Comments
Abstract
Rationale: Hypercapnia may affect the outcome of sepsis. Very few clinical studies conducted in noncritically ill patients have investigated the effects of hypercapnia and hypercapnic acidemia in the context of sepsis. The effect of hypercapnia in critically ill patients with sepsis remains inadequately studied.
Objectives: To investigate the association of hypercapnia with hospital mortality in critically ill patients with sepsis.
Methods: This is a retrospective study conducted in three tertiary public hospitals. Critically ill patients with sepsis from three intensive care units between January 2011 and May 2019 were included. Five cohorts (exposure of at least 24, 48, 72, 120, and 168 hours) were created to account for immortal time bias and informative censoring. The association between hypercapnia exposure and hospital mortality was assessed with multivariable models. Subgroup analyses compared ventilated versus nonventilated and pulmonary versus nonpulmonary sepsis patients.
Results: We analyzed 84,819 arterial carbon dioxide pressure measurements in 3,153 patients (57.6% male; median age was 62.5 years). After adjustment for key confounders, both in mechanically ventilated and nonventilated patients and in patients with pulmonary or nonpulmonary sepsis, there was no independent association of hypercapnia with hospital mortality. In contrast, in ventilated patients, the presence of prolonged exposure to both hypercapnia and acidemia was associated with increased mortality (highest odds ratio of 16.5 for ⩾120 hours of potential exposure; P = 0.007).
Conclusions: After adjustment, isolated hypercapnia was not associated with increased mortality in patients with sepsis, whereas prolonged hypercapnic acidemia was associated with increased risk of mortality. These hypothesis-generating observations suggest that as hypercapnia is not an independent risk factor for mortality, trials of permissive hypercapnia avoiding or minimizing acidemia in sepsis may be safe.
The mortality and morbidity of patients with sepsis admitted to intensive care units (ICUs) remains high. Hypercapnia may influence the pathophysiology of sepsis (1, 2) with effects on immunity (3), cardiac output, and tissue oxygen delivery (4, 5). The impact of hypercapnia and hypercapnic acidemia have been mostly studied in animal experiments (6–9). Very few clinical studies, of which most were conducted in noncritically ill patients, have investigated the effects of hypercapnia and hypercapnic acidosis in the context of sepsis (10, 11).
In animal models of sepsis, studies of hypercapnia have reported contradictory effects. When hypercapnia was investigated in conjunction with acidemia (hypercapnic acidemia), it was protective from sepsis-induced lung injury (6, 7). However, buffering of hypercapnic acidemia (buffered hypercapnia) worsened lung injury (8).
In noncritically ill patients (10–13), hypercapnia has been associated with increased mortality. However, in a small observational study of mechanically ventilated patients with community-acquired pneumonia, hypercapnia was associated with better survival (14). Thus, the net effect of hypercapnia in critically ill patients with sepsis remains inadequately studied or understood (14–19). Accordingly, we aimed to investigate the hypothesis that there would be a consistent independent association of hypercapnia and hypercapnic acidemia with hospital mortality in critically ill patients with sepsis.
The study was approved by all the three human research ethics committees (HRECs) of the participating hospitals (Peninsula Health HREC approval number: QA/53646/PH-2019; Austin HREC approval number: Audit 19/Austin/56; Alfred HREC approval number: 292/19).
This was a retrospective study with data collected from three tertiary public hospitals in Victoria, Australia.
All adult patients (⩾18 years) admitted to the ICU with a diagnosis of sepsis in the participating hospitals from January 2011 to May 2019 were included. The final dataset considered in the analysis included patients staying in the ICU for at least 24 hours. Patients who were discharged from the ICU or died before 24 hours were excluded from analysis.
Baseline demographic, severity of illness, diagnostic, and outcome data were extracted from local ICU databases (for submission to the Australia and New Zealand Intensive Care Society Adult Patient Database) (20). We obtained all arterial blood gases available from ICU admission until ICU discharge or death from the electronic health records of the study hospitals.
The primary outcome was all-cause in-hospital mortality.
The overall cohort of the study was adjusted to create cohorts with longer ICU length of stay and consequently longer potential windows of exposure. This approach was chosen because duration of potential exposure might provide information on a possible dose effect and increase the ability to assess the effect of dose on outcome. However, the opportunity to be exposed to hypercapnia while in an ICU is time dependent and is affected by time to ICU discharge and time to death. As longer time in an ICU indicates greater illness severity and thus greater mortality, such immortal time bias represents a major confounder. Thus, we reasoned that creating time of exposure cohorts (see below) and assessing outcomes in such cohorts would be a logical and reasonable approach to achieving an immortal time bias independent assessment.
Thus, to explore the duration of the potential exposure period for arterial carbon dioxide pressure (PaCO2), we considered five exposure windows:
| • | At least 24 hours’ exposure (Cohort 0–1): patients discharged or dead before 24 hours were excluded (the start cohort); | ||||
| • | At least 48 hours’ exposure (Cohort 0–2): patients discharged or dead before 48 hours were excluded; | ||||
| • | At least 72 hours’ exposure (Cohort 0–3): patients discharged or dead before 72 hours were excluded; | ||||
| • | At least 120 hours’ exposure (Cohort 0–5): patients discharged or dead before 120 hours were excluded; and | ||||
| • | At least 168 hours’ exposure (Cohort 0–7): patients discharged or dead before 168 hours were excluded. | ||||
Hypercapnia was defined as a time-weighted average PaCO2 (defined as the area under the PaCO2 time curve) equal to or higher than 45 mm Hg. Time-weighted average PaCO2 for each cohort was derived from PaCO2 measurements obtained only during the period for a given cohort.
Time-weighted average PaCO2 was further assessed according to three categories: 1) 35–45 mm Hg (normal reference range); 2) 45–55 mm Hg (mild hypercapnia); and 3) more than 55 mm Hg (severe hypercapnia). We chose these thresholds because there is no consistent grading of hypercapnia and these cut-points had already been assessed in other studies (21–24).
