The five evaluated cohorts indicate the possibilities and limits of measuring Hb non-invasively. Overall, the difference between the highest and lowest Hb values was quite similar between the two technologies but, as shown many times before, there were wide limits of agreement at the baseline. Nevertheless, changes in Hb measured by the pulse oximeter often followed quite well the trends yielded by invasive measurement. However, this was not a universal finding. The non-invasive technique showed good precision but underestimated the magnitude of the invasive Hb changes in the presence of pre-surgical or surgical stress. By contrast, the precision was poor but the accuracy good in the volunteer studies.
Surgical setting versus non-stressed volunteers
The clinical question of whether erythrocytes should be transfused was most real in Study 4 where measurements were performed during major surgery. The blood loss during the data collection amounted to almost 1 l, and the patients received both Ringer’s and 20% albumin. Norepinephrine infusion was used to maintain the arterial pressure. The hemodilution was twice as large as that seen in any of the other studies.
The non-invasive Hb followed the trends observed in the invasive Hb quite well, with an overall r2 as high as 0.72 despite occasional outliers that were sometimes due to a low perfusion index. One should note that 0.72 is the mean linearity from individual analysis of each patient, whereas the r2 yielded by point-wise comparisons based on the pooled data from all patients in Study 4 was considerably lower, only 0.27. Importantly, the magnitude of the hemodilution changes was underestimated, and it usually amounted to only half of the invasive value (Fig. 3d).
The underestimation of the hemodilution by the non-invasive approach is shared by Study 3, which was performed in patients awaiting major open abdominal surgery. The fact that the two clinical studies show the same pattern — namely, a too small hemodilution indicated by the non-invasive Hb and, hence, a low slope factor — suggests that adrenergic stress limits the magnitude of the Hb changes detected by the non-invasive technology, albeit without distorting the trend over time.
Studies 1 and 2 were both performed in a similar way in volunteers, but using different amounts of crystalloid fluid. Both studies had slopes with a median value close to 1.0, and the non-invasive hemodilution was approximately 20% greater than that obtained with the invasive Hb. Hence, the relationship between the two modes of measurement was opposite to that obtained in the perioperative setting, which supports the notion that data obtained in volunteers cannot be uncritically assumed to be valid in stressful situations.
By contrast, the performance of the non-invasive technology with oral administration of fluid was disappointing. Here, non-invasive measurement usually showed no change at all, or a trend opposite to that obtained by the invasive Hb. However, the average hemodilution obtained by invasive sampling averaged only 3%, which then seems too small to overcome the between-subject variability of the non-invasive Hb technique to any reasonable degree.
Another issue is whether non-invasive monitoring can be used for the study of fluid population kinetics. For this purpose, the Hb data had to be converted to plasma dilution, which attains values that are approximately twice as large as the hemodilution. Previous studies have used non-invasive Hb for kinetic analysis in a two-stage approach, but with limited success [2, 12]. Here, all appropriate curves were analyzed in one single run, which offers a more stable output.
The analysis shown in Fig. 6 reveals that fitting the commonly used two-volume kinetic model to the present data produced more outliers and less precise estimates of the model parameters when based on the non-invasive as compared to the invasive technology. In our fictitious fluid experiment, the difference between the modes of measuring Hb became apparent during the distribution phase but not during infusion.
Present report versus the literature
Many previous studies have examined the methodological bias by making point-wise comparisons between invasive and non-invasive measurements of Hb, usually without separating intra- and inter-individual differences. Some studies show a bias of 1–2% [1,2,3,4], while studies in cardiac and emergency care report underestimations of the invasive Hb of as much as 12–18 g L− 1 by the non-invasive technique [5,6,7]. These results support our observation that changes in non-invasive Hb become too small in stressed settings. The precision has consistently been poor, with 95% limits of agreement being in the range of 40–50 g L− 1 [3, 8], or even wider [4,5,6,7]. The conclusion in a meta-analysis of 32 studies from 2016 confirmed that the mean difference between the two modes of measuring Hb is small, while the range between the 95% limits of agreement is too wide to allow clinical decisions to be based on them [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17].
Our review does not focus on point-wise comparisons; rather, it emphasizes a series of Hb changes, which we feel is more informative about the value of this monitoring in the perioperative setting. Here, hemodilution curves derived by the two modes of measuring Hb most often showed quite similar trends, but the many successful cases (high r2) were occasionally interrupted by curves showing poor agreement, so as to be misleading. Some of these were “switches” remain unexplained. The most stable indications were obtained in the surgical setting, where r2 was below 0.25 in only 2/23 patients, although the non-invasive Hb systematically underestimated the hemodilution. The scattering was aggravated in the presence of a very low perfusion index. By contrast, Chang et al. , found the accuracy of the Radical-7 to be good even when the perfusion index is below 1.0.
Limitations of this analysis include that the invasive Hb was measured in venous blood while the non-invasive method is believed to represent arterial or capillary Hb. However, the two sites of measurement should not differ much in terms of Hb concentration . The five series of Hb experiments were collected under different conditions, which are described in detail elsewhere, but these differences do form the basis of the present conclusions [11,12,13,14,15,16]. Data were collected over a period of almost 10 years, during which the algorithm for calculating non-invasive Hb is claimed to have been improved. However, more recent work on Hb during spinal surgery and liver transplantation [10, 19] show almost identical accuracy and precision for the non-invasive Hb as we reported in 2010 . These authors also studied the concordance between changes in invasive and non-invasive Hb during surgery [10, 19]. However, we chose to present our results differently, as we obtained data at very short intervals, usually only 5–10 min, and up to 25 times per experiment.