Association of perioperative transfusion of fresh frozen plasma and outcomes after cardiac surgery.

BACKGROUND: Fresh frozen plasma (FFP) transfusion is used to manage coagulopathy and bleeding in cardiac surgery patients despite uncertainty about its safety and effectiveness. METHODS: We performed a propensity score matched analysis of the Australian and New Zealand Society of Cardiac and Thoracic Surgeons National Cardiac Surgery Database including patients from 39 centres from 2005 to 2018. We investigated the association of perioperative FFP transfusion with mortality and other clinical outcomes. RESULTS: Of 119,138 eligible patients, we successfully matched 13,131 FFP recipients with 13,131 controls. FFP transfusion was associated with 30-day mortality (odds ratio (OR), 1.41; 99% CI, 1.17-1.71; p < .0001), but not with long-term mortality (hazard ratio (HR), 0.92; 99% CI, 0.85-1.00; p = .007, Holm-Bonferroni α = 0.0004). FFP was also associated with return to theatre for bleeding (OR, 1.97; 99% CI, 1.66-2.34; p < .0001), prolonged intubation (OR, 1.15; 99% CI, 1.05-1.26; p < .0001) and increased chest tube drainage (Mean difference (MD) in mL, 131; 99% CI, 120-141; p < .0001). It was also associated with reduced postoperative creatinine levels (MD in g/L, -6.33; 99% CI, -10.28 to -2.38; p < .0001). CONCLUSION: In a multicentre, propensity score matched analysis, perioperative FFP transfusion was associated with increased 30-day mortality and had variable associations with secondary clinical outcomes.

Platelet versus fresh frozen plasma transfusion for coagulopathy in cardiac surgery patients.

BACKGROUND: Platelets (PLTS) and fresh frozen plasma (FFP) are often transfused in cardiac surgery patients for perioperative bleeding. Their relative effectiveness is unknown. METHODS: We conducted an entropy-weighted retrospective cohort study using the Australian and New Zealand Society of Cardiac and Thoracic Surgeons National Cardiac Surgery Database. All adults undergoing cardiac surgery between 2005-2021 across 58 sites were included. The primary outcome was operative mortality. RESULTS: Of 174,796 eligible patients, 15,360 (8.79%) received PLTS in the absence of FFP and 6,189 (3.54%) patients received FFP in the absence of PLTS. The median cumulative dose was 1 unit of pooled platelets (IQR 1 to 3) and 2 units of FFP (IQR 0 to 4) respectively. After entropy weighting to achieve balanced cohorts, FFP was associated with increased perioperative (Risk Ratio [RR], 1.63; 95% Confidence Interval [CI], 1.40 to 1.91; P<0.001) and 1-year (RR, 1.50; 95% CI, 1.32 to 1.71; P<0.001) mortality. FFP was associated with increased rates of 4-hour chest drain tube output (Adjusted mean difference in ml, 28.37; 95% CI, 19.35 to 37.38; P<0.001), AKI (RR, 1.13; 95% CI, 1.01 to 1.27; P = 0.033) and readmission to ICU (RR, 1.24; 95% CI, 1.09 to 1.42; P = 0.001). CONCLUSION: In perioperative bleeding in cardiac surgery patient, platelets are associated with a relative mortality benefit over FFP. This information can be used by clinicians in their choice of procoagulant therapy in this setting.

Platelet Transfusion in Cardiac Surgery: An Entropy-Balanced, Weighted, Multicenter Analysis.

