INTRODUCTION
Correct estimation of intraoperative blood loss is a key challenge for clinicians. Apart from massive bleeding, the indication for transfusion is often determined by the hemoglobin (Hb) level, although this only provides an approximate indication of blood loss after sufficient volume substitution. Despite this, recommendations and guidelines for transfusion contain Hb levels as transfusion triggers (1). It is therefore of major importance to be able to estimate blood loss correctly intraoperatively. In clinical practice, blood loss is estimated visually by physicians (V-EBL) or gravimetrically (G-EBL). Contamination of sponges and canisters with other fluids (e.g., amniotic fluids) is difficult to detect and therefore leads to misjudgments of the actual volume of blood lost (2). It is known that the V-EBL method results in equally under- and overestimated values of blood loss (3). Smaller amounts of blood loss are known to be estimated more precisely than larger amounts (4–6). Blood collected in canisters is easier to estimate than blood-soaked sponges or clothes (7, 8). An innovative option for estimating blood loss is a smartphone application (Triton) developed by Gauss Surgical Inc (Los Altos, Calif). By capturing images of sponges, the application is able to calculate the estimated blood and Hb loss. The calculation of blood loss is based on the preoperative Hb level and is performed by colorimetric image correction and analysis (herein called “colorimetric estimated blood loss,” (C-EBL)). The application uses computer vision algorithms for the calculation of hemoglobin concentration. The used surgical sponges are subjected to photographic analysis. From the preprocessed image, the software extracts sets of geometric features and characteristics at the pixel level (9). Proprietary mathematical models, which cluster these characteristics to already known Hb mass values, are used. This algorithm takes special account of fluctuations in the saline solution, volume absorbed on a sponge acting as a diluent, and changes the appearance of sponge color (9). The device must be validated with a color QR code before each use at its location (Fig. 1). This ensures that different lighting conditions do not influence the measurement results. The application registers when it has been moved or when the lighting conditions have changed. Then a new validation is necessary and takes only a few seconds.
;)
Fig. 1: Colorimetric blood loss estimation. The application app is installed on a smartphone. The smartphone can be attached to infusion stands, for example. The device is validated by means of a color QR code (A and B) before use on site. Revalidation is necessary if the lighting conditions or location change. The diluted whole blood mixtures were applied to the sponges and measured with different techniques (C). The used sponges are held in front of the device (D). The device detects the color differences of the configured sponges and calculates the blood loss based on the algorithm and the preoperative hemoglobin value (E+F).
“The measurements were obtained under bright lighting conditions similar to conditions in the operating room (Median 882 Lux). Lighting conditions were measured with a luxmeter (TFA Dostmann LM37 Luxmeter, TFA Dostmann GmbH & Co. KG, Wertheim-Reicholzheim, Germany).” The application has already been tested in vitro(10, 11) and is capable of measuring Hb with acceptable accuracy in blood samples with a wide range of Hb levels and dilutions (9, 12). The application provides real-time information on blood loss and has the potential to improve the management of bleeding patients and hemotherapy (13–17). To our knowledge, there is no study that directly compares a colorimetric blood loss estimate to the existing clinical practice of visual and gravimetric blood loss estimation in a simulated scenario with known blood volumes as a reference.
PATIENTS AND METHODS
This study was approved by the local Ethics Committee at the University Hospital Frankfurt, Goethe University (Ref: 163/19). The study was previously registered on ClinicalTrials.gov (NCT04245995) and performed in accordance with the Declaration of Helsinki. The date of conductance was February 5, 2020. All participants provided written consent to participate.
