One of the most difficult tasks in clinical medicine is the assessment of intravascular volume status (1). This assessment is usually made by evaluating the patient's traditional vital signs, including heart rate (HR), blood pressure (BP), respiratory rate (RR), and oxygen saturation (SpO2), along with physical examination and laboratory studies. Unfortunately, traditional vital signs are notoriously unreliable due to the body's many compensatory mechanisms, which serve to maintain these vital signs during loss of up to 30% to 40% of total blood volume (2, 3). As a result, unrecognized volume loss during the compensatory phase of hemorrhage can quickly lead to poor tissue perfusion, progressive acidosis and sudden, unexpected hemodynamic decompensation.
Individual-specific changes in vital sign waveforms are introduced by the body's many compensatory mechanisms. These mechanisms work together during the early stages of hemorrhage to maintain hemodynamic stability, but unfortunately, cannot be seen or trended by physicians. In 2007, we were aware of prior studies showing that photoplethysmogram (PPG) waveforms change shape in the setting of acute volume loss (4–7). At that time, we hypothesized that the powerful data analytics techniques we originally developed for autonomous robot navigation in outdoor unstructured environments (8) could be used to analyze large, noisy physiological waveform datasets and continuously assess changes in central volume status in humans. Our approach was enabled by work at the US Army Institute of Surgical Research where a human model of acute blood loss to the point of decompensation was developed using lower body negative pressure (LBNP) (9). Our solution employed state-of-the-art computational methods (10–13) to find vital sign waveform alterations in the PPG that represent the physiological mechanisms of hemodynamic compensation. PPG waveforms were recorded from healthy volunteers during LBNP experiments, which redistribute blood volume from the upper body to the pelvis and lower extremities to simulate hemorrhage from normovolemia to decompensation (SBP <80 mm Hg). These waveform data were used to develop a “smart” algorithm, called the compensatory reserve index (CRI), that can identify small changes to hundreds of photoplethysmogram waveform features which correlate with intravascular volume loss, recognizable by computer technology but too subtle to appreciate or trend with the human eye (14, 15).
The total volume of blood loss an individual can tolerate before collapse varies depending on each person's unique ability to compensate. The CRI algorithm is designed to estimate an individual's current proportion of total tolerable volume loss, as defined by the following quantity: CRI = 1 − [BLV/BLVHDD]. BLV represents the current blood loss volume of the subject and BLVHDD is the blood loss volume at which the subject will experience hemodynamic decompensation (defined in LBNP experiments as a systolic blood pressure <80 mm Hg, loss of vision, or discomfort resulting in subject termination). CRI estimates this value by analyzing waveform features within a sliding 30-heartbeat window, and compares them to a library of waveform features originally obtained in the above LBNP experiments. By matching waveform features in the current subject to subjects from the LBNP experiments, CRI is able to determine when a patient will experience hemodynamic collapse in near real time. With beat-to-beat recalculation of CRI, this value can be trended over time. CRI values range from 1 to 0, and can be thought of as a percentage (100–0%) of physiologic reserve remaining, where “1” represents supine normovolemia and “0” implies hemodynamic decompensation (Fig. 1). Values between “1” and “0” indicate the proportion of reserve available to compensate for further volume loss.
The CRI algorithm has subsequently been validated in healthy volunteers donating one unit of blood, and undergoing stepwise removal and replacement of up 1.3 L of blood (16–18). In these studies, we showed that CRI could quickly, reliably, and noninvasively detect relatively small to moderate volumes of blood loss in healthy adults, whereas other vital sign parameters, including HR, systolic BP (SBP), cardiac output, and stroke volume, were unable to reliably detect the same volumes of blood loss. Further, when withdrawn blood was re-infused in study subjects, CRI increased and returned to baseline (18). These results prompted a prospective clinical trial to examine performance of the CRI algorithm in traumatically injured adolescent and adult patients. Our goal was to determine whether or not the CRI algorithm could reliably detect acute blood loss and continued blood loss in the setting of trauma. Doing so would validate that the CRI algorithm effectively monitors volume loss in patient populations subject to sympathetic stimuli including those experiencing injury and pain.