To understand the combined effect of pH and PaCO2, we further classified patients into three groups based on a combination of time-weighted average PaCO2 and time-weighted average pH. The groups were 1) normocapnia (PaCO235–45 mm Hg) and normal pH (7.35–7.45); 2) compensated hypercapnia (normal pH [7.35–7.45] with elevated PaCO2 [PaCO2 > 45 mm Hg]); and 3) hypercapnic acidemia (PaCO2 > 45 mm Hg and pH < 7.35) (25). Patients who had hypocapnia or alkalemia were excluded from this analysis.
We reported all continuous data as medians (quartile 25%–quartile 75%) and categorical data as numbers and percentages.
In all analyses, five models were fitted, with one for each cohort. We applied mixed-effect logistic regression models, with center as random effect and year of ICU admission (as a categorical variable) as fixed effect. The covariates used to further adjust the model were selected based on clinical relevance and rationale. They included 1) acute physiology and chronic health evaluation III score; 2) age; 3) sex; 4) respiratory diagnosis at ICU admission; 5) presence of chronic respiratory disease; 6) use of mechanical ventilation; 7) time-weighted average lactate levels; 8) highest serum creatinine during ICU stay; 9) coefficient of variation of PaCO2 (defined as the standard deviation divided by the mean of PaCO2 over the specific time period); 10) amplitude of variation in PaCO2 (defined as the maximum minus the minimum PaCO2 over the specific time period); and 11) lowest pH during ICU admission.
We modeled age nonlinearly with restricted cubic splines according to evidence obtained from the data. We included time-weighted average lactate levels after log transformation. We entered all continuous variables after standardization to improve convergence of the models. The odds ratios represent the increase in one standard deviation of the variable.
The analyses were further conducted independently in two different groups, according to the use or not of mechanical ventilation (owing to hypothesis of different impact of PaCO2 according to the use of mechanical ventilation) and the presence or not of pulmonary sepsis.
We conducted all analyses in R v.3.6.3 (R Foundation) (26). To increase robustness, we used a two-sided P value of less than 0.01 to indicate statistical significance.
We identified 3,687 patients with sepsis and with 84,819 PaCO2 measurements. After exclusions, we included 3,153 patients in the main cohort of patients with at least 24 hours of ICU length of stay (Figure E1 in the online supplement). As described in the Methods, we then divided this overall cohort into five cohorts based on the duration of exposure (Table 1). Most patients were male (57.6%), with a median age of 62.5 (51.8–75.3) years, and 30.2% received mechanical ventilation. The most common comorbidity was immunosuppression, and the main diagnosis was sepsis from nonurinary sources. The median ICU and hospital length of stay values were 2.8 (1.6–5.9) days and 12.9 (6.7–25.9) days, respectively, and hospital mortality was 17.5%. Details of the number and frequency of PaCO2 measurements are shown in Table E1.
| Cohort 0–1 (N = 3,153) | Cohort 0–2 (N = 2,418) | Cohort 0–3 (N = 1,751) | Cohort 0–5 (N = 1,017) | Cohort 0–7 (N = 732) | |
|---|---|---|---|---|---|
| Age, years | 65.2 (51.8–75.3) | 65.1 (51.6–75.1) | 63.9 (50.8–74.2) | 62.1 (49.1–71.8) | 61.2 (48.2–70.4) |
| Sex, male, n (%) | 1,814 (57.6) | 1,418 (58.6) | 1,041 (59.5) | 624 (61.4) | 456 (62.3) |
| Body mass index, k/m2 | 25.9 (22.5–30.2) | 26.0 (22.6–30.4) | 26.0 (22.6–30.6) | 26.0 (22.7–31.0) | 26.3 (22.9–31.3) |
| APACHE III score, Day 1 | 65.0 (50.0–82.0) | 67.0 (53.0–84.0) | 71.0 (56.2–87.0) | 76.0 (60.0–92.0) | 77.0 (62.0–93.2) |
| Diagnosis, n (%) | |||||
| Sepsis - nonurinary source | 1,193 (37.8) | 901 (37.3) | 628 (35.9) | 343 (33.7) | 239 (32.7) |
| Septic shock - nonurinary source | 762 (24.2) | 604 (25.0) | 483 (27.6) | 309 (30.4) | 234 (32.0) |
| Bacterial pneumonia | 714 (22.6) | 569 (23.5) | 429 (24.5) | 261 (25.7) | 186 (25.4) |
| Urinary sepsis | 241 (7.6) | 151 (6.2) | 76 (4.3) | 25 (2.5) | 14 (1.9) |
| Urinary septic shock | 115 (3.6) | 86 (3.6) | 51 (2.9) | 26 (2.6) | 17 (2.3) |
| Viral pneumonia | 77 (2.4) | 66 (2.7) | 55 (3.1) | 38 (3.7) | 32 (4.4) |
| Gastrointestinal inflammatory | 36 (1.1) | 26 (1.1) | 15 (0.9) | 7 (0.7) | 3 (0.4) |
| Parasitic pneumonia | 15 (0.5) | 15 (0.6) | 14 (0.8) | 8 (0.8) | 7 (1.0) |
| Coexisting disorders, n (%) | |||||
| Diabetes | 374 (14.1) | 295 (14.3) | 219 (14.5) | 116 (13.4) | 77 (12.4) |
