Secondary Logo

Journal Logo

Technology, Computing, and Simulation: Review Article

A Review of Signal Processing Used in the Implementation of the Pulse Oximetry Photoplethysmographic Fluid Responsiveness Parameter

Addison, Paul S. PhD

Author Information
doi: 10.1213/ANE.0000000000000392
  • Free


Hemodynamic optimization through intravascular volume expansion is commonly used for the critically ill patient. Fluid is administered with the expectation that it will increase cardiac preload and cardiac output significantly; however, the response may be variable. The key question therefore is: before the administration of the fluid bolus, how can we identify patients who would respond to intravascular volume expansion? Reversible fluid challenge maneuvers, including passive leg raising or anti-Trendelenburg/Trendelenburg positional changes, may be used to provide a short-term fluid shift which, through an associated change in preload, allows a transient hemodynamic test to identify the patient’s operating point on the Frank–Starling (F–S) curve.1 The measurement of stroke volume (SV) changes due to reversible fluid challenge maneuvers is, however, time consuming, disrupts work flows, and may not be the standard of care, or even possible, in certain clinical settings. Because of these disadvantages, much attention has centered on the respiratory modulation of the cardiovascular system. This modulation may be thought of as a once-per-breath reversible challenge where the respiratory change in intrathoracic pressure and corresponding blood flow causes a fluctuation in preload and hence a cyclical variation in SV known as SV variation (SVV). We may use SVV to tell us where the patient is on the F–S curve. This is depicted in Figure 1. For a patient operating at low preload on the steep part of the F–S curve, the respiratory-induced preload modulation causes a marked modulation in SV (i.e., SVV). This situation is depicted at point A in Figure 1A as SVVA. Conversely, at higher preload values, as the curve flattens out to a plateau, corresponding respiratory fluctuations in preload cause a much reduced SVV. This situation is depicted at point B in Figure 1A as SVVB. SVV, therefore, effectively tells the clinician where on the F–S curve the patient is operating. Hence, it infers what gains in SV potentially could be made should fluids be administered and the patient is pushed up the F–S curve to an optimal operating point (nominally marked “O” in the schematic of Figure 1B). These potential gains in SV (ΔSV) are depicted in Figure 1B for each of the starting points A and B. Thus, from Figure 1, A and B, we see that large values of SVV correspond to potentially large gains in SV on the administration of fluid. A fuller account including the many subtleties associated with the concept is provided elsewhere.2

Figure 1
Figure 1:
Respiratory fluctuations on the Frank–Starling curve. A, Respiratory-induced preload fluctuations generate localized stroke volume variations (SVVs). B, Location on the curve implying potential gains in stroke volume (ΔSV) relative to optimal operating point (O).

Instead of measuring SV change as an indicator of fluid responsiveness, which may in practice be difficult to measure easily and continuously in many patient populations, the respiratory effect on the arterial blood pressure waveform has been considered a useful surrogate.2 Many measures of the respiratory modulation in the arterial waveform are, in fact, possible including pulse pressure variation (PPV), systolic pressure variation (SPV), systolic pressure changes due to changes in peak respiratory ventilator pressures,3 or during imposed apnea of short duration.4 Respiratory variation in the arterial blood pressure waveform is a good indicator of response to fluid loading in the mechanically ventilated patient,5,6 and many monitors now compute the PPV parameter for display on the screen. The use of PPV to indicate the volemic status of a patient is increasingly widespread in practice and thus has been the focus of much attention in this area.

PPV is, however, an invasive parameter requiring an arterial line, whereas a measure based on the pulse oximeter signal (the photoplethysmogram or “pleth”) would provide an entirely noninvasive technology. This is the main driver of the current interest in this area. Studies reporting the use of the pleth as a measure of fluid responsiveness are dominated by the ΔPOP and pulse variability index (PVI®) parameters. PVI is a proprietary algorithm of the Masimo Corporation (Masimo Corporation, Irvine, CA), and, as such, there is little detail in the literature regarding the signal processing aspects of its implementation. In contrast, ΔPOP is a method that many research groups have implemented from first principles and where its correlation with PPV is well documented and its algorithmic details have been described. This review therefore focuses on the signal processing aspects of ΔPOP as reported in the literature.


Definition of ΔPOP

ΔPOP has been suggested by Cannesson et al.7 as a parameter that can be extracted from the pleth which correlates with PPV. They defined ΔPOP as the “respiratory variation in pulse oximetry plethysmographic (POP) waveform amplitude.”

ΔPOP is defined by Cannesson et al.7 as follows:

where AMP is the amplitude of the cardiac pulse waveforms in the pleth and

. ΔPOP is usually expressed as a percentage. This is illustrated in Figure 2 where the cardiac pulse component of the pleth is shown being modulated by respiratory activity. The maximum, minimum, and mean values of the cardiac pulse amplitudes over each respiratory cycle are used to derive ΔPOP. Note that the equation for ΔPOP can be traced back in the literature to Michard et al.6,8 where the same mathematical formulation is used to determine PPV (from an a-line signal).

Figure 2
Figure 2:
Deriving ΔPOP from the pleth. AMP = amplitude of the cardiac pulse waveforms in the pleth.

Origins and Context of ΔPOP

Early articles by Partridge9 and Shamir et al.10 considered the use of respiratory modulations in the pleth as measures of fluid responsiveness. Partridge9 calculated a “pulse waveform variation” from the pleth: a measure analogous to SPV from a continuous blood pressure signal. Shamir et al.10 also considered an analogous measure to SPV, which they named SPVplet. In addition, they considered dUpplet and dDownPlet defined as the height of the maximal and minimal pleth pulse peaks during respiration measured from the constant peak value during apnea.