Institutional review boards
This study was conducted under a protocol reviewed and approved by the Colorado Multi-Institutional Review Board (COMIRB) and the Office of Research Protection (ORP), U.S. Army Medical Research and Materiel Command (USAMRMC). Clinical and waveform data collection were carried out under a waiver of consent. Waiver of consent was used for all patients, because the clinical conditions studied involved traumatic and potentially life-threatening injuries. Patients were considered unable to consent based on the nature of their injuries and family members were considered distressed. Consent was later obtained if possible from the subject and/or family members, once the patient was stabilized. If the patient's condition did not stabilize or a family member was not available, a post-card seeking consent was placed at the patient's bedside at the end of the data collection period.
Data collection and inclusion/exclusion criteria
Continuous noninvasive PPG waveform data collection from 50 acutely injured, possibly bleeding patients. Investigators were present in the Denver Health Emergency Department Thursday–Sunday from 7 PM to 5 AM to enroll consecutive eligible patients from October 17, 2013 to February 2, 2014. Eligible patients were 15 to 89 years of age admitted to the emergency department with evidence of blunt or penetrating trauma, and remained eligible if treated in the operating room or surgical intensive care unit. All enrolled patients were categorized as either an “alert” or “activation” according to established Denver Health Emergency Department criteria (Table 1).
Patients were considered ineligible if they were pregnant, objected to participation at any time, were or became incarcerated, or were transferred from the emergency department to the ward. Once a patient was determined eligible, an adhesive pulse-oximeter finger sensor (Nonin Medical Inc, Plymouth, MN) was placed on the patient's index, middle, or ring finger, on the side opposite the blood pressure cuff. The finger sensor was attached to a DataOx monitor (Flashback Technologies Inc, Boulder, CO). These small, lightweight data collection devices are composed of a Nonin OEM III pulse oximeter, processor, memory, Bluetooth radio and battery. DataOx devices continuously recorded and time stamped each patient's PPG waveforms. Due to the limited number of devices, patients were enrolled each night until all devices were in use; at that point no more patients could be enrolled until the next evening.
Data collection was carried out over a 24-h period, unless the patient was incarcerated or transferred out of the emergency department to the ward, at which point it was stopped. Demographic, clinical, and treatment information was prospectively collected in parallel with waveform data collection. Once waveform data collection was complete for each patient, the waveform data were off-loaded from each DataOx for later analysis. Recordings were retrospectively analyzed to generate CRI estimates using the CipherOx CRI system (V2.0.2, Flashback Technologies Inc). Demographic, clinical, and treatment information was entered into a RedCap database by hour of treatment. All data were available for each patient except where noted otherwise.
For data analysis purposes patients were categorized into three groups based on their estimated blood loss (EBL) as determined by the first author: active bleeding, defined as EBL >500 mL by the first author (known femur or pelvic fracture, known solid organ injury ≥grade 3, and/or intraoperative EBL of >500 mL); indeterminate bleeding (known fractures other than femur or pelvis, grade 1–2 solid organ injuries, intraoperative EBL <500 mL); or not actively bleeding (no evidence of bleeding, or minor bleeding <100 mL). Values between bleeding and non-bleeding patients were compared using two-tailed Student t test and Chi-square analysis. Receiver-operating characteristic area under the curve analysis was performed with identification of cut-off values that maximized the sensitivity and specificity for the identification of bleeding.
There were 50 patients prospectively enrolled, 3 of whom were excluded from data analysis due to incomplete data collection/device malfunction. Of the remaining 47 patients, there were 12 patients who were categorized as actively bleeding (estimated blood loss >500 mL), 3 who were indeterminate for bleeding (estimated blood loss between 100 mL and 500 mL), and 32 who were not actively bleeding (estimated blood loss <100 mL). For simplicity of comparison, the three patients who were classified as indeterminate were also excluded from further analysis. All further analysis is on the 44 patients who were classified as either bleeding or not bleeding. Demographics and clinical characteristics of these patients are detailed in Table 2. Actively bleeding patients tended to be younger than non-bleeding patients (27.5 ± 7 years vs. 35.9 ± 12 years, P = 0.03), had more long bone and pelvic fractures (33.3% vs. 0%, P = 0.001), more often had a positive FAST examination (41.7% vs. 3.1%, P = 0.001), more often had emergent operative intervention (91.7% vs. 15.6%, P <0.001), more often required mechanical ventilation (75% vs. 15.6%, P <0.001), more often were admitted to the intensive care unit (100% vs. 15.6%, P <0.001), and had longer length of stay in the intensive care unit (median length of stay 5.5 days vs. 0 days, P <0.001). Actively bleeding patients were also more likely to receive blood products in the first hour of treatment (25% vs. 3.1%, P = 0.03), and received a larger volume of crystalloid resuscitation (2.3L vs. 1.3L, P = 0.007) (Table 3).
