score data from Emory and Grady, RSOFA, CVSOFA, and SOFA scores had 52.7% (46 381 encounters), 52.9% (46 538 encounters), and 52.7% (46 381 encounters) missingness, respectively. Missingness for Spo2 variability and other variables are described in eTable 4 and eTable 5 in the Supplement.

    ABG Measurement Frequency by Race and Ethnicity

    There were differences in the likelihood of receiving an ABG measurement during a hospital encounter that varied by race and ethnicity, with the White subgroup most likely to receive an initial ABG measurement despite similar RSOFA and CVSOFA scores: White patients, 85 872 of 1 532 492 encounters (5.6%); Asian patients, 3249 of 95 813 (3.4%); Black patients, 49 053 of 1 781 868 (2.8%); and Hispanic patients, 3426 of 179 617 (1.9%) (eTable 6 in the Supplement). Although following the first ABG, subsequent ABG frequencies were similar across racial and ethnic subgroups (eTable 7 in the Supplement). A table comparing patients with and without ABG measurements across races and ethnicities appears in eTable 3 in the Supplement.

    Comparison of Spo2 With Sao2 by Race and Ethnicity

    The White subgroup was used as a reference because it had the highest prevalence in the data set. At most Spo2 values, there were statistically significant differences in true oxygen saturation levels between White patients compared with those in other racial and ethnic subgroups, although these differences were small in magnitude (evaluated using means and medians) (eTable 5 in the Supplement). Similar results were obtained when using all ABGs measured during a patient’s hospitalization (268 904) instead of the first ABG measurement only (87 971) (eFigure 1 in the Supplement). This finding was robust to adjustment for sex, age, and RSOFA and CVSOFA scores (eFigures 2-5 in the Supplement).

    Hidden Hypoxemia by Race and Ethnicity

    Hidden hypoxemia occurred across all racial and ethnic subgroups, assessed using the first ABG measurement with an Spo2 level greater than 88%. Patients self-identified as Black had higher Sao2 variability for any given Spo2 value, as evidenced by a larger IQR (eg, median [IQR] Sao2 at Spo2 of 88%, Black patients: 90.10% [10.13]; White patients, 90.00% [9.10]). There was a varying incidence of hidden hypoxemia among racial and ethnic group in descending order: Black: 1789 (6.9%); Hispanic: 145 (6.0%); Asian: 94 (4.9%); White: 2831 (4.9%) (P < .001) (Figure 2; eTable 8 and eTable 9 in the Supplement).

    Organ Dysfunction at Time of First ABG Measurement and 24 Hours Later

    Across all racial and ethnic groups statistically significant, although clinically small, differences in baseline organ dysfunction (SOFA, CVSOFA, RSOFA) were present at the time of the first ABG measurements between patients with and without hidden hypoxemia (Table 2). Furthermore, across all racial and ethnic groups, patients with hidden hypoxemia subsequently experienced greater organ dysfunction than patients without hidden hypoxemia, as evidenced by higher CVSOFA and SOFA scores measured 24 hours after the ABG measurement was drawn (mean [SE] CVSOFA: 1.48 [0.03] vs 1.25 [0.01]; mean [SE] SOFA: 7.2 [0.1] vs 6.29 [0.02]) (Table 2). This difference persisted even when adjusted for age, sex, and SOFA score (data not shown).

    Clinical Outcomes

    Across all racial and ethnic groups, patients with hidden hypoxemia had higher in-hospital mortality than patients without hidden hypoxemia (Table 2). The difference was significant for Black and White patients (eg, Black patients: 369 [21.1%] vs 3557 [15.0%]; P < .001). This association persisted even when adjusted for age, sex, SOFA score. However, there was no differences in length of stay for patients with and without hidden hypoxemia when considering survivors only (Table 2).

