The association of near-infrared spectroscopy-derived tissue oxygenation measurements with sepsis syndromes, organ dysfunction and mortality in emergency department patients with sepsis

Study design and overall approach

We conducted a prospective, multicenter, observational study of a convenience sample of patients who presented to the ED of one of three large, urban, tertiary care facilities. Three specific cohorts of patients (septic shock (the "SHOCK" cohort), sepsis without shock (the "SEPSIS" cohort) and uninfected controls) were assembled, encompassing a spectrum of sepsis severity. We collected pertinent demographic and clinical covariates as well as initial StO₂% and NIRS-derived measurements in response to VOT testing. Then we analyzed the association and predictive ability of the NIRS measurements with our outcomes of interest. The study was approved by the ethics committees of each of the hospitals.

Assembly of cohorts

Three distinct cohorts of patients were enrolled. The SHOCK cohort had to meet the American College of Chest Physicians/Society of Critical Care Medicine criteria for septic shock, specifically (1) suspected infection, (2) fulfillment of two or more of the criteria for systemic inflammatory response syndrome (SIRS) (temperature > 100.4°F or < 96.8°F, heart rate > 90 beats/minute, respiratory rate > 20 breaths/minute or partial pressure of carbon dioxide < 32 mmHg, white blood cell count > 12,000/μL or < 4,000/μL or > 10% bands) and (3) hypotension despite adequate fluid resuscitation (systolic blood pressure (SBP) < 90 mmHg after 20 mL/kg crystalloids) [7]. The SEPSIS cohort had to meet the inclusion criteria of suspected infection, two or more SIRS criteria (see above) and no refractory hypotension. The third cohort comprised uninfected ED control patients who met the criteria of no suspected infection, no SIRS criteria met and no evidence of hypoperfusion. The control patients were age-matched (by decade) as well as sex- and race-matched to the SHOCK cohort.

A common set of exclusion criteria were applied to all patient cohorts, which included any of the following: age < 18 years, pregnancy, established "Do Not Resuscitate" orders prior to enrollment, acute traumatic or burn injury (primary diagnosis), acute cerebrovascular event (primary diagnosis), acute coronary syndrome (primary diagnosis), acute pulmonary edema (primary diagnosis), cardiac dysrhythmia (primary diagnosis), acute and active gastrointestinal bleeding (primary diagnosis), acute drug overdose (primary diagnosis), requirement for immediate surgery and inability to obtain written informed consent. Clinical management at each institution is in agreement with the Surviving Sepsis Campaign guidelines.

Demographic and clinical covariates

We collected demographic variables (age, sex and race), comorbidities (cerebrovascular disease, chemotherapy, congestive heart failure, chronic obstructive pulmonary disease, dementia, diabetes, HIV or AIDS, Hodgkin's disease, intravenous drug use, leukemia, liver disease, myocardial infarction, non-Hodgkin's lymphoma, peripheral vascular disease, renal disease, splenectomy, steroid use, ulcer disease, transplant, presence of any malignancy and residence in a nursing home), vital signs data (temperature, blood pressure, heart rate, respiratory rate and oxygen saturation) and pertinent laboratory data (serum lactate, complete blood count, chemistry panels, markers of coagulation and liver function tests).

NIRS assessment

The InSpectra StO₂ Tissue Oxygenation Monitor (model 650; Hutchinson Technology, Hutchinson, MN, USA) with probes spaced at 15 mm was utilized to obtain StO₂ measurements. The measurements were taken at the thenar eminence during the resuscitation phase. Following a minimum initial five-minute stabilization period, we assessed the initial StO₂ measurement and then performed a VOT procedure using an automated tourniquet (Delfi Tourniquet System; Delfi Medical Innovations, Inc, Vancouver, BC, Canada), which was insufflated to 50 mmHg over the patient's SBP for a period of three minutes. After three minutes, the cuff was quickly removed. The subsequent StO₂ tracing was analyzed offline to record the following NIRS-derived metrics (see Figure 1) (1) StO₂ initial, the baseline StO₂ recorded after a five-minute stabilization period; (2) StO₂ occlusion, the steady-state rate of occlusion (StO₂%/second), represented by the descending slope during the ischemic period; and (3) StO₂ recovery, the steady-state recovery slope during the reoxygenation phase after the tourniquet was released. The StO₂ measurements were imported into a Microsoft Excel software file (Microsoft Corporation, Redmond, WA, USA), and the slopes were derived by (1) drawing a best-fit line for the steady-state slope for the respective metric and (2) calculating the slope.