Initial CRI versus traditional metrics of blood loss
The average initial CRI was calculated over the first 5 min of data collection and was compared to the prehospital HR and SBP since these values are used to triage patients, the initial HR and SBP in the trauma bay, the initial shock index (SI), and labs obtained within the first hour of admission including base deficit, lactate, hemoglobin, and hematocrit (Table 4). The average initial CRI for bleeding patients was 0.17 (95% CI 0.13–0.22), which was significantly lower compared with non-bleeding patients (0.56, 95% CI 0.49–0.62, P <0.001). The prehospital SBP, initial SBP, initial shock index, hemoglobin, and hematocrit were also significantly different between actively bleeding and non-bleeding patients (Table 4).
Receiver-operating characteristic area under the curve (ROC AUC) analysis was performed on each metric, using a cut-off value that maximized sensitivity and specificity for identifying actively bleeding patients. Using a threshold value of 0.21, the receiver-operating characteristic area under the curve for CRI was 0.97, yielding a sensitivity of 0.83 and a specificity of 0.97 for identifying acutely bleeding patients (Fig. 2). Alternatively, a threshold value of 0.37 had a sensitivity of 1.00, but a specificity of 0.81. CRI had the highest AUC of all metrics examined (Table 5). Metrics including lactate and initial heart rate also had relatively high sensitivity but had lower specificity compared with CRI. Metrics including prehospital HR, prehospital SBP, initial SBP, initial SI, base deficit, hemoglobin, and hematocrit all had specificity >90%, but all had lower sensitivity compared with CRI.
Changes in CRI versus fluid and blood administration
To test the ability of CRI to identify ongoing blood loss under fluid administration, the rate of change of CRI over the hour following fluid delivery was calculated for crystalloid and blood products for bleeding and non-bleeding patients. In actively bleeding patients, CRI on average decreased the hour after administration of each liter of crystalloid (CRI −0.02, 95% CI −0.05 to 0.01), and with each unit of blood product (CRI −0.04, 95% CI −0.08 to −0.01). In non-bleeding patients, CRI on average increased the hour after each liter of crystalloid (CRI 0.05, 95% CI −0.05 to 0.14), and after each unit of blood product (CRI 0.003, 95% CI −0.26 to 0.26). There was no statistically significant difference in CRI between bleeding and non-bleeding patients the hour after infusion of 1 L of crystalloid (P = 0.60). This value did, however, approach significance for the change in CRI the hour after infusion of one unit of blood product (P = 0.06). These data are graphically represented in Figure 3, demonstrating that the average change in CRI after fluid and blood administration had a narrower range in bleeding patients compared with non-bleeding patients.
To examine these trends over longer periods of time, we visually compared trends in CRI to fluid and blood product administration. Despite multiple factors that can alter the patient's compensatory factors (basal tone, heart rate, sedation, pain), it graphically appeared that CRI accurately reflected the patient's clinical volume status. Some examples of this are shown in Figures 4 and 5.
The patient in Figure 4 was taken to the operating room emergently for an anterior abdominal gunshot wound. After laparotomy, he was found to have no injuries and did not have more than minimal bleeding. CRI reflected his fluid responsiveness, increasing to normal values after 3 L of crystalloid. The patient's low initial CRI was likely a reflection of ethanol intoxication and dehydration.
The patient in Figure 5 presented with a gunshot wound to the left upper quadrant of the abdomen, and was also emergently taken to the operating room. The patient was found to have large volume hemoperitoneum, a left kidney injury requiring nephrectomy, and several enterotomies requiring bowel resection. The estimated blood loss from this procedure was 5 L. CRI increased significantly once the source of bleeding was identified and treated.