    Across all racial and ethnic subgroups, patients with hidden hypoxemia had a significantly higher mean (SE) serum creatinine level (1.96 [0.04] mg/dL vs 1.69 [0.01] mg/dL [to convert to micromoles per liter, multiply by 88.4]; P < .001) and serum lactate level (3.15 [0.09] mg/dL vs 2.66 [0.02] mg/dL [to convert to millimoles per liter, multiply by 0.111]; P < .001) before the ABG measurement than patients without hypoxemia. Patients with hidden hypoxemia continued to maintain significantly higher mean (SE) serum creatinine (1.86 [0.05] mg/dL vs 1.63 [0.01] mg/dL; P < .001) and lactate (2.83 [0.14] mg/dL vs 2.83 [0.14] mg/dL; P < .001) values after the ABG measurements. When comparing values before and after the ABG, serum lactate demonstrated a smaller mean (SE) decrease among patients with hidden hypoxemia overall (−0.54 [0.12] vs −0.79 [0.03]; P < .001) and in all racial and ethnic groups except Asian patients. However, the difference in serum creatinine levels before and after the ABG measurement was not consistent across race and ethnicity and hidden hypoxemia status.

    Hidden Hypoxemia and RSOFA Score

    The likelihood of hidden hypoxemia increased with higher RSOFA values and was highest among patients with the highest RSOFA scores at baseline (ie, at the time of ABG) (eTable 10 in the Supplement). The presence of hidden hypoxemia, irrespective of RSOFA score, was associated with higher risk of mortality (eTable 11 in the Supplement).

    Risk of Hidden Hypoxemia by Race and Ethnicity

    Hidden hypoxemia occurred across racial and ethnic groups, but the risk differed among subgroups when calculated with a risk-threshold of 5%. White patients had a 5% risk of hidden hypoxemia at an Spo2 of 94%, whereas this 5% risk occurred at a higher Spo2 for Black (97%), Hispanic (97%), and Asian (95%) patients. The risk of hidden hypoxemia at an SpO2 of 93% to 96% is 6.5% in White patients, increasing to 6.6% for Hispanic patients and up to 10.9% for Black and Asian patients (a 68% higher relative risk). In conjunction with eTable 8 in the Supplement, which shows more granular data, a clinician could select their threshold for an acceptable risk of hidden hypoxemia by Spo2 target by selecting the highest Spo2 with a fixed risk of hidden hypoxemia. To ensure a risk of hidden hypoxemia of less than 10%, Spo2 should be greater than 93% among Asian patients (mean [SE] risk, 11.0% [5.0]), 96% among Black patients (mean [SE] risk, 8.2% [1.0]); 92% among Hispanic patients (mean [SE] risk, 18.4% [6.0]), and 93% among White patients (mean [SE] risk, 10.6% [1.0]) (Table 3).

    Discussion

    To our knowledge, this study is the first to characterize the prevalence of hidden hypoxemia (Sao2 <88%, but Spo2 ≥88%) by race and ethnicity in hospitalized patients’ first ABG measurements in their hospitalization across 5 large US databases. Racial and ethnic disparities in the incidence of hidden hypoxemia in the hospital are worrying because low oxygen saturation levels, when undetected, can lead to complications in the short and long term.

    This analysis demonstrates that all racial and ethnic groups experienced discrepancies in pulse oximetry resulting in hidden hypoxemia; however, it was more prevalent in Asian, Black, and Hispanic patients than White patients. There is a risk of hidden hypoxemia at all Spo2 values, and this risk increases as Spo2 approaches 88%. When hidden hypoxemia occurs, despite similar organ dysfunction scores at the time of the ABG measurement, it was associated with increased in-hospital mortality, increased short-term future organ dysfunction (ie, SOFA, RSOFA, and CVSOFA scores), and increased laboratory findings (ie, lactate and creatinine levels), even when adjusting for covariates, such as age, sex, and SOFA score. These continue to hold true when restricted to ABGs with carboxyhemoglobin and methemoglobin levels less than 2% (eTable 12 in the Supplement). Furthermore, the presence of hidden hypoxemia was independently associated with increased mortality for all RSOFA values, suggesting that hidden hypoxemia and RSOFA are complementary. Finally, to maintain a risk of hidden hypoxemia less than 10%, each race and ethnicity would have a different Spo2 threshold: Asian, 93%; Black, 95%; Hispanic, 92%, White, 93%) (Table 3; eTable 9 in the Supplement).

    As this analysis was restricted to the first ABG measurement in a hospitalization, the clinician could not have been aware of hidden hypoxemia prior to the ABG test. It is therefore unknowable how long a patient was truly hypoxemic before their ABG measurement, although the clinician would be aware of hypoxemia once they had the ABG results. Despite similar organ dysfunction scores at the time of the ABG measurement, patients with hidden hypoxemia had greater laboratory abnormalities (ie, for lactate and creatinine levels) before the ABG measurement that persisted for at least 24 hours, suggesting that these patients may have more severe illness. This study was not designed to assess causality; it is both plausible that the patient’s illness could be causative of hidden hypoxemia (and thus be a marker of dysfunction) and that hidden hypoxemia for an unknown (perhaps prolonged) duration resulted in worse organ dysfunction.