Outcomes

We examined the association of StO₂ parameters in relation to three patient-oriented outcomes: (1) presence of shock, as defined above, assessed at the time of enrollment; (2) in-hospital mortality, defined as vital signs status at hospital discharge; and (3) organ dysfunction at 24 hours assessed on the basis of the SOFA scores calculated at the time of enrollment and 24 hours later [6]. Consistent with prior publications, we defined organ dysfunction as a SOFA score ≥ 2, which was our primary outcome of interest. The use of a threshold SOFA score ≥ 2 for an ill patient has previously been established [8–10]. All patients in the control group with missing SOFA scores at 24 hours (discharge was the primary reason for missing data) were assumed to have a SOFA score < 2. For patients enrolled with a history of chronic renal insufficiency or end-stage renal disease, the renal SOFA score was not included in the total SOFA score.

Data analysis

Descriptive statistics (means, standard deviations, medians or proportions with percentiles) were reported for demographics, clinical characteristics, vital signs and laboratory values stratified by the three cohorts. We compared mean (or median) values for the StO₂ parameters of interest (initial, ischemic slope and reperfusion slope) of the three different groups using the Wilcoxon two-sample test. We used a Bonferroni correction to address multiple testing; thus, for this analysis, Cronbach's α < 0.017 was considered significant. Next, we compared the StO₂ parameters of interest for the outcomes of in-hospital mortality and SOFA scores ≥ 2 at 24 hours. To assess the diagnostic accuracy for the StO₂ parameters as a predictor of outcomes, we constructed receiver operating characteristic curves (ROCs) and calculated the area under the curve (AUC) along with the 95% confidence intervals (95% CIs). We used multivariate logistic regression models to obtain adjusted estimates for age, serum lactate and SBP and to identify the StO₂ parameters and adjustor variables with the strongest independent associations with outcomes by using a stepwise backward elimination technique with forward examination of parameters eliminated after final model selection. Throughout the analysis we used serum lactate level as a comparison predictor.

Sample size calculation

Our study was powered on the ability of our anticipated best StO₂ readout (StO₂ recovery slope) to discriminate the SEPSIS cohort from the SHOCK cohort. Based on previous studies, assuming mean changes in slope of 2.3 ± 1.3 for the SHOCK group and 3.2 ± 1.4 for the SEPSIS group, with a power of 90% and Cronbach's α set at 0.05, we calculated that approximately 60 patients per group were needed [11]. We also enrolled 50 uninfected controls as comparators for comparisons between controls and the sepsis groups.

Patient characteristics

Differentiation of SHOCK, SEPSIS and control cohorts

The mean values for the three main StO₂ parameters of interest, StO₂ initial, ischemic slope and reperfusion slope, were calculated and compared for the different levels of sepsis syndrome and for controls upon enrollment (Table 2 and Figures 2 through 4). The mean values for all three parameters in the SHOCK patients were significantly different from those in the SEPSIS patients. However, the comparison between the SHOCK patients and controls (who were age- and sex-matched) revealed that only the upslope mean was significantly different in the SHOCK vs control cohorts. Conversely, the initial and ischemic slope means, but not the recovery slope mean, were significantly different between the SEPSIS and control groups.

StO2 parameters	Groups (N= 168)			P values
	SHOCK ( n = 58)	SEPSIS ( n = 60)	Control ( n = 50)	SHOCK vs SEPSIS	SHOCK vs control	SEPSIS vs control
Initial, %	76 ± 9 (78)	82 ± 7 (82)	79 ± 7 (79)	< 0.001*	0.23	< 0.02
Occlusion, %/second	-10.2 ± 5.2 (-9.6)	-13.1 ± 4.4 (-13.1)	-11.3 ± 4.5 (-10.8)	< 0.002*	0.21	< 0.03
Recovery, %/second	2.4 ± 1.6 (2.0)	3.9 ± 1.7 (4.1)	3.8 ± 1.7 (3.8)	< 0.001*	< 0.001*	0.52

^aSHOCK: the septic shock cohort (systolic blood pressure < 90 mmHg after fluid challenge); SEPSIS: the sepsis without shock cohort; Control: emergency department patients without infection; StO₂: tissue oxygen saturation; *Wilcoxon two-sample test; statistically significant using P = 0.017 with the Bonferroni correction. Data are means ± SD (medians).