A 2004 abstract by Poli de Figueiredo et al.11 contains details of a trial performed to test the hypothesis that “pulse wave variation by pulse oximetry during mechanical ventilation adequately reflects pulse pressure variation by invasive arterial lines.” In the reported study of 31 critically ill patients, a “significant correlation” (R = 0.88) was found between PPV and “ΔPleth” (a measure by Poli de Figueiredo et al.11 with its formulation equivalent to PPV). Further, the investigators determined the ability of the ΔPleth measure to differentiate a threshold of 13% for PPV, finding a sensitivity of 93% and specificity of 95%. Although in abstract form, this work covered all the main elements and essentially set the scene for this type of pleth-based analysis. However, although the work by Poli de Figueiredo et al.11 appeared earlier, it is the study by Cannesson et al.7 of 2005 that is repeatedly cited and Cannesson’s group that has published many times over the intervening years (Cannesson et al.7,12–17) and is foremost in promoting the use of the parameter. Around the same time, Natalini et al.4 considered a number of pleth-based modulation parameters including ΔDownpleth, ΔUppleth, Δsyspleth, and Δpulsepleth: the latter being their name for a ΔPOP measure. Since these early works, many articles have now appeared concerning the use of ΔPOP as a measure of fluid responsiveness.

ΔPOP as a Fluid Responsiveness Parameter

The standard method used to determine the suitability for ΔPOP as a surrogate for PPV is to quantify the relationship between them. This may be illustrated by a scatter plot, used in this context by Cannesson et al.,7 where ΔPOP is plotted against PPV. A best fit line may be drawn and various statistics computed from the data illustrating the degree to which the two measures correlate, including the R value and corresponding P value. The plot of ΔPOP against PPV from Cannesson et al.7 is shown in Figure 3. The results from all such studies reported in the literature are listed in the fourth column of Table 1, where it may be seen that they range from R = 0.91 (P < 0.001) to R = 0.05 (P < 0.15). The lower value is an anomalously poor result and should be viewed in the context that the second lowest R value found was 0.48 and 10 of the 13 values in the list are >0.78. All results, including this markedly poorer result of Landsverk et al.,20 are discussed in more detail later in this text. Note that the R correlation statistic is the main statistic cited in the discussion herein as it is common to all articles considered. However, a range of analysis methods have been used including sensitivity and specificity pairs, areas under the curve (AUCs), Bland-Altman plots, and polar plots. These are not dealt with in detail herein, and their appropriateness for use in such analysis is in itself a worthy topic of investigation.

Figure 3
Figure 3:
ΔPOP against pulse pressure variation (PPV) from Cannesson et al.7
Table 1
Table 1:
Reported Performance of ΔPOP Against PPV in the Literature

It should be noted that the method used in the construction of the scatter plots affects the resulting values of R. For example, in our own work, we have found that plotting the average values on a per-subject basis (as is regularly performed in the literature) may increase the R value. Hence, results reported should be interpreted in the context of the method used to produce them. Cannesson et al.14 echo this sentiment arguing that for such studies “the way in which the data are recorded, analyzed, and reported should be standardized in order to avoid potential confounding factors.”

We may also determine other measures, such as sensitivity and specificity and AUC values corresponding to the substitution of ΔPOP for PPV as a fluid responsiveness measure. This is achieved by setting a threshold for PPV (ranging between 8.8% and 15% for the studies considered here4,7,11,12,15,19,21,22,25,33), and finding the sensitivity and specificity pairs for varying ΔPOP thresholds (i.e., computing the receiver operator curve). We may select the optimal ΔPOP threshold from a predefined criterion such as maximizing the sensitivity + specificity summation (which is often used in the literature, although other criteria may be defined). The sensitivity, specificity, and AUC values are given in columns 6 to 8 in Table 1 for those studies where they were computed.

The above analysis may be performed when a blood pressure signal and pleth signal are both available. However, it is important to note that it only quantifies the relationship between them, i.e., to consider ΔPOP as a substitute for PPV, and is not a test of ΔPOP directly as a fluid responsiveness measure. To test the ability of ΔPOP to predict fluid responsiveness itself, a study incorporating a fluid challenge is required, where the sensitivity and specificity and AUC for predicting response to the challenge may be calculated. A few groups have performed this including Cannesson et al.,7 who found that ΔPOP can predict fluid responsiveness in cardiac surgery patients; Feissel et al.,18 who found ΔPOP to be as accurate as PPV in predicting fluid responsiveness in mechanically ventilated intensive care unit (ICU) patients; and Hoiseth et al.,25 who concluded that its diagnostic value was relatively poor when used with abdominal surgery patients. Although these results are mixed, as will be seen from this literature review, much depends on the implementation of the parameter.


In this section, an overview is presented of the implementations of the ΔPOP algorithm reported in the literature. This was conducted by searching the MEDLINE database for relevant studies using the following keywords: (“fluid responsiveness” OR “volume responsiveness” OR “preload dependence” OR “preload responsiveness” OR “hypovolemia”) AND (pulse oximeter OR pulse oximetry OR photoplethysmog* OR plethysmog* OR ΔPOP OR POP). All articles containing correlation information between ΔPOP and PPV were selected. (Note that all varieties of “ΔPOP” were included in the search: ΔPOP, Delta-POP, DPOP.) Bibliographies of all selected articles were then searched for other relevant articles, including reviews, comments, and replies to comments. The main output of the review of the searched literature is presented in Table 2.

