Method for preoperative prediction of need for intraoperative hemotransfusion in planned abdominal surgery

Abstract

FIELD: anesthesiology and resuscitation.SUBSTANCE: patient gender, presence of coronary heart disease, presence of cancer, hemoglobin level, ASA class, characteristics of the planned surgical intervention are determined before surgery. The risk of intraoperative blood transfusion requirements are determined according to the original formula.EFFECT: method allows for preoperative prediction need for intraoperative blood transfusion in planned abdominal surgery.1 cl, 1 tbl, 2 ex

Description

[0001] Intraoperative blood transfusions, i.e. transfusion of one or more units of donor red blood cells directly during surgery, are relatively rare - in one study, the incidence was about 1.8% (Meier J, Filipescu D, Kozek-Langenecker S, Llau Pitarch J, Mallett S, Martus P, Matot I; ETPOS Collaborators. Intraoperative transfusion practices in Europe: the European Transfusion Practice and Outcome Study (ETPOS). British Journal of Anaesthesia. 2016;116(2):255–261. doi:10.1093 / bja / aev456), with 59% of all blood transfusions initiated not by laboratory parameters, but in response to physiological triggers (arterial hypotension - 55.4%, tachycardia - 30.7%). In a larger sample, which also included the immediate postoperative period, the proportion of patients who received at least one unit of packed red blood cells reached 25.44% (dos Santos Goiabeira L, Silva Meireles S, Silva Leocadio AS, et al.Intraoperative and immediate postoperative transfusion: clinical-hematological profile of transfused patients in a university hospital.Transfusion Clinique et Biologique. 2024;31(1):23–29. doi:10.1016 / j.tracli.2024.03.003).

[0002] The need for blood transfusion is a proven predictor of perioperative complications - even transfusion of just one or two units of packed red blood cells is associated with a 29% increased risk of 30-day mortality (OR 1.29; 95% CI 1.03-1.62) and an increased likelihood of respiratory, septic, wound, and thromboembolic complications compared with patients who did not receive blood transfusion (Glance LG, Dick AW, Mukamel DB, et al. Association between intraoperative blood transfusion and mortality and morbidity in patients undergoing noncardiac surgery. Anesthesiology. 2011;114(2):283–292. doi:10.1097 / ALN.0b013e318204cea1). Blood management programs based on preoperative correction of anemia, minimization of blood loss and a multidisciplinary approach demonstrate a reduction in transfusion volumes by 30–40%, a reduction in the duration of hospitalization and a decrease in the incidence of complications and mortality (Goodnough LT. Patient blood management as the standard of care. Hematology.American Society of Hematology Education Program. 2019;2019(1):583–589. doi:10.1182 / asheducation-2019.1.583). However, the implementation of such programs requires systems for predicting and identifying high-risk patients.

[0003] Modern analytical models can significantly improve the accuracy of predicting the need for intraoperative transfusions: for example, the TRANSFUSE model, developed and validated on a cohort of 816,618 patients, achieved an AUC of 0.93, outperforming existing risk assessment tools (Montefiore Einstein Medical Center TRANSFUSE Model Developers. A risk model to predict intraoperative blood transfusion: development and validation of the Transfusion Forecast Utility for Surgical Events (TRANSFUSE) score. JAMA Network Open. 2025;8(4):e2832906. doi:10.1001 / jamanetworkopen.2025.2832906). Other models developed using machine learning methods have also shown high accuracy in predicting blood transfusions in specific populations, identifying key predictors (operative time, D-dimer level, age) (Chen C, Wang Y, Yang X, et al. Machine learning–based prediction of intraoperative red blood cell transfusion in aortic valve replacement surgery. Clinical Laboratory.2024;70:xx–xx. doi:10.7754 / Clin.Lab.2023.230930). However, these methods are developed for specific surgical interventions.

