Pulse Oximetry and Race in Critically Ill Adults

Seitz, Kevin P.; Wang, Li; Casey, Jonathan D; Markus, Shannon A.; Jackson, Karen E.; Qian, Edward T; Self, Wesley H.; Rice, Todd W.; Semler, Matthew W.

doi:10.1097/cce.0000000000000758

Cited by 17 publications

(15 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…First, we hypothesized that inaccuracies in pulse oximetry contributed at least partially to the racial differences in predictive performance of the XGB model, given that pulse oximetry has been consistently shown to be less accurate in Black patients (16)(17)(18), and even the ROX index also had slightly worse performance in Black patients (although not statistically significant). However, removing pulse oximetry variables from the XGB model did not reduce the racial bias, suggesting that potential inaccuracies in pulse oximetry or vital signs did not completely account for the discrepancies.…”

Section: Discussionmentioning

confidence: 99%

Racial Differences in Accuracy of Predictive Models for High-Flow Nasal Cannula Failure in COVID-19

Yang,

Gregory,

Robichaux

et al. 2024

Critical Care Explorations

View full text Add to dashboard Cite

OBJECTIVES: To develop and validate machine learning (ML) models to predict high-flow nasal cannula (HFNC) failure in COVID-19, compare their performance to the respiratory rate-oxygenation (ROX) index, and evaluate model accuracy by self-reported race. DESIGN: Retrospective cohort study. SETTING: Four Emory University Hospitals in Atlanta, GA. PATIENTS: Adult patients hospitalized with COVID-19 between March 2020 and April 2022 who received HFNC therapy within 24 hours of ICU admission were included. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Four types of supervised ML models were developed for predicting HFNC failure (defined as intubation or death within 7 d of HFNC initiation), using routine clinical variables from the first 24 hours of ICU admission. Models were trained on the first 60% (n = 594) of admissions and validated on the latter 40% (n = 390) of admissions to simulate prospective implementation. Among 984 patients included, 317 patients (32.2%) developed HFNC failure. eXtreme Gradient Boosting (XGB) model had the highest area under the receiver-operator characteristic curve (AUROC) for predicting HFNC failure (0.707), and was the only model with significantly better performance than the ROX index (AUROC 0.616). XGB model had significantly worse performance in Black patients compared with White patients (AUROC 0.663 vs. 0.808, p = 0.02). Racial differences in the XGB model were reduced and no longer statistically significant when restricted to patients with nonmissing arterial blood gas data, and when XGB model was developed to predict mortality (rather than the composite outcome of failure, which could be influenced by biased clinical decisions for intubation). CONCLUSIONS: Our XGB model had better discrimination for predicting HFNC failure in COVID-19 than the ROX index, but had racial differences in accuracy of predictions. Further studies are needed to understand and mitigate potential sources of biases in clinical ML models and to improve their equitability.

show abstract

Section: Discussionmentioning

confidence: 99%

Racial Differences in Accuracy of Predictive Models for High-Flow Nasal Cannula Failure in COVID-19

Yang,

Gregory,

Robichaux

et al. 2024

Critical Care Explorations

View full text Add to dashboard Cite

show abstract

“…Polfer et al (28) found that Black patients with limb ischemia were significantly less commonly identified compared with White patients (23.3% vs. 92.9%). Conventional pulse oximetry is also less sensitive to hypoxemia in Black individuals (29–32). Neoplastic CCC was more common in the White cohort.…”

Section: Discussionmentioning

confidence: 99%

Sociodemographic Factors are Associated with Care Delivery and Outcomes in Pediatric Severe Sepsis