Table 2
Table 2:
Reported Details of Signal Processing Implementations

“Global” and “Local” Filtering

A number of authors describe screening collected signals to remove entire patient data records from subsequent analysis. This manual “global” filtering may be through predefined exclusion criteria or removal of a whole record from the analysis due to the acquired signal being of a quality deemed too poor for use. For example, researchers have excluded subjects due to low perfusion, poor waveform, unstable pleths, presence of arrhythmias, spontaneous breathing activities, bronchospasm, signs of excessive intravascular volume, and left ventricular dysfunction. The manual selection of optimal subsegments of signal before analysis has been reported by many authors. Such “local” manual filtering is defined in one of two ways: either in terms of local exclusion criteria, for example, signal containing spontaneous respiratory efforts, arrhythmias, “disturbances,” or the observation of “poor quality signal”; or inclusion criteria, for example, using only segments containing a stable or distinct photoplethysmographic waveform or adequate perfusion. A special case of local filtering is that used to cope with signal gain changes. These often manifest as localized high-frequency, high-amplitude signal transients. If an algorithm where several computed values are being combined to produce the reported parameter (i.e., the one displayed on the device screen) is used, then measures need to be put in place to mitigate transient values occurring in the buffer at the point where the gain change occurred. A number of authors note that they consider the automatic gain function during the computation of ΔPOP: some disabling the gain or performing measurements of ΔPOP when no gain change is detected. A full account for the global and local filtering reported in the literature is provided in columns 2 to 4 of Table 2.

Acquisition and Processing Details

The methods of extraction and manipulation of signal information to compute ΔPOP are reported (in varying detail) in all searched studies. The location at which the signal was interrogated and the averaging period used in the computation of the ΔPOP parameter vary greatly across studies. In addition, some authors report a single value per subject, while others report multiple values computed at various locations within the temporal record of the pleth. Table 2, columns 5 to 7, contains details of the acquisition and processing techniques used. This is split into details of the analysis period used, the averaging used, and further details of the analysis undertaken.


Algorithm Implementation

The algorithm used for the computation of ΔPOP varies widely in the reported literature, as does the detail of explanation of its implementation. Cannesson et al.7,12,13,15 stated that the reported ΔPOP was calculated by averaging over 3 respiratory cycles. This is the only reported detail of the signal processing undertaken during these studies. It is not clear whether this was a single value (or if multiple 3-cycle averages were further averaged), or when this single value was taken in relation to the period of data acquisition. Pereira De Souza Neto et al.24 and Chandler et al.26 also report simply that ΔPOP was computed over 3 respiratory cycles without further context. Hoiseth et al.25 used 3 respiratory cycles averaged as 1 observation but computed these for 20 consecutive cycles. Whereas the same group in a later article34 stated that the reported ΔPOP was computed over 10 respiratory cycles and that these were calculated in a custom-made program in LabVIEW (National Instruments, Austin, TX).

Westphal et al.22 state that ΔPOP and PPV were “determined as the average of three respiratory cycles over one minute.” This ambiguous statement is interpreted by the author to mean that the best 3 respiratory cycles within the 1-minute acquisition period were used to determine ΔPOP. The analysis of pressure and plethysmographic waveforms was performed “off-line on a personal computer.” It appears that Westphal et al.22 manually interpreted these data for the calculation because it is stated that one of the limitations was the “lack of scales in the screen.” Pizov et al.23 used a printer module to record both the arterial and plethysmographic waveforms which were scanned at a resolution of 300 dpi and all the measured points were verified by 2 independent researchers. Chandler et al.,26 in 2012, stated that “Data were recorded for at least 1 min during which the subject was not stimulated and any drug infusion rates were kept constant”; however, they averaged over only 3 respiratory cycles. Cannesson et al.13,15 stated that POP waveform amplitude was measured on a beat-to-beat basis as the vertical distance between peak and preceding valley trough in the waveform and was “expressed in pixels.”

Whereas the investigators cited above report on a single representative averaged value of ΔPOP per subject (over 3, 5, or 10 cycles), three studies report on the computation of multiple values of the parameter over significantly longer periods. These are described in more detail below:

Landsverk et al.20 were the first group to report numerous values of ΔPOP per subject and hence began to probe the intrapatient behavior of the parameter. They reported that the PPV and ΔPOP values were computed “automatically” over each “manually” chosen respiratory cycle and then averaged over 3 respiratory cycles until 70 pairs of PPV/ΔPOP values were obtained. However, it is also stated that the pleth was measured in millimeters and that the interobserver variability in calculating ΔPOP and PPV was considered by having it evaluated independently by two of the authors in 4 patients. This would appear to indicate that the “automation” mentioned in fact relates to the manual analysis method and is not meant to be indicative of an automated computing process of any real sophistication. It should also be noted that the work by Landsverk et al.20 considered sedated ICU patients in whom significant autonomic function may be present. They also only included data where PPV is <13%, thus is at the lower end of the scale of percentage modulations and as such may be more inherently susceptible to noise. (No reason for restricting the analysis to such low PPV values is given in the article.)

Hoiseth et al.25 studied 25 patients during open abdominal surgery. One thousand three hundred sixty values of ΔPOP and the corresponding respiratory changes in pulse pressure were calculated from 4080 respiratory cycles in 68 recording periods. However, Hoiseth et al.25 performed significant manual filtering of the data before the determination of ΔPOP including manual delimitation of the respiratory cycle using a thorax impedance signal and manual verification of the minimum and maximum values of POP and pulse pressure before their use in computing ΔPOP and PPV. Respiratory cycles with “arrhythmias” or “disturbances” were also omitted manually. In addition, corresponding PPV and ΔPOP values from 3 consecutive respiratory cycles were averaged as 1 observation before subsequent calculations. This manual filtering may explain the significantly better correlation obtained by the study by Hoiseth et al.25 compared with the earlier study by Landsverk et al.20 (R = 0.78 vs 0.05).

Interestingly, Hoiseth et al.25 further explored intraindividual variability in the parameters by studying 60 respiratory cycles during the surgery (approximately 5 minutes of data). This they stated represented a “realistic window of decision.” They pointed out that most other studies only have 3 to 5 respiratory cycles worth of data (around 15–20 seconds). They show examples where stable parameter values over these longer registration periods provide positive prediction of outcomes.