[0004] Methods for forecasting intraoperative blood transfusion requirements allow for advance assessment of the required volumes of components across various groups, ensuring timely procurement planning and accurate inventory distribution, significantly reducing the risk of shortages in critical situations and overspending during overstocking. Accurate forecasts optimize blood management, reducing material and financial losses, lowering costs, and increasing the efficiency of blood transfusion services.

[0005] An analogue of the proposed method is the method of preoperative assessment of the risk of blood transfusion (Feng Y, Xu Z, Sun X, et al. (Feng Y, Xu Z, Sun X, Wang D, Yu Y. Machine learning for predicting preoperative red blood cell demand. Transfusion Medicine. 2021; 1–9. https: / / doi.org / 10.1111 / tme.12794). The authors conducted a retrospective analysis of 130,996 records of patients operated on in the period 2011–2017, studying 77 preoperative characteristics: demographics, laboratory test results, blood coagulation parameters, blood gases, surgical parameters, etc. To build the model, the data were randomly divided into 80% training cohort and 20% testing cohort, after which seven machine learning algorithms and classical logistic regression were compared.

[0006] The main disadvantages of the method:

[0007] 1. Retrospective single-center study, the model needs to be validated using data from other institutions.

[0008] 2. Does not exceed the experience of physicians in some clinical situations.

[0009] 3. Requires large amounts of data and computing resources.

[0010] 4. Low interpretability.

[0011] 5. Mixed surgical population

[0012] Another analogue of the method for assessing the risk of blood transfusion is the method proposed by Lee SM, Lee G, Kim TK, et al. (Lee SM, Lee G, Kim TK, Le T, Hao J, Jung YM, Park CW, Park JS, Jun JK, Lee HC, Kim D. Development and Validation of a Prediction Model for the Need for Massive Transfusion During Surgery Using Intraoperative Hemodynamic Monitoring Data. JAMA Netw Open. 2022 Dec 1;5(12):e2246637. doi: 10.1001 / jamanetworkopen.2022.46637. PMID: 36515949; PMCID: PMC9856486.). The model was developed using machine learning methods based on the analysis of data on 17,986 interventions; external validation was performed on 494 patients (AUROC was 0.943).

[0013] Factors used in assessing the risk of intraoperative blood transfusion include basic demographics and clinical status of the patient - age, gender, comorbidities (hypertension, diabetes mellitus, etc.), preoperative hemoglobin and other routine laboratory parameters, as well as dynamic (intraoperative) signals: area under the blood pressure curve in each cardiac cycle, hematocrit measured intraoperatively, hemoglobin oxygen saturation by pulse oximetry, ST-segment change on ECG.

[0014] Main disadvantages:

[0015] 1. Intraoperative parameters are required.

[0016] 2. Dependence on high-quality signals from the monitor.

[0017] 3. Infrastructure requirements: software installation and a stable high-frequency connection between the monitors and the server are required.

[0018] 4. Single- and two-center validation.

[0019] 5. Transfusion type limitation: Only massive red blood cell transfusion (≥3 units) is predicted.

[0020] 6. Risk of "signal fatigue": Frequent alerts at a low threshold may lead to the signal being ignored.

[0021] The closest analogue of the proposed method is the method developed by Alonso-Tuñón O, Bertomeu-Cornejo M, Castillo-Cantero I, et al. (Alonso-Tuñón O, Bertomeu-Cornejo M, Castillo-Cantero I, Borrego-Domínguez JM, García-Cabrera E, Bejar-Prado L, Vilches-Arenas A. Development of a Novel Prediction Model for Red Blood Cell Transfusion Risk in Cardiac Surgery. Journal of Clinical Medicine. 2023; 12(16):5345. https: / / doi.org / 10.3390 / jcm12165345). The authors conducted a retrospective cohort study of 1,234 adult patients who underwent elective cardiac surgery between 2017 and 2019 at a single center. Of these, 59.4% received at least one unit of packed red blood cells. More than 20 potential factors were considered, including demographics, anthropometric parameters, comorbidities, the EuroSCORE I score, surgical parameters (type of procedure, cardiopulmonary bypass time, complications), preoperative hemoglobin, and others.