| Chronic respiratory disease | 205 (6.5) | 162 (6.7) | 118 (6.7) | 63 (6.2) | 47 (6.4) |
| Liver cirrhosis | 130 (4.3) | 110 (4.7) | 81 (4.8) | 49 (4.9) | 33 (4.6) |
| Immune disease | 262 (8.3) | 206 (8.5) | 160 (9.1) | 96 (9.4) | 72 (9.8) |
| Immunosuppression | 535 (17.0) | 417 (17.2) | 320 (18.3) | 199 (19.6) | 144 (19.7) |
| Hepatic failure | 27 (0.9) | 23 (1.0) | 19 (1.1) | 14 (1.4) | 12 (1.6) |
| Use of mechanical ventilation | 952 (30.2) | 783 (32.4) | 640 (36.6) | 443 (43.6) | 345 (47.1) |
| Vital signs during first 24 h of ICU admission | |||||
| Highest temperature, °C | 37.3 (36.7–38.1) | 37.4 (36.8–38.2) | 37.4 (36.8–38.2) | 37.5 (36.9–38.4) | 37.5 (37.0–38.4) |
| Highest heart rate, bpm | 110.0 (95.0–127.0) | 110.0 (95.0–130.0) | 110.0 (96.0–130.0) | 115.0 (100.0–132.5) | 115.0 (100.0–135.0) |
| Lowest MAP, mm Hg | 64.0 (58.0–70.0) | 63.0 (58.0–69.0) | 63.0 (58.0–69.0) | 63.0 (58.0–69.0) | 63.0 (57.0–69.0) |
| Highest respiratory rate, mpm | 26.0 (22.0–31.0) | 26.0 (22.0–32.0) | 26.0 (21.0–32.0) | 26.0 (20.0–32.0) | 26.0 (20.0–32.0) |
| Laboratory tests during first 24 h of ICU admission | |||||
| Highest creatinine, μmol/L | 114.0 (73.0–195.0) | 119.0 (74.2–202.0) | 127.5 (76.0–216.0) | 132.0 (80.0–223.0) | 138.5 (81.2–245.5) |
| Lowest platelets, cells/mm3 | 167.0 (105.0–237.0) | 168.0 (101.0–240.0) | 168.0 (96.5–243.0) | 168.0 (90.0–245.0) | 166.0 (84.0–243.0) |
| Bilirubin, mmol/L | 13.0 (8.0–24.0) | 14.0 (8.0–25.0) | 14.0 (9.0–26.0) | 14.0 (9.0–29.0) | 15.0 (9.0–30.0) |
| Clinical outcomes | |||||
| ICU length of stay, days | 2.8 (1.6–5.9) | 3.8 (2.4–7.6) | 5.2 (3.5–10.3) | 9.0 (6.1–14.5) | 11.7 (8.4–17.4) |
| Hospital length of stay, days | 12.9 (6.7–25.9) | 15.1 (8.4–29.0) | 18.2 (10.0–32.6) | 22.4 (14.3–39.1) | 25.6 (16.2–41.8) |
| Hospital mortality, n (%) | 553 (17.5) | 428 (17.7) | 358 (20.4) | 248 (24.4) | 198 (27.0) |
Hypercapnia and severe hypercapnia were more common in ventilated patients and those with pulmonary sepsis (Figure E2). The time-weighted average PaCO2 and all elements of CO2 exposure for different time-based cohorts and PaCO2 measurements strata are presented in Table E2. The time-weighted PaCO2 value increased with time of exposure to ICU from 38.3 (33.3–45.0) mm Hg in cohort 0–1 to 43.4 (38.6–54.8) mm Hg in cohort 0–7. However, the change in time-weighted average PaCO2. or pH measurements did not show a significant difference between survivors and nonsurvivors (Figure E3).
Overall, after adjusting for key confounders, including baseline severity of illness, hypercapnia was not associated with increased hospital mortality in any time-based cohorts (Table 2). However, time-weighted blood lactate levels, acute physiology and chronic health evaluations III score, and the coefficient of variation for PaCO2 showed a consistent association with increased odds ratios for mortality (Table 2).
| Cohort 0–1 (N = 3,153) Number of Deaths = 553 (17.5) | Cohort 0–2 (N = 2,418) Number of Deaths = 428 (17.7) | Cohort 0–3 (N = 1,751) Number of Deaths = 358 (20.4) | Cohort 0–5 (N = 1,017) Number of Deaths = 248 (24.4) | Cohort 0–7 (N = 732) Number of Deaths = 198 (27.0) | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Odds Ratio (95% CI) | P Value | Odds Ratio (95% CI) | P Value | Odds Ratio (95% CI) | P Value | Odds Ratio (95% CI) | P Value | Odds Ratio (95% CI) | P Value | |
| Hypercapnia (>45 mm Hg) | 0.97 (0.65–1.45) | 0.890 | 0.95 (0.63–1.43) | 0.806 | 0.91 (0.59–1.38) | 0.643 | 0.92 (0.53–1.59) | 0.764 | 0.98 (0.50–1.94) | 0.965 |
| APACHE III | 2.82 (2.30–3.47) | <0.001 | 2.11 (1.73–2.58) | <0.001 | 1.76 (1.44–2.16) | <0.001 | 1.49 (1.18–1.89) | 0.001 | 1.31 (1.00–1.71) | 0.047 |
| Age* | 1.18 (0.58–2.40) | 0.655 | 1.39 (0.67–2.90) | 0.375 | 1.48 (0.68–3.22) | 0.323 | 1.45 (0.61–3.43) | 0.396 | 3.25 (1.07–9.86) | 0.037 |
| Sex, male | 1.38 (0.99–1.91) | 0.054 | 1.24 (0.88–1.74) | 0.217 | 1.16 (0.82–1.66) | 0.404 | 1.27 (0.81–1.98) | 0.295 | 1.11 (0.67–1.83) | 0.693 |
| Respiratory disease | 0.70 (0.45–1.09) | 0.113 | 0.59 (0.37–0.93) | 0.023 | 0.57 (0.35–0.92) | 0.021 | 0.53 (0.30–0.95) | 0.032 | 0.52 (0.27–0.99) | 0.046 |
| Chronic respiratory disease | 2.08 (1.27–3.42) | 0.004 | 1.95 (1.14–3.35) | 0.016 | 1.87 (1.03–3.40) | 0.039 | 1.95 (0.91–4.16) | 0.086 | 1.44 (0.60–3.49) | 0.417 |
| Mechanical Ventilation | 1.23 (0.80–1.90) | 0.343 | 1.25 (0.80–1.94) | 0.331 | 1.02 (0.64–1.63) | 0.925 | 0.96 (0.55–1.69) | 0.883 | 0.77 (0.40–1.47) | 0.427 |
| TW average lactate, mmol/L | 1.88 (1.42–2.51) | <0.001 | 2.37 (1.66–3.39) | <0.001 | 2.57 (1.73–3.83) | <0.001 | 3.02 (1.74–5.24) | <0.001 | 4.14 (2.04–8.42) | <0.001 |