Hengy et al.28 reported on the interrogation of continuous recordings of ΔPOP and PPV for 43 patients undergoing abdominal surgery. Data were collected for 4.1 ± 2.0 hours per patient. This resulted in more than 125,000 respiratory cycles for analysis. Hengy et al.28 excluded 3 patients due to unusable pleths or intraoperative occurrence of an arrhythmia. Of the remaining data, individual respiratory cycles were excluded manually by investigators for arterial or pleth signal noise, arrhythmias, spontaneous breathing, and arterial line flushes. This accounted for 5.7% of the respiratory cycles. Furthermore, approximately one-quarter of the data (22%) was discarded through the use of “automatic software analysis.” The only detail given by the authors of this automatic analysis is that it is “homemade software using classical methodology to further eliminate such abnormal data.” This combined manual and “homemade” automated filtering methodology appears to be caught somewhere between the completely manual methodology of many authors (itself very useful in determining underlying relationships through manual parsing of the good data by visual inspection) and the fully automated algorithm required for a robust, continuous determination of ΔPOP in the clinical setting. It is suspected that this may account for the “weak correlation” between ΔPOP and PPV reported by the authors. However, there was still a marked increase in correlation obtained over the study results by Landsverk et al.20 (i.e., R = 0.79 vs 0.05).

Hengy et al.28 went on to perform multivariate analysis of the data and found that PPVs >12% (used to indicate preload dependence in the study) significantly increased the degree of correlation. (This might also explain the poor results of Landsverk et al.20 who considered data only for PPV <13% in their analysis.) This increase in correlation at larger PPV values is an important point which impinges on such studies. It highlights the effect noise has on computing a representative ΔPOP and/or PPV value for those patients exhibiting smaller respiratory modulations, where noise may dominate and cause both significant spreading and decorrelation of the data.

Hengy et al.28 stated that the preceding 6 data points for both ΔPOP and PPV are used to test the latest calculation. In their method, the latest value of either of these parameters is considered, and if it differs from the mean of the preceding 6 points by 2.5 SD, it is not used in the analysis. This manual method appears to be the only reference made in the literature to the removal of outliers during the derivation of ΔPOP.

The articles by the Landsverk, Hoiseth, and Hengy groups where computation of multiple values of the ΔPOP parameter over significant time scales are to be commended because they provide a deeper insight into the overall behavior of the parameter and are a significant first step toward considering all the data within a fully automated implementation. However, such analysis has its own potential pitfall of some significance where the computational methods are not yet robust enough to deal with the multifaceted nature of the signal noise. By including even a few spurious points, which would be accounted for either using a carefully controlled (e.g., blinded) manual method or a fully honed, robust automated method, significantly degraded results may be generated.

It is the author’s view that care has to be taken when interpreting results where manual inspection of physical tracing of signals (in millimeters or pixels) are used to calculate subtle signal differences because small percentage errors especially around predefined threshold values (e.g., 15% or so) may significantly affect the outcome of the analysis. Landsverk et al.20 reported a markedly poor correlation between ΔPOP and PPV for ICU patients, reporting R = 0.05, P = 0.15: the study with by far the poorest R value reported in the literature. However, inspection of the ΔPOP–PPV correlation plot contained within Figure 2 of the article by Landsverk et al.20 shows it to be noticeably different than those produced in the work of others (e.g., Cannesson et al.,7,12,13 etc.), with a diverse spread of data points exhibited with no obvious underlying or unifying interpatient relationship. Even on an intrapatient basis, most of the individual patient data groupings in the Landsverk et al.20 plot appear quite random in nature, which the author suspects may be the result of excessive noise still present in the data.

Reported Physiological Influences on the Pleth and Probe Position

Researchers in the field are, in general, very aware of the particular difficulties associated with interpreting the pleth waveform, and a number of researchers have made specific comments on the nature of the pleth in relation to the underlying physiology and how this relates to the difficulties associated with the extraction of useful information. Natalini et al.4 recognized that ventilation-induced photoplethysmographic changes are caused by blood volume changes both in the arterial and venous bed and further that the variation in photoplethysmographic waveform amplitude depends on the “intravascular pulse pressure as well as on the distensibility of the vascular wall.” Hengy et al.28 state that “the distensibility or compliance could vary over anesthesia, depending on microvascular tone changes induced by anesthetic events such as temperature, sympathetic nervous system activity, or ischemia/reperfusion-induced changes.” Feissel et al.19 commented on the influence of vasomotor tone on the plethysmographic waveform especially in patients receiving norepinephrine. Landsverk et al.20 state that changes in vasomotor tone strongly influence the waveform and that, therefore, they would “anticipate an impairment of the respiratory variations in the pulse oximetry photoplethysmographic waveform signal from the patients with low laser Doppler flow values, indicating vasoconstriction.” The effect of vasotone on the pleth is also discussed by Cannesson et al.,15 Westphal et al.,22 Landsverk et al.,20 and Hoiseth et al.25 and the review by Antonsen and Kirkebøen29 where, in the latter reference, it is stated that “Variations in total peripheral resistance and vasomotor tone increase under the influence of general anesthesia, with vasoactive drugs, with site of measurement, and with physiological responses such as inflammation, pain, fear, and body temperature. This may lead to inaccuracy of the photoplethysmography signal.” Cannesson et al.12 stated that “The raw plethysmographic signal is much more variable. Density will be a function of tissue (nonchanging signal) and blood (changing signal), and changing density will be a function of changing blood.” They also noted that “the blood density change is determined by both perfusion pressure and vasomotor tone.”