    The effects of hypoxia can be organized by the duration a patient experiences hypoxia. Brief, acute episodes of hypoxia have been associated with electrocardiogram changes14 and, in healthy patients, brief cognitive impairment without sustained long-term cognitive changes.15 However, hypoxia has been associated with increased oxidative stress, reactive-oxygen species, angiogenesis, hypoxia-inducible factors, and systemic and vascular inflammation with endothelial dysfunction.16-18 Critically ill patients may have impaired tolerance of these changes,17 and hypoxia can result in kidney injury and lactic acidosis.19-21

    Hidden hypoxemia was prevalent across all racial and ethnic subgroups, but it disproportionately affected certain groups because pulse oximeters are not tested or calibrated on an adequate number of individuals with varying skin pigmentation. Since 2013, the FDA has required that the test sample for pulse oximeters include at least 15% people with diverse skin pigmentation, including 2 individuals with darkly pigmented skin.22 However, this sampling does not reflect the United States 2010 census of Asian (5.9% of population), Black (13.4%), Hispanic (18.5%), and White (60.1%) individuals.23,24 Going forward, population differences will only increase in relevance as the United States becomes more racially and ethnically diverse.25 Furthermore, although older studies questioned the accuracy of pulse oximetry in critical illness, sample sizes were too small to examine the issue of skin color and race and ethnicity in meaningful detail (eFigure 6 in the Supplement).3-7

    The results of this study highlight several societal and medical issues. First, pulse oximeters are inadequately tested or calibrated in hospitalized patients, despite often being the intended population. Second, a 2% accuracy range (4% total) is too wide at low blood oxygenation levels, and third, pulse oximeters are insufficiently tested across different racial and ethnic groups prior to approval by the FDA. As the results of this study demonstrate, this combination has unintended negative health outcomes. By providing a data-driven approach to identify hidden hypoxemia using pulse oximetry, this study is a step toward greater health care equity.

    While a short-term solution to hidden hypoxemia may be to more frequently sample ABG values, such a strategy is invasive and inefficient.26 If anything, greater ABG sampling is merely a stopgap to cover the use of imperfect medical technology. The important message is that health care devices, like predictive algorithms and medications, must be designed more inclusively to achieve comparable measurement accuracy irrespective of race and ethnicity. As noted in previous studies,1,27,28 pulse oximetry devices are not reliably accurate and do not capture blood oxygenation readings equally across different skin colors. In the meantime, prudent clinicians should note the Spo2 reading at the time the ABG is drawn to accurately identify any Spo2-to-Sao2 discrepancy once the ABG result is reported.

    It is important to be cognizant of the patient population in which pulse oximeters used in critical care are validated. There is a need for more transparency in the labeling of all patient care devices, including the detailed characteristics of groups on which they were evaluated. To further achieve more equitable health outcomes, we call for reinforced testing and recalibration of health care devices—across all target patient populations.

    Limitations

    This retrospective EHR analysis has inherent limitations that were systematically addressed. First, Sao2-Spo2 pairs combine measurements that are not always collected simultaneously. Analysis was thus restricted to Spo2 values recorded in the 5 minutes preceding the ABG test. Shock and critical illness, in conjunction with other comorbid conditions (eg, peripheral arterial disease, diabetes), may further affect pulse oximetry accuracy and need further characterization for proper adjustment for confounding beyond CVSOFA score.

    There was high missingness, especially in SOFA scores and laboratory values, that was accounted for with missing flags during regression analysis; SOFA scores were only calculated for eICU-CRD, MIMIC-III, and MIMIC-IV data and were not calculated for Emory or Grady data. Multiple imputation methods could improve robustness. Additionally, the EHR data did not record Spo2 signal quality or pulse oximeter brand, leading to unclear knowledge of Spo2 accuracy and homogeneity.