| Base excess, mEq/L | 1.23 (1.01–1.48) | 0.036 | 1.42 (1.15–1.75) | 0.001 | 1.41 (1.13–1.76) | 0.003 | 1.25 (0.94–1.65) | 0.125 | 1.18 (0.85–1.63) | 0.315 |
| Creatinine, μmol/L | 0.87 (0.73–1.03) | 0.106 | 0.96 (0.80–1.15) | 0.644 | 0.98 (0.81–1.19) | 0.835 | 0.97 (0.76–1.24) | 0.828 | 1.00 (0.75–1.33) | 0.993 |
| Amplitude of PaCO2, mm Hg | 1.06 (0.81–1.38) | 0.671 | 0.81 (0.60–1.10) | 0.176 | 0.93 (0.69–1.26) | 0.653 | 0.77 (0.52–1.15) | 0.205 | 0.83 (0.52–1.32) | 0.427 |
| CV of PaCO2, % | 0.99 (0.75–1.31) | 0.950 | 1.63 (1.20–2.21) | 0.002 | 1.53 (1.12–2.10) | 0.008 | 1.77 (1.18–2.68) | 0.006 | 1.86 (1.14–3.02) | 0.013 |
| FIO2 | 1.28 (1.07–1.53) | 0.007 | 1.27 (1.05–1.55) | 0.015 | 1.16 (0.94–1.44) | 0.163 | 1.07 (0.83–1.37) | 0.608 | 1.09 (0.82–1.44) | 0.552 |
Subgroup analyses comparing ventilated and nonventilated patients in time-based cohorts and according to strata of PaCO2 values are shown in Table 3. In ventilated as well as nonventilated patients, hypercapnia was not significantly associated with increased hospital mortality. However, hypocapnia was significantly associated with an increased odds ratio for mortality (P < 0.001) in ventilated patients with 120 hours of exposure (Table 3). The relationship between PaCO2 and mortality in the largest time cohort is presented in Figure 1, which shows the highest risk of mortality with hypocapnia, the lowest risk in the 35–50 mm Hg range, and an increased risk for up to 70 mm Hg, beyond which there was no further increase in the risk of mortality. Similar patterns were present in the other time-based cohorts (Figures E4–E7). In the subgroups of patients who had pulmonary sepsis or nonpulmonary sepsis, hypercapnia was also not associated with increased risk of hospital mortality (Table E3).
| Exposure Time | Cutoff | No Mechanical Ventilation | Mechanical Ventilation | P Value for Interaction | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Deaths n (%) | Patients N | Odds Ratio (95% CI) | P Value | Deaths n (%) | Patients N | Odds Ratio (95% CI) | P Value | |||
| At least 24 h exposure | <30 | 63 (22.3) | 282 | 2.11 (1.02–4.35) | 0.043 | 34 (34.0) | 100 | 1.83 (0.88–3.81) | 0.108 | 0.769 |
| 30–34 | 68 (15.1) | 451 | 1.46 (0.78–2.72) | 0.237 | 37 (21.1) | 175 | 0.89 (0.48–1.66) | 0.712 | ||
| 35–45 | 91 (12.6) | 722 | 1 (Reference) | — | 85 (21.3) | 399 | 1 (Reference) | — | ||
| 45–55 | 38 (16.7) | 227 | 1.20 (0.54–2.66) | 0.659 | 37 (30.1) | 123 | 0.96 (0.48–1.94) | 0.919 | ||
| >55 | 45 (18.5) | 243 | 1.62 (0.69–3.84) | 0.271 | 20 (26.0) | 77 | 0.93 (0.38–2.29) | 0.880 | ||
| At least 48 h exposure | <30 | 35 (17.8) | 196 | 1.07 (0.47–2.46) | 0.869 | 18 (27.7) | 65 | 1.90 (0.81–4.43) | 0.139 | 0.365 |
| 30–34 | 46 (14.5) | 317 | 1.55 (0.79–3.03) | 0.198 | 32 (22.2) | 144 | 0.98 (0.52–1.87) | 0.963 | ||
| 35–45 | 81 (15.7) | 515 | 1 (Reference) | — | 76 (21.1) | 359 | 1 (Reference) | — | ||
| 45–55 | 36 (17.4) | 207 | 0.98 (0.45–2.16) | 0.969 | 35 (29.2) | 120 | 1.79 (0.92–3.49) | 0.088 | ||
| >55 | 42 (16.7) | 252 | 0.96 (0.41–2.27) | 0.928 | 13 (19.7) | 66 | 1.36 (0.56–3.30) | 0.496 | ||
| At least 72 h exposure | <30 | 20 (16.8) | 119 | 1.12 (0.46–2.73) | 0.795 | 11 (27.5) | 40 | 1.75 (0.64–4.74) | 0.273 | 0.387 |
| 30–34 | 35 (21.6) | 162 | 1.14 (0.53–2.45) | 0.743 | 20 (21.0) | 95 | 1.03 (0.50–2.13) | 0.928 | ||
| 35–45 | 72 (18.1) | 398 | 1 (Reference) | — | 74 (23.2) | 319 | 1 (Reference) | — | ||
| 45–55 | 37 (20.5) | 180 | 1.14 (0.54–2.43) | 0.728 | 36 (31.8) | 113 | 1.59 (0.81–3.13) | 0.176 | ||
| >55 | 40 (23.0) | 174 | 1.61 (0.69–3.75) | 0.271 | 5 (9.1) | 55 | 0.55 (0.17–1.76) | 0.314 | ||
| At least 120 h exposure | <30 | 11 (27.5) | 40 | 1.18 (0.34–4.05) | 0.795 | 11 (61.1) | 18 | 11.54 (2.97–44.760 | < 0.001 | 0.168 |
| 30–34 | 15 (18.7) | 80 | 0.41 (0.13–1.260 | 0.118 | 10 (17.5) | 57 | 1.00 (0.38–2.59) | 0.998 | ||
| 35–45 | 62 (25.4) | 244 | 1 (Reference) | — | 55 (24.5) | 224 | 1 (Reference) | — | ||
| 45–55 | 31 (28.4) | 109 | 0.34 (0.55–3.26) | 0.525 | 29 (31.5) | 92 | 2.61 (1.16–5.86) | 0.020 | ||
| >55 | 14 (22.9) | 61 | 0.59 (0.15–2.27) | 0.440 | 7 (15.5) | 45 | 1.23 (0.37–4.14) | 0.739 | ||
| At least 168 h exposure | <30 | 1 (16.7) | 6 | 1.19 (0.07–20.97) | 0.907 | 4 (44.4) | 9 | 3.39 (0.60–19.23) | 0.168 | 0.823 |
| 30–34 | 10 (30.3) | 33 | 1.03 (0.31–3.37) | 0.962 | 12 (28.6) | 42 | 1.42 (0.55–3.71) | 0.471 | ||
| 35–45 | 36 (28.8) | 125 | 1 (Reference) | — | 53 (28.8) | 184 | 1 (Reference) | — | ||
| 45–55 | 17 (27.9) | 61 | 1.24 (0.42–3.65) | 0.690 | 20 (27.8) | 72 | 2.60 (1.02–6.65) | 0.046 | ||
