Cumulative oxygen deficit is a novel predictor for the timing of invasive mechanical ventilation in COVID-19 patients with respiratory distress

Introduction

Coronavirus disease 2019 (COVID-19) has spread all over the world since its first outbreak in Wuhan, China in December 2019 (Wang et al., 2020; Novel Coronavirus Pneumonia Emergency Response Epidemiology Team, 2020). The fatality rate was reported to be around 5% all over the world (Phua et al., 2020). A substantial number of patients (19%) infected with the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) will develop respiratory distress and acute lung injury (Wu & McGoogan, 2020; Ruan et al., 2020). Respiratory support becomes important for this type of severe patients (Yang et al., 2020). The surviving sepsis guideline of critically ill COVID-19 patients recommended use of oxygen supplementation to maintain pulse oximetry >90%, followed by non-invasive mechanical ventilation (NIV), high-flow nasal canula (HFNC), invasive mechanical ventilation (IMV) and extracorporeal membrane oxygenation (ECMO). However, there is no specific recommendations for the timing of transition from non-invasive support to IMV, and the recommendations are largely based on expert opinions. For example, the guideline recommends “close monitoring for worsening of respiratory status, and early intubation in a controlled setting if worsening occurs” (Alhazzani et al., 2020). This recommendation is based on best practice statement and there is no data on when IMV should be initiated. In clinical practice, the judgement of “worsening” is subjective and varied substantially between different institutions and physicians. The timing of initiation of IMV is not standardized and is mainly determined by subjective judgement. On the one hand, IMV is able to reverse catastrophic hypoxemia and maintain tissue oxygenation, which is life-saving for COVID-19 patients with severe hypoxemia. On the other hand, IMV can cause ventilator-induced lung injury (Herasevich et al., 2011; Cressoni et al., 2016), and patients on IMV usually require large dose of sedatives, analgesics and even neuromuscular blockades (Jakob et al., 2012; Bellani et al., 2016; Chang et al., 2020). These drugs have significant adverse effects (Barr et al., 2013; Murray et al., 2016). Thus, it is difficult to determine the appropriate timing of IMV.

In our experience, we proposed that the timing of transition from non-invasive oxygenation to IMV should consider both the magnitude of hypoxemia and the duration of the hypoxemia. Thus, we developed a novel marker called Cumulative Oxygen Deficit (COD) to reflect both dimensions. By using a single predictor, we reduce a two-dimension feature to a one-dimension parameter that is comparable among different patients. In our study, we hypothesized that the COD before IMV could be a better biomarker than PaO₂ to predict survival outcome.

Methods

Calculation of cumulative oxygen deficit

For patients with IMV, COD was calculated before the use of IMV. Figure 1 is a sample patient used to illustrate the calculation of COD: $\mathrm{C}\mathrm{O}\mathrm{D}\left(\mathrm{m}\mathrm{m}\mathrm{H}\mathrm{g}\cdot \mathrm{d}\mathrm{a}\mathrm{y}\right)=80\times \left(t_5-t_1\right)-{\sum }_{i=1}^4\left(x_{i+1}+x_i\right)\cdot \left(t_{i+1}-t_i\right)/2$, where x_i is the value of PaO₂ measured in mmHg, and t_i is the time at which x_i is measured. The low end value of PaO₂ was 80 mmHg in our hospital and this value is also physiologically reasonable that the oxygen saturation will not continue to rise above this reference value (Collins et al., 2015). Thus, the COD accounted for both magnitude and duration of hypoxemia before IMV. We hypothesized that the longer a patient was on hypoxemia before IMV, the worse of the survival outcome. On the other hand, the outcome would be not so bad if hypoxemia was immediately corrected with IMV even if the magnitude of hypoxemia is large.

