Evaluation of risk scores in predicting perioperative blood transfusions in adult cardiac surgery

Predicting the Need for Blood Transfusions in Cardiac Surgery: A Comparison between Machine Learning Algorithms and Established Risk Scores in the Brazilian Population

INTRODUCTION

Blood transfusion is a common practice in cardiac surgery, despite its well-known negative effects. To mitigate blood transfusion-associated risks, identifying patients who are at higher risk of needing this procedure is crucial. Widely used risk scores to predict the need for blood transfusions have yielded unsatisfactory results when validated for the Brazilian population.

METHODS

In this retrospective study, machine learning (ML) algorithms were compared to predict the need for blood transfusions in a cohort of 495 cardiac surgery patients treated at a Brazilian reference service between 2019 and 2021. The performance of the models was evaluated using various metrics, including the area under the curve (AUC), and compared to the commonly used Transfusion Risk and Clinical Knowledge (TRACK) and Transfusion Risk Understanding Scoring Tool (TRUST) scoring systems.

RESULTS

The study found that the model had the highest performance, achieving an AUC of 0.7350 (confidence interval [CI]: 0.7203 to 0.7497). Importantly, all ML algorithms performed significantly better than the commonly used TRACK and TRUST scoring systems. TRACK had an AUC of 0.6757 (CI: 0.6609 to 0.6906), while TRUST had an AUC of 0.6622 (CI: 0.6473 to 0.6906).

CONCLUSION

The findings of this study suggest that ML algorithms may offer a more accurate prediction of the need for blood transfusions than the traditional scoring systems and could enhance the accuracy of predicting blood transfusion requirements in cardiac surgery patients. Further research could focus on optimizing and refining ML algorithms to improve their accuracy and make them more suitable for clinical use.

Development of a Novel Prediction Model for Red Blood Cell Transfusion Risk in Cardiac Surgery

BACKGROUND

Cardiac surgery is a complex and invasive procedure that often requires blood transfusions to replace the blood lost during surgery. Blood products are a scarce and expensive resource. Therefore, it is essential to develop a standardized approach to determine the need for blood transfusions in cardiac surgery. The main objective of our study is to develop a simple prediction model for determining the risk of red blood cell transfusion in cardiac surgery.

METHODS

Retrospective cohorts of adult patients who underwent cardiac surgery between 2017 and 2019 were studied to identify hypothetical predictors of blood transfusion. Finally, a multivariable logistic regression model was developed to predict the risk of transfusion in cardiac surgery using the AUC and the Hosmer-Lemeshow goodness-of-fit test.

RESULTS

We included 1234 patients who underwent cardiac surgery. Of the entire cohort, 875 patients underwent a cardiac procedure 69.4% [CI 95% (66.8%; 72.0%)]; 119 patients 9.6% [CI 95% (8.1%; 11.4%)] underwent a combined procedure, and 258 patients 20.9% [CI 95% (18.7; 23.2)] underwent other cardiac procedures. The median perioperative hemoglobin was 13.0 mg/dL IQR (11.7; 14.2). The factors associated with the risk of transfusion were age > 60 years OR 1.37 CI 95% (1.02; 1.83); sex female OR 1.67 CI 95% (1.24; 2.24); BMI > 30 OR 1.46 (1.10; 1.93); perioperative hemoglobin < 14 OR 2.11 to 51.41 and combined surgery OR 3.97 CI 95% (2.19; 7.17). The final model shows an AUC of 80.9% for the transfusion risk prediction [IC 95% (78.5-83.3%)]; p < 0.001].

CONCLUSIONS

We have developed a model with good discriminatory ability, which is more parsimonious and efficient than other models.

A transfusion risk stratification score to facilitate quality management in cardiopulmonary bypass

BACKGROUND

Our previous showed that a blood management program in the cardiopulmonary bypass (CPB) department, reduced red blood cell (RBC) transfusion and complications, but assessing transfusion practice solely based on transfusion rates was insufficient. This study aimed to design a risk stratification score to predict perioperative RBC transfusion to guide targeted measures for on-pump cardiac surgery patients.

