Prediction of Plasma Hemoglobin Concentration by Near-Infrared Spectroscopy

Bleeding is inevitable during most surgical procedures and estimation of the amount of bleeding is an important part of an anesthesiologists' daily work. Anesthesiologists tend to depend on their own educated guess as an index for estimating the amount of blood loss. The plasma hemoglobin concentration (Hb) represents a standard measurement that is used to estimate blood loss and the current status of oxygen carrying capacity of arterial blood. Measurement of Hb is costly and takes a few minutes. At times, time-delay can interfere with patient management.

Research on a noninvasive in vivo estimation of Hb is very limited. In addition, the utility of any methods studied frequently requires unfamiliar instrument (1-3) or is applicable to very specific situations (4-6). We have assessed the Hb value as a factor that influences near-infrared spectroscopy referred to as cerebral oximetry, and abbreviated as rSO₂ during a series of clinical investigations using cerebral oximetry. rSO₂ is a well known important variable used to predict oxygen balance in the brain. We postulated that rSO₂ would indicate Hb levels when the cerebral metabolic rate and cerebral blood flow were assumed to be constant.

Our attempt to predict Hb noninvasively was based on a simple assumption. The values of rSO₂ represent regional cerebral field oxygen saturation; these values can be confounded by several factors such as arterial oxygen saturation (SaO₂), arterial carbon dioxide tension (PaCO₂), and Hb levels. We have evaluated the potential usefulness of rSO₂ to predict Hb levels by controlling and excluding the confounders. This study was designed to test our assumption that changes in the rSO₂ can predict changes in the Hb level when PaO₂ and PaCO₂ are constant, and to provide a mathematical model that demonstrates the correlation of rSO₂ with Hb levels during general anesthesia.

After approval from the Institutional Ethics Committee, we recruited thirty, American Society of Anesthesiologists (ASA) physical status I or II, adult patients (Table 1) who were scheduled for elective spine surgery, with an expectation of a moderate degree of intraoperative bleeding (20-30% of estimated blood volume). Patients with cardiovascular, endocrine or cerebral disease were excluded. General anesthesia was induced with intravenous administration of propofol (1.5 mg/kg) and vecuronium (0.1 mg/kg) and inhalation of isoflurane via a face mask. The trachea was intubated and ventilation was mechanically controlled. Anesthesia was maintained with isoflurane and 50% oxygen with air and intermittent administration of vecuronium. An emitter-sensor couplet of a cerebral oximeter was attached to the right forehead.

Table 1
Patient characteristics

• Click for larger image

Blood loss was estimated by naked eyes. The amount of homologous blood transfusion was not regulated but performed by in-hospital protocols. Average 2-4 repeats of blood samplings in a patient for the measurement of Hb and PaCO₂ were performed by anesthesiologist's subjective judgment at irregular periods. Full arterial oxygen saturation was confirmed at every period. At the time of skin closure, the amount of blood loss was guessed by an experienced anesthesiologist, unaware of the purpose of this study. Measured Hb and PaCO₂ were recorded alongside of rSO₂, esophageal temperature (T_ESO), potency of vapor anesthetic (ExpISO; expired concentration of isoflurane), mean arterial pressure (MAP) and anesthetic exposure duration expressed by minutes (Tmin).

Hb, MAP, T_ESO, rSO₂, Exp_ISO, PaCO₂, and estimated blood loss (EBL) were recorded (Table 2). Hb was tightly regressed with three variables. Resultant regression equation was proved as follows (p=0.000, r²=0.809):

Table 2
Basic measurements during surgical procedures

• Click for larger image

Hb=+8.580+0.238 · rSO₂-0.338 · PaCO₂-0.004 · Tmin (Eq. A)

Ignoring the minute contribution of Tmin (substituting Tmin with mean Tmin), equation A can be simplified as follows:

Hb=+8.332+0.238 · rSO₂-0.338 · PaCO₂ (Eq. B), (Fig. 1)

Fig. 1
Multilinear regression analysis of Hb by rSO₂ and PaCO₂. Regression equation was Hb=+8.332+0.238 · rSO₂-0.338 · PaCO₂ (p=0.000, r²=0.809). Changes of rSO₂ and PaCO₂ reflected the changes of Hb.