[0022] The final model included five readily available predictors:

[0023] Preoperative hemoglobin category (<11; 11–11.9; 12–12.9; 13–13.9; ≥14 g / dL)

[0024] Operation type (isolated or combined)

[0025] BMI ≥ 30 kg / m 2

[0026] Gender (women)

[0027] Age ≥ 60 years old.

[0028] The discriminatory ability of the model was assessed by the area under the ROC curve (AUC = 0.809; 95% CI 0.785–0.833; p<0.001).

[0029] Main disadvantages

[0030] 1. Single-center retrospective study - limited external validity: the model needs to be confirmed in other cohort data.

[0031] 2. Limited discrimination – AUC 0.81 gives incomplete confidence in the prediction for borderline cases.

[0032] 3. Categorical processing of hemoglobin - separation by hemoglobin levels.

[0033] 4. Designed for cardiac surgery only - cannot be extrapolated to abdominal surgery.

[0034] TASK: to improve the accuracy of predicting the need for intraoperative blood transfusion in planned abdominal surgery.

[0035] Technical result – the proposed method will make it possible to identify patients with a high risk of intraoperative blood transfusion during abdominal operations by determining preoperative risk factors that reliably influence the occurrence of this unfavorable outcome, which is necessary for optimizing the perioperative period in these patients.

[0036] The essence of the proposed method is to determine before surgery the patient’s gender (F1), the presence of coronary heart disease (CHD) (F2), oncological disease (F3), hemoglobin level (F4), ASA class (F5 and F6), characteristics of the planned surgical intervention (F7, F8 and F9), and then determine the risk of developing the need for intraoperative blood transfusion according to the formula: K = 0.126+0.354×F1−0.419×F2+0.984×F3−F4×0.041+0.711×F5+0.732×F6−1.331×F7+1.142×F8−1.132×F9, where F1 is taken as 1 if the gender is male and as 0 if female; F2 is taken as 1 if there is coronary heart disease, and as 0 if not; F3 is taken as 1 if there is oncological disease, and as 0 if not; F4 is taken equal to the hemoglobin level in g / L; F5 is taken equal to 1 if the patient corresponds to ASA class 2, and equal to 0 if to another class; F6 is taken equal to 1 if the patient corresponds to ASA class 3, and equal to 0 if to another class;F7 is taken equal to 1 if the type of planned surgical intervention is laparoscopic surgery, and equal to 0 if the type of surgery is another; F8 is taken equal to 1 if the type of planned surgical intervention is open surgery on the organs of the upper abdominal cavity and equal to 0 if the type of surgery is another; and F9 is taken equal to 1 if the type of planned surgical intervention is surgery on the abdominal wall and equal to 0 if the type of surgery is another, and if the value of K ≥ -3.66, the patient is classified as a high-risk group for the need for intraoperative blood transfusion, if the value of K is less than -3.66 - as a low-risk group.

[0037] The proposed method is justified as follows. A total of 11,478 patients undergoing abdominal surgery with an initial physical status of ASA classes 1-3 were examined. Preoperative factors were recorded: patient gender, hemoglobin level in g / L, the presence of coronary artery disease, and oncological disease. The type of planned surgical intervention (laparoscopic surgery, open surgery on lower abdominal organs, open surgery on upper abdominal organs, abdominal wall surgery, gynecological surgery, urological surgery, or surgery on abdominal vessels) and physical status according to the ASA classification (class 1, 2, or 3) were determined. Blood transfusions were recorded during anesthesia.

[0038] Logistic regression analysis showed that the type of planned surgery, patient gender, hemoglobin level in g / L, the presence of coronary artery disease, cancer, and physical status according to the ASA classification statistically significantly affect the risk of needing intraoperative blood transfusion (Table 1).