Results

Study population

A total of 111 patients met the inclusion criteria and were included for analysis. No patients were excluded due to COPD, pregnancy CVA and paralysis. There was no patient being excluded from the participating hospitals. There were 80 patients who did not need IMV, and 31 patients required IMV during hospitalization. Patients with IMV had substantially lower PaO₂ (62 (49, 89) vs. 90.5 (68, 125.25) mmHg; p < 0.001), higher pH (7.44 (7.38, 7.47) vs. 7.40 (7.35, 7.43); p = 0.006), higher serum lactate (2.5 (1.7, 3.1) vs. 1.7 (1.1, 2.85) mmol/L, p < 0.036) and higher COD (−6.87 (−29.36, 52.38) vs. −231.68 (−1040.78, 119.83) mmHg ∙ day) than patients without IMV during hospitalization (Table 1). These variables were reported as the first value during hospitalization. The time courses of the transition from different types of respiratory support are shown in Fig. 2. It is noted that larger proportion of patients required IMV in the non-survivors.

Table 1:

Comparison between patients with and without invasive mechanical ventilation.

Variables	Total (n = 111)	No IMV (n = 80)	IMV (n = 31)	p
Age, Median (IQR)	57 (45.5, 68.5)	56 (40, 67.25)	64 (54.5, 70.5)	0.076
Sex, male (%)	60 (55)	39 (50)	21 (68)	0.143
Comorbidities
Respiratory system, n (%)	18 (16)	13 (16)	5 (16)	1
Cardiovascular system, n (%)	41 (37)	29 (36)	12 (39)	0.983
Smoking history, n (%)	2 (2)	1 (1)	1 (3)	0.482
PaCO2 (mmHg), Median (IQR)	43 (39, 47.5)	44 (41, 48)	39 (36, 45)	0.051
PaO2 (mmHg), Median (IQR)	82 (60, 120)	90.5 (68, 125.25)	62 (49, 89)	<0.001
pH, Median (IQR)	7.42 (7.35, 7.44)	7.40 (7.35, 7.43)	7.44 (7.38, 7.47)	0.006
BE (mmol/l), Median (IQR)	2.4 (−2.05, 5.1)	2.35 (−2.57, 5.1)	2.4 (0.2, 5.2)	0.342
Lactate (mmol/l), Median (IQR)	1.9 (1.3, 3)	1.7 (1.1, 2.85)	2.5 (1.7, 3.1)	0.036
PF (mmHg), Median (IQR)	224.32 (106.14, 348.02)	250.05 (194.14, 363.64)	111.25 (71.98, 200.27)	<0.001
CRP (mg/dl), Median (IQR)	8.9 (2.75, 53.2)	5.5 (1.78, 35.83)	19.5 (9.85, 78.05)	<0.001
Lymphocyte count (mmHg·day), Median (IQR)	0.78 (0.28, 0.86)	0.78 (0.43, 0.89)	0.43 (0.28, 0.66)	0.005
COD (mmHg·day), Median (IQR)	−50.38 (-827.91, 72.67)	−231.68 (−1040.78, 119.83)	−6.87 (−29.36, 52.38)	0.018
COD, n (%)				<0.001
<0 (mmHg·day)	56 (50)	49 (61)	7 (23)
0–30 ($\times {10}^9/L$)	14 (13)	3 (4)	11 (35)
30–50 (mmHg·day)	9 (8)	5 (6)	4 (13)
>50 (mmHg·day)	32 (29)	23 (29)	9 (29)
Hospital LOS (days), Mean ± SD	26.55 ± 11.78	26.1 ± 11.35	27.7 ± 12.97	0.548
Mortality, n (%)	22 (20)	13 (16)	9 (29)	0.211

DOI: 10.7717/peerj.10497/table-1

Note:

LOS, length of stay; SD, standard deviation; COD, cumulative oxygen deficit; CRP, C-reactive protein; IQR, interquartile range; BE, base excess.

