Introduction

Pulse oximetry is used in many emergency and critical care settings to monitor arterial oxygenation and to guide adjustments of FIO2. Pulse oximetry devices use different techniques to process the signal [1, 2, 3] which may result in a different performance when challenged by artifacts or low perfusion [4, 5, 6, 7]. Although the effect of low perfusion on pulse oximetry readings has been studied using exposure to hypothermia, vasoconstriction, and hemorrhagic hypotension [5, 8, 9, 10, 11, 12], only limited and conflicting data are available on the effect of sepsis on the accuracy of pulse oximetry readings [13, 14]. An unexpected and thus secondary finding of our previous studies in septic animals with low perfusion was that SpO2 measurements may under- or overestimate SaO2 considerably [15, 16].

The objective of this study was to test the hypothesis that progressively deteriorating perfusion would affect accuracy of pulse oximetry measurements using the latest available techniques in an animal model of emerging sepsis. Further, we wanted to evaluate whether the perfusion index can serve as a useful marker of impaired peripheral perfusion to alert the clinician that accuracy of pulse oximetry readings may be affected.

Materials and methods

All animals were cared for according to current German laws on the protection of animals and to NIH guidelines for the care and use of laboratory animals and all experiments were approved by the appropriate government agencies. Thirty-seven anesthetized and ventilated adult New Zealand white rabbits received Escherichia coli by tracheal instillation to induce pneumonia/sepsis and were supported for 8 h as described previously in detail [15]. Arterial hemoglobin oxygen saturation (SpO2) was simultaneously measured with a Radical Masimo SET, software version 4.1.0.1 (Masimo. Irvine, Calif., USA) equipped with an LNOP-Neo Sensor, and a Nellcor N-395 Oxismart XL, software version 1.6.2.0 pulse oximeter equipped with a Sensor type D-YS (Tyco Healthcare/Mallinckrodt, St. Louis, Mo., USA). After closely shaving both forelegs the pulse oximeter sensors were randomly assigned to one foreleg each and switched hourly. Sensor sites were shielded against ambient light using an opaque cover.

Using pulse oximetry a variable amount of light is absorbed by pulsating arterial flow (AC), and a constant amount of light is absorbed by nonpulsating blood and tissue (DC). The pulsating signal indexed against nonpulsating signal and expressed as ratio is commonly referred to as the “perfusion index” = ACx100/DC. It has been suggested as a marker of poor peripheral perfusion in critically ill patients [17, 18].

Arterial hemoglobin O2 saturation (SaO2) as measured by CO oximetry (Omni 3, Roche Diagnostics, Graz, Austria) was drawn every 30 min to calculate individual bias (defined as SpO2 – SaO2) for each device in each animal at each time point. From all absolute bias values of any given animal during the first 4 h an individual median bias was calculated and compared to the corresponding median bias value derived from data obtained from the second 4 h of the experimental period in each animal. Differences in paired, continuous variables were analyzed by using the two-tailed paired t test or Wilcoxon signed rank test where appropriate. Differences in proportions were analyzed by using the χ2 test (including Yates' correction). Repeated measurements across time were analyzed with repeated-measures analysis of variance on ranks. Differences with a p value less than 0.05 were considered statistically significant. Values presented are mean ± SD or median (range). Primary outcome measure in this study was the absolute bias value (SpO2 − SaO2). In a previous study the average absolute bias value comparing the first 4 h with the following 4 h increased [15], and the standard deviation of this change was 1.64. A power calculation based on this data set revealed that 37 animals would provide a power of 0.9 to obtain a significant test result, if an increase in the absolute value of bias by 1.0 is present, corresponding to an approximate tripling of the average bias comparing the two study periods.

Results

Comparing the first with the second half of the experimental period, the individual absolute value of bias calculated from individual measurements increased with progressively deteriorating hemodynamics with both devices: from a median of 0.10% (range 0.10–0.65%) to one of 1.1% (0.10–14.6%) with Nellcor (p < 0.001) and from 0.10% (0.10–0.65%) to 1.3% (0.1–5.7%) with Masimo (p < 0.001). Bias values increased over time for both devices, indicating that accuracy is affected upon time with progressing sepsis (detailed data available upon request). Individual bias values exceeded the ± 3% error limit given for conditions of low perfusion by the manufacturers of pulse oximetry devices designed for clinical use in human subjects in 10.9% of SpO2/SaO2 measurements with the Nellcor Oxismart XL (66/603) vs. 11.2% of those with the Masimo SET (67/600) and in 21.4% of measurements during the second half of the experimental period (4–8 h) with the Nellcor Oxismart XL (66/309) vs. 22.6% of those with the Masimo SET (66/305).