The primary objective of this study was to evaluate the clinical applicability and accuracy of a novel mobile application using colorimetric image correction and analysis for blood loss estimation. Dilutions within were randomly selected in the present study (see supplemental material for an illustration of the method of preparation, https://links.lww.com/SHK/B96). Whole blood donations (provided by the German Red Cross, Institute of Transfusions Medicine, Goethe University, Frankfurt, Frankfurt, Germany) were diluted with full electrolyte solutions (Sterofundin ISO, B. Braun, Melsungen, Germany) in various concentrations (Hb level between 9.5 g/dL and 12.1 g/dL). The mixture of whole blood and fluids produced by dilution was defined as the reference blood loss (RBL). After the preparation of the RBL, an additional full electrolyte solution was added to simulate typical dilution effects (such as those caused by irrigation, infusion therapy or ascites (Hb level in sample blood volume: 4.9 g/dL–6.2 g/dL). Whole blood donations were treated with Citratephosphate-dextrose stabilizer solution in the blood donation service to inhibit blood clotting. No other anticoagulants were used. Only blood group system ABO-equivalent whole blood was used for the simulations. Hb concentration and hematocrit were measured by blood gas analysis (Radiometer ABL800 Flex, Radiometer GmbH, Krefeld, Germany) before and after dilution. White sponges (Curity Lap Sponge Sterile 18" × 18", Covidien, Dublin, Ireland) were used to simulate conditions in the operating room. The sponges were prepared with a predetermined volume of the blood samples. Four scenarios were created, and each scenario was independent of the others. The study participants estimated the blood loss visually for the sponges in each scenario and recorded their estimates in a case report form (CRF) within 1.5 min. A blinded member of the study team, who was unaware of the RBL in the scenarios, measured the blood loss in the sponges with the application (C-EBL) and separately with the gravimetric method (G-EBL). Using C-EBL and G-EBL, not only the blood volumes per scenario but also for each individual sponge were determined. This is technically easy to implement, but for physicians it is unusual in a clinical setting and cannot be implemented within the timeframe of the simulation study. Screenshots were taken of each individual measurement with the results of the application. The measurements were obtained under bright lighting conditions similar to conditions in the operating room (Median 882 Lux). Lighting conditions were measured with a luxmeter (TFA Dostmann LM37 Luxmeter, TFA Dostmann GmbH & Co. KG, Wertheim-Reicholzheim, Germany). Measurements sponges were carried out at the same location with consistent lightning conditions for comparability.
STATISTICS
Statistical analysis was performed using IBM SPSS Statistics (version 26, IBM, Armonk, NY) and R (version 3.5.1, R Foundation for Statistical Computing, Vienna, Austria).
The descriptive results for RBL, C-EBL, G-EBL, and V-EBL were expressed as the means (±standard deviations and/or ± standard errors) or medians (with interquartile ranges [IQRs]) as appropriate. Additionally, C-EBL, G-EBL, and V-EBL were analyzed for correlation with RBL by Spearman rank correlation coefficient Rho (with a P value of 0.05 or less considered statistically significant). Scatter plots (including the corresponding regression line) were performed to illustrate this. Furthermore, Bland-Altman analysis, which recognized a graphical method for comparing measurement methods, was applied. Additionally, histograms of the differences between the estimated and reference blood volumes were constructed.
RESULTS
A total of 53 participants (anesthesiologists) completed the scenarios to determine the V-EBL. All 53 CRFs were complete and hence valid for inclusion in the analysis. The majority of participants were anesthetic trainees (52%), followed by consultants (25%) and senior consultants (23%). The median clinical work experience was 3 years for the anesthetic trainees, 7 years for the consultants, and 15 years for the senior consultants.
Visually estimation
In the visual method an overestimation was common (86%). The V-EBL in the sponges per scenario correlated moderately with the reference blood volume (Spearman correlation coefficient, Rho = 0.52; P < 3.7×10−16). Scatterplots of visually estimated and reference values, as well as the corresponding linear regression lines (Sponges: regression coefficient = 1.85; P < 2×10−16) are shown (Fig. 2). In the Bland–Altman plot, the mean difference is displayed by the middle line, and the upper and lower 95% limits of agreement are shown by the upper and lower lines (Fig. 3). The median visual estimation of blood volume per scenario was 250 mL (IQR 150 mL–412.5 mL) (Table 1). Histograms of the differences between the estimated and reference values are presented in Figure 4. Overestimation by V-EBL reached up to 1,685% in the sponges per scenario. A median overestimation by 133.0 mL (IQR 33.0 mL–283.0 mL) in sponges per scenario was observed. The maximum underestimation in the sponges corresponded to 33% of the reference value.
