Obesity has emerged as one of the leading health concerns over the past century (1). In the United States, for the years 2013 and 2014, the overall obesity prevalence was 37.7% (2). Moreover, for adult obesity and severe obesity, the prevalence will continue to increase (3). Recent studies estimate by 2030 nearly 1 in 2 adults will have obesity (4, 5). At the same time, obese patients are over-represented in primary care (6). The similar circumstance is also reflected in critically ill patients where a recent study reported a prevalence of around 20% for the year 2012 (7, 8). Moreover, obese patients represent a specific population that seemed to be associated with high morbidity and increased resource utilization, who require an adapted ICU management (7–10).

The provision of supplemental oxygen is a ubiquitous intervention for mechanically ventilated patients. However, excessive oxygen could also be injurious (11, 12). British Thoracic Society (BTS) recommends, for morbidly obese patients who are at the risk of hypoventilation, even without evidence of coexistent obstructive sleep apnea, targeted oxygen saturation of 88–92% should be titrated (grade D) (13). Two small trials conducted in morbidly obese patients showed exposure to hyperoxia caused elevated partial pressure of carbon dioxide (Paco₂) and hypoventilation (14, 15). However, for patients with milder obesity, the literature is really sparse, especially concerning the impact in the critical setting. Moreover, the BTS or Thoracic Society of Australia and New Zealand guidelines did not cover the area of ICU (13, 16). Given the epidemiologic and specific pathophysiologic changes observed in obese patients, one could hypothesize that the obesity status could be an important confounder in the relationship between supplemental oxygen and outcomes. To the best of our knowledge, no study has specifically evaluated the relationship between supplemental oxygen and hospital mortality outcome in obese critically ill patients with mechanical ventilation.

The main aim of this study was to analyze the influence of the obesity status on the relationship between oxygen saturation and the hospital mortality in mechanically ventilated patients. The hypothesis was the impact of the supplemental oxygen on mortality in non-obese and obese critically ill patients with mechanical ventilation may be different.

Figure 1 showed the flow chart of the study. After exclusion, a total of 25,100 patients met our inclusion criteria, of which 10,564 (42%) were obese patients and 14,534 (58%) were non-obese patients. The median BMI in the total population was 28.3 (IQR 23.9–34.4) kg/m², of which the non-obese was 25 (IQR 22–27) kg/m² and the obese was 36 (IQR 33–42) kg/m². The median age was 65 (IQR 54–75) years, 13,632 (54%) was male. 4,242 (17%) patients were non-survivors at hospital discharge. Demographics and baseline characteristics between the obese and non-obese were presented in Table 1.

In non-obese patients, the median age was 67 (IQR 55–78) and the hospital mortality was 18%. BMI of non-obese patients was higher in survivors than in non-survivors (25 vs. 24, p < 0.001). Non-obese patients with comorbidities of hypertension or chronic renal insufficiency had higher hospital mortality (Table 1). Non-obese survivors had a higher percentage of TWM-Spo₂ of 99–100% (Table 2).

In obese patients, the median age was 63 (IQR 53–72) and hospital mortality was 15%, which was lower than non-obese patients (p < 0.001). Obese survivors had higher BMI than non-survivors (36 vs. 35, p = 0.014). Obese patients had a higher percentage of hypertension, diabetes mellitus, chronic heart failure, chronic obstructive pulmonary disease, and chronic renal insufficiency than the non-obese. Interestingly, obese survivors had a higher percentage of diabetes mellitus (33% vs. 30%, p = 0.009). Obese survivors had lower percentage of TWM-Spo₂ of 99–100% (11% vs. 14%) and higher percentage of TWM-Paco₂ > 45 mmHg (30 vs. 26%) (Table 2). Obese patients had lower PTS-Spo₂ of 99–100% than non-obese patients (39.8% vs. 54.8%, p < 0.001) (Table 3).

The relationship between TWM-Spo₂ categories and hospital mortality, which was assessed by the multivariable logistic models after adjusting confounders, was comparing with 94–98% category, < =88% category (OR 3.572; CI [2.343, 5.455]; p < 0.001) and 89–93% category (OR 1.514; CI [1.343, 1.706]; p < 0.001) both had higher odds ratio for hospital mortality, while 99–100% category was not associated with hospital mortality (Figure 2A and Supplementary Table 1). After multivariable analysis, PTS-Spo₂ of < =88% (OR 1.445; CI [1.356, 1.541]; p < 0.001; per 10% increase) and PTS-Spo₂ of 89–93% (OR 1.080; CI [1.052, 1.108]; p < 0.001; per 10% increase) had higher odds ratio for hospital mortality, PTS-Spo₂ of 94–98% (OR 0.968; CI [0.955, 0.981]; p < 0.001; per 10% increase) had lower odd ratio for hospital mortality, and PTS-Spo₂ of 99–100% (OR 0.997; CI [0.986, 1.009]; p = 0.636; per 10% increase) was not associated with high risk of hospital mortality (Figure 2B and Supplementary Tables 2–5).