Independent association of COD and survival outcome in IMV patients

Cumulative Oxygen Deficit was independently associated with survival outcome in multivariable Cox regression model. As compared to patients with COD < 0, patients with COD from 0 to 30 $\mathrm{m}\mathrm{m}\mathrm{H}\mathrm{g}\cdot \mathrm{d}\mathrm{a}\mathrm{y}$ were not more likely to die, whereas those with COD > 30 $\mathrm{m}\mathrm{m}\mathrm{H}\mathrm{g}\cdot \mathrm{d}\mathrm{a}\mathrm{y}$ had higher risk of fatality (HR: 3.79, 95% CI [2.57–16.93]; p = 0.037), and those with COD > 50 $\mathrm{m}\mathrm{m}\mathrm{H}\mathrm{g}\cdot \mathrm{d}\mathrm{a}\mathrm{y}$ were 10 times more likely to die (HR: 10.45, 95% CI [1.28–85.37]; p = 0.029). The time from admission to intubation, PaO₂ and lymphocyte count were not associated with survival outcome (Table 2). The time-dependent AUCs of COD, PaO₂ and the time from admission to intubation are shown in Fig. 3. It showed that COD had consistently higher AUCs from day 14 to 21. In other words, COD was the best predictor after day 14. Table 3 shows factors associated with IMV.

Table 2:

Multivariable Cox regression model to explore the independent association of cumulative oxygen deficit and hazard of death in patients with invasive mechanical ventilation.

	HR [95% CI]	p
COD (< 0 as reference)	1
0–30	1.45 [0.98–3.47]	0.052
30–50	3.79 [2.57–16.93]	0.037
>50	10.45 [1.28–85.37]	0.029
Time from admission to intubation	0.78 [0.59–1.03]	0.078
Age	1.20 [1.00–1.44]	0.056
Sex (female as reference)	9.84 [0.43–225.59]	0.152
PaCO2	0.85 [0.60–1.19]	0.344
PaO2	1.00 [0.97–1.03]	0.939
pH	0.70 [0.44–1.13]	0.144
BE	1.89 [0.70–5.08]	0.206
CRP	0.98 [0.94–1.01]	0.233
Lymphocyte count	2.21 [0.04–115.48]	0.695

DOI: 10.7717/peerj.10497/table-2

Note:

COD, cumulative oxygen deficit; CRP, C-reactive protein; IQR, interquartile range; BE, base excess; HR, hazard ratio; CI, confidence interval.

Table 3:

Factors associated with use of IMV.

Variables	HR [95% CI]	p
PaO2 (for every 10-mmHg increase)	0.91 [0.84–0.99]	0.022
PaCO2	1.02 [0.98–1.06]	0.337
Age	1.02 [0.99–1.04]	0.235
pH (for every 0.01 increase)	1.02 [0.96–1.09]	0.565
CRP	1.00 [0.99–1.01]	0.824
Lymphocyte count	0.27 [0.09–0.81]	0.020
Lactate	1.56 [1.17–2.08]	0.003

DOI: 10.7717/peerj.10497/table-3

Note:

CRP, C-reactive protein; HR, hazard ratio; CI, confidence interval.

Discussion

The study developed a novel biomarker COD which considered both magnitude and duration of hypoxemia, to assist the timing of IMV in patients with COVID-19. In patients with IMV during hospitalization, COD before intubation was a strong predictor of survival outcome. Patients with COD > 30 $\mathrm{m}\mathrm{m}\mathrm{H}\mathrm{g}\cdot \mathrm{d}\mathrm{a}\mathrm{y}$, which is equivalent to a persistent hypoxemia with PaO₂ of 60 mmHg for 1.5 days, are more likely to die during hospitalization. Patients in crowded hospital during COVID-19 pandemic were more likely to experience this situation. The time dependent AUCs of COD were substantially higher than that of the PaO₂ or the time from admission to intubation alone. Clinical implication of this finding is that we need to consider both the magnitude and duration of hypoxemia before IMV is considered. Long duration of mild hypoxemia, which is prevalent in clinical practice under NIV, may be dangerous for COVID-19 patients.