Fig. 2: Scatter plots of estimated and reference values. Scatter plots of Colorimetric (C-EBL), Gravimetric (G-EBL), and Visually (V-EBL) estimated and reference (RBL) values, as well as the corresponding linear regression lines are presented. C-EBL indicates colorimetric estimated blood loss; G-EBL, gravimetrically estimated blood loss; RBL, reference blood loss; V-EBL, visually estimated blood loss.
Fig. 3: Bland–Altman plots of estimated and reference values. Bland–Altman plots of Colorimetric (C-EBL), Gravimetric (G-EBL), and Visually (V-EBL) estimated and reference (RBL) values are presented. The mean difference is displayed by the middle line, and the upper and lower 95% limits of agreement are shown by the upper and lower lines. C-EBL indicates colorimetric estimated blood loss; G-EBL, gravimetrically estimated blood loss; RBL, reference blood loss; V-EBL, visually estimated blood loss.
Table 1 -
Visual, gravimetric, and colorimetric methods compared with the reference volume
|
Method |
RBL |
V-EBL |
G-EBL |
C-EBL |
| Sponges per scenario (mL) |
Median (first quartil; third quartil)IQR |
103.0 (86; 162.8)76.8 |
250.0 (150; 412.5)262.5 |
135.5∗ (114; 189.5)74.5 |
134.0† (103.0; 205.5)102.5 |
|
mean (±SD ± SE) |
145.8 (± 90.4 ± 6.2) |
356.6 (± 313.9 ± 21.6) |
168.0 (±123 ± 61.5) |
174.5 (±126.2 ± 63.1) |
|
Rho; P value |
|
0.52; P < 3.7 × 10−16
|
Rho = 1; P = 0.083 |
Rho = 1; P = 0.083 |
| Single sponges (mL) |
Median (first quartil; third quartil)IQR |
21.5 (14.75; 33.0)IQR = 18.25 |
Not performed |
26.5 (15.25; 38)IQR = 22.75 |
25.0 (20; 35.5)IQR = 15.5 |
|
mean (±SD ± SE) |
22.4 (± 8.1 ± 1.6) |
Not performed |
25.9 (±11.9 ± 2.3) |
27.1 (± 10.3 ± 2.0) |
|
Rho; P value |
|
Not performed |
0.725; P = 2.784 × 10−05
|
0.774; P = 3.533×10−06
|
Results of visual (V-EBL), gravimetric (G-EBL), and colorimetric (C-EBL) methods compared with the reference volume (RBL) during simulated blood scenarios.
∗The values of the individual sponges were calculated per scenario.
†The system provides the total volume per scenario.C-EBL indicates colorimetric estimated blood loss; G-EBL, gravimetrically estimated blood loss; RBL, reference blood loss; V-EBL, visually estimated blood loss; IQR, interquartile range.
Fig. 4: Differences between estimated and reference values. Histograms of the differences between the Colorimetric (C-EBL), Gravimetric (G-EBL), and Visually (V-EBL) estimated and reference (RBL) values are presented. C-EBL indicates colorimetric estimated blood loss; G-EBL, gravimetrically estimated blood loss; RBL, reference blood loss; V-EBL, visually estimated blood loss.
Gravimetric estimation
Gravimetric blood loss estimation in single sponges correlated strongly with the RBL (Spearman correlation coefficient, Rho = 0.73; P = 2.78×10−05). The median estimated blood loss by the G-EBL method in the single sponges was 26.5 (IQR 15.3 mL–38 mL), while in the sponges per scenario it was 135.5 mL (IQR 114 mL–189.5 mL) (Table 1). Scatterplots of gravimetric estimated and reference values, as well as the corresponding linear regression lines (single sponges: regression coefficient = 1.15; P = 1.86×10−06) are shown (Fig. 2). Overestimation was observed in 58% of the single sponges, while underestimation was observed in 42%. A median overestimation by 4.0 mL (IQR −2.0 mL to 6.8 mL) in single sponges and 32.5 mL (IQR 10.8 mL–44.0 mL) in sponges per scenario was observed. The maximum underestimation for the sponges was 74% of the reference value.