The multivariable models including the interaction between PTS-Spo₂ and obesity status (non-obese and obese) showed that there was a significant interaction between PTS-Spo₂ and obesity status (Table 3). While the PTS-Spo₂ of 99-100% was not significantly associated with hospital mortality outcome (OR 1.004; 95% CI 0.990-1.018; p = 0.597; per 10% increase) in non-obese patients, it was associated with increased risk of mortality in obese patients (OR 1.028; 95% CI 1.010-1.046; p = 0.002; per 10% increase) (Figure 3A). While the PTS-Spo₂ of 89-93% was significantly associated with increased risk of hospital mortality outcome (OR 1.089; 95% CI 1.051-1.128; p <0.001; per 10% increase) in non-obese patients, it was not associated with hospital mortality in obese patients (OR 1.022; 95% CI 0.985-1.060; p = 0.235; per 10% increase) (Figure 3B). Supplementary Figure 3 showed the interaction between PTS-Spo₂ of < = 88% and 94-98% and obesity status (obese and non-obese).

The results of our study provide evidence to support our hypothesis. The impact of supplemental oxygen on hospital mortality was different in non-obese and obese critically ill patients with mechanical ventilation. Spo₂ of 99–100% was associated with a higher hospital mortality for obese patients but was not statistically significant for non-obese patients. Spo₂ of 94–98% was associated with decreased risk of hospital mortality outcome for both obese and non-obese patients. Spo₂ of 89–93% was associated with higher mortality for non-obese patients, but not statistically significant for obese patients. Spo₂ of < =88% was associated with an increased risk of hospital mortality outcome in both non-obese patients and obese patients. This is the first study to assess the relationship between supplemental oxygen and mortality in the specific population of obese critically ill patients with mechanical ventilation. The results of the present study showed that supplemental oxygen therapy should consider the obesity status into account.

The Oxygen-ICU randomized clinical trial showed among critically ill patients with an ICU length of stay of 72 h or longer, conservative therapy (Spo₂ 94–98%) vs. conventional therapy (Spo₂ 97–100%) resulted in lower ICU mortality (11). However, ICU-ROX trial found in adults undergoing mechanical ventilation in the ICU, conservative oxygen therapy (Spo₂ 91–96%), as compared with usual oxygen therapy (Spo₂ ≥91%), did not significantly affect the number of ventilator-free days or mortality (21). In the present study, compared with the TWM-Spo₂ 94–98% category, TWM-Spo₂ of 99–100% was not associated with hospital mortality. However, when considering the dose of Spo₂, PTS-Spo₂ of 94–98% was associated with lower hospital mortality in overall patients. The results of our study were consistent with the two RCTs.

There were some explanations of the differences observed between the non-obese and obese groups. First, obese patients can cause hypoxemia by decreasing lung volumes and may be at risk for hyperoxia-induced hypercapnia (22, 23). Two small trials conducted in morbidly obese patients with obesity hypoventilation syndrome (OHS) showed an elevated baseline Paco₂. After a 20-min exposure to hyperoxia, the patients had a mean increase in Paco₂ of about 4.4 mmHg (14, 15). However, whether hyperoxia-induced hypercapnia occurs in patients with milder obesity and without hypoventilation is unclear. Denault et al. found in obese patients with mean BMI 34 kg/m² after cardiac surgery, there was no clinically important increase in Paco₂ associated with higher Spo₂ values (24). Second, obese patients had higher rate of comorbidities, like COPD, chronic heart failure, diabetes mellitus, of which hyperoxia could exacerbate the outcome (25, 26). In the present study, percentage of COPD or heart failure was higher in obese than non-obese patients. Interestingly, in obese patients, survivors had a higher percentage of diabetes mellitus than non-survivors, which was not the case in the non-obese patients. Vught LA et al. also found diabetes was not associated with adjusted 90-day mortality risk in critically ill patients admitted with sepsis (27). Third, hyperoxia may result in nitric oxide depletion and induction of oxidative stress on the adipose tissue, which may exacerbate the critical illness (28).