STUDY DESIGN AND METHODS

We analyzed data from 42,435 adult cardiac patients. Eight predictors were entered into the final model including age, sex, anemia, New York Heart Association classification, body surface area, cardiac surgery history, emergency surgery, and surgery type. We then simplified the score to an integer-based system. The area under the receiver operating characteristic curve (AUC), Hosmer-Lemeshow goodness-of-fit test, and a calibration curve were used for its performance test. The score was compared to existing scores.

RESULTS

The final score included eight predictors. The AUC for the model was 0.77 (95% CI, 0.76-0.77) and 0.77 (95% CI, 0.76-0.78) in the training and test set, respectively. The calibration curves showed a good fit. The risk score was finally grouped into low-risk (score of 0-13 points), medium-risk (14-19 points), and high-risk (more than 19 points). The score had better predictive power compared to the other two existing risk scores.

DISCUSSION

We developed an effective risk stratification score with eight variables to predict perioperative RBC transfusion for on-pump cardiac surgery. It assists perfusionists in proactively preparing blood conservation measures for high-risk patients before surgery.

Systematic review highlights high risk of bias of clinical prediction models for blood transfusion in patients undergoing elective surgery

BACKGROUND

Blood transfusion can be a lifesaving intervention after perioperative blood loss. Many prediction models have been developed to identify patients most likely to require blood transfusion during elective surgery, but it is unclear whether any are suitable for clinical practice.

STUDY DESIGN AND SETTING

We conducted a systematic review, searching MEDLINE, Embase, PubMed, The Cochrane Library, Transfusion Evidence Library, Scopus, and Web of Science databases for studies reporting the development or validation of a blood transfusion prediction model in elective surgery patients between January 1, 2000 and June 30, 2021. We extracted study characteristics, discrimination performance (c-statistics) of final models, and data, which we used to perform risk of bias assessment using the Prediction model risk of bias assessment tool (PROBAST).

RESULTS

We reviewed 66 studies (72 developed and 48 externally validated models). Pooled c-statistics of externally validated models ranged from 0.67 to 0.78. Most developed and validated models were at high risk of bias due to handling of predictors, validation methods, and too small sample sizes.

CONCLUSION

Most blood transfusion prediction models are at high risk of bias and suffer from poor reporting and methodological quality, which must be addressed before they can be safely used in clinical practice.

Machine learning-based prediction of massive perioperative allogeneic blood transfusion in cardiac surgery

BACKGROUND

Massive perioperative allogeneic blood transfusion, that is, perioperative transfusion of more than 10 units of packed red blood cells (pRBC), is one of the main contributors to perioperative morbidity and mortality in cardiac surgery. Prediction of perioperative blood transfusion might enable preemptive treatment strategies to reduce risk and improve patient outcomes while reducing resource utilisation. We, therefore, investigated the precision of five different machine learning algorithms to predict the occurrence of massive perioperative allogeneic blood transfusion in cardiac surgery at our centre.

OBJECTIVE

Is it possible to predict massive perioperative allogeneic blood transfusion using machine learning?

DESIGN

Retrospective, observational study.

SETTING

Single adult cardiac surgery centre in Austria between 01 January 2010 and 31 December 2019.

PATIENTS

Patients undergoing cardiac surgery.

MAIN OUTCOME MEASURES

Primary outcome measures were the number of patients receiving at least 10 units pRBC, the area under the curve for the receiver operating characteristics curve, the F1 score, and the negative-predictive (NPV) and positive-predictive values (PPV) of the five machine learning algorithms used to predict massive perioperative allogeneic blood transfusion.

RESULTS

A total of 3782 (1124 female:) patients were enrolled and 139 received at least 10 pRBC units. Using all features available at hospital admission, massive perioperative allogeneic blood transfusion could be excluded rather accurately. The best area under the curve was achieved by Random Forests: 0.810 (0.76 to 0.86) with high NPV of 0.99). This was still true using only the eight most important features [area under the curve 0.800 (0.75 to 0.85)].