Colorimetric estimation
The C-EBL measurements in the single sponges correlated strongly with the RBL (Spearman correlation coefficient, Rho = 0.77; P = 3.53×10−06). Scatterplots of colorimetric estimated and reference values, as well as the corresponding linear regression lines (single sponges: regression coefficient = 1.03; P = 6.23×10−07) are shown (Fig. 2). The median estimated blood loss by the C-EBL method for single sponges was 25 (IQR 20 mL–35.5 mL), while for the sponges per scenario it was 134.0 mL (IQR 103.0 mL–205.5 mL) (Table 1). Overestimation occurred in an average of 76% of the single sponges, while underestimation was observed in 23%. A median overestimation by 4 mL (IQR 1.3 mL–7.5 mL) in single sponges and 31 mL (IQR 17.0 mL–42.8 mL) per scenario was observed. The maximum underestimation for the sponges was 94% of the reference value.
DISCUSSION
Our study demonstrated a strong correlation between the blood loss estimated by the device and the reference blood volume in sponges. In comparison with the visual estimation and gravimetric methods, the colorimetric blood loss estimation with the application proved to be superior, with a correlation coefficient of 0.78 (C-EBL single sponges) versus 0.73 (G-EBL single sponges) and 0.52 (V-EBL sponges per scenario).
KEY FINDINGS
- 1. The use of visual estimation must be strongly discouraged, as it shows major discrepancies.
- 2. Gravimetric measurements provide more accurate results than visual estimation; however, the conversion from 1 mg to 1 mL of blood may result in inaccurate results, especially with high irrigation/ dilution, ascites, or amniotic fluid.
- 3. Colorimetric measurement provides the most accurate measurement results and provides real-time accurate results independent of dilution.
What are the challenges in blood loss estimation? Particularly for protracted bleeding situations, the realization of a relevant blood loss is necessary for early initiation of therapy to avoid hemodynamic instability and shock. The inaccuracy of visual estimation by clinicians is well known, but it is still the most common method of estimating blood loss in an intraoperative setting (18). One of the main reasons for this is the easy implementation and availability of this method. Many confounders increase the complexity of correct perioperative blood loss estimation, however. Regarding surgical aspects, blood loss falsely appears to be higher when blood is mixed with other fluids such as infusion solution, irrigation fluid, ascites, or amniotic fluid to the human eye. The lack of regular feedback on the real amount of blood loss may be an explanation for the missing learning effect. The use of visual tools such as pictograms or illustrated guidelines can help clinicians visually estimate blood loss contained in surgical sponges and other surgical materials more correctly (19–22). Education and continuous training can improve estimation and raise awareness of the error-proneness of visual estimation (7, 23–26). In our study, visual estimation correlated moderately with the reference blood volume. Especially in the estimation of blood volume in the sponges per scenario (total estimated blood loss from single sponges), a relevant overestimation was observed.
The quick and easy calculation of blood loss based on various formulas and laboratory parameters can provide an approximate classification of bleeding. However, a delayed reaction of laboratory and vital parameters can lead to an underestimation of acute blood loss. The assumption of normovolemia can also lead to a bias in the calculation. Therefore, these methods can only serve as an orientation in the intraoperative phase.
The gravimetric method involves weighing blood-contaminated objects and materials and calculating the blood loss by subtracting the dry weights. This allows better detection of blood loss than visual estimation. The conversion of 1 g to 1 mL of blood, which is frequently used in the literature and our study, is only an approximation of the actual value but can be helpful as a guide (26–28). In our simulation setting, we found a strong correlation between the gravimetric method and the reference blood volume in the sponges. However, gravimetric measurements are prone to errors due to other collected fluids, such as irrigation or other body fluids.
Currently, the indication for transfusion is often determined by the Hb level, in addition to massive bleeding, although this provides an inaccurate indication of blood loss only after sufficient volume substitution. Despite this fact, this limit is reflected in the recommendations or guidelines for transfusion, whereby the actual blood loss should have a greater impact on the decision to transfuse, taking into account the individual physiological transfusion trigger.
