Overt and occult hypoxemia in patients hospitalized with novel coronavirus disease 2019

Abstract

Background: Progressive hypoxemia is the predominant mode of deterioration in COVID-19. Among hypoxemia measures, the ratio of the partial pressure of arterial oxygen to the fraction of inspired oxygen (P/F ratio) has optimal construct validity but poor availability because it requires arterial blood sampling. Pulse oximetry reports oxygenation continuously, but occult hypoxemia can occur in Black patients because the technique is affected by skin color. Oxygen dissociation curves allow non-invasive estimation of P/F ratios (ePFR) but this approach remains unproven.

Research question: Can ePFRs measure overt and occult hypoxemia?

Study design and methods: We retrospectively studied COVID-19 hospital encounters (n=5319) at two academic centers (University of Virginia [UVA] and Emory University). We measured primary outcomes (death or ICU transfer within 24 hours), ePFR, conventional hypoxemia measures, baseline predictors (age, sex, race, comorbidity), and acute predictors (National Early Warning Score (NEWS) and Sepsis-3). We updated predictors every 15 minutes. We assessed predictive validity using adjusted odds ratios (AOR) and area under receiver operating characteristics curves (AUROC). We quantified disparities (Black vs non-Black) in empirical cumulative distributions using the Kolmogorov-Smirnov (K-S) two-sample test.

Results: Overt hypoxemia (low ePFR) predicted bad outcomes (AOR for a 100-point ePFR drop: 2.7 [UVA]; 1.7 [Emory]; p<0.01) with better discrimination (AUROC: 0.76 [UVA]; 0.71 [Emory]) than NEWS (AUROC: 0.70 [UVA]; 0.70 [Emory]) or Sepsis-3 (AUROC: 0.68 [UVA]; 0.65 [Emory]). We found racial differences consistent with occult hypoxemia. Black patients had better apparent oxygenation (K-S distance: 0.17 [both sites]; p<0.01) but, for comparable ePFRs, worse outcomes than other patients (AOR: 2.2 [UVA]; 1.2 [Emory], p<0.01).

Interpretation: The ePFR was a valid measure of overt hypoxemia. In COVID-19, it may outperform multi-organ dysfunction models like NEWS and Sepsis-3. By accounting for biased oximetry as well as clinicians’ real-time responses to it (supplemental oxygen adjustment), ePFRs may enable statistical modelling of racial disparities in outcomes attributable to occult hypoxemia.

Figure 1:. Evaluation of the construct validity of operational markers of hypoxemia in hypothetical clinical scenarios

Construct validity of any marker of hypoxemia is the extent to which that marker accurately reflects the clinical construct of hypoxemia. This figure examines the construct validity of five operational markers of hypoxemia (rows) in common clinical scenarios (columns). In each scenario (column), two records of a patient’s oxygenation are compared (Record A on left, Record B on right). The first row titled “clinical acumen” describes a clinically sensible conclusion that a clinician might draw by comparing the two records. For example, in Scenario 2, a clinician will likely conclude that the two records do not represent any meaningful change in the severity of hypoxemic respiratory failure (row 1, column 2). Rather, Record B (SpO₂ of 91% on 2LPM of oxygen) might simply reflect the fact that a clinician initiated supplemental oxygen in response to Record A (SpO₂ of 85% on room air). Each of the subsequent rows describes the conclusion based solely on comparing a particular marker of hypoxemia. For example, if one solely compared SpO₂ in Scenario 2 (row 2, column 2), the conclusion would be that Record A reflects significantly more severe hypoxemia than Record B (SpO₂ of 85% v/s 91%). Considering the varying range of each marker, we used the following cutoffs to determine a “significantly more/less hypoxemia”: any difference ≥ 1 for NEWS (range 0 to 5), any difference ≥ 2 for SpO₂ (range 85 to 100) and supplemental oxygen flow rate (range 0 to 15 LPM), and any difference ≥ 50 for S/F ratio (range 85 – 476) and P/F ratio (range 50 – 632). A cell is shaded green when there is agreement between the marker of hypoxemia and clinical acumen; and it is shaded red when there is disagreement. This figure illustrates the advantages of estimated P/F ratios over other markers – it is the only marker to agree with clinical acumen in all scenarios. We were unable to conceptualize any scenario where P/F ratio would be inferior to other markers. (RA = Room Air; LPM = liters per minute)

Figure 2:. Discrimination of estimated P/F ratio for clinical deterioration in patients with COVID-19

This figure compares the Area Under the Receiver Operator Characteristic curve (AUROC) of multivariable logistic regression models for clinical deterioration (transfer to ICU or mortality within 24 hours) from COVID-19. The blue boxes show the AUROC for a model and the yellow boxes show p-values from pairwise comparison (DeLong’s test). Results from UVA are on the left and those from Emory are on the right. The baseline risk model used age, sex, race, Charlson Comorbidity Index, and pre-infection baseline Sequential Organ Failure Assessment (SOFA) score as predictors (baseline SOFA was only available at UVA). The model for the each criterion was created by adding that criterion to the baseline risk predictors. The estimated P/F ratio (ePFR) had optimal model discrimination, and it outperformed NEWS and Sepsis-3 (acute rise in SOFA score at UVA and total SOFA in Emory) models.

Figure 3:. Characterizing the impact of racially biased pulse oximetry measurements

Panels A - C show the Empirical Cumulative Distribution Functions for SpO₂, S/F ratio, and ePFR respectively. This figure depicts the results from UVA. Corresponding results from Emory are shown in eFigure 2. Race is encoded by color (red - Black patients, blue - others). The separation in SpO₂ distributions was narrow (being minimal at SpO₂ < 92%), suggesting an equitable clinician effort to prevent oxygen desaturation. Yet, the separation in S/F ratio and ePFR distributions was wide at all values. This suggests that, on average, clinicians were achieving their SpO₂ targets with lower FiO₂ settings in Black patients (eFigure 4). For comparable S/F ratio and ePFR values, outcomes were worse for Black patients than others (Panel E-F). Together, these findings reveal that clinicians were likely undertreating hypoxemia due to an overestimation of SpO₂. Significantly, this disparity remained undetected when the SpO₂ was studied instead of S/F ratio or ePFR (Panel D). To make the plots directly comparable despite the varying scales of the hypoxemia measures, we used SpO₂ values ranging from 85% to 100% and the corresponding range from a minimum S/F ratio 85 and ePFR 50 (representing a SpO₂ of 85% on 100% FiO₂) to a maximum S/F ratio 476 and ePFR 633 (representing a SpO₂ of 100% on room air). To smoothen the ECDFs, we converted SpO₂ from integer to continuous by adding uniformly distributed noise (+/− 0.5% with a maximum SpO₂ of 100%). To calculate the rate of clinical deterioration at a particular level, we used a window centered at that level with width equal to one standard deviation (2.5 for SpO₂, 100 for S/F ratio and 120 for ePFR). The dashed horizontal lines (Panels D-F) mark the rate of clinical deterioration in the entire dataset (1.85%).