SUPPLEMENTAL MATERIAL

ICM-2007-01169

Central Venous to Mixed Venous Blood Oxygen and Lactate Gradients are Associated with Outcome in Critically Ill Patients

Possible physiological mechanisms resulting in ∆SO2 and ∆[Lac]

∆SO2 and ∆[Lac] are likely produced by the mixing of superior vena cava (SVC) blood with blood from the inferior vena cava (IVC) and the coronary sinus (CS). The separate influence of IVC and CS blood streams in determining the magnitude and sign of ∆SO2 and ∆[Lac] depends on their respective flows and concentrations of O2 and lactate. Given its greater flow, the IVC probably exerts a greater influence in determining ∆SO2 and ∆[Lac] than CS. Should that be the case, positive DSO2 values would indicate greater O2 extraction (ERO2) by lower body organs relative to those of the upper body1a resulting in SsvcO2 > SivcO2. Similarly, a positive D[Lac] would indicate lesser lactate production by lower body organs, resulting in [Lac]svc > [Lac]ivc. Both conditions are compatible with active aerobic metabolism.

Negative values for DSO2 would result from decreased lower body ERO2, such as in sepsis, or conversely, by a relative increase in upper body ERO2, such as it might occur with increased work of breathing2a. Both situations are indicative of clinical distress and would result in SsvcO2 < SivcO2. Likewise, negative D[Lac] values would be consonant with greater release of lactate by lower body tissues, such as it might occur with mesenteric ischemia, resulting in [Lac]svc < [Lac]ivc.

Although its blood flow is much lower than that of the IVC, coronary venous blood (including the CS and Thebesian system) also can contribute to the development of DSO2 and D[Lac]. This is likely to occur under conditions in which SO2 and [Lac] in the IVC and SVC are approximately equal. This was demonstrated in a study conducted in resting individuals undergoing right heart catheterization for the diagnosis of pulmonary hypertension3a. Although these patients had similar SO2 and [Lac] in the SVC and IVC, development of ΔSO2 and Δ[Lac] occurred in the right atrium, the site of entry of CS blood.

The response of the aerobic heart to increased mechanical demand is to augment O2 consumption with lactate as its preferred metabolic substrate4a. The result is an increase in myocardial O2 and lactate extraction with CS having the lowest SO2 and [Lac] of any venous blood 5a. Accordingly, DSO2 and D[Lac] could reflect alterations in myocardial metabolism, with positive values for DSO2 and D[Lac] associated with increased myocardial extraction. Conversely, negative values would suggest decreased myocardial O2 extraction or even the release of lactate into the coronary circulation. More likely, however, DSO2 and D[Lac] result from SVC blood mixing with IVC and CS blood in varying proportions.

ADDITIONAL FIGURES

Figure 1a. Distribution of Post-Op and Septic patients according to survivors and decedents based on Final DSO2 % and ∆[Lac] mmol·L-1 values, positive or negative. Both patient groups had greater proportions of survivors with positive DSO2 or D[Lac] values (Fisher Exact test, p < 0.05 for all groups).

Figure 2a. Number of patients remaining in the study at each successive sample period and corresponding changes in ∆[Lac] and ∆SO2 for the Septic and Post-Op groups. Error bars have been omitted for the sake of clarity, but are included in the Table shown below.


To determine if differences existed between survivors and decedents in the timing of samples, we counted the number of patients remaining in each group at each successive sample period. Although more samples per patient were obtained in the Sepsis group, not a surprising finding given their longer ICU stay, we found no differences in the timing of samples between survivors and decedents in either group (log rank test).

Also shown in the figure above are the mean values for ∆[Lac] and ∆SO2 for each sample period for up to 42 hours in the Septic group and 18 hours in the Post-Op group. During these times at least four patients remained in each group. In Septic patients, ∆[Lac] and ∆SO2 became increasingly negative at each successive sample period in decedents but remained positive in survivors. The same pattern was observed for ∆[Lac] in Post-Op patients (p<0.05 by ANOVA), but not for ∆SO2. Although suggestive of different behavior, positive ∆SO2 and ∆ [Lac] values in survivors and the opposite in decedents, meaningful statistical comparison of these curves by ANOVA is possibly misleading, given the few patients remaining towards the end of the observation period.