CONCLUSION

Machine learning models may provide clinical decision support as to which patients to focus on for perioperative preventive treatment in order to preemptively reduce massive perioperative allogeneic blood transfusion by predicting, which patients are not at risk.

TRIAL REGISTRATION

Johannes Kepler University Ethics Committee Study Number 1091/2021, Clinicaltrials.gov identifier NCT04856618.

The association of thrombin generation with bleeding outcomes in cardiac surgery: a prospective observational study

BACKGROUND

Cardiac surgery with cardiopulmonary bypass (CPB) is associated with coagulopathic bleeding. Impaired thrombin generation may be an important cause of coagulopathic bleeding but is poorly measured by existing hemostatic assays. We examined thrombin generation during cardiac surgery, using calibrated automated thrombography, and its association with bleeding outcomes.

METHODS

We conducted a prospective observational study in 100 patients undergoing cardiac surgery with CPB. Calibrated automated thrombography parameters were expressed as a ratio of post-CPB values divided by pre-CPB values. The association of thrombin generation parameters for bleeding outcomes was compared with conventional tests of hemostasis, and the outcomes of patients with the most severe post-CPB impairment in thrombin generation (≥ 80% drop from baseline) were compared with the rest of the cohort.

RESULTS

All 100 patients were included in the final analysis, with a mean age of 63 (12) yr, 31 (31%) female, and 94 (94%) undergoing bypass and/or valve surgery. Post-CPB, peak thrombin decreased by a median of 73% (interquartile range [IQR], 49-91%) (P < 0.001) and total thrombin generation, expressed as the endogenous thrombin potential (ETP), decreased 56% [IQR, 30-83%] (P < 0.001). In patients with ≥ 80% decrease in ETP, 21% required re-exploration for bleeding compared with 7% in the rest of the cohort (P = 0.04), and 48% required medical or surgical treatment for hemostasis compared with 27% in the rest of the cohort (P = 0.04).

CONCLUSIONS

Thrombin generation is significantly impaired by CPB and associated with higher bleeding severity. Clinical studies aimed at the identification and treatment of patients with impaired thrombin generation are warranted.

Transfusion requirement prediction score for patients undergoing cardiac surgery: An experience from a tertiary care set-up from South India

BACKGROUND

Prediction of transfusion requirement is part of preoperative management in a surgical case. We aimed to develop one such tool for patients undergoing cardiac surgery.

METHODS

A retrospective study for a period of 3 years was done to develop the scoring tool, Transfusion Requirement Prediction Score for Cardiac Surgery (TRPS), and internal validation was done prospectively. The primary outcome was administration of allogenic red cell units to the patients during perioperative period. The outcome is dichotomized as controls and cases based on the number of Red Blood Cell units received. Independent variables were chosen based on statistical significance and clinical judgement. Receiver operating characteristic curve was used to obtain the cut-off for each independent variable, odds ratio, and regression coefficients were used to assign the score. All patients with a cumulative score below the cut-off value were categorised as 'low risk' and above the cut off as 'high risk' group.

RESULTS

During the study period, out of 602 patients, 345 met the inclusion criteria (controls: 175; cases: 170). Six variables such as age (more than 58 years), gender (female), bypass time (more than 148 min), haemoglobin (less than 12.5 g/dL), ejection fraction (less than 57%), and history of warfarin prophylaxis were chosen to develop the score. The total score value of 5 was chosen as the cut-off for the two risk groups. It predicted blood utilisation with a strength of 68% sensitivity and 79% specificity. On internal validation, the score was observed to have an accuracy of 70%.

CONCLUSION

The TRPS is a simple reliable and handy tool with high accuracy.

Bleeding risk stratification in coronary artery surgery: the should-not-bleed score

BACKGROUND

An estimated 20% of allogeneic blood transfusions in the United States are associated with cardiac surgery. It is estimated that 11% of red cell resources were used for transfusion support of patients undergoing coronary artery bypass grafting (CABG) with a documented wide variability in transfusion rate (7.8 to 92.8%). To address the issue of unnecessary transfusions within the CABG population, we developed a model to predict which patients are at low risk of bleeding for whom transfusion treatment might be considered unnecessary. Herein we present our "SHOULD-NOT-BLEED-SCORE" application developed for the Windows® software platform which is based on our previous research.

