Results
A total of 174 patients with COVID-19 required ABG sampling in our hospital wards during the study period. After the exclusion of 57 patients, a total of 117 patients’ data was evaluated (Figure 1). The mean age of study patients was 69.4 ± 12.0 years, and 65% (n:76) were male. Of all study patients, 97 (82.9%) had at least one comorbid disease. The computed tomography findings of thorax were compatible with COVID-19 pneumonia in 110 (94%) patients. The patients were hospitalized at the median 4th [2-6] day, and the ABG samples were analyzed at the median 11th [8-15] day after symptom onset. Among the patients, 98.3% received anticoagulant, and 45.3% received antiaggregant agents during hospitalization.
Twenty-nine (24.8%) patients were transferred to the intensive care units and 14 (12.0%) died.
The median SpO2 and SaO2 levels of the patients were 88% [84-88] and 91.8% [88.3-94.4], respectively. In 10 out of 117 patients, SpO2 levels were higher than SaO2 (mean difference 1.1±0.7%). We categorized the patients into two groups; in whom the difference between SaO2 and SpO2 was ≤4% (acceptable difference group) and >4% (large difference group). In 59 patients (50.4%), the difference between SpO2 and SaO2 measurements was greater than 4% (large difference), and within this group, all SaO2 levels measured were higher than SpO2. The baseline features and comorbid conditions of these two groups were given in Table 1.
Patients with large difference have higher neutrophil count d-dimer, ferritin, fibrinogen and C-reactive protein (CRP) level than the patients with an acceptable difference (Table 2). To determine the effect of clinical and laboratory parameters on large difference risk, a binary logistic regression analysis was employed; revealing that increased d-dimer, fibrinogen, ferritin level and decreased lymphocyte count were significantly associated with large difference between SpO2 & SaO2 (Table 3).
We performed receiver operating characteristic (ROC) curve analyses to determine cut-off values for ferritin, fibrinogen and lymphocyte count that would predict large difference between SpO2 and SaO2. The best cut-off value was 4.8 g/dL (area under curve-AUC: 0.761, 95% CI: 0.674–0.848, p < 0.001, sensitivity:71%; specificity:73%) for fibrinogen, 228 g/dL (AUC: 0.813, 95% CI: 0.734–0.892, p < 0.001, sensitivity:86%; specificity:57%) for ferritin, and 1,04 x 103/mm3 (AUC: 0.806, 95% CI: 0.722–0.890, p < 0.001, sensitivity:86%; specificity:70%) for lymphocyte count.
Bland-Altman analysis comparing SpO2 with SaO2 within the total study group demonstrated the negative bias (mean difference) of 4.02% with an SD of 2.65 (precision) and the limits of agreement of −9.22% to 1.17% (Figure 2). This indicates that the SpO2 underestimated SaO2 value by an average 4.02 ± 2.65% and limits of agreement were clinically important since SpO2 could be measured 9% lower or 1% greater than actual arterial oxygen saturation. Also, we performed Bland-Altman analyses to calculate mean differences and precision separately in subgroups of patients with increased and normal serum ferritin, with increased and normal serum fibrinogen; normal and decreased lymphocyte count. The results showed a significant increase of the bias in patients with high fibrinogen level, high ferritin level and low lymphocyte count compared those with normal values (Figure 3, Figure 4 and Figure 5, respectively). Within all subgroups, we showed that SpO2 underestimated SaO2. These finding suggested that the reliability of pulse oximeter is further reduced in patients with high fibrinogen, high ferritin and low lymphocyte count.