This study is a post-hoc analysis of the trial Albumin Italian Outcome Sepsis (ALBIOS), prospectively registered on 30 June 2008 - NCT00707122.14 The ALBIOS trial was a multicenter, open-label, randomized, controlled trial (RCT) that enrolled 1810 patients with severe sepsis or septic shock. Adult patients who met the clinical criteria for severe sepsis or septic shock within the previous 24 h were screened for eligibility criteria and enrolled in the ALBIOS trial. Patients were randomly assigned to receive either 20% albumin and crystalloid solution or crystalloid solution alone for fluid replacement from randomization until day 28 or ICU discharge. All the other treatments were administered according to the standard guidelines of the treatment of septic patients. The study found that albumin replacement in addition to crystalloids, as compared to crystalloids alone, did not improve the survival at 28 or 90 days14,15

For this post-hoc analysis, patients who had died within 48 h after randomization were excluded as well as patients whose data was incomplete regarding partial pressure of arterial oxygen (PaO2) values at all timepoints. Recorded values of PaO2 expressed in millimeters of mercury (mmHg) were collected at different time points during the first 48 h. Patients were stratified into two groups according to their weighted average PaO2 levels, the cut-off value was established at 100 mmHg as in previous literature5,16: patients with average PaO2 in the first 48 h (PaO2 0–48 h)>100 mmHg were assigned to the hyperoxemia group. Conversely patients with PaO2 0–48h≤100 mmHg were included in the normoxemia group.

The primary outcome was 90-day mortality. Secondary outcomes were 28-day mortality, ICU mortality, duration of mechanical ventilation, length of stay in ICU, incidence of renal-replacement therapy and Acute Kidney Injury (AKI) defined according to the Kidney Disease: Improving Global Outcomes (KDIGO) criteria,17 days to suspension of vasopressor or inotropic agents, length of stay in ICU, and the percentage of resolution of primary and secondary infections.

Patients were dichotomized based on average PaO2≤100 mmHg or >100 mmHg per hour for the first 48 h after randomization. PaO2 was available at baseline, 6, 24 and 48 h, The area under the curve was calculated and divided by 48 to have the average PaO2 value per hour.

Continuous variables were expressed as means +/- Standard Deviation (SD) when normally distributed, as median and interquartile range [IQR] when non-normally distributed. Categorical data were reported as frequencies and proportions. The Pearson Chi2 test, one-way ANOVA test, or Kruskal-Wallis test were used to examine differences between the two groups.

The associations between the groups and mortality at 90 days were presented by Kaplan-Meier. The Log-Rank test was applied to compare the survival distribution of the groups. Cox regression analysis was performed to estimate the hazard-ratio associated with the risk of mortality, adjusted for the variables selected considering baseline unbalances that were statistically and clinically relevant and independent from each other: type of admission, mechanical ventilation, shock (according to Sepsis-2 definition18), chronic renal failure and Simplified Acute Physiology Score II (SAPS II).

Five subgroup analyses were performed. In the first analysis, patients with hypoxemia (PaO2≤60 mmHg) at baseline were excluded, as done in previous literature in order to avoid the possible confounding effect of extremely low values of PaO25. The second subgroup analysis only focused on septic patients admitted to ICU after emergency and elective surgery, to evaluate the possible benefit of hyperoxia in this specific cohort of patients. Two other sub-group analyses were performed: one excluding patients with lung as primary site of infection and another one only including patients with lung as primary site of infection, to observe how the infection site would influence survival. Lastly, we performed a subgroup analysis including patients who underwent mechanical ventilation.

The ALBIOS trial included 1810 patients diagnosed with severe sepsis or septic shock. After the exclusion of 174 patients who died in the first 48 h, and 4 patients due to incomplete PaO2 values, 1632 patients were included in this analysis (Supplementary Fig. 1). Information on the characteristics of excluded patients can be found in Supplementary Table 1. Baseline characteristics of the cohort included are shown in Table 1.

Patients were divided into two groups according to PaO2 values during the first 48 h since randomization. Six hundred and sixty-one patients were assigned to the normoxemia group with PaO2 0–48h≤100 mmHg, while 971 patients were allocated to the hyperoxemia group with PaO2 0–48h>100 mmHg. No significant differences were observed in the distribution of patients in the two study groups in relation to the study treatment (20% albumin and crystalloid solution or crystalloid solution alone for fluid replacement). The proportion of patients mechanically ventilated was 10% greater in the hyperoxemia group. Abdominal and urinary tract infections were more often the primary site of infection in patients with hyperoxemia. Conversely, lung was more often the primary site of infection in patients with normoxemia. The proportion of patients with septic shock was greater in the hyperoxemia group, see Table 1.

At baseline (first time point, at randomization) PaO2 values were similar between the two groups and differed significantly at 6 and 24 h from randomization (Fig. 1).