Landsverk et al.20 draw attention to oscillatory components “slower than those of the heartbeat and respiration” present in the pleth and that these “slow oscillations” are related to the “sympathetic nervous system and local vascular control mediated from the vascular wall, known as vasomotion.” They also state that “Sedative and anesthetic drugs impair these oscillations.” Westphal et al.22 state that extracorporeal circulation may interfere with results because it elicits “an inflammatory response, reduces vascular tonus and increases endothelial permeability.” Westphal et al.22 also cite “movement artifacts, peripheral vasoconstriction and cutaneous pigments” as limitations attributed to dynamic indicators of cardiovascular responsiveness.

The physiological processes that both enable and confound the measurement of ΔPOP therefore appear well understood in this space and link to the wider literature on the causes of erroneous pleth components including vasotone, vasomotion, posture, patient motion, temperature, metabolic state, pain, drug administration, lung compliance, upper airway obstruction, heart rate, respiratory rate, the venous blood component, arrhythmia, and ventilator pressure and flow settings. These are well documented more fully in the work of others.27,35–45

Probe position is also an important factor in acquiring good data. Cannesson et al.12 attached the probe to the index of either right or left hand and wrapped it to prevent outside light from interfering with the signal. Chandler et al.26 also describe covering the fingers to exclude light. Natalini et al.4 obtained a photoplethysmographic waveform by applying the pulse oximeter probe to a finger or toe. Cannesson et al.16 commenting on an article by Landsverk et al.20 stated that “alternative sites of measurement (such as ear or forehead) may improve the accuracy of POP, because these sites do not present the same sensitivity to changes in vasomotor tone as compared with the finger. In addition, these sites of measurement also impact ΔPOP itself up to 10-fold.” Hoiseth et al.25 suggested the earlobe as an alternative site because it “might lead to less distortion by sympathetic nervous activity and thus less variability of ΔPOP.” Pereira De Souza Neto et al.24 stated that ΔPOP variations “depend on the site of measurements.” “For example, the POP waveform can be up to 10 times stronger in the head when compared with the finger. Further studies evaluating the ear, forehead, and finger signal for fluid responsiveness prediction in the current setting may be of interest.” Again, these comments in the cited studies link with a wider literature available on the subject.46–53

Investigator Comments on Signal Processing Requirements

There is generally an acute awareness of the shortcomings of the signal processing reported in the literature, and many authors take the opportunity in the discussion section of their own work, or sometimes in appraisals of the work of others, to critique the methodology used and suggest improvements.

A number of authors draw attention to the already preprocessed nature of the pleth with which they are working. The situation is aptly summed up by Cannesson et al.14 where it is stated that the pleth is “a highly processed signal, and that only the raw waveform can display consistent respiratory variations.” Landsverk et al.20 echo the sentiment commenting that the commercial pulse oximeter used in their study had “filters built in,” hence the analog output signal they used was therefore not a “raw signal.” They therefore could not exclude the possibility that the respiratory variations may have been altered by the preprocessing of the device. Cannesson et al.15 suggest that visual analysis of the respiratory variations in the waveform is unreliable, since the amplitude of the curve “is constantly processed and smoothed by most of the devices commercially available.”

Hoiseth et al.25 used a “commercially available pulse oximeter, downloading a ‘raw’ signal.” But they further state that it is unknown to them “if details in the generation and processing of this signal influence the results, and, importantly, if other pulse oximeters might give different results.” Cannesson et al.17 commenting on the article by Hengy et al.28 state that the pleth waveform is “highly processed and contains many different pieces of information (stroke volume, vasomotor tone, venous signal)” which, according to Hengy et al.,28 may explain the poor agreement between PPV and ΔPOP in this study. However, Cannesson et al.17 suggest that “a rich signal does not mean that it cannot be interpreted” and draw attention to the much noisier ΔPOP signal (compared with PPV) in the article by Hengy et al.28 and state that “ a more sophisticated processing of the trend (filtering and smoothing) could eventually make the signal more relevant,” which “emphasizes the importance of using more advanced signal-processing algorithms when analyzing the plethysmographic waveform than when studying the arterial pressure signal.” This echoes a sentiment of Cannesson et al.14 where they mention “the digital signal processing that is required to make accurate measurements in clinical conditions that were otherwise impossible (e.g., patient motion, low perfusion, electrical interference).” Shelley et al.54 further commenting on the article by Hengy et al.28 suggested that, in agreement with Cannesson et al.,17 an improved digital signal processing is required to allow for the correct isolation of the underlying signals. In addition, however, a better understanding of the effect of the respiratory-induced venous modulation component is cited as another key factor necessary for the interpretation of the signal.

Cannesson et al.16 take issue with the signal processing undertaken by Landsverk et al.,20 commenting that it is “an example of how important the specifics of the algorithm used to determine ΔPOP can be” and go on to suggest the use of high-pass filtering to mitigate against vasomotor tone artifact. Cannesson et al.16 also stated that when “clinical monitors are used as research tools, subtle differences in proprietary software may have a profound impact on the results.” Landsverk et al.21 replied that a high-pass filter could remove the slower oscillations from the original signal. Landsverk et al.21 have also asserted their belief that because the photoplethysmogram is more complex than the invasive blood pressure curve, “an algorithm used in the photoplethysmogram should reflect this complexity.” Cannesson et al.12 recognized the need for the “automated calculation of ΔPOP.” Chandler et al.,26 in a study of children,32 also echo this sentiment: “To use this index clinically would require an automated method of calculation,” whereas Feissel et al.19 suggest standardizing the way in which the “data are recorded and analyzed.”

In the meta-analysis and review by Sandroni et al.,30 the authors summarize the potential issue in interpretation of results by various groups caused by device-dependent behaviors when they state that they cannot exclude the possibility that the “accuracy of the plethysmographic indices of fluid responsiveness as reported in our review could have been affected by filtering techniques” and “Moreover, the use of different filtering techniques in different devices may have represented an undetected source of heterogeneity in our meta-analysis.” In critiquing the work of Natalini et al.,4 Feldman55 states that the impact of the work is limited without complete knowledge of the manner in which the pleth is processed. He also points out that whether results can be applied to similar devices from other manufacturers depends on the details for the signal processing algorithms which are typically neither standardized nor reported by the manufacturer. Feldman55 argues that until manufacturers publish algorithm details and standards emerge, the “generalized knowledge” derived from using clinical monitors as a scientific instrument will not be generalizable.