The results of the current study suggest that compared with non-obese patients, excessive oxygen may be injurious for obese patients. However, targeted oxygenation was still unknown. BTS guideline recommends for morbidly obese patients (BMI > 40 kg/m₂), a target saturation of 88–92% should be titrated to maintain (grade D) (13). However, the literature was really sparse. In the present study, for obese patients with mechanical ventilation, Spo₂ of < =88% may be associated with poor outcome, 89–93% was not associated with mortality, 94–98% was associated with lower mortality, and 99–100% was associated with higher mortality. The results suggested Spo₂ 99–100% was not appropriate for obese patients. On the other hand, in obese patients, non-survivors had a higher percentage of TWM-Pao₂ of 120–300 mmHg and >300 mmHg, which also suggested hyperoxia may be detrimental for obese patients.

For overall and non-obese patients, PTS-Spo₂ of 89–93% was associated with increased hospital mortality in the present study. The contribution of modest hypoxemia to mortality is unknown. The recommendation of oxygen supply by guidelines mainly based on the normal ranges of Spo₂ (13, 16). A North American study found Spo₂ at sea level for adults with two standard deviation range is 94–98% (29). However, the standard reference ranges obtained from healthy volunteers may differ from the analogous range generated from data of ICU patients (30). Another larger observational study of over 37,000 patients admitted to four acute medical admissions units found that median Spo₂ was 98% (IQR 97–99%) for young adults and 96% (IQR 95–98%) for old adults (31). They also found for patients with initial Spo₂ values of 89, 90, 91, 92, and 93% had higher mortality and the 95% confidential intervals did not overlap with those initial Spo₂ values of ≥95% (31). However, clearly, there were limitations to the raw results without adjusting for confounders.

The present study has several other strengths. First, a total of 25,100 patients was extracted from ICUs in 166 hospitals, which allowed for the adjustment of multiple confounding factors, making the findings more generalizable. Second, the total observations of Spo₂ were 6,925,863, the average observations of Spo₂ were 276 for each patient, accounting for the time of 95.8% in the first 24 h, which made the results robust. Third, we calculated both the TWM-Spo₂ and PTS-Spo₂, which considered the dose of Spo₂.

However, several obvious limitations should be mentioned. First, the retrospective design in nature was subjected to the inherent limitations. This study design may only show statistical associations and but not causality between PTS-Spo₂ and hospital mortality. Second, although the multivariable logistic regression to was used to adjust for potential confounders, the potential confounding factors (e.g., altitude, smoking status et al.) which were not included in the analysis, could lead to biased results. Third, some data could be invalidated due to the acquisition directly from the bedside vital sign monitors. Some of the data had not been checked and verified by a bedside care provider, i.e., the measurements may be noisy and may not reflect the true status. However, the Spo₂ values recorded by the nurse and observed by the monitor at the same time were highly correlated (Supplementary Figure 4). Fourth, we only extracted the data for the first 24 h, which could not represent the entire course of stay in ICU. Fifth, we arbitrarily classified the Spo₂ into four categories, < =88%, 89–93, 94–98%, and 99–100%, the rationality may need further research. Sixth, although Spo₂ is widely used for clinical practice for continuously assessing the oxygenation in critically ill patients, blood gasses analyses are still the golden standard, especially in assessing the severity of hypoxemia and other gas exchange anomalies. Spo₂ may not display the real oxygenation status. Finally, we excluded patients without APACHE IV score, whose diagnoses were mainly burns and certain organ transplantation (17, 32), which may cause biases.

The impact of supplemental oxygen on hospital mortality was different in non-obese and obese critically ill patients with mechanical ventilation. Spo₂ of 99–100% was associated with higher mortality for obese patients, while Spo₂ of 89–93% was associated with higher mortality for non-obese patients.

Publicly available datasets were analyzed in this study. This data can be found here: eicu-crd.mit.edu.

The eICU was exempt from Institutional Review Board (IRB) approval due to the retrospective design, lack of direct patient intervention, and the security schema, for which the re-identification risk was certified as meeting safe harbor standards by Privacert (Cambridge, MA) (Health Insurance Portability and Accountability Act Certification no. 1031219-2). Due to the HIPAA compliant de-identification in this database, our institutional IRB requirement was waived.

DZ and TL conceived this study and reviewed and modified the final manuscript. DZ extracted the data. TL, DZ, DW, and CW designed and performed the statistical analyses. DZ, TL, QL, RZ, and DZ wrote the first draft of the manuscript. All authors read, critically reviewed, and approved the final manuscript.

This study was supported by Youth Top-notch Talent Program of Beijing Tongren Hospital.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.