    There may be a selection bias in acquiring ABG measurements. Given similar SOFA scores at the time of testing, White patients were significantly more likely to receive the criterion-standard test. It is plausible that there was selection bias with underdetection of hidden hypoxemia among Asian, Black, and Hispanic patients. The disparities in clinical outcomes may be underestimated if more ABG tests were performed in these racial and ethnic subgroups. Additionally, these retrospective analyses reflect associations; future studies should be designed to assess causality.

    Conclusions

    In this study, all racial and ethnic subgroups experienced high variability in arterial oxygen saturation for fixed pulse oximetry levels, with a greater discrepancy in patients self-reporting as Asian, Black, and Hispanic than White. Small but statistically significant differences in the bias of Spo2 measurements (vs true Sao2 measurements) were associated with increased incidence of hidden hypoxemia (Sao2 <88% despite Spo2 ≥88%). Although demographically and clinically similar to patients without hypoxemia at baseline ABG measurement, those with hidden hypoxemia had higher rates of organ dysfunction 24 hours later and higher in-hospital mortality. Validation of all health technologies, including pulse oximetry, must be performed across a wider range of patient populations to avoid perpetuating harm from miscalibration.22

    Back to top
    Article Information

    Accepted for Publication: August 25, 2021.

    Published: November 3, 2021. doi:10.1001/jamanetworkopen.2021.31674

    Correction: This article was corrected on February 9, 2022, to fix errors in eTable 9 and the text.

    Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2021 Wong AKI et al. JAMA Network Open.

    Corresponding Author: An-Kwok Ian Wong, MD, PhD, Division of Pulmonary, Allergy, and Critical Care Medicine, Duke University, Three Genome Court, PO Box 103050, Durham, NC 27710 (med@aiwong.com).

    Author Contributions: Dr Wong had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Drs Lough and Celi are co–senior authors.

    Concept and design: Wong, Charpignon, Kim, Carvalho, Holder, Buchman, Lough, Celi.

    Acquisition, analysis, or interpretation of data: Wong, Charpignon, Kim, Josef, de Hond, Fojas, Tabaie, Liu, Mireles-Cabodevila, Kamaleswaran, Madushani, Adhikari, Steyerberg, Lough, Celi.

    Drafting of the manuscript: Wong, Charpignon, Kim, Fojas, Mireles-Cabodevila, Lough, Celi.

    Critical revision of the manuscript for important intellectual content: Wong, Charpignon, Kim, Josef, de Hond, Tabaie, Liu, Mireles-Cabodevila, Carvalho, Kamaleswaran, Madushani, Adhikari, Holder, Steyerberg, Buchman, Lough, Celi.

    Statistical analysis: Wong, Charpignon, Kim, Josef, de Hond, Fojas, Tabaie, Madushani, Adhikari, Steyerberg, Celi.

    Administrative, technical, or material support: Wong, Liu.

    Supervision: Kamaleswaran, Adhikari, Holder, Buchman, Celi.

    Conflict of Interest Disclosures: Dr Wong reported holding equity and management roles in Ataia Medical outside the submitted work. Dr Josef reported receiving grants from the National Institutes of Health during the conduct of the study. Dr Madushani reported receiving grants from the National Institute on Drug Abuse outside the submitted work. Dr Holder reported receiving grants from the National Institutes of Health during the conduct of the study and receiving speakers’ fees from Baxter international outside the submitted work. Dr Buchman reported that Emory University received funding from the US government for Dr Buchman’s role as senior advisor and IPA to the Division of Research, Innovation, and Ventures within BARDA; furthermore, he reported serving as editor in chief of Critical Care Medicine and receiving grants from Henry M. Jackson Foundation. No other disclosures were reported.

    Funding/Support: Dr Wong is supported by grant 2T32GM095442 from the National Institute of General Medical Sciences and the clinical and translational science award pilot informatics grant by the National Center for Advancing Translation Science under award UL1TR002378. Dr Kamaleswaran is supported by grant 17267 from the Michael J. Fox Foundation. Dr Josef is supported by grant 2T32GM095442 from the National Institute of General Medical Sciences. Dr Tabaie was funded by the Surgical Critical Care Initiative (SC2i), the Department of Defense’s Defense Health Program Joint Program Committee 6/Combat Casualty Care (grants USUHS HT9404-13-1-0032 and HU0001-15-2-0001). Dr Holder is supported by the National Institute for General Medical Sciences under award number K23GM137182 and the National Center for Advancing Translational Sciences of the National Institutes of Health under award number UL1TR002378. Dr Buchman is supported by the Society of Critical Care Medicine and the Biomedical Advanced Research and Development Authority. Dr Celi is funded by the National Institute of Health through grant NIBIB R01 EB017205.