Humans have a number of survival compensatory mechanisms, which allow them to tolerate up to 30% to 40% of circulating blood volume loss before changes in traditional vital signs become frankly apparent. There is also substantial variation in the response to blood loss from patient to patient and from study to study, making it difficult to establish cut-off values (19). Traditional vital signs are also not specific to volume loss and can be abnormal for a variety of reasons (20). Thus, discrimination between bleeding and non-bleeding patients is difficult using these metrics alone. Physical examination findings, such as changes in skin turgor and capillary refill, are late findings and offer little to no diagnostic value in the assessment of acutely injured adults (21). Laboratory values such as hemoglobin, hematocrit, lactate, and base deficit can also be used as surrogates to assess circulating blood volume. The accuracy of hemoglobin and hematocrit is limited in patients who have received significant crystalloid resuscitation (22). While techniques have been described to discriminate been true anemia and hemodilution, they require additional tests using specific equipment (22). Base deficit is a rapidly and widely available serum laboratory marker of systemic acidosis that increases with hypoxemia and/or shock. In trauma settings the degree of base deficit correlates with blood transfusion requirement, risk of multiorgan failure, and mortality (23, 24). Unfortunately, base deficit is nonspecific and can rise due to any derangement causing metabolic acidosis, including, but not limited to intravascular volume loss. Serum lactate has similarly been used as a marker of acute blood loss (25), but is also nonspecific (26). Furthermore, intoxication and chronic alcohol abuse, which are relatively common among the traumatically injured, are known causes of lactic acidosis (27). These facts underscore the importance of identifying new physiological metrics, which can reliably detect volume loss in traumatically injured patients.
In this prospective clinical study, we collected continuous PPG waveform data from trauma patients from the time of arrival through resuscitation and/or operative intervention, and used this data to calculate a novel metric of hemodynamic reserve called CRI. CRI was then compared to traditional vital signs and the laboratory measures of volume loss referenced above. The change in CRI after fluid and blood administration was also examined. Results from this study support the primary hypothesis that CRI provides increased sensitivity and excellent specificity for the detection of acute blood loss. CRI decreased despite fluid and blood administration in patients who were actively bleeding, and increased in those who were not. CRI also increased with appropriate volume resuscitation once the source of bleeding was identified and stopped. Consistent with our previous observations and the observations of others, traditional vital signs and laboratory studies had limited utility during initial evaluation. Our secondary hypothesis, that injury and pain do not alter the fundamental features of the waveforms that were used to build the CRI algorithm, is also supported by our results. These findings are important, because the physiological response of an injured patient who experiences blood loss could arguably differ from that of the research subjects on which the algorithms were built.
Many new and sophisticated parameters have recently been developed for the evaluation of fluid status and fluid responsiveness (1, 28, 29). Static parameters have not performed well, and while dynamic parameters are better at predicting fluid responsiveness, their clinical use has not been widely accepted possibly because of difficultly in application and/or interpretation. Parameters such as stroke volume variation and pulse pressure variation are limited to the evaluation of mechanically ventilated patients and often require an invasive monitoring device. Ideal parameters for monitoring fluid status and responsiveness to fluids are noninvasive and can function in spontaneously breathing patients. Another example of one such parameter is near-infrared spectroscopy (NIRS), which utilizes the near-infrared light spectrum to penetrate several centimeters into human tissue (30). NIRS demonstrates decreased muscle oxygenation in central hypovolemia and can discriminate between patients at various levels of shock, but may still be inferior to traditional SBP (31).
A number of metrics for monitoring volume loss using the PPG signal have been proposed. Its ability to demonstrate the interaction between cardiac pulsation, arterial/venous pressure and peripheral vascular tone has led many to attempt to characterize subtle changes in the circulation, which are not otherwise apparent (32). Most studies have focused on the beat-to-beat variation of PPG waveforms. Various types of analyses of PPG waveform variability can detect small volume blood loss in spontaneously breathing patients without appreciable changes in HR or BP (6, 33). The pleth variability index (PVI) has been shown to predict fluid responsiveness in mechanically ventilated adult patients (34, 35). Unfortunately, PVI shows a considerable degree of inter-subject variability, limiting its use for distinguishing between hypovolemic and non-hypovolemic subjects based on a single measurement (32). Thus, like many other parameters used to detect hypovolemia, PVI must be trended over time to give meaningful information.