METHODS

This study is aimed to develop a user-friendly application that stratifies patients with respect to bleeding risk. The statistical model we used in our previous research was focused on detection of CABG patients at low risk of bleeding. The rationale behind such an approach was to identify a CABG patient subgroup at low risk of bleeding. By identifying patients at low risk of bleeding we can define a subgroup of patients for whom transfusion treatment might be considered unnecessary. We developed a Windows platform application based on risk modelling which we previously calculated for 1426 patients undergoing elective CABG from January 2010 to January 2018.

RESULTS

The SHOULD-NOT-BLEED-SCORE risk score is developed for the Windows software platform. A mathematical model that is based on multivariate analysis was used for app development. The variables that entered the scoring system were: Age; Body Mass Index; Chronic Renal Failure; Preoperative Clopidogrel Exposure; Preoperative Red Blood Cells Count; Preoperative Fibrinogen Level; Preoperative Multiplate ASPI test area under the curve (AUC) units. The SHOULD-NOT-BLEED-SCORE identifies/predicts patients without a risk for excessive bleeding with strong discriminatory performance (Receiver Operating Curve (ROC) analysis AUC 72.3%, p < 0.001).

CONCLUSION

The SHOULD-NOT-BLEED risk scoring application may be useful in the preoperative risk screening process. The clinical and economic burden associated with unnecessary transfusions may be adequately addressed by a preoperative scoring system detecting patients at low risk of bleeding for whom transfusion treatment might be considered unnecessary.

Predictors of perioperative bleeding in left ventricular assist device implantation

Study objective

Early bleeding is a common source of morbidity associated with left ventricular assist device (LVAD) implantation. Our objective was to identify potential predictors of peri-implant bleeding.

Methods

We conducted a retrospective cohort study of LVAD implants at our institution between January 2010 and November 2018. A total of 210 patients were included. Data were collected for the duration of implant hospitalization, including perioperative invasive hemodynamics, echocardiography and operative details, antiplatelet and anticoagulant use, bleeding events and blood product use, and thromboembolic events. Peri-operative bleeding was defined as a transfusion requirement of >4 units of packed red blood cells in the intraoperative and first 7 days postoperative period, or a major 7-day post-implant overt bleeding event requiring procedural intervention.

Results

Perioperative bleeding occurred in 32% of patients and required surgical re-exploration in 9%. Multivariable logistic regression analysis identified history of previous sternotomy (OR 2.63, 95% CI 1.29 to 5.35, p-value 0.008), preoperative glomerular filtration rate <60 ml/min (OR 2.58, 95% CI 1.34 to 4.94, p-value 0.004), preoperative right atrial pressure >13 mm Hg (OR 2.36, 95% CI 1.19 to 4.67, p-value 0.014) and concomitant tricuspid valve repair (OR 2.48, 95% CI 1.23 to 5.01, p-value 0.011) as independent predictors of perioperative bleeding. In-hospital thromboembolic events occurred in 5% of patients, but there were no significant predictors for them.

Conclusions

Elevated right atrial pressure appears to be a reversible risk factor for early bleeding that should be targeted during pre-implant optimization of LVAD candidates.

Evaluation of a novel Cardiac Peri-Operative Transfusion Trigger Scoring system in patients with coronary artery disease

BACKGROUND

A simple and accurate scoring system to guide perioperative blood transfusion in patients with coronary artery disease (CAD) undergoing cardiac surgery is lacking. The trigger point for blood transfusions for these patients may be different from existing transfusion guidelines. This study aimed to evaluate the safety and efficacy of a new scoring strategy for use in guiding transfusion decisions in patients with CAD.