There is therefore a clear consensus in the literature that using commercial oximeters for such studies comes with device-specific, inbuilt signal filtering which may not be optimal for the determination of a fluid responsiveness parameter in the research setting.


As stated in the Introduction, although there are a number of techniques,4,9,10 studies reporting the use of the pleth as a measure of fluid responsiveness are dominated by the ΔPOP and PVI parameters. PVI is a proprietary algorithm, and hence, there is very little information available concerning its underlying signal processing. A few authors have, however, directly compared the two measures24,26,34 with little difference in performance found between them in terms of prediction of fluid responsiveness or correlation with PPV. In addition, systematic reviews of the literature concerning studies using either ΔPOP or PVI have been undertaken by Antonsen and Kirkebøen29 and Sandroni et al.30 In the latter review, the meta-analysis performed by Sandroni et al.30 led to the conclusion that ΔPOP and PVI were equally effective in predicting fluid responsiveness in ventilated adult patients in sinus rhythm, whereas these plethysmographic indices probably have limited use in patients with spontaneous breathing activity.


The determination of a clinically useful physiological parameter is a distinctly nontrivial task. Aside from the core algorithm at the heart of the parameter (e.g., the ratio of ratios used in pulse oximetry for the determination of oxygen saturation, the determination of cyclical modulation for respiration rate, the modulation strength for fluid responsiveness, etc.), a sophisticated algorithmic infrastructure is required to take the raw biosignal, process it, present it to the core algorithm, and then apply further processing to the output to produce a value with the integrity necessary for display on the screen of a medical monitoring device.36 This algorithmic infrastructure generally includes a number of preprocessing and postprocessing code modules, as well as alarm management systems, hardware interface routines (including signal acquisition processes), and often involves thousands of lines of computer code. These advanced filtering and logical decision-making processes are inherent in all monitoring devices. The core computation embodying the basic method (e.g., ΔPOP of equation [1]) may, in fact, be a very small fraction (<1%) of the total lines of code. However, for the development of a fully automated algorithm capable of coping with the extremes of data characteristics in the clinical environment, significantly more processing is required.

Details of the signal processing methods implemented by the various groups are often scant and/or ambiguous. This may, in part, be driven by the clinical leanings of the authors and/or the intended audience. However, in general, investigators appear to understand the shortcomings of their signal processing methods but do not possess the means to overcome them: not least because they often do not have access to the raw signal, but instead start with a pleth which may not be optimally filtered for the task in hand and therefore not optimized for the extraction of the respiratory modulations in the pleth waveform. Often, the analysis is performed manually where a few optimal respiratory cycles are selected for interrogation. In these cases, the final results are generally good, showing significant correlation between ΔPOP and PPV. In fact, it is a technique always used by the author when first getting to grips with the analysis of a biosignal. Manual preprocessing often produces excellent results, which may significantly aid the rapid development of an automated method: it may also quickly highlight the futility of a proposed technique.

It is interesting to note that, in their review of the ΔPOP literature, Antonsen and Kirkebøen29 state that “…photoplethysmography shows best results in standardized conditions, during short registration periods, and in homogenous groups of pre- and postoperative patients.” Contrary to this, it is this author’s view that longer registration periods should be used to optimize results and that the finding of Antonsen and Kirkebøen29 is, in fact, a direct result of the difficulties experienced by research groups in implementing a fully automated robust algorithm. As a result, investigators resort to simply manually selecting the best data over short periods (from a consideration of longer segments). It is precisely this “hand-picking” of the quality data for analysis that is so difficult to automate fully. The author suspects that a lack of access to, and expertise with, proprietary technical systems required to produce a robust automated algorithm, with the associated pre- and postprocessing sophistication, is the reason that one has not yet appeared in the literature for the ΔPOP parameter. This may account for the inferior results reported for attempts to deal with large amounts of data which, in turn, lends itself to the erroneous conclusion that short registration periods provide better results.

One group (Hoiseth et al.25) has provided a few ad hoc examples of the ability to determine the stability of the computed ΔPOP parameter by examining longer-term signal segments (approximately 5 minutes), and another group (Hengy et al.28) advocated the removal of outliers by checking the difference between the latest point and the previous 6 values. This may be taken to its logical conclusion where longer signal segments, involving a significant number of respiratory cycles, are considered over which the calculated values are both subjected to the outlier rejection and smarter averaging. In summary, it is this author’s belief that the use of more signal information over significantly longer periods of time but with the emphasis on both the quality and temporal relevance of the information would appear the most fruitful way forward in the development of a robust, fully automated ΔPOP algorithm.


Name: Paul S. Addison, PhD.

Contribution: The author is the sole contributor to the manuscript.

Attestation: Paul S. Addison approves the final manuscript.

Conflicts of Interest: The author is an employee of Covidien (UK) Ltd.

This manuscript was handled by: Maxime Cannesson, MD, PhD.