We believe that measuring the magnitude of the reserve to compensate for blood loss using our algorithm was more sensitive than other metrics likely because it reflected the integrated capacity of regulatory mechanisms to maintain adequate perfusion and oxygen delivery to the tissue. By discovering associations and understanding patterns within vital sign waveforms, current and future analytical tools will have the potential to improve care, save lives, and lower health care costs. These algorithms in turn will become the foundation for the next generation of algorithms, which will enable powerful compact models for estimation, prediction, and control of medical care. This study does have limitations, including the lack of prehospital waveform data, small sample size, and estimation of blood loss volumes. Despite these limitations, it did, however, demonstrate superior sensitivity when compared with traditional vital sign and laboratory values in predicting clinically significant blood loss. It also suggested that CRI could identify continued blood loss in the face of fluid resuscitation. These results support a growing body of literature suggesting that the CRI has promise as a clinically meaningful indicator of hemorrhage and a noninvasive method to monitor fluid resuscitation. A multicenter phase II clinical trial to examine real-time results from the CRI algorithm and its impact on clinical management and survival is warranted.
The authors thank the physicians and nurses at Denver Health Medical Center; their cooperation and assistance in waveform and clinical data collection were invaluable in carrying out this study.
1. Marik PE, Cavallazzi R, Vasu T, Hirani A. Dynamic changes in arterial waveform derived variables and fluid responsiveness in mechanically ventilated patients: a systematic review of the literature. Crit Care Med
2009; 37 9:2642–2647.
2. Brasel KJ, Guse C, Gentilello LM, Nirula R. Heart rate: is it truly a vital sign? J Trauma
2007; 62 4:812–817.
3. King RW, Plewa MC, Buderer NM, Knotts FB. Shock index as a marker for significant injury in trauma patients. Acad Emerg Med
1996; 3 11:1041–1045.
4. Partridge BL. Use of pulse oximetry as a noninvasive indicator of intravascular volume status. J Clin Monit
1987; 3 4:263–268.
5. Shamir M, Eidelman LA, Floman Y, Kaplan L, Pizov R. Pulse oximetry plethysmographic waveform during changes in blood volume. Br J Anaesth
1999; 82 2:178–181.
6. Gesquiere MJ, Awad AA, Silverman DG, Stout RG, Jablonka DH, Silverman TJ, Shelley KH. Impact of withdrawal of 450 ml of blood on respiration-induced oscillations of the ear plethysmographic waveform. J Clin Monit Comput
2007; 21 5:277–282.
7. McGrath SP, Ryan KL, Wendelken SM, Rickards CA, Convertino VA. Pulse oximeter plethysmographic waveform changes in awake, spontaneously breathing, hypovolemic volunteers. Anesth Analg
2011; 112 2:368–374.
8. Grudic GZ, Mulligan J. Outdoor Path Labeling Using Polynomial Mahalanobis Distance
. Proceedings: Robotics: Science and Systems II, August 2006. University of Pennsylvania, Philadelphia, PA.
9. Cooke WH, Ryan KL, Convertino VA. Lower body negative pressure as a model to study progression to acute hemorrhagic shock in humans. J Appl Physiol
1985; 96 4:1249–1261.
10. Strohmann T, Grudic G. A formulation for minimax probability machine regression. Advances in Neural Information Processing Systems 15
2003; Cambridge, MA: MIT Press, 769–776.
11. Strohmann T, Belitski A, Grudic G, DeCoste D. Thrun S, Saul LK, Schölkopf B. Sparse greedy minimax probability machine classification. MIT Press, Advances in Neural Information Processing Systems 16
. Cambridge, MA: 2004.
12. Bohte S, Breitenbach M, Grudic G. Nonparametric classification with polynomial MPMC cascades. International Conference on Machine Learning
13. Breitenbach M, Grudic G. Clustering through ranking on manifolds. International Conference on Machine Learning
14. Convertino VA, Moulton SL, Grudic GZ, Rickards CA, Hinojosa-Laborde C, Gerhardt RT, Blackbourne LH, Ryan KL. Use of advanced machine-learning techniques for non-invasive monitoring of hemorrhage. J Trauma
2011; 71 (1 Suppl):S25–S32.
15. Moulton SL, Mulligan J, Grudic GZ, Convertino VA. Running on empty? The compensatory reserve index. J Trauma Acute Care Surg
2013; 75 6:1053–1059.
16. Stewart CL, Mulligan J, Grudic GZ, Convertino VA, Moulton S. Detection of low volume blood loss: the compensatory reserve index versus traditional vital signs
. J Trauma Acute Care Surg
2014; 77 6:892–898.