The perfusion index, obtained from the Masimo SET pulse oximeter showed a wide variation and increased across time (Fig. 1, upper panel), indicating increasing pulsatility in the peripheral vascular bed toward the end of the experimental period, i.e., at a time when absolute bias values increased as well. However, when individual bias values are plotted along with the corresponding perfusion index values (Fig. 1, lower panel), large bias values seem to coincide more with low perfusion index values. Whereas bias values with the Nellcor N-395 were observed in both directions, the Radical overestimated oxygen saturation during episodes with low perfusion as indicated by a lower perfusion index (Fig. 1, lower panel). Bias exceeding the ± 3% error limit was more common with a perfusion index less than 0.5 than with a perfusion index of 0.5 or greater (Nellcor: 22/129, 17.1%, vs. 44/474, 9.3%, p < 0.05; Masimo: 21/126, 16.7%, vs. 46/474, 9.7%, p < 0.05), suggesting that the perfusion index may be used as a marker for situations of low perfusion at risk for increased bias. Sensitivity, specificity, and positive and negative predictive values for two different perfusion index cutoffs to detect bias exceeding the ± 3% error limit are shown in Table 1. These data show that sensitivity and positive predictive values with a cutoff of 0.5 are quite low. Using a cutoff of 1.0 increases sensitivity but causes a considerable loss of specificity.

Fig. 1
figure 1

Perfusion index (for definition see text) across time (upper panel) and relationship of bias to perfusion index (lower panel). Upper panel Data are median (range) derived from all animals. Perfusion index increased across time (p < 0.001, repeated-measures analysis of variance on ranks). Lower panel Relationship of bias (defined as SpO2 – SaO2) to perfusion index. Each dot refers to one measured bias value with its corresponding perfusion index value in one given animal at one given time point. Note: Large bias values seem to coincide with low perfusion index values

Full size image
Table 1 Sensitivity, specificity, positive (PPV) and negative predictive values (NPV) of the perfusion index to detect bias exceeding the ± 3% error limit using two different cutoffs: percentages
Full size table

Discussion

The findings of this study confirm that individual bias values may be affected considerably with emerging sepsis even when using more recently available pulse oximeters, as observed in our previous study using older techniques [15]. A false SpO2 reading may lead to inappropriate interventions such as increases or decreases in FIO2 or ventilator pressures, which may expose the subject on mechanical ventilation to unnecessary hypoxemic or hyperoxic injury or barotrauma. The fact that more than 20% of SpO2 measurements during the second half of the experimental period were beyond the ± 3% error limit in conditions of low perfusion given by the manufacturers of pulse oximetry devices for clinical use may not be reassuring for the clinician.

We evaluated the perfusion index as a measure of peripheral pulsatility to detect the subject at risk for increased bias of pulse oximetry readings. The median perfusion index increased across time, suggesting increasing pulsatility in the peripheral vascular bed and decreasing peripheral vascular resistance, which would be in agreement with findings from other investigators, who found an increased bias in adults with sepsis and low systemic vascular resistance [13]. However, in this present study large bias values seemed to coincide with low perfusion index values. We found a higher percentage of bias beyond the ± 3% error limit when perfusion index was below 0.5, suggesting that the perfusion index indeed may be used as a tool in situations of low peripheral perfusion at risk for increased bias of pulse oximetry readings. Using a perfusion index less than 0.5 as a cutoff, specificity was approx. 80% in both devices tested, whereas sensitivity was only 33% and 31% and the positive predictive value was approx. only 17% for both devices. If an arterial blood gas and/or CO oximetry were obtained in that particular situation, 83% of measurements would not show increased bias. Using a perfusion index less than 1.0 as a cutoff would increase sensitivity as expected, but causes a considerable decrease in specificity and positive predictive values. Other investigators have evaluated the relationship of the perfusion index to clinical signs of poor perfusion and found that a value of below 1.4 was a useful indicator of low perfusion in critically ill adult patients [17] and 1.24 or less in severely ill neonates [18]. Unfortunately, neither study provides data regarding bias of pulse oximetry readings to be compared with ours. Based on our data the preliminary conclusion may be drawn that the clinician faced with a perfusion index less than 0.5 may decide to obtain an arterial blood gas and/or CO oximetry to confirm or disprove the pulse oximetry reading, especially if the patient is septic and/or other clinical findings suggest poor peripheral perfusion. Another approach may be to use a reflectance pulse oximetry device placed in the esophagus, as bias seems to be reduced in comparison to peripheral measurements during adverse hemodynamic conditions in severely ill adult patients [19]. Whereas bias values had a more symmetrical distribution using the Nellcor, the Masimo Radical overestimated SaO2 during episodes of low perfusion. This finding may be important for the clinician, and we speculate that the different performance of the devices may be related to signal processing.

In summary, our findings show that the bias of pulse oximetry readings is considerably affected during experimental sepsis with both devices tested. Increased bias was more common with a low perfusion index. However, because of limited sensitivity, specificity, and positive predictive value the perfusion index is not a useful tool as a marker for increased risk of bias. Therefore intermittent blood gas measurements or CO oximetry may be warranted in septic patients with severely affected hemodynamics and respiratory failure.