    Role of the Funder/Sponsor: The funders 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.

    References
    1.
    Sjoding  MW, Dickson  RP, Iwashyna  TJ, Gay  SE, Valley  TS.  Racial bias in pulse oximetry measurement.   N Engl J Med. 2020;383(25):2477-2478. doi:10.1056/NEJMc2029240PubMedGoogle ScholarCrossref
    2.
    US Food and Drug Administration. Pulse oximeter accuracy and limitations: FDA safety communication. February 19, 2021. Accessed March 22, 2021. https://www.fda.gov/medical-devices/safety-communications/pulse-oximeter-accuracy-and-limitations-fda-safety-communication
    3.
    Seguin  P, Le Rouzo  A, Tanguy  M, Guillou  YM, Feuillu  A, Mallédant  Y.  Evidence for the need of bedside accuracy of pulse oximetry in an intensive care unit.   Crit Care Med. 2000;28(3):703-706. doi:10.1097/00003246-200003000-00017PubMedGoogle ScholarCrossref
    4.
    Van de Louw  A, Cracco  C, Cerf  C,  et al.  Accuracy of pulse oximetry in the intensive care unit.   Intensive Care Med. 2001;27(10):1606-1613. doi:10.1007/s001340101064PubMedGoogle ScholarCrossref
    5.
    Perkins  GD, McAuley  DF, Giles  S, Routledge  H, Gao  F.  Do changes in pulse oximeter oxygen saturation predict equivalent changes in arterial oxygen saturation?   Crit Care. 2003;7(4):R67. doi:10.1186/cc2339PubMedGoogle ScholarCrossref
    6.
    Wilson  BJ, Cowan  HJ, Lord  JA, Zuege  DJ, Zygun  DA.  The accuracy of pulse oximetry in emergency department patients with severe sepsis and septic shock: a retrospective cohort study.   BMC Emerg Med. 2010;10:9. doi:10.1186/1471-227X-10-9PubMedGoogle ScholarCrossref
    7.
    Singh  AK, Sahi  MS, Mahawar  B, Rajpurohit  S.  Comparative evaluation of accuracy of pulse oximeters and factors affecting their performance in a tertiary intensive care unit.   J Clin Diagn Res. 2017;11(6):OC05-OC08. doi:10.1183/1393003.congress-2017.PA2127PubMedGoogle Scholar
    8.
    Vandenbroucke  JP, Von Elm  E, Altman  DG,  et al.  Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): explanation and elaboration; translation to Russian.   Digital Diagnostics. 2021;2(2):1-10. doi:10.17816/dd70821Google ScholarCrossref
    9.
    Pollard  T, Johnson  A, Raffa  J, Celi  LA, Badawi  O, Mark  R. eICU Collaborative Research Database. PhysioNet. April 15, 2019. Accessed February 15, 2020. https://physionet.org/content/eicu-crd/2.0/
    10.
    Pollard  TJ, Johnson  AEW, Raffa  JD, Celi  LA, Mark  RG, Badawi  O.  The eICU Collaborative Research Database, a freely available multi-center database for critical care research.   Sci Data. 2018;5:180178. doi:10.1038/sdata.2018.178PubMedGoogle Scholar
    11.
    Goldberger  AL, Amaral  LAN, Glass  L,  et al.  PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals.   Circulation. 2000;101(23):E215-E220. doi:10.1161/01.cir.101.23.e215PubMedGoogle ScholarCrossref
    12.
    Johnson  AEW, Pollard  TJ, Shen  L,  et al.  MIMIC-III, a freely accessible critical care database.   Sci Data. 2016;3:160035. doi:10.1038/sdata.2016.35PubMedGoogle Scholar
    13.
    Vincent  J-L, Moreno  R, Takala  J,  et al; the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine.  The SOFA (Sepsis-Related Organ Failure Assessment) score to describe organ dysfunction/failure.   Intensive Care Med. 1996;22(7):707-710. doi:10.1007/s001340050156PubMedGoogle ScholarCrossref
    14.
    Coustet  B, Lhuissier  FJ, Vincent  R, Richalet  JP.  Electrocardiographic changes during exercise in acute hypoxia and susceptibility to severe high-altitude illnesses.   Circulation. 2015;131(9):786-794. doi:10.1161/CIRCULATIONAHA.114.013144PubMedGoogle ScholarCrossref