Number of patients sampled at different times during the study and corresponding mean ± SEM values for ∆SO2% and ∆[Lac] mM/L

Sepsis
Survivors
Time (Hrs) / No. Patients / % of Total / ∆SO2 / SEM / ∆[Lac] / SEM
0 / 12.0 / 100 / 1.2 / 1.4 / 0.14 / 0.06
6 / 12.0 / 100 / 0.2 / 1.2 / -0.03 / 0.06
12 / 11.0 / 92 / 1.0 / 1.1 / 0.02 / 0.08
18 / 10 / 83 / 0.7 / 1.9 / 0.01 / 0.06
24 / 9 / 75 / 2.0 / 2.1 / 0.17 / 0.07
30 / 7 / 58 / 1.2 / 1.4 / -0.10 / 0.12
36 / 6 / 50 / 1.1 / 2.2 / 0.06 / 0.04
42 / 4 / 33 / 3.3 / 4.3 / 0.03 / 0.06
Decedents
Time (Hrs) / No. Patients / % of Total / ∆SO2 / SEM / ∆ [Lac] / SEM
0 / 20 / 100 / 1.3 / 1.3 / 0.00 / 0.14
6 / 20 / 100 / 1.3 / 1.4 / -0.06 / 0.22
12 / 20 / 100 / 0.8 / 1.2 / 0.09 / 0.22
18 / 18 / 90 / -0.9 / 1.4 / 0.13 / 0.12
24 / 16 / 80 / 0.4 / 1.7 / -0.09 / 0.19
30 / 9 / 45 / -0.5 / 0.8 / -0.18 / 0.28
36 / 7 / 35 / -0.7 / 0.9 / -0.15 / 0.16
42 / 6 / 30 / -1.8 / 0.4 / -0.26 / 0.16
Post-Op
Survivors
Time (Hrs) / No. Patients / % of Total / ∆SO2 / SEM / ∆[Lac] / SEM
0 / 67 / 100 / 3.1 / 0.5 / -0.03 / 0.06
6 / 67 / 100 / 3.6 / 0.5 / 0.08 / 0.04
12 / 59 / 88 / 4.4 / 0.5 / 0.03 / 0.07
18 / 51 / 76 / 3.3 / 0.5 / 0.13 / 0.03
Decedents
Time (Hrs) / No. Patients / % of Total / ∆SO2 / SEM / ∆[Lac] / SEM
0 / 7 / 100 / 1.7 / 3.0 / -0.15 / 0.10
6 / 7 / 100 / 5.6 / 2.6 / -0.11 / 0.11
12 / 6 / 86 / 4.6 / 2.0 / -0.09 / 0.07
18 / 5 / 71 / 4.4 / 3.2 / -0.25 / 0.24

Figure 3a. DSO2 as a function of ∆[Lac] for Initial and Final measurements separated according to survival

A sense for the clinical evolution of the patients can be gained from these graphs. The numbers within each quadrant denote the number of patients. DSO2 - ∆[Lac] distributions in survivors and decedents differed both for Initial measurements (comparing plots A and C; p<0.05) and Final measurements (comparing plots B and D; p<0.001). Initial DSO2 - ∆[Lac] distribution differed from Final in survivors (comparing plots A and B; p < 0.01) as data migrated from other quadrants to the right upper quadrant of Plot B where DSO2 ≥ 0 and ∆[Lac] ≥ 0. There were no differences between Initial and Final DSO2 - ∆[Lac] distributions in decedents (comparing plots C and D), although there was a preponderance of deceased patients with negative DSO2 and ∆[Lac].

Figure 4a. Final ∆SO2 values plotted against its sensitivity and specificity as predictor of mortality for all patients in the study. The greatest Youden’s J statistic, shown by the vertical dashed line, was ΔSO2 = -0.7, a value indistinguishable from 0.


POSSIBLE CONFOUNDING FACTORS

Differences in Post-Op and Sepsis status.

To control for the Post-Op or Sepsis status of the patients, we performed the following post-hoc logistic regression model testing whether ΔSO2 and Δ[Lac] had significant effects on the likelihood of survival (postop=1 for post-operation and postop=0 for sepsis). The coefficient for postop shows a p value of 0.0721. Since the estimated coefficient is positive, it means that the post-operation group seems to have higher likelihood of survival than the sepsis group.