1. Guyton AC, Hall JE Textbook of Medical Physiology. 201012th ed Philadelphia: Elsevier:110
2. Cannesson M, Aboy M, Hofer CK, Rehman M. Pulse pressure variation: where are we today? J Clin Monit Comput 2011;25:45–56
3. Preisman S, Kogan S, Berkenstadt H, Perel A. Predicting fluid responsiveness in patients undergoing cardiac surgery: functional haemodynamic parameters including the Respiratory Systolic Variation Test and static preload indicators. Br J Anaesth. 2005;95:746–55
4. Natalini G, Rosano A, Franceschetti ME, Facchetti P, Bernardini A. Variations in arterial blood pressure and photoplethysmography during mechanical ventilation. Anesth Analg. 2006;103:1182–8
5. Monnet X, Teboul JL. Assessment of volume responsiveness during mechanical ventilation: recent advances. Crit Care. 2013;17:R217
6. Michard F, Boussat S, Chemla D, Anguel N, Mercat A, Lecarpentier Y, Richard C, Pinsky MR, Teboul JL. Relation between respiratory changes in arterial pulse pressure and fluid responsiveness in septic patients with acute circulatory failure. Am J Respir Crit Care Med. 2000;162:134–8
7. Cannesson M, Besnard C, Durand PG, Bohé J, Jacques D. Relation between respiratory variations in pulse oximetry plethysmographic waveform amplitude and arterial pulse pressure in ventilated patients. Crit Care. 2005;9:R562–8
8. Michard F, Chemla D, Richard C, Wysocki M, Pinsky MR, Lecarpentier Y, Teboul JL. Clinical use of respiratory changes in arterial pulse pressure to monitor the hemodynamic effects of PEEP. Am J Respir Crit Care Med. 1999;159:935–9
9. Partridge BL. Use of pulse oximetry as a noninvasive indicator of intravascular volume status. J Clin Monit. 1987;3:263–8
10. Shamir M, Eidelman LA, Floman Y, Kaplan L, Pizov R. Pulse oximetry plethysmographic waveform during changes in blood volume. Br J Anaesth. 1999;82:178–81
11. Poli de Figueiredo LF, Silva E, Rocha e Silva M, Westphal GA, Caldeira M. Pulse oximetry wave respiratory variations for the assessment of volume status in patients under mechanical ventilation. Crit Care. Med 2004;32:A96
12. Cannesson M, Attof Y, Rosamel P, Desebbe O, Joseph P, Metton O, Bastien O, Lehot JJ. Respiratory variations in pulse oximetry plethysmographic waveform amplitude to predict fluid responsiveness in the operating room. Anesthesiology. 2007;106:1105–11
13. Cannesson M, Desebbe O, Hachemi M, Jacques D, Bastien O, Lehot JJ. Respiratory variations in pulse oximeter waveform amplitude are influenced by venous return in mechanically ventilated patients under general anaesthesia. Eur J Anaesthesiol. 2007;24:245–51
14. Cannesson M, Desebbe O, Lehot JJ. Comment on “Plethysmographic dynamic indices predict fluid responsiveness in septic ventilated patients” by Feissel et al. Intensive Care Med. 2007;33:1853
15. Cannesson M, Delannoy B, Morand A, Rosamel P, Attof Y, Bastien O, Lehot JJ. Does the Pleth variability index indicate the respiratory-induced variation in the plethysmogram and arterial pressure waveforms? Anesth Analg. 2008;106:1189–94
16. Cannesson M, Awad AA, Shelley K. Oscillations in the plethysmographic waveform amplitude: phenomenon hides behind artifacts. Anesthesiology. 2009;111:206–7
17. Cannesson M, Manach YL. Noninvasive hemodynamic monitoring: no high heels on the farm; no clogs to the opera. Anesthesiology. 2012;117:937–9
18. Feissel M, Teboul JL, Merlani P, Badie J, Faller JP, Bendjelid K. Plethysmographic dynamic indices predict fluid responsiveness in septic ventilated patients. Intensive Care Med. 2007;33:993–9
19. Feissel M, Teboul JL, Bendjelid K. Reply to the comment by Drs. Cannesson et al. Intensive Care Med. 2007;33:1854
20. Landsverk SA, Hoiseth LO, Kvandal P, Hisdal J, Skare O, Kirkeboen KA. Poor agreement between respiratory variations in pulse oximetry photoplethysmographic waveform amplitude and pulse pressure in intensive care unit patients. Anesthesiology. 2008;109:849–55
21. Landsverk SA, Hoiseth LO, Kirkeboen KA. Reply on oscillations in the plethysmographic waveform amplitude: phenomenon hides behind artifacts. Anaesthesiology. 2009;111:207–8
22. Westphal GA, Silva E, Gonçalves AR, Caldeira Filho M, Poli-de-Figueiredo LF. Pulse oximetry wave variation as a noninvasive tool to assess volume status in cardiac surgery. Clinics (Sao Paulo). 2009;64:337–43
23. Pizov R, Eden A, Bystritski D, Kalina E, Tamir A, Gelman S. Arterial and plethysmographic waveform analysis in anesthetized patients with hypovolemia. Anesthesiology. 2010;113:83–91
24. Pereira de Souza Neto E, Grousson S, Duflo F, Ducrex C, Joly H, Convert J, Mottolese C, Dailler F, Cannesson M. Predicting fluid responsiveness in mechanically ventilated children under general anaesthesia using dynamic parameters and transthoracic echocardiography. Br J Anaesth. 2011;106:856–64
25. Hoiseth L, Hoff IE, Skare O, Kirkeboen KA, Landsverk SA. Photoplethysmographic and pulse pressure variations during abdominal surgery. Acta Anaesthesiol Scand. 2011;55:1221–30
26. Chandler JR, Cooke E, Petersen C, Karlen W, Froese N, Lim J, Ansermino JM. Pulse oximeter plethysmograph variation and its relationship to the arterial waveform in mechanically ventilated children. J Clin Monit Comput. 2012;26:145–51
27. Chung E, Cannesson M. Commentary article on “Using non invasive dynamic parameters of fluid responsiveness in children: there is still much to learn”. J Clin Monit Comput. 2012;26:153–5