    15.
    Bickler  PE, Feiner  JR, Lipnick  MS, Batchelder  P, MacLeod  DB, Severinghaus  JW.  Effects of acute, profound hypoxia on healthy humans: implications for safety of tests evaluating pulse oximetry or tissue oximetry performance.   Anesth Analg. 2017;124(1):146-153. doi:10.1213/ANE.0000000000001421PubMedGoogle ScholarCrossref
    16.
    Dewan  NA, Nieto  FJ, Somers  VK.  Intermittent hypoxemia and OSA: implications for comorbidities.   Chest. 2015;147(1):266-274. doi:10.1378/chest.14-0500PubMedGoogle ScholarCrossref
    17.
    MacIntyre  NR.  Tissue hypoxia: implications for the respiratory clinician.   Respir Care. 2014;59(10):1590-1596. doi:10.4187/respcare.03357PubMedGoogle ScholarCrossref
    18.
    Colgan  SP, Campbell  EL, Kominsky  DJ.  Hypoxia and mucosal inflammation.   Annu Rev Pathol. 2016;11:77-100. doi:10.1146/annurev-pathol-012615-044231PubMedGoogle ScholarCrossref
    19.
    Jansen  TC, van Bommel  J, Schoonderbeek  FJ,  et al; LACTATE study group.  Early lactate-guided therapy in intensive care unit patients: a multicenter, open-label, randomized controlled trial.   Am J Respir Crit Care Med. 2010;182(6):752-761. doi:10.1164/rccm.200912-1918OCPubMedGoogle ScholarCrossref
    20.
    Kruse  O, Grunnet  N, Barfod  C.  Blood lactate as a predictor for in-hospital mortality in patients admitted acutely to hospital: a systematic review.   Scand J Trauma Resusc Emerg Med. 2011;19:74. doi:10.1186/1757-7241-19-74PubMedGoogle ScholarCrossref
    21.
    Scholz  H, Boivin  FJ, Schmidt-Ott  KM,  et al.  Kidney physiology and susceptibility to acute kidney injury: implications for renoprotection.   Nat Rev Nephrol. 2021;17(5):335-349. doi:10.1038/s41581-021-00394-7PubMedGoogle ScholarCrossref
    22.
    US Food and Drug Administration. Pulse oximeters—premarket notification submissions [510(k)s]: guidance for industry and Food and Drug Administration staff. March 2013. Accessed April 2, 2021. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/pulse-oximeters-premarket-notification-submissions-510ks-guidance-industry-and-food-and-drug
    23.
    US Census Bureau. 2017 National population projections tables: main series. Accessed April 2, 2021. https://www.census.gov/data/tables/2017/demo/popproj/2017-summary-tables.html
    24.
    US Census Bureau. QuickFacts: United States. Accessed April 4, 2021. https://www.census.gov/quickfacts/fact/table/US/PST045219
    25.
    Ghosh  I. Visualizing the US population by race. December 28, 2020. Accessed April 2, 2021. https://www.visualcapitalist.com/visualizing-u-s-population-by-race/
    26.
    Blum  FE, Lund  ET, Hall  HA, Tachauer  AD, Chedrawy  EG, Zilberstein  J.  Reevaluation of the utilization of arterial blood gas analysis in the intensive care unit: effects on patient safety and patient outcome.   J Crit Care. 2015;30(2):438.e1-438.e5. doi:10.1016/j.jcrc.2014.10.025PubMedGoogle ScholarCrossref
    27.
    Feiner  JR, Severinghaus  JW, Bickler  PE.  Dark skin decreases the accuracy of pulse oximeters at low oxygen saturation: the effects of oximeter probe type and gender.   Anesth Analg. 2007;105(6)(suppl):S18-S23. doi:10.1213/01.ane.0000285988.35174.d9PubMedGoogle ScholarCrossref
    28.
    Bickler  PE, Feiner  JR, Severinghaus  JW.  Effects of skin pigmentation on pulse oximeter accuracy at low saturation.   Anesthesiology. 2005;102(4):715-719. doi:10.1097/00000542-200504000-00004PubMedGoogle ScholarCrossref
    ×