Model Fit Statistics

Intercept

Intercept and

Criterion Only Covariates

AIC 122.301 69.084

SC 124.965 109.035

-2 Log L 120.301 39.084

Testing Global Null Hypothesis: BETA=0

Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 81.2173 14 <.0001

Score 61.5794 14 <.0001

Wald 14.7899 14 0.3927

Analysis of Maximum Likelihood Estimates

Standard Wald

Parameter DF Estimate Error Chi-Square Pr > ChiSq

Intercept 1 -14.1337 4.6008 9.4372 0.0021

Postop 1 3.1214 1.7354 3.2352 0.0721

Age 1 0.8120 1.1022 0.5428 0.4613

Apache_II 1 2.3701 1.7154 1.9091 0.1671

ΔSO2 1 2.9602 1.3516 4.7968 0.0285

Δ[Lac] 1 2.0570 1.1057 3.4612 0.0628

Δ[Glu] 1 2.1188 1.2910 2.6938 0.1007

SsvcO2 1 0.3082 1.5125 0.0415 0.8385

SvO2 1 -0.8450 1.6653 0.2575 0.6118

[Lac] 1 5.0264 1.6808 8.9433 0.0028

[Glu] 1 3.5252 2.1228 2.7578 0.0968

CI 1 1.3311 1.2024 1.2255 0.2683

MAP 1 3.8285 1.7472 4.8017 0.0284

ERO2 1 0.8801 1.5050 0.3420 0.5587

pHa 1 1.1088 1.2752 0.7559 0.3846

Odds Ratio Estimates

Point 95% Wald

Effect Estimate Confidence Limits

Postop 22.678 0.756 680.441

Age 2.253 0.260 19.535

Apache_II 10.699 0.371 308.631

ΔSO2 19.301 1.365 272.928

Δ[Lac] 7.822 0.896 68.307

Δ[Glu] 8.321 0.663 104.486

SsvcO 1.361 0.070 26.379

SvO2 0.430 0.016 11.233

[Lac] 152.381 5.653 >999.999

[Glu] 33.959 0.530 >999.999

CI 3.785 0.359 39.959

MAP 45.993 1.498 >999.999

ERO2 2.411 0.126 46.054

pHa 3.031 0.249 36.899

The coefficient of ΔSO2 was statistically significant (p-value = .0285) even after controlling for the post-operation/sepsis status. Furthermore, the p-value for the coefficient of Δ[Lac] was .0628. Although not statistically significant at the p < 0.05 level, this value for p does indicate an association between Δ[Lac] and the likelihood of survival. In summary, these results imply the association between the delta measures and the likelihood of survival even after controlling for the post-operation/sepsis status.

Collection of data from different centers.

Treatment was likely to differ among participating centers, and probably among clinicians in each center. We did not dictate specific therapeutic strategies and could not control for this variable but we doubt that treatment heterogeneity significantly affected our findings, given the observational nature of the study.

In order to explore if the variation across centers was substantial, we ran a mixed-effect logistic regression model below. The ordinary logistic regression models the logit link function of the probability of the target event occurring as a linear function of the predictors as follows:

where g(∙) is the logit link function and π(x) is the probability of survival for the patients with the predictor values, x (x is a vector of such values). Here, β’s are coefficients of the predictors that are unknown fixed numbers.

Instead of treating all the coefficients as fixed numbers, we treated the intercept term, β0, to be randomly distributed as a normal distribution across centers. In other words, we consider each center may possibly have a unique value of β0; hence, the five centers in the data also have possibly different values of β0. By applying the mixed-effect logistic regression, we estimated the variance of β0 and the standard error of its estimate and made a rough judgment of the degree of variation across centers. The estimated variance of β0 is 5.0147 and its standard error is 5.7821. Although it is not a rule of thumb or is not supported by any scholarly research, an estimate that is smaller than its standard error puts a doubt that the parameter value to be estimated is substantial in size. In this context, the variation across centers appears to have been negligible.

Different blood gas and lactate analyzers

Blood gas and lactate analyzers from different manufacturers were used at the participating centers. To avoid discrepancies in metabolite measurement among centers, equipment was calibrated according to standard laboratory procedure prior to analyzing the blood samples. To minimize bias and relative error we measured individual blood samples twice for SO2 and three times for [Lac]6a.

Monitoring time.

Another possible confounding issue is that average monitoring time was relatively short (30.9 hours); however, this is in keeping with current practice of minimizing the duration of PAC usage.

Pulmonary artery catheter placement

Other than the chest radiograph taken after placement of the PAC, we had no way of verifying the location of the proximal port in the superior vena cava. Although it is possible that in some individuals this port may have migrated from the SVC to the right atrium, there is no reason to believe that this created a systematic bias in DSO2 and ∆[Lac] in our sample of patients.

Intravenous solutions containing lactate.

Administering solutions with different lactate concentrations to different subjects may theoretically produce heterogeneous effects on serum lactate levels. Lactated Ringer’s solution was not used in the Sepsis group. In the Post-Op group, one liter of lactated Ringer’s was used preoperatively in the priming of the extracorporeal circulation, but this solution was not used during the postoperative period. Since the first blood sample was taken after the patient was off the bypass pump, the effect of the priming fluid on the ∆[Lac] should have been minimal.