| >55 | 37 (28.0) | 132 | 0.57 (0.12–2.78) | 0.490 | 7 (18.9) | 37 | 0.82 (0.19–3.50) | 0.786 | ||
Figure 1. Local regression (locally estimated scatterplot smoothing) curve assessing the impact of time-weighted (TW) average arterial carbon dioxide pressure (PaCO2) and hospital mortality in the time-based cohort for the first 24 hours. Dots represent the actual hospital mortality for each PaCO2 level, and the line is the locally estimated scatterplot smoothing curve with 95% confidence interval. The size of each dot or bubble reflects the number of patients in a given 2 mm Hg PaCO2 range. The smoothing curves used a bandwidth of 0.75, a polynomial regression with 1 degree of freedom, and a tricubic weight function.
[More] [Minimize]As sensitivity analysis, we also assessed the independent association of hypercapnia (PaCO2 > 45 mm Hg) and severe hypercapnia (PaCO2 > 55 mm Hg) in the subgroups of patients according to the use of mechanical ventilation or the presence of pulmonary sepsis. These analyses did not show an independent association of hypercapnia or severe hypercapnia with hospital mortality in patients with or without mechanical ventilation or pulmonary sepsis (Tables E4 and E5). In addition, there was no change in PaCO2 and mortality over the 8 years of observation (Figure 2).
Figure 2. Time-weighted average arterial carbon dioxide pressure (PaCO2) and hospital mortality over time. Circles are mean values, and error bars are 95% confidence interval.
[More] [Minimize]The adjusted associations of hypercapnia with hospital mortality according to the presence or absence of acidemia are shown in Table 4. In mechanically ventilated patients, the presence of prolonged exposure to both hypercapnia and acidemia was associated with increased mortality (highest odds ratio of 16.52 for ⩾120 hours of potential exposure; P = 0.007). In patients with nonpulmonary sepsis, hypercapnic acidemia at 72 hours of exposure to ICU treatment was associated with increased mortality (Table E6).
| Exposure time | No Mechanical Ventilation | Mechanical Ventilation | P Value for Interaction | |||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Group | Deaths n (%) | Patients N | Odds Ratio (95% CI) | P Value | Deaths n (%) | Patients N | Odds Ratio (95% CI) | P Value | ||
| At least 24 h exposure | Normocapnia and normal pH | 15 (6.9) | 218 | 1 (Reference) | — | 19 (23.2) | 82 | 1 (Reference) | — | 0.692 |
| Compensated hypercapnia | 23 (17.2) | 134 | 0.15 (0.01–3.17) | 0.221 | 14 (28.6) | 49 | 1.09 (0.25–4.74) | 0.908 | ||
| Hypercapnic acidemia | 22 (36.7) | 60 | 2.76 (0.22–35.12) | 0.433 | 13 (61.9) | 21 | 4.05 (0.47–35.05) | 0.204 | ||
| At least 48 h exposure | Normocapnia and normal pH | 36 (12.7) | 283 | 1 (Reference) | — | 28 (19.3) | 145 | 1 (Reference) | — | 0.330 |
| Compensated hypercapnia | 21 (10.8) | 195 | 0.65 (0.14–3.07) | 0.583 | 22 (28.9) | 76 | 3.19 (0.94–10.75) | 0.062 | ||
| Hypercapnic acidemia | 22 (33.3) | 66 | 1.98 (0.46–8.44) | 0.356 | 12 (46.1) | 26 | 6.22 (1.59–24.34) | 0.009 | ||
| At least 72 h exposure | Normocapnia and normal pH | 42 (16.9) | 249 | 1 (Reference) | — | 34 (21.0) | 162 | 1 (Reference) | — | 0.081 |
| Compensated hypercapnia | 26 (15.3) | 170 | 0.48 (0.13–1.70) | 0.254 | 23 (27.7) | 83 | 2.35 (0.67–8.230 | 0.181 | ||
| Hypercapnic acidemia | 30 (46.1) | 65 | 1.34 (0.31–5.91) | 0.695 | 11 (47.8) | 23 | 9.73 (2.16–43.79) | 0.003 | ||
| At least 120 h exposure | Normocapnia and normal pH | 37 (22.4) | 165 | 1 (Reference) | 35 (24.5) | 143 | 1 (Reference) | — | 0.030 | |
| Compensated hypercapnia | 18 (20.9) | 86 | 0.88 (0.16–4.77) | 0.887 | 15 (20.5) | 73 | 1.89 (0.36–9.89) | 0.451 | ||
| Hypercapnic acidemia | 12 (40.0) | 30 | 0.05 (0.00–1.60) | 0.091 | 9 (69.2) | 13 | 16.52 (2.13–128.20) | 0.007 | ||
| At least 168 h exposure | Normocapnia and normal pH | 21 (23.3) | 90 | 1 (Reference) | — | 33 (28.0) | 118 | 1 (Reference) | — | 0.796 |
| Compensated hypercapnia | 31 (24.6) | 126 | 3.83 (0.11–136.20) | 0.461 | 19 (30.1) | 63 | 3.77 (0.94–15.04) | 0.060 | ||
| Hypercapnic acidemia | 12 (44.4) | 27 | 1.67 (0.04–62.83) | 0.782 | 4 (50.0) | 8 | 4.53 (0.61–33.58) | 0.139 | ||
The ventilator management data was available for a cohort of patients receiving ventilation for more than 7 days in the study units (Table E7). It shows the common use of pressure support ventilation, a median tidal volume of 7.5 (7.1–8.5) mL/kg predicted body weight (PBW), respiratory rate of 18 (15–20) breaths/minute and positive end expiratory pressure (PEEP) of 7 (5–9) cm H2O (Table E7).