A possible implementation could be for the sponges to be scanned by the nurse directly through the system. The current calculated blood loss is displayed to the team. With the gravimetric method, the sponges must be weighed individually, and the blood volume recalculated each time. In everyday clinical practice, this is not feasible. The new system provides the anesthetists and surgeons with information about the current blood loss. In case of long-term operations with a high number of sponges used, the blood loss can be taken into account in real time during treatment. Real-time monitoring allows early measures of patient blood management such as optimization of coagulation management. Thus, unnecessary transfusions can be avoided. The authors see further advantages in the fact that the system does not deliver incorrect values due to high dilution as already shown in vitro(11). This is the case with gravimetric measurement. Further development of the system with automatic alarm when the predefined critical blood loss is exceeded seems to be reasonable. A connection in a new tablet-based anesthesia documentation system using artificial intelligence, which is currently being developed in our research group, should be the next step and considered for future applications.
Limitation
Since the simulation is artificial by nature, the Hawthorne effect must be considered. This effect describes the fact that participants change their behavior by participating as study subjects. A comparable situation cannot be created in the operating room given the need for a known volume of blood as a reference. Finally, the application is only suitable for measuring blood loss on white surgical sponges.
CONCLUSION
Although visual estimation is frequently used for blood loss, deviations in this measurement can be serious. The blood loss measured with the application correlated strongly with the reference blood volume. Currently, visual blood loss estimation is used for most operations. Colorimetric blood loss estimation provides reliable and easily accessible real-time information on blood loss. This could improve perioperative monitoring in an aspect where we have been rather limited so far. The main advantages of the system are the independence from dilution effects, which can have a great influence on the gravimetric measurement, and the fully automatic real-time calculation of blood loss.
Further research is needed to assess the clinical impact on transfusion strategies using a novel electronic blood loss estimation device.
Acknowledgments
The authors thank the anesthetists of the University Hospital Frankfurt for their support. The authors thank the donors of blood products that, although not suitable for transfusion, could ultimately serve a scientific purpose.
REFERENCES
1. Kozek-Langenecker SA, Ahmed AB, Afshari A, Albaladejo P, Aldecoa C, Barauskas G, De Robertis E, Faraoni D, Filipescu DC, Fries D, et al. Management of severe perioperative bleeding: guidelines from the European Society of Anaesthesiology: first update 2016.
Eur J Anaesthesiol 34 (6):332–395, 2017.
2. Meiser A. Quantification of blood loss. How precise is visual estimation and what does its accuracy depend on?
Der Anaesthesist 50:13–20, 2001.
3. Schorn MN. Measurement of blood loss: review of the literature.
J Midwifery Womens Health 55 (1):20–27, 2010.
4. Al-Kadri HM, Dahlawi H, Al Airan M, Elsherif E, Tawfeeq N, Mokhele Y, Brown D, Tamim HM. Effect of education and clinical assessment on the accuracy of post partum blood loss estimation.
BMC Pregnancy Childbirth 14:110, 2014.
5. Ram GG, Suresh P, Vijayaraghavan PV. Surgeons often underestimate the amount of blood loss in replacement surgeries.
Chin J Traumatol 17 (4):225–228, 2014.
6. Parayre I, Riviere O, Debost-Legrand A, Lemery D, Vendittelli F. Reliability of student midwives’ visual estimate of blood loss in the immediate postpartum period: a cross-sectional study.
Int J Nurs Stud 52 (12):1798–1803, 2015.
7. Mbachu II, Udigwe GO, Ezeama CO, Eleje GU, Eke AC. Effect of on-site training on the accuracy of blood loss estimation in a simulated obstetrics environment.
Int J Gynaecol Obstet 137 (3):345–349, 2017.
8. Buckland SS, Homer CS. Estimating blood loss after birth: using simulated clinical examples.
Women Birth 20 (2):85–88, 2007.
9. Sharareh B, Woolwine S, Satish S, Abraham P, Schwarzkopf R. Real time intraoperative monitoring of blood loss with a novel tablet application.
Open Orthop J 9:422–426, 2015.
10. Konig G, Waters JH, Javidroozi M, Philip B, Ting V, Abbi G, Hsieh E, Tully G, Adams G. Real-time evaluation of an image analysis system for monitoring surgical hemoglobin loss.
J Clin Monit Comput 32 (2):303–310, 2018.
11. Konig G, Holmes AA, Garcia R, Mendoza JM, Javidroozi M, Satish S, Waters JH. In vitro evaluation of a novel system for monitoring surgical hemoglobin loss.
Anesth Analg 119 (3):595–600, 2014.