Many studies have been conducted to address the question on whether NIV should be used for patients with pulmonary/direct ARDS, but the results are conflicting (Chawla et al., 2020). NIV was not associated with improved mortality or length of stay, compared with patients who were intubated without trying NIV in a cohort of Middle East Respiratory Syndrome (MERS) patients. Furthermore, most patients (92.4%) who had tried NIV were eventually managed with IMV (Alraddadi et al., 2019). However, this was a retrospective study and the initiation of IMV was not standardized prospectively. Our study indicated that large COD can be harmful and the correction of COD with IMV might be beneficial. This could be explained by potential adverse effects of NIV including large tidal volumes and injurious transpulmonary pressures (Brochard et al., 2014). These adverse effects of NIV could be avoided by using IMV. For example, protective ventilation strategy can be performed with IMV (Zhang et al., 2015; Fan, Brodie & Slutsky, 2018), but it is impossible under NIV. Furthermore, the use of NIV or HFNC can delay IMV, leading to emergency or more unstable intubations (Brochard, 2003). Thus, IMV should be considered as early as possible if the COD reaches 30 $\mathrm{m}\mathrm{m}\mathrm{H}\mathrm{g}\cdot \mathrm{d}\mathrm{a}\mathrm{y}$, without trying NIV or HFNC to delay intubation.

Arterial partial oxygen pressure and its derivatives such as PF ratio are well established risk factor for mortality outcome in patients with ARDS. Thus, PF ratio is used to classify ARDS patients into mild, moderate and severe cases (ARDS Definition Task Force et al., 2012). However, this risk classification system considers only the magnitude of hypoxemia (Cartotto et al., 2016; Dai et al., 2019). Our results suggest that the duration of hypoxemia can be equally important. A strength of our study was that all measurements of PaO₂ were collected longitudinally, allowing for the calculation of the area under the PaO₂-day curve to derive a novel predictor. Our analysis focused on patients with IMV and found that the predictive performance for survival outcome of COD before intubation was substantially better than PaO₂ or the time from admission to intubation. The latter two indices are the two components of COD. The combination of the two indices substantially improves the predictive discrimination for mechanically ventilated patients. Although direct causal inference that the use of IMV to reduce COD can improve survival outcome cannot be established with current analysis, our result identified a modifiable risk factor for survival outcome. It is reasonable to deduce that reducing COD as early as possible with IMV can be beneficial.

Several limitations should be acknowledged in the study. First, the study was retrospective in design, and many unmeasured confounders may exist to influence the choice of respiratory supports (Uddin et al., 2016). The presence of such unmeasured confounders will compromise the effectiveness of the propensity score matching procedure. Second, the use of NIV or HFNC was completely at the discretion of the treating physician. There was no standard protocol in participating hospitals. Thus, it is difficult to determine whether the use of NIV or HFNC could benefit COVID-19 induced ARDS. Third, for patients without IMV, we calculated the COD across all days of hospitalization. This could be biased because the time-dimension was longer than the IMV group. However, since non-IMV group generally did not have oxygen deficit across hospital stay, the COD was substantially lower than the IMV group. Finally, we only included broad categories of comorbidity burden in our analysis (i.e., respiratory system, cardiovascular system), because the retrospective design of the study did not allow detailed information for the calculation of the Elixhauser’s comorbidity index. It is well known that Elixhauser’s comorbidity index is a good quantity for risk stratification of hospitalized patients (Elixhauser et al., 1998). However, this index is designed to work with ICD-9-CM codes in administrative database, which is not applicable to data collected in retrospective studies.

Conclusions

In conclusion, the study developed a novel predictor COD, which considered both magnitude and duration of hypoxemia, to assist the timing of IMV in patients with COVID-19. We suggest IMV should be the preferred ventilatory support once the COD reaches 30 $\mathrm{m}\mathrm{m}\mathrm{H}\mathrm{g}\cdot \mathrm{d}\mathrm{a}\mathrm{y}$, as mortality increases beyond this value.

Supplemental Information