[0039] Table 1

[0040] Logistic Regression Equation for the Model

[0041] Factor β OR p Gender: Male 0,354 1,425 0,0047 Hemoglobin (g / L) -0,041 0,96 0,0 Oncological disease 0,984 2,675 0,0 coronary heart disease -0,419 0,658 0,012 ASA III 0,732 2,08 0,0484 ASA II 0,711 2,036 0,0493 Open operations of the upper abdominal cavity 1,142 3,133 0,0 Laparoscopic surgery -1,331 0,264 0,0 Abdominal wall surgeries -1,132 0,323 0,0052

[0042] K = 0.126+0.354×F1−0.419×F2+0.984×F3−F4×0.041+0.711×F5+0.732×F6−1.331×F7+1.142×F8−1.132×F9

[0043] The numerical coefficients were identified and substantiated through logistic regression analysis and ROC analysis. F1– patient’s gender (taken as 1 if the gender is male and as 0 if female), F2– presence of coronary heart disease (taken as 1 if present and as 0 if absent), F3– presence of oncological disease (taken as 1 if present and as 0 if absent), F4– hemoglobin level in g / l, F5– physical status corresponding to ASA class 2 (taken as 1 if the physical status corresponds to ASA class 2 and as 0 if it does not correspond), F6– physical status corresponding to ASA class 3 (taken as 1 if the physical status corresponds to ASA class 3 and as 0 if it does not correspond), F7– laparoscopic surgery (taken as 1 if the planned surgery is laparoscopic and as 0 if it is other types of surgical intervention), F8– open surgery on the organs of the upper abdominal cavity (taken as for 1,if the planned operation is an open operation on the organs of the upper abdominal cavity, and 0 if it is other types of surgical intervention), F9 - surgery on the anterior abdominal wall (taken as 1 if the planned operation is an operation on the anterior abdominal wall, and 0 if it is other types of surgical intervention),

[0044] If the K value is greater than or equal to -3.66, the patient is classified as a high-risk group requiring intraoperative blood transfusion; if the K value is less than -3.66, the patient is classified as a low-risk group.

[0045] The conducted ROC analysis confirmed the good prognostic value of the developed method (the area under the ROC curve was 0.894). The sensitivity and specificity for the obtained cutoff point were 87.0% and 77.4%, respectively.

[0046] The proposed method can be used in practical healthcare, does not require special conditions for implementation, is time-saving, and easy to implement.

[0047] 1. A 67-year-old man with gastric cancer, anemia (hemoglobin = 90 g / L), and concomitant chronic renal failure. Open right hemicolectomy is planned. ASA class III.

[0048] The obtained values ​​are substituted into the developed equation:

[0049] K = 0.126+0.354×1−0×0.419+0.984×1−90×0.041+0+0.732−1.331×0+1.142×F8−1.132×0=−0.352

[0050] Because K is greater than -3.66, the patient is classified as high-risk. According to the model, the risk of requiring intraoperative blood transfusion is 41%. Two doses of red blood cell suspension were administered during the surgery.

[0051] 2. A 42-year-old woman with no serious underlying medical conditions, Hb = 140 g / L, is scheduled for laparoscopic cholecystectomy for chronic calculous cholecystitis.

[0052] The obtained values ​​are substituted into the developed equation:

[0053] K = 0.126 + 0.354 × 0 - 0 × 0.419 + 0.984 × 0 - 140 × 0.041 + 0.711 × 0 + 0.732 × 0 - 1.331 × 1 + 1.142 × 0 - 1.132 × 0 = - 6.945

[0054] Since K is less than -3.66, the patient is classified as low-risk. According to the model, the probability of blood transfusion is ≈ 0.1%. Blood transfusion was not required.

[0055] Thus, the use of the proposed method allows predicting the need for intraoperative blood transfusion.