Mortality prediction

For the mortality outcomes, we assessed the StO₂ parameters obtained in the ED, as well as serum lactate, SBP and age. The recovery slopes for patients who died were significantly lower (mean ± SD: 1.7 ± 1.5 vs StO₂%: 3.7%/second; P < 0.0001), with impaired oxygen recovery observed among the nonsurvivors (Table 3). Similarly, the ischemic slope was less steep, showing decreased oxygen consumption during the vasoocclusion phase of the VOT (mean ± SD: -8.8 ± 5.1 vs StO₂%: -12.0 ± 4.7%/second; P < 0.002). Both of these metrics are postulated to represent impaired microcirculation and a reduced capacity to exchange and deliver oxygen. Initial StO₂% did not differ significantly between the survivors and nonsurvivors, nor did the mortality rate differ when StO₂ was stratified as < 80% or ≥ 80% (15% vs 15%; P = 1.0). The AUC as a predictor of mortality was 0.81 (95% confidence interval: 0.71 to 0.91) for the recovery slope, 0.70 (0.57 to 0.83) for the ischemic slope and 0.56 (0.43 to 0.69) for the initial slope. Serum lactate levels were also increased in the nonsurvivors compared to the survivors (4.7 ± 2.7 vs 1.9 ± 1.4 mmol/L; P < 0.001), with an AUC of 0.85 for serum lactate. The ROCs are shown in Figure 5. The multivariable logistic regression model used to determine independent predictors of mortality included the age, serum lactate, SBP and StO₂ parameters. Using both forward and backward selection techniques, the lactate and recovery slopes were retained in the model as the strongest predictors of in-hospital mortality, regardless of cohort. These two variables yielded an AUC for model discrimination of 0.88.

Parameter	Died (n= 25)	Survived (n= 143)	P value*	AUC (95% CI)
Serum lactate, mmol/L	4.7 ± 2.7 (4.2) (n = 24)	1.9 ± 1.4 (1.5) (n = 128)	< 0.001	0.85 (0.76 to 0.93)
Systolic blood pressure, mmHg	105 ± 30 (102)	122 ± 30 (123)	0.004	0.68 (0.56 to 0.80)
Mean age, years	69 ± 14 (72)	62 ± 18 (67)	0.100	0.60 (0.49 to 0.72)
StO2
Initial (%)	76 ± 11 (79)	79 ± 7 (80)	0.341	0.56 (0.43 to 0.69)
Occlusion (%/minute)	-8.8 ± 5.1 (-8.2)	-12.0 ± 4.7 (-11.9)	0.002	0.70 (0.57 to 0.83)
Recovery (%/second)	1.7 ± 1.5 (1.3)	3.7 ± 1.7 (3.8)	< 0.001	0.81 (0.71 to 0.91)

^aAUC: area under the curve; 95% CI: 95% confidence interval; StO₂: tissue oxygen saturation; *Wilcoxon two-sample test or Fisher's exact test, as appropriate. Data are means ± SD (medians).

Organ dysfunction at 24 hours

Next, we assessed the StO₂ parameters for their ability to predict organ dysfunction, defined a priori as SOFA score ≥ 2 at 24 hours. The initial and occlusion StO₂ metrics, as well as serum lactate, SBP and age, were significantly abnormal in patients with SOFA scores ≥ 2 at 24 hours compared to those with SOFA scores < 2 (Tables 4 and 5). The StO₂ occlusion slope did not show a statistically significant difference between the groups. The AUC for the groups were initial slope, 0.61 (0.52 to 0.70); ischemic slope, 0.57 (0.48 to 0.66); recovery slope 0.68 (0.59 to 0.76); and lactate slope 0.69 (0.61 to 0.78). The ROCs are shown in Figure 6. We also examined the correlation between the NIRS parameters at initial presentation and the total SOFA score and found a correlation between StO₂ initial slope (Spearman's ρ correlation coefficient = -0.18; P < 0.04), occlusion slope (Spearman's ρ correlation coefficient = 0.21; P < 0.02) and recovery slope (Spearman's ρ correlation coefficient = -0.35; P < 0.001).