28. Hengy B, Gazon M, Schmitt Z, Benyoub K, Bonnet A, Viale JP, Aubrun F. Comparison between respiratory variations in pulse oximetry plethysmographic waveform amplitude and arterial pulse pressure during major abdominal surgery. Anesthesiology. 2012;117:973–80
29. Antonsen LP, Kirkebøen KA. Evaluation of fluid responsiveness: is photoplethysmography a noninvasive alternative? Anesthesiol Res Pract. 2012;2012:617380
30. Sandroni C, Cavallaro F, Marano C, Falcone C, De Santis P, Antonelli M. Accuracy of plethysmographic indices as predictors of fluid responsiveness in mechanically ventilated adults: a systematic review and meta-analysis. Intensive Care Med. 2012;38:1429–37
31. Desebbe O, Cannesson M. Using ventilation-induced plethysmographic variations to optimize patient fluid status. Curr Opin Anaesthesiol. 2008;21:772–8
32. Gan H, Cannesson M, Chandler JR, Ansermino JM. Predicting fluid responsiveness in children: a systematic review. Anesth Analg. 2013;117:1380–92
33. Natalini G, Rosano A, Taranto M, Faggian B, Vittorielli E, Bernardini A. Arterial versus plethysmographic dynamic indices to test responsiveness for testing fluid administration in hypotensive patients: a clinical trial. Anesth Analg. 2006;103:1478–84
34. Hoiseth LO, Hoff IE, Myre K, Landsverk A, Kirkeboen KA. Dynamic variables of fluid responsiveness during pneumoperitoneum and laparoscopic surgery. Acta Anaesthesiol Scand. 2012:777–86
35. Delerme S, Renault R, Le Manach Y, Lvovschi V, Bendahou M, Riou B, Ray P. Variations in pulse oximetry plethysmographic waveform amplitude induced by passive leg raising in spontaneously breathing volunteers. Am J Emerg Med. 2007;25:637–42
36. Addison PS, Watson JN, Mestek ML, Mecca RS. Developing an algorithm for pulse oximetry derived respiratory rate (RR(oxi)): a healthy volunteer study. J Clin Monit Comput. 2012;26:45–51
37. Alian AA, Shelley KH. Respiratory physiology and the impact of different modes of ventilation on the photoplethysmographic waveform. Sensors (Basel). 2012;12:2236–54
38. Allen J, Frame JR, Murray A. Microvascular blood flow and skin temperature changes in the fingers following a deep nspiratory gasp. Physiol Meas. 2002;23:365–73
39. Hamunen K, Kontinen V, Hakala E, Talke P, Paloheimo M, Kalso E. Effect of pain on autonomic nervous system indices derived from photoplethysmography in healthy volunteers. Br J Anaesth. 2012;108:838–44
40. Knorr-Chung BR, McGrath SP, Blike GT. Identifying airway obstructions using photoplethysmography (PPG). J Clin Monit Comput. 2008;22:95–101
41. Nilsson L, Johansson A, Kalman S. Macrocirculation is not the sole determinant of respiratory induced variations in the reflection mode photoplethysmographic signal. Physiol Meas. 2003;24:925–37
42. Reisner A, Shaltis PA, McCombie D, Asada HH. Utility of the photoplethysmogram in circulatory monitoring. Anesthesiology. 2008;108:950–8
43. Selvaraj N, Jaryal AK, Santhosh J, Deepak KK, Anand S. Influence of respiratory rate on the variability of blood volume pulse characteristics. J Med Eng Technol. 2009;33:370–5
44. Shah A, Shelley KH. Is pulse oximetry an essential tool or just another distraction? The role of the pulse oximeter in modern anesthesia care. J Clin Monit Comput. 2013;27:235–42
45. Shelley KH. Photoplethysmography: beyond the calculation of arterial oxygen saturation and heart rate. Anesth Analg. 2007;105:S31–6
46. Addison PS, Watson JN. Enhanced measurement of respiratory modulations using a flexible pulse oximeter probe. January 15–18, 2014 Orlando, FL Paper presented at: Society of Technology in Anesthesia, 2014 Annual Meeting
47. Awad AA, Ghobashy MA, Ouda W, Stout RG, Silverman DG, Shelley KH. Different responses of ear and finger pulse oximeter wave form to cold pressor test. Anesth Analg. 2001;92:1483–6
48. Awad AA, Stout RG, Ghobashy MA, Rezkanna HA, Silverman DG, Shelley KH. Analysis of the ear pulse oximeter waveform. J Clin Monit Comput. 2006;20:175–84
49. Desgranges FP, Desebbe O, Ghazouani A, Gilbert K, Keller G, Chiari P, Robin J, Bastien O, Lehot JJ, Cannesson M. Influence of the site of measurement on the ability of plethysmographic variability index to predict fluid responsiveness. Br J Anaesth. 2011;107:329–35
50. Jablonka DH, Awad AA, Stout RG, Silverman DG, Shelley KH. Comparing the effect of arginine vasopressin on ear and finger photoplethysmography. J Clin Anesth. 2008;20:90–3
51. Mou L, Gong Q, Wei W, Gao B. The analysis of transesophageal oxygen saturation photoplethysmography from different signal sources. J Clin Monit Comput. 2013;27:365–370
52. Nilsson L, Goscinski T, Kalman S, Lindberg LG, Johansson A. Combined photoplethysmographic monitoring of respiration rate and pulse: a comparison between different measurement sites in spontaneously breathing subjects. Acta Anaesthesiol Scand. 2007;51:1250–7
53. Shelley KH, Jablonka DH, Awad AA, Stout RG, Rezkanna H, Silverman DG. What is the best site for measuring the effect of ventilation on the pulse oximeter waveform? Anesth Analg. 2006;103:372–7
54. Shelley KH, Alian AA, Shelley AJ. Role of the photoplethysmographic waveform in the care of high-risk surgical patients. Anesthesiology. 2013;118:1479–1480
55. Feldman JM. Can clinical monitors be used as scientific instruments? Anesth Analg. 2006;103:1071–2
© 2014 International Anesthesia Research Society