METHODS

A multicenter randomized controlled trial was conducted at three third-level grade-A hospitals from January 2015 to May 2018. Data of 254 patients in a Cardiac Peri-Operative Transfusion Trigger Score (cPOTTS) group and 246 patients in a group receiving conventional evaluation of the need for transfusion (conventional group) were analysed. The requirements for transfusion and the per capita consumption of red blood cells (RBCs) were compared between groups.

RESULTS

Baseline characteristics of the two groups were comparable. Logistic regression analyses revealed no significant differences between the two groups in primary outcomes (1-year mortality and perioperative ischemic cardiac events), secondary outcomes (shock, infections, and renal impairment), ICU admission, and ICU stay duration. However, patients in the cPOTTS group had significantly shorter hospital stays, lower hospital costs, lower utilization rate and lower per capita consumption of transfused RBCs than controls. Stratified analyses revealed no significant differences between groups in associations between baseline characteristics and perioperative ischemic cardiac events, except for hemofiltration or dialysis and NYHA class in I.

CONCLUSIONS

This novel scoring system offered a practical and straightforward guideline of perioperative blood transfusion in patients with CAD. Trial registration chiCTR1800016561(2017/7/19).

Predictiveness of different preoperative risk assessments for postoperative bleeding after coronary artery bypass grafting surgery

Aim:

Postoperative bleeding is a significant cause of morbidity and mortality in patients undergoing cardiac surgery. Studies have been conducted, and guidelines have been published regarding patient blood management and aiming to prevent blood loss in the perioperative period. Various bleeding risk assessments were developed for preoperative period. We aimed to examine the correlations of scoring systems in the literature with the amount of postoperative bleeding in patients undergoing first time coronary artery bypass graft surgery, and to show the most suitable preoperative bleeding risk assessment for coronary artery bypass graft patients.

Methods:

The study included 550 consecutive patients who underwent coronary artery bypass graft operation. The inclusion criteria were considered as patients to be older than 18 years old and to undergo elective or emergent myocardial revascularization using cardiopulmonary bypass. All variables required for scoring systems were recorded. The initial results of the study were determined as the amount of chest tube drainage, the use of blood products, the change in hematocrit level, reoperation due to bleeding, duration of ventilation, duration of intensive care unit stay, and hospital stay. Mortality which occurred during first 30 days after operation was considered as operative mortality. Operative mortality was accepted as the primary endpoint. Secondary endpoints were massive bleeding and high amount of transfusion.

Results:

Data were obtained from a series of 550 consecutive patients treated with isolated coronary artery bypass graft. It was seen that PAPWORTH and WILL-BLEED risk assessments responded better for E-CABG grade 2 and 3 bleeding compared to other risk assessments. TRACK, TRUST, and ACTA-PORT scales were found to have low ability to distinguish patients with E-CABG bleeding grade 2 and 3.

Conclusion:

Predicting postoperative bleeding and transfusion rates with preoperative risk scores in patients undergoing coronary artery bypass graft surgery will provide valuable information to physicians for establishing a proper patient blood management protocol and this will decrease excessive transfusions, unnecessary reoperations as well as improve postoperative outcomes.

Prediction of perioperative transfusions using an artificial neural network

BACKGROUND

Accurate prediction of operative transfusions is essential for resource allocation and identifying patients at risk of postoperative adverse events. This research examines the efficacy of using artificial neural networks (ANNs) to predict transfusions for all inpatient operations.

METHODS

Over 1.6 million surgical cases over a two year period from the NSQIP-PUF database are used. Data from 2014 (750937 records) are used for model development and data from 2015 (885502 records) are used for model validation. ANN and regression models are developed to predict perioperative transfusions for surgical patients.

RESULTS

Various ANN models and logistic regression, using four variable sets, are compared. The best performing ANN models with respect to both sensitivity and area under the receiver operator characteristic curve outperformed all of the regression models (p < .001) and achieved a performance of 70-80% specificity with a corresponding 75-62% sensitivity.

CONCLUSION

ANNs can predict >75% of the patients who will require transfusion and 70% of those who will not. Increasing specificity to 80% still enables a sensitivity of almost 67%. The unique contribution of this research is the utilization of a single ANN model to predict transfusions across a broad range of surgical procedures.