• Click for larger image

Standardized residuals between Hb and predicted Hb were 0.086 ± 1.100 gm/dL (p=0.443) and were regressed by Hb as follows (p=0.000, r2=0.20):

sR=-2.7705+0.2492 Hb

However, goodness-of-fit test for above equation was failed (p=0.0274). Lowess was performed (Fig. 2). Regression equation of lowess'ed pairs of residuals (ssR) and Hb (sHb) was proved as follows (p=0.000, r²=0.14):

Fig. 2
Standardized residuals vs. Hb before (triangle-up symbols and dotted line) and after (round symbols and straight line) lowess technique. A window is placed about χ (Hb); data points that lie inside the window are weighted so that nearby points get the weight and a robust weighted regression is to predict y value at χ. The parameter, controlling the window's size, is set at 0.2.

• Click for larger image

ssR=0.9730-0.1032 · sHb (Eq. C)

Test statistic for goodness-of-fit of equation C was 0.6877 (Fig. 3).

Fig. 3
Scatter plot of standardized residuals (predicted Hb-Hb) and Hb before (a) and after (b) lowess techniques. Lowess'ed data points were fitted linearly more successfully, which revealed the existence of hemoglobin dependence of prediction of Hb by rSO₂. Predicted Hb tends to underestimate Hb at higher Hb, while overestimate Hb at lower Hb.

• Click for larger image

In this study, we evaluated whether rSO₂ could predict the true or measured Hb levels using a prediction equation that was compared with clinically relevant levels of measured Hb. We successfully derived a mathematical equation that showed a correlation of Hb and rSO₂. Arterial carbon dioxide tension and anesthetic duration also affected the predicted Hb levels. Other potential variables that could affect the equation were excluded or included by stepwise selection. By the analysis of the residuals according to the measured Hb levels, the predicted Hb level by equation A had a tendency to underestimate the Hb level when the measured Hb levels were high, and to overestimate the Hb level when the hematocrit was low. Interestingly, our results are consistent with the classical concept of an inverse correlation between cerebral blood flow and hematocrit in animals (8, 9) and humans (10). For the level of measured Hb that appears to be cut-off point for overestimation was 9.4 gm/dL (0.9730/0.1032 from equation C). We predict that 1) low Hb below 9.4 gm/dL causes the cerebral vessels to dilate and raises the rSO₂, and 2) high Hb over 9.4 gm/dL may impede cerebral blood flow and lowers rSO₂ and thereby lowers the predicted Hb level.

Arterial carbon dioxide tension is another important variable that affects the result of our equation, and also a potential confounding factor, which made Dullenkopf et al. (11) fail to prove the relationship of rSO₂ and Hb. We believe that an anesthesiologist can manage the arterial carbon dioxide tension within 5 mmHg, which affects Hb level ± 1.6 gm/dL (5 × 0.338 from equation A).

Anesthetic duration also affects the equation. Most vapor anesthetics increase cerebral blood flow (CBF) initially but decrease CBF to a steady state near pre-exposure levels in animals (12). This finding was not evident in humans (13). Our results show that prolonged exposure of isoflurane caused small increases of rSO₂. This finding is not consistent with previous reports (12, 13). Possible explanation includes the inevitable failure of statistical independence of time-effect; the measured Hb decreases as a function of time in every patient. However, our result confined the time-effect within 0.004; this would suggest that the rSO₂ increased spontaneously up to +1% during 180 min of anesthesia despite a steady state of Hb; therefore, the time effect can be ignored in clinical decisions.

Our findings are compatible with the basic understanding on cerebral physiology and rheology. However, we do not suggest that our equation should be used to predict Hb levels for everyday anesthesia procedures; but rather provide additional information to better understand noninvasive prediction of Hb levels. Potential bias should be eliminated by further studies with a controlled study design and larger clinical database.

In conclusion, the value of near-infrared spectroscopy, rSO₂ as an accurate, easy-to-read, noninvasive, real-time, and continuous predictor of Hb levels in clinical anesthesia has been demonstrated in this study. PaCO₂ and anesthetic duration should also be considered as important factors for predicting Hb by rSO₂. The Hb levels tend to be overestimated at lower Hb levels with the model presented here. Users of this model should keep in mind the assumptions of this design which includes a normal cerebral vasculature and blood pressure.