In a detailed multicenter study involving more than 3,000 critically ill patients with sepsis and more than 84,000 PaCO2 measurements, we explored the association between hypercapnia and hospital mortality. We found that exposure to hypercapnia was common and occurred in most mechanically ventilated patients with sepsis. Overall, however, after adjustment for key confounders, isolated hypercapnia was not independently associated with hospital mortality in the time-exposure cohorts we studied. In contrast, the coefficient of variation for PaCO2 appeared to have a significant relationship with increased risk of death.
Hypercapnia has physiological effects on all organ systems (19), including the immune system (27). Thus, it may affect the outcome of sepsis. Data on hypercapnia and hypercapnic acidemia in sepsis, however, are limited to animal experiments and physiological outcomes (6, 8, 9, 28, 29) studied with short duration of hypercapnia exposure. Moreover, such studies have reported contradictory findings (6, 9, 29). These factors limit the applicability of such preclinical findings to clinical practice.
Clinical data on hypercapnia in critically ill patients are variable, with some studies suggesting hypercapnia to be safe or even beneficial (30, 31) and other studies reporting harm (11, 23–25, 32). In patients with acute respiratory distress syndrome, hypercapnia has been associated with increased risk of hospital mortality in two studies (23, 25), whereas permissive hypercapnia has been reported to achieve outstanding outcomes in such patients (30). Additionally, in patients with acute cerebral injury, such as following cardiac arrest, hypercapnia on the day of admission was associated with better outcomes (22, 24, 33). In patients with sepsis, the data are limited to noncritically ill patients (11, 13, 14, 34). Moreover, all studies are small, limited to community-acquired pneumonia, and have only used a single measurement (on admission to hospital) of PaCO2 from arterial or venous blood gases (13, 14) to define hypercapnia. Thus, to the best of our knowledge, no study has ever comprehensively assessed the association of hypercapnia in critically ill patients with sepsis.
Hypercapnia, when associated with academia, has been associated with increased risk of hospital mortality in mechanically ventilated patients and in patients with nonpulmonary sepsis. Acidosis has been shown to be an independent predictor for mortality in patients with sepsis irrespective of the cause of acidosis (35).
The coefficient of variation for PaCO2 has been shown to be a consistent predictor of mortality in our study. This appears to be similar to other clinical conditions such as hyperglycemia where variability has been shown to be associated with increased mortality, likely reflecting physiological instability (36).
It is unclear from our study what the reason for increased risk of hospital mortality associated with coefficient of variation of PaCO2 might be. However, the rapid correction of hypercapnia with extracorporeal membrane oxygenation has been shown to be associated with increased incidence of neurological complications, suggesting that the variation of PaCO2 is an important marker of possible harm (37).
In our study, hypocapnia was significantly associated with an increased risk of hospital mortality in ventilated patients with 120 hours of exposure to ICU treatment. Although this could be a marker of illness severity, it is likely that hypocapnia has several other pathophysiological consequences that could adversely affect patients’ outcome. These include a hypocapnia-mediated shift in the oxygen dissociation curve to the left, leading to reduced oxygen delivery to tissues, as well as worsening ventilation perfusion mismatching, bronchoconstriction, reduced collateral ventilation, and reduced pulmonary parenchymal compliance, leading to impaired gas exchange (38, 39).
Our findings imply that, in critically ill patients with sepsis, exposure to hypercapnia is frequent. However, they imply that, after adjustment for key confounders, isolated hypercapnia per se is not associated with mortality. In contrast, hypercapnic acidemia, hypocapnia, and the coefficient of variation for PaCO2 showed a significant association with mortality, likely reflecting their possible roles as markers of illness severity. Finally, our findings imply probable equipoise for a pilot feasibility randomized controlled trial of permissive hypercapnia with our acidemia versus usual care in ventilated patients with sepsis.
Our study has several strengths. It is the first study to investigate the independent association of hypercapnia with hospital mortality in critically ill patients with sepsis. In addition, it included more than 3,000 critically ill patients with sepsis from three different hospitals, thus providing a degree of external validity for similar centers in high-income countries and representing a 40-fold increase in the data available on this issue. Because of such size, our study was at relatively low risk of type II error. With over 84,000 measurements of PaCO2, this study addressed the limitations of previous studies on hypercapnia, which only used a single value measured on the day of admission (11, 13, 25). Moreover, exposure of hypercapnia was studied over time, thus providing data on time of exposure and well as severity. Patients had PaCO2 measured at a median time interval of 185 minutes. Such close measurements provide clear insights into the association of exposure to PaCO2 during the patient’s entire ICU stay and hospital mortality instead of a single measurement obtained during the first 24 hours (11, 13, 25). Finally, the observation that isolated hypercapnia is not associated with mortality in multiple time-based cohorts and in patients with or without mechanical ventilation or pulmonary sepsis has clinical and trial design implications.