Parameter	SOFA score ≥ 2 at 24 hours#			AUC (95% CI)
	Yes ( n = 65)	No ( n = 100)	P value*
Serum lactate, mmol/L	3.1 ± 2.2 (2.1)	1.7 ± 1.5 (1.5)	< 0.001	0.69 (0.61 to 0.78)
Systolic blood pressure	107 ± 27 (104)	128 ± 30 (126)	< 0.001	0.71 (0.63 to 0.79)
Mean age, years	67 ± 17 (71)	61 ± 18 (64)	0.020	0.61 (0.52 to 0.70)
StO2
Initial (%)	77 ± 9 (79)	80 ± 7 (81)	< 0.02	0.61 (0.52 to 0.70)
Occlusion (%/minute)	-10.7 ± 4.9 (-10.6)	-12.1 ± 4.8 (-11.6)	0.14	0.57 (0.48 to 0.66)
Recovery (%/second)	2.7 ± 1.9 (2.0)	3.8 ± 1.7 (3.9)	< 0.001	0.68 (0.59 to 0.76)

^aSOFA: Sequential Organ Failure Assessment; AUC: area under the curve; 95% CI: 95% confidence interval; StO₂: tissue oxygen saturation. ^#For patients in the control group in which the SOFA score at 24 hours was not measured, a SOFA score < 2 was assumed. For patients with a history of chronic renal insufficiency or end-stage renal disease, the total SOFA score used did not include the renal portion of the score. *Wilcoxon two-sample test or Fisher's exact test, as appropriate. Data are means ± SD (medians).

Parameter included	Parameter retained (Yes/No (P value))b
	Outcomes
	SOFA ≥ 2 at 24 hoursc	In-hospital mortality
Age, years	No (0.086)	No (0.089)
Serum lactate	Yes (< 0.001)	Yes (< 0.001)
Systolic blood pressure	Yes (0.014)	No (0.571)
StO2
Initial	Yes (0.046)	No (0.128)
Occlusion	No (0.199)	No (0.669)
Recovery	Yes (0.037)	Yes (0.003)
Multivariate model AUC (95% CI)	0.79 (0.72 to 0.87)	0.88 (0.80 to 0.95)
Log (p/(1 - p))	5.6651 + 0.5135*Lactate - 0.0173*SBP - 0.0529*Initial - 0.2465*Recovery	-1.7582 + 0.5084*Lactate - 0.5271*Recovery
Odds ratios (95% CI)
Serum lactate (1 mEq/L increase)	1.7 (1.2 to 2.2)	1.7 (1.3 to 2.2)
SBP (10 mmHg decrease)	1.2 (1.0 to 1.4)	NA
Initial (10% decrease)	1.7 (1.0 to 2.9)	NA
Upslope (1%/second decrease)	1.3 (1.0 to 1.6)	1.7 (1.2 to 2.4)

^aSOFA: Sequential Organ Failure Assessment; StO₂: tissue oxygen saturation; AUC: area under the curve; 95% CI: 95% confidence interval; SBP: systolic blood pressure; NA: not applicable. ^bStepwise backward elimination technique used. The P value in parentheses after "Yes" is the P value in the resultant model. The P value after "No" is the P value when the eliminated parameter was included back into the final model for reevaluation after elimination. ^cFor patients in the control group in which the SOFA score at 24 hours was not measured, a SOFA score < 2 was assumed. For patients with a history of chronic renal insufficiency or end-stage renal disease, the total SOFA score used did not include the renal portion of the score.

Sensitivity analysis

We chose to use the entire study population and included the uninfected control population in our analyses to assess mortality outcomes, as well as organ dysfunction at 24 hours, to take advantage of our full data set. However, one could argue against this approach and make the case for including only patients who fulfilled the minimum inclusion criteria for the definition of sepsis. Thus, we performed a subsequent sensitivity analysis in which we limited the analysis to the 118 patients enrolled from the SEPSIS and SHOCK groups. The results of this analysis were very similar to those of our primary analyses (Additional file 1 online data supplement).