Fig. 1.

Box-plot by patient group with distribution of PaO2 values for the first 48 h.

The figure assesses the distribution of PaO2 for different timepoints during the first 48 h for patients divided in two groups according to PaO2 levels: in red patients with normoxiemia (PaO2 0–48≤100 mmHg) and in blue patients with hyperoxiemia (PaO2 0–48>100 mmHg). The horizontal line in the boxes indicates the median, the top and bottom of the box the interquartile range, and I bars the 5th and 95th percentile range.

(0.08MB).

Concerning the primary outcome, 344 (35.4%) patients in the hyperoxemia group vs. 236 (35.7%) in the normoxemia group died (p = 0.909, Table 2) within 90 days from randomization. Fig. 2 shows the Kaplan-Meier curves for survival up to day 90 according to study treatment that do not differ significantly (p = 0.81). The restricted cubic splines in Fig. 3 show the trend of the hazard ratio of 90-day mortality by arterial oxygen levels. After adjusting for type of admission, mechanical ventilation, SAPS II, presence of shock and chronic renal failure, hyperoxemia was not significantly associated with 90-day mortality (HR 0.87 (CI [95%] 0.736–1.028, p = 0.102)) (Table 3). Using log-transformed PaO2 as continuous variable in the model did not change the results (Supplementary Table 2).

Fig. 2.

90-day survival curve and life table of patients.

In red patients with normoxiemia (PaO2 0–48≤100 mmHg) and in blue patients with hyperoxiemia (PaO2 0–48>100 mmHg). The colored area indicates the Confidence Interval (CI) while lines indicate the Hazard Ratio (HR).

(0.08MB).

Fig. 3.

Restricted cubic splines for 90-day mortality by arterial oxygen levels.

In blue the cubic splines curve; in red the reference: hazard ratio=1.

(0.07MB).

Mortality at 28 days was similar in the two groups. AKI, proportion of patients who underwent renal-replacement therapy, timing of suspension of vasopressor or inotropic agents, resolution of primary infection and ICU mortality were not significantly different between the study groups. Patients with normoxemia underwent mechanical ventilation for a shorter period of time and had a shorter length of stay in ICU, compared to the hyperoxemia group (Table 2).

After excluding 112 patients with a PaO2≤60 mmHg at baseline, the adjusted association between log-transformed continuous PaO2 and 90-day mortality was not significant (Supplementary Table 3).

In the cohort of septic patients admitted after emergency or elective surgery (n = 718), hyperoxemia was not associated with lower or higher 90-day mortality (Supplementary Table 4A). Secondary outcomes did not differ between hyperoxemia and normoxemia groups (Supplementary Table 5).

In the cohort of patients without respiratory infections as primary source (n = 969), hyperoxemia was not significantly associated with 90-day mortality both in the unadjusted and adjusted analysis (Supplementary Table 6A). No significant adjusted association was found between log-transformed continuous PaO2 and 90-day mortality in this group (Supplementary Table 6B). Conversely, the subgroup analysis including patients who had the lung as primary site of infection (n = 663) showed a reduced risk of 90-day mortality in patients with hyperoxemia both in the unadjusted and adjusted analysis (Supplementary Table 7A). This finding was confirmed by the adjusted analysis of the log-transformed continuous PaO2 (Supplementary Table 7B).

Lastly, in the cohort of patients mechanically ventilated (n = 1290) hyperoxemia was not significantly associated with 90-day mortality (Supplementary Table 8A and 8B).

Strengths of the analysis include the high quality of the trial design and of the generated dataset. Moreover, to the best of our knowledge this post-hoc analysis includes the largest cohort of septic patients available. Unlike other studies, we decided to stratify the patients based upon PaO2 values rather than SpO2. This allowed us to avoid the risk of overestimating the effect of eventual brief episodes of rapid and marked desaturation. Lastly, patients’ stratification based on weighted average PaO2 during the first 48 h avoids the possible mis-categorization of patients that could result from the evaluation of a single value. Indeed, as shown in Fig. 1, PaO2 values were similar between the two groups at baseline and differed significantly at 6 and 24 h.

Our study has limitations, mostly inherent to the nature of the study, which is a post-hoc analysis of a previous trial that was not designed to assess the effect of different levels of oxygenation on patient outcomes. So, due to the study design, residual confounding is likely to remain in the adjusted analysis for the primary outcome. Patients’ stratification was based on weighted average values of PaO2 during the first 48 h using the four timepoints collected in the original dataset: weighting on a more granular assessment would improve the categorization in the study groups. Moreover, we could not calculate the possible cumulative effect of exposure to oxygen supplementation over time (after 48 h), as was done in other studies on mechanically ventilated critically ill patients.19 Finally, co-interventions used at the time of patients’ enrollment during the ALBIOS trial may differ from current clinical practice.