We acknowledge several limitations. First, our study is retrospective in design with the inherent limitations of such studies. However, the data used for analysis were collected prospectively, were numerical and objective, and were not subject to investigator bias. Second, our study did not investigate the association of hypercapnia with physiological outcomes. However, such physiological outcomes are not patient centered and carry limited clinical value. We did not have granular data on other variables that could influence the outcome of patients with sepsis, including the use of antibiotics, source of sepsis and infections with multiresistant organisms (40), or detailed data on ventilator settings that may have influenced PaCO2. However, we have presented data on the general practice of mechanical ventilation in the three study units for patients receiving prolonged ventilation as how they would be applied to the 7-days cohort. We reason that it would be logical to expect that such treatment would have been applied to all patients irrespective of the presence or absence of hypercapnia.
Overall, in patients with sepsis, isolated hypercapnia does not appear to be independently associated with increased mortality, irrespective of mechanical ventilation or pulmonary sepsis. Hypercapnic acidemia was associated with increased risk of mortality in mechanically ventilated patients and in patients with nonpulmonary sepsis between 48 and 120 hours of exposure. These findings have clinical implications and contradict our study hypothesis. Moreover, they imply that there may be equipoise for a pilot feasibility randomized controlled trial of permissive hypercapnia avoiding or minimizing acidemia compared with usual PaCO2 management in ventilated patients with sepsis and that such a trial might carry an acceptable expectation of safety.
| 1. | Casalino-Matsuda SM, Nair A, Beitel GJ, Gates KL, Sporn PH. Hypercapnia inhibits autophagy and bacterial killing in human macrophages by increasing expression of Bcl-2 and Bcl-xL. J Immunol 2015;194:5388–5396. |
| 2. | Curley G, Contreras MM, Nichol AD, Higgins BD, Laffey JG. Hypercapnia and acidosis in sepsis: a double-edged sword? Anesthesiology 2010;112:462–472. |
| 3. | Curley G, Hayes M, Laffey JG. Can ‘permissive’ hypercapnia modulate the severity of sepsis-induced ALI/ARDS? Crit Care 2011;15:212. |
| 4. | Akça O. Carbon dioxide and tissue oxygenation: is there sufficient evidence to support application of hypercapnia for hemodynamic stability and better tissue perfusion in sepsis? Intensive Care Med 2008;34:1752–1754. |
| 5. | Wang Z, Su F, Bruhn A, Yang X, Vincent JL. Acute hypercapnia improves indices of tissue oxygenation more than dobutamine in septic shock. Am J Respir Crit Care Med 2008;177:178–183. |
| 6. | Chonghaile MN, Higgins BD, Costello J, Laffey JG. Hypercapnic acidosis attenuates lung injury induced by established bacterial pneumonia. Anesthesiology 2008;109:837–848. |
| 7. | Higgins BD, Costello J, Contreras M, Hassett P, O’ Toole D, Laffey JG. Differential effects of buffered hypercapnia versus hypercapnic acidosis on shock and lung injury induced by systemic sepsis. Anesthesiology 2009;111:1317–1326. |
| 8. | Nichol AD, O’Cronin DF, Howell K, Naughton F, O’Brien S, Boylan J, et al. Infection-induced lung injury is worsened after renal buffering of hypercapnic acidosis. Crit Care Med 2009;37:2953–2961. |
| 9. | O’Croinin DF, Nichol AD, Hopkins N, Boylan J, O’Brien S, O’Connor C, et al. Sustained hypercapnic acidosis during pulmonary infection increases bacterial load and worsens lung injury. Crit Care Med 2008;36:2128–2135. |
| 10. | Akça O, Kurz A, Fleischmann E, Buggy D, Herbst F, Stocchi L, et al.; Hypercapnia Trial Investigators. Hypercapnia and surgical site infection: a randomized trial. Br J Anaesth 2013;111:759–767. |
| 11. | Sin DD, Man SF, Marrie TJ. Arterial carbon dioxide tension on admission as a marker of in-hospital mortality in community-acquired pneumonia. Am J Med 2005;118:145–150. |
| 12. | Afessa B, Greaves WL, Frederick WR. Pneumococcal bacteremia in adults: a 14-year experience in an inner-city university hospital. Clin Infect Dis 1995;21:345–351. |
| 13. | Vonderbank S, Gibis N, Schulz A, Boyko M, Erbuth A, Gürleyen H, et al. Hypercapnia at hospital admission as a predictor of mortality. Open Access Emerg Med 2020;12:173–180. |
| 14. | Lee JH, Ryu YJ, Chun EM, Chang JH. Outcomes and prognostic factors for severe community-acquired pneumonia that requires mechanical ventilation. Korean J Intern Med (Korean Assoc Intern Med) 2007;22:157–163. |
| 15. | Akça O, Doufas AG, Morioka N, Iscoe S, Fisher J, Sessler DI. Hypercapnia improves tissue oxygenation. Anesthesiology 2002;97:801–806. |
| 16. | Brower RG, Shanholtz CB, Fessler HE, Shade DM, White P Jr, Wiener CM, et al. Prospective, randomized, controlled clinical trial comparing traditional versus reduced tidal volume ventilation in acute respiratory distress syndrome patients. Crit Care Med 1999;27:1492–1498. |
| 17. | Otulakowski G, Kavanagh BP. Hypercapnia in acute illness: sometimes good, sometimes not. Crit Care Med 2011;39:1581–1582. |