17. Nadler R, Convertino VA, Gendler S, Lending G, Lipsky AM, Cardin S, Lowenthal A, Glassberg E. The value of noninvasive measurement of the compensatory reserve index in the monitoring and triage of patients experiencing minimal blood loss. Shock
2014; 42 2:93–98.
18. Convertino VA, Howard JT, Hinojosa-Laborde C, Cardin S, Batchelder P, Mulligan J, Grudic GZ, Moulton SL, MacLeod DB. Individual-specific beat-to-beat trending of significant human blood loss: the compensatory reserve. Shock
2015; 44 (Supp 1):27–32.
19. Pacagnella RC, Souza JP, Durocher J, Perel P, Blum J, Winikoff B, Gülmezoglu AM. A systematic review of the relationship between blood loss and clinical signs. PLoS One
2013; 8 3:e57594.
20. Tisherman SA, Barie P, Bokhari F, Bonadies J, Daley B, Diebel L, Eachempati SR, Kurek S, Luchette F, Carlos Puyana J, et al. Simon clinical practice guideline: endpoints of resuscitation. J Trauma
2004; 57 4:898–912.
21. McGee S, Abernathy WB, Simel DL. The rational clinical examination. Is this patient hypovolemic? JAMA
1999; 281 11:1022–1029.
22. Van PY, Riha GM, Cho SD, Underwood SJ, Hamilton GJ, Anderson R, Ham LB, Schreiber MA. Blood volume analysis can distinguish true anemia from hemodilution in critically ill patients. J Trauma
2011; 70 3:646–651.
23. Davis JW, Parks SN, Kaups KL, Gladen HE, O’Donnell-Nicol S. Admission base deficit predicts transfusion requirements and risk of complications. J Trauma
1996; 41 5:769–774.
24. Rutherford EJ, Morris JA Jr, Reed GW, Hall KS. Base deficit stratifies mortality and determines therapy. J Trauma
1992; 33 3:417–423.
25. Moomey CB Jr, Melton SM, Croce MA, Fabian TC, Proctor KG. Prognostic value of blood lactate, base deficit, and oxygen-derived variables in an LD50 model of penetrating trauma. Crit Care Med
1999; 27 1:154–161.
26. van der Beek A, de Meijer PH, Meinders AE. Lactic acidosis: pathophysiology, diagnosis and treatment. Neth J Med
2001; 58 3:128–136.
27. Allison MG, McCurdy MT. Alcoholic metabolic emergencies. Emerg Med Clin North Am
2014; 32 2:293–301.
28. Auler JO, Galas F, Hajjar L, Santos L, Carvalho T, Michard F. Online monitoring of pulse pressure variation to guide fluid therapy after cardiac surgery. Anesth Analg
2008; 106 4:1201–1206.
29. Benington S, Ferris P, Nirmalan M. Emerging trends in minimally invasive haemodynamic monitoring and optimization of fluid therapy. Eur J Anaesthesiol
2009; 26 11:893–905.
30. Scheeren TW, Schober P, Schwarte LA. Monitoring tissue oxygenation by near infrared spectroscopy (NIRS): background and current applications. J Clin Monit Comput
2012; 26 4:279–287.
31. Soller BR, Zou F, Ryan KL, Rickards CA, Ward K, Convertino VA. Lightweight noninvasive trauma monitor for early indication of central hypovolemia and tissue acidosis: a review. J Trauma Acute Care Surg
2012; 73 (2 Suppl 1):S106–S111.
32. Chan GSH, Middleton PM, Lovell NH. Photoplethysmographic variability analysis in critical care—current progress and future challenges. Conf Proc IEEE Eng Med Biol Soc
33. Middleton PM, Chan GS, O’Lone E, Steel E, Carroll R, Celler BG, Lovell NH. Spectral analysis of finger photoplethysmographic waveform variability in a model of mild to moderate haemorrhage. J Clin Monit Comput
2008; 22 5:343–353.
34. Cannesson M, Desebbe O, Rosamel R, Delannoy B, Robin J, Bastien O, Lehot JJ. Pleth variability index to monitor the respiratory variations in the pulse oximeter plethysmographic waveform amplitude and predict fluid responsiveness in the operating theatre. Br J Anesth
2008; 101 2:200–206.
35. 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 9:1429–1437.