Webb,

Evans,

Smith

et al. 2024

Critical Care Explorations

View full text Add to dashboard Cite

IMPORTANCE: Sepsis is a leading cause of morbidity and mortality in the United States and disparate outcomes exist between racial/ethnic groups despite improvements in sepsis management. These observed differences are often related to social determinants of health (SDoH). Little is known about the role of SDoH on outcomes in pediatric sepsis. OBJECTIVE: This study examined the differences in care delivery and outcomes in children with severe sepsis based on race/ethnicity and neighborhood context (as measured by the social vulnerability index). DESIGN, SETTING, AND PARTICIPANTS: This retrospective, cross-sectional study was completed in a quaternary care children’s hospital. Patients 18 years old or younger who were admitted between May 1, 2018, and February 28, 2022, met the improving pediatric sepsis outcomes (IPSO) collaborative definition for severe sepsis. Composite measures of social vulnerability, care delivery, and clinical outcomes were stratified by race/ethnicity. MAIN OUTCOMES AND MEASURES: The primary outcome of interest was admission to the PICU. Secondary outcomes were sepsis recognition and early goal-directed therapy (EGDT). RESULTS: A total of 967 children met the criteria for IPSO-defined severe sepsis, of whom 53.4% were White/non-Hispanic. Nearly half of the cohort (48.7%) required PICU admission. There was no difference in illness severity at PICU admission by race (1.01 vs. 1.1, p = 0.18). Non-White race/Hispanic ethnicity was independently associated with PICU admission (odds ratio [OR] 1.35 [1.01–1.8], p = 0.04). Although social vulnerability was not independently associated with PICU admission (OR 0.95 [0.59–1.53], p = 0.83), non-White children were significantly more likely to reside in vulnerable neighborhoods (0.66 vs. 0.38, p < 0.001). Non-White race was associated with lower sepsis recognition (87.8% vs. 93.6%, p = 0.002) and less EGDT compliance (35.7% vs. 42.8%, p = 0.024). CONCLUSIONS AND RELEVANCE: Non-White race/ethnicity was independently associated with PICU admission. Differences in care delivery were also identified. Prospective studies are needed to further investigate these findings.

show abstract

“…Racial Differences in Pulse Oximetry. Recent studies in both adult and pediatric patients have demonstrated that pulse oximeters overestimate the true arterial oxygen saturation in non-White patients (18,19). Given our use of OSI in this study, we conducted a sensitivity analysis.…”

Section: Batch Correctionmentioning

confidence: 99%

Comparison of 16 Pediatric Acute Respiratory Distress Syndrome-Associated Plasma Biomarkers With Changing Lung Injury Severity*

Williams

Jones

Yunger

et al. 2023

Pediatric Critical Care Medicine

View full text Add to dashboard Cite

OBJECTIVES: Pediatric acute respiratory distress syndrome (PARDS) is a source of substantial morbidity and mortality in the PICU, and different plasma biomarkers have identified different PARDS and ARDS subgroups. We have a poor understanding of how these biomarkers change over time and with changing lung injuries. We sought to determine how biomarker levels change over PARDS course, whether they are correlated, and whether they are different in critically ill non-PARDS patients.DESIGN: Two-center prospective observational study. SETTING:Two quaternary care academic children's hospitals PATIENTS: Subjects under 18 years of age admitted to the PICU who were intubated and met the Second Pediatric Acute Lung Injury Consensus Conference-2 PARDS diagnostic criteria and nonintubated critically ill subjects without apparent lung disease. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS:Plasma samples were obtained on study days 1, 3, 7, and 14. The levels of 16 biomarkers were measured using a fluorometric bead-based assay. Compared with non-PARDS subjects, on day 1 PARDS subjects had increased concentrations of tumor necrosis factor-alpha, interleukin (IL)-8, interferon-γ, IL17, granzyme B, soluble intercellular adhesion molecule-1 (sICAM1), surfactant protein D, and IL18 but reduced matrix metalloproteinase 9 (MMP-9) concentrations (all p < 0.05). Day 1 biomarker concentrations and PARDS severity were not correlated. Over PARDS course, changes in 11 of the 16 biomarkers positively correlated with changing lung injury with sICAM1 (R = 0.69, p = 2.2 × 10 -16 ) having the strongest correlation. By Spearman rank correlation of biomarker concentrations in PARDS subjects, we identified two patterns. One had elevations of plasminogen activator inhibitor-1, MMP-9, and myeloperoxidase, and the other had higher inflammatory cytokines.CONCLUSIONS: sICAM1 had the strongest positive correlation with worsening lung injury across all study time points suggesting that it is perhaps the most biologically relevant of the 16 analytes. There was no correlation between biomarker concentration on day 1 and day 1 PARDS severity; however, changes in most biomarkers over time positively correlated with changing lung injury. Finally, in day 1 samples, 7 of the 16 biomarkers were not significantly different between PARDS and critically ill non-PARDS subjects. These data highlight the difficulty of using plasma biomarkers to identify organ-specific pathology in critically ill patients.

show abstract

Pulse Oximetry and Race in Critically Ill Adults

Cited by 17 publications

References 9 publications

Racial Differences in Accuracy of Predictive Models for High-Flow Nasal Cannula Failure in COVID-19

Racial Differences in Accuracy of Predictive Models for High-Flow Nasal Cannula Failure in COVID-19

Sociodemographic Factors are Associated with Care Delivery and Outcomes in Pediatric Severe Sepsis

Comparison of 16 Pediatric Acute Respiratory Distress Syndrome-Associated Plasma Biomarkers With Changing Lung Injury Severity*

Contact Info

Product

Resources

About