12. Konig G, Waters JH, Hsieh E, Philip B, Ting V, Abbi G, Javidroozi M, Tully GW, Adams G. In vitro evaluation of a novel image processing device to estimate surgical blood loss in suction canisters.
Anesth Analg 126 (2):621–628, 2018.
13. Doctorvaladan SV, Jelks AT, Hsieh EW, Thurer RL, Zakowski MI, Lagrew DC. Accuracy of blood loss measurement during cesarean delivery.
AJP Rep 7 (2):e93–e100, 2017.
14. Thurer R. Accurate measurement of intraoperative blood loss improves prediction of postoperative hemoglobin levels.
J Anesthesia Clin Res 8 (7):1–7, 2017.
15. Bernal N. Accurate measurement of intraoperative blood loss during wound excision leads to more appropriate transfusion and reduced blood utilization.
J Anesthesia Clin Res 8 (11):1–6, 2017.
16. Holmes AA, Konig G, Ting V, Philip B, Puzio T, Satish S, Waters JH. Clinical evaluation of a novel system for monitoring surgical hemoglobin loss.
Anesth Analg 119 (3):588–594, 2014.
17. Rubenstein AF, Block M, Zamudio S, Douglas C, Sledge S, Tully G, Thurer RL. Accurate assessment of blood loss during cesarean delivery improves estimation of postoperative hemoglobin.
Am J Perinatol 36:434–439, 2018.
18. Piekarski F, Wunderer F, Raimann FJ, Neef V, Peuckert M, Brenner T, Grottke O, Grünewald M, Belt K, Renner J, Thal S, et al. Quantification of intraoperative blood loss: results of a multi-center survey and overview of current methods for the quantification of blood losses.
Anästhesiologie Intensivmedizin 61:110–116, 2020.
19. Merlin MA, Alter SM, Raffel B, Pryor PW 2nd. External blood loss estimation using the MAR method.
Am J Emerg Med 27 (9):1085–1090, 2009.
20. Cheerranichanunth P, Poolnoi P. Using blood loss pictogram for visual blood loss estimation in cesarean section.
J Med Assoc Thai 95 (4):550–556, 2012.
21. Yeung CY, Yim WW, Chan SY, Lo RSL, Leung LY, Hung KKC, Graham CA. Improvement of blood loss volume estimation by paramedics using a pictorial nomogram: a developmental study.
Injury 48 (12):2693–2698, 2017.
22. Wilcox L, Ramprasad C, Gutierrez A, Oden M, Richards-Kortum R, Sangi-Haghpeykar H, Gandhi M. Diagnosing postpartum hemorrhage: a new way to assess blood loss in a low-resource setting.
Matern Child Health J 21 (3):516–523, 2017.
23. Dildy GA 3rd, Paine AR, George NC, Velasco C. Estimating blood loss: can teaching significantly improve visual estimation?
Obstet Gynecol 104 (3):601–606, 2004.
24. Sukprasert M, Choktanasiri W, Ayudhya NI, Promsonthi P, Prasertsawat PO: Increase accuracy of visual estimation of blood loss from education programme.
J Med Assoc Thai 89 suppl 4:S54–59, 2006.
25. Toledo P, McCarthy RJ, Burke CA, Goetz K, Wong CA, Grobman WA. The effect of live and web-based education on the accuracy of blood-loss estimation in simulated obstetric scenarios.
Am J Obstet Gynecol 202 (4):400.e1–400.e5, 2010.
26. Khadilkar SS, Sood A, Ahire P. Quantification of peri-partum blood loss: training module and clot conversion factor.
J Obstet Gynaecol India 66: (suppl 1): 307–314, 2016.
27. Atukunda EC, Mugyenyi GR, Obua C, Atuhumuza EB, Musinguzi N, Tornes YF, Agaba AG, Siedner MJ. Measuring post-partum haemorrhage in low-resource settings: the diagnostic validity of weighed blood loss versus quantitative changes in hemoglobin.
PLoS One 11 (4):e0152408, 2016.
28. Ambardekar S, Shochet T, Bracken H, Coyaji K, Winikoff B. Calibrated delivery drape versus indirect gravimetric technique for the measurement of blood loss after delivery: a randomized trial.
BMC Pregnancy Childbirth 14:276, 2014.