| 18. | Pugin J, Dunn-Siegrist I, Dufour J, Tissières P, Charles PE, Comte R. Cyclic stretch of human lung cells induces an acidification and promotes bacterial growth. Am J Respir Cell Mol Biol 2008;38:362–370. |
| 19. | Tiruvoipati R, Botha JA, Pilcher D, Bailey M. Carbon dioxide clearance in critical care. Anaesth Intensive Care 2013;41:157–162. |
| 20. | Stow PJ, Hart GK, Higlett T, George C, Herkes R, McWilliam D, et al.; ANZICS Database Management Committee. Development and implementation of a high-quality clinical database: the Australian and New Zealand Intensive Care Society Adult Patient Database. J Crit Care 2006;21:133–141. |
| 21. | Diehl A, Burrell AJC, Udy AA, Alexander PMA, Rycus PT, Barbaro RP, et al. Association between arterial carbon dioxide tension and clinical outcomes in venoarterial extracorporeal membrane oxygenation. Crit Care Med 2020;48:977–984. |
| 22. | Eastwood GM, Schneider AG, Suzuki S, Peck L, Young H, Tanaka A, et al. Targeted therapeutic mild hypercapnia after cardiac arrest: a phase II multi-centre randomised controlled trial (the CCC trial). Resuscitation 2016;104:83–90. |
| 23. | Nin N, Muriel A, Peñuelas O, Brochard L, Lorente JA, Ferguson ND, et al.; VENTILA Group. Severe hypercapnia and outcome of mechanically ventilated patients with moderate or severe acute respiratory distress syndrome. Intensive Care Med 2017;43:200–208. |
| 24. | Tiruvoipati R, Pilcher D, Botha J, Buscher H, Simister R, Bailey M. Association of hypercapnia and hypercapnic acidosis with clinical outcomes in mechanically ventilated patients with cerebral injury. JAMA Neurol 2018;75:818–826. |
| 25. | Tiruvoipati R, Pilcher D, Buscher H, Botha J, Bailey M. Effects of hypercapnia and hypercapnic acidosis on hospital mortality in mechanically ventilated patients. Crit Care Med 2017;45:e649–e656. |
| 26. | R Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2020 [accessed 2021 Feb 5]. Available from: https://www.R-project.org/. |
| 27. | Kimura D, Totapally BR, Raszynski A, Ramachandran C, Torbati D. The effects of CO2 on cytokine concentrations in endotoxin-stimulated human whole blood. Crit Care Med 2008;36:2823–2827. |
| 28. | Ni Chonghaile M, Higgins BD, Costello JF, Laffey JG. Hypercapnic acidosis attenuates severe acute bacterial pneumonia-induced lung injury by a neutrophil-independent mechanism. Crit Care Med 2008;36:3135–3144. |
| 29. | Norozian FM, Leoncio M, Torbati D, Meyer K, Raszynski A, Totapally BR. Therapeutic hypercapnia enhances the inflammatory response to endotoxin in the lung of spontaneously breathing rats. Crit Care Med 2011;39:1400–1406. |
| 30. | Hickling KG, Walsh J, Henderson S, Jackson R. Low mortality rate in adult respiratory distress syndrome using low-volume, pressure-limited ventilation with permissive hypercapnia: a prospective study. Crit Care Med 1994;22:1568–1578. |
| 31. | Kregenow DA, Rubenfeld GD, Hudson LD, Swenson ER. Hypercapnic acidosis and mortality in acute lung injury. Crit Care Med 2006;34:1–7. |
| 32. | Barnes T, Zochios V, Parhar K. Re-examining permissive hypercapnia in ARDS: a narrative review. Chest 2018;154:185–195. |
| 33. | Eastwood GM, Tanaka A, Bellomo R. Cerebral oxygenation in mechanically ventilated early cardiac arrest survivors: the impact of hypercapnia. Resuscitation 2016;102:11–16. |
| 34. | Laserna E, Sibila O, Aguilar PR, Mortensen EM, Anzueto A, Blanquer JM, et al. Hypocapnia and hypercapnia are predictors for ICU admission and mortality in hospitalized patients with community-acquired pneumonia. Chest 2012;142:1193–1199. |
| 35. | Wernly B, Heramvand N, Masyuk M, Rezar R, Bruno RR, Kelm M, et al. Acidosis predicts mortality independently from hyperlactatemia in patients with sepsis. Eur J Intern Med 2020;76:76–81. |
| 36. | Krinsley JS. Glycemic variability: a strong independent predictor of mortality in critically ill patients. Crit Care Med 2008;36:3008–3013. |
| 37. | Cavayas YA, Munshi L, Del Sorbo L, Fan E. The early change in PaCO2 after extracorporeal membrane oxygenation initiation is associated with neurological complications. Am J Respir Crit Care Med 2020;201:1525–1535. |
| 38. | Curley G, Laffey JG, Kavanagh BP. Bench-to-bedside review: carbon dioxide. Crit Care 2010;14:220. |
| 39. | Laffey JG, Kavanagh BP. Hypocapnia. N Engl J Med 2002;347:43–53. |
| 40. | Sanderson M, Chikhani M, Blyth E, Wood S, Moppett IK, McKeever T, et al. Predicting 30-day mortality in patients with sepsis: An exploratory analysis of process of care and patient characteristics. J Intensive Care Soc 2018;19:299–304. |
Author Contributions: Concept: R.T., D.P., M.B., and R.B. Design: R.T., A.S.N., D.P., M.B., and R.B. Analysis: A.S.N. and M.B. Interpretation: R.T., A.S.N., M.Y., N.M., J.W., S.G., D.P., M.B., and R.B. Manuscript preparation: R.T., A.S.N., and R.B. Manuscript revision and approval: R.T., A.S.N., M.Y., N.M., J.W., S.G., D.P., M.B., and R.B.
This article has an online supplement, which is accessible from this issue’s table of contents at www.atsjournals.org.
Author disclosures are available with the text of this article at www.atsjournals.org.