BACKGROUND: Platelet transfusion is common in cardiac surgery, but some studies have suggested an association with harm. Accordingly, we investigated the association of perioperative platelet transfusion with morbidity and mortality. METHODS: We conducted a retrospective analysis of prospectively collected data from the Australian Society of Cardiac and Thoracic Surgeons National Cardiac Surgery Database. We included consecutive adults from 2005 to 2018 across 40 centers. We used inverse probability of treatment weighting via entropy balancing to investigate the association of perioperative platelet transfusion with our 2 primary outcomes, operative mortality (composite of both 30-day and in-hospital mortality) and 90-day mortality, as well as multiple other clinically relevant secondary outcomes. RESULTS: Among 119,132 eligible patients, 25,373 received perioperative platelets and 93,759 were considered controls. After entropy balancing, platelet transfusion was associated with reduced operative mortality (odds ratio [OR], 0.63; 99% confidence interval [CI], 0.47-0.84; P < .0001) and 90-day mortality (OR, 0.66; 99% CI, 0.51-0.85; P < .0001). Moreover, it was associated with reduced odds of deep sternal wound infection (OR, 0.57; 99% CI, 0.36-0.89; P = .0012), acute kidney injury (OR, 0.84; 99% CI, 0.71-0.99; P = .0055), and postoperative renal replacement therapy (OR, 0.71; 99% CI, 0.54-0.93; P = .0013). These positive associations were observed despite an association with increased odds of return to theatre for bleeding (OR, 1.55; 99% CI, 1.16-2.09; P < .0001), pneumonia (OR, 1.26; 99% CI, 1.11-1.44; P < .0001), intubation for longer than 24 hours postoperatively (OR, 1.13; 99% CI, 1.03-1.24; P = .0012), inotrope use for >4 hours postoperatively (OR, 1.14; 99% CI, 1.11-1.17; P < .0001), readmission to hospital within 30 days of surgery (OR, 1.22; 99% CI, 1.11-1.34; P < .0001), as well as increased drain tube output (adjusted mean difference, 89.2 mL; 99% CI, 77.0 mL-101.4 mL; P < .0001). CONCLUSIONS: In cardiac surgery patients, perioperative platelet transfusion was associated with reduced operative and 90-day mortality. Until randomized controlled trials either confirm or refute these findings, platelet transfusion should not be deliberately avoided when considering odds of death.

Development and Validation of Scores to Predict Prolonged Mechanical Ventilation after Cardiac Surgery.

OBJECTIVES: To optimize the early prediction of prolonged postoperative mechanical ventilation after cardiac surgery (>24 hours postoperatively). DESIGN: The authors performed a retrospective analysis. SETTING: The Australian and New Zealand Society of Cardiac and Thoracic Surgeons (ANZSCTS) database was utilized. PARTICIPANTS: All patients included in the ANZSCTS database between January 2015 and December 2018 were analyzed. INTERVENTIONS: No interventions were performed in this observational study. MEASUREMENTS AND MAIN RESULTS: A previously developed model was modified to allow retrospective risk calculation and model assessment (Modified Hessels score). The database was split into development and validation sets. A new risk model was developed using forward and backward stepwise elimination (ANZ-PreVent score). The authors assessed 48,382 patients, of whom 5004 (10.3%) were ventilated mechanically for >24 hours post-operatively. The Modified Hessels score demonstrated good performance in this database, with a c-index of 0.78 (95% CI 0.77-0.78) and a Brier score of 0.08. The newly developed ANZ-PreVent score demonstrated better performance (validation cohort, n = 12,229), with a c-index of 0.84 (95% CI 0.83-0.85) (p < 0.0001) and a Brier score of 0.07. Both scores performed better than the severity of illness scores commonly used to predict outcomes in intensive care. CONCLUSIONS: The authors validated a modified version of an existing prediction score and developed the ANZ-PreVent score, with improved performance for identifying patients at risk of ventilation for >24 hours. The improved score can be used to identify high-risk patients for targeted interventions in future randomized controlled trials.

Paying attention to cardiac surgical risk: An interpretable machine learning approach using an uncertainty-aware attentive neural network.

Machine learning (ML) is increasingly applied to predict adverse postoperative outcomes in cardiac surgery. Commonly used ML models fail to translate to clinical practice due to absent model explainability, limited uncertainty quantification, and no flexibility to missing data. We aimed to develop and benchmark a novel ML approach, the uncertainty-aware attention network (UAN), to overcome these common limitations. Two Bayesian uncertainty quantification methods were tested, generalized variational inference (GVI) or a posterior network (PN). The UAN models were compared with an ensemble of XGBoost models and a Bayesian logistic regression model (LR) with imputation. The derivation datasets consisted of 153,932 surgery events from the Australian and New Zealand Society of Cardiac and Thoracic Surgeons (ANZSCTS) Cardiac Surgery Database. An external validation consisted of 7343 surgery events which were extracted from the Medical Information Mart for Intensive Care (MIMIC) III critical care dataset. The highest performing model on the external validation dataset was a UAN-GVI with an area under the receiver operating characteristic curve (AUC) of 0.78 (0.01). Model performance improved on high confidence samples with an AUC of 0.81 (0.01). Confidence calibration for aleatoric uncertainty was excellent for all models. Calibration for epistemic uncertainty was more variable, with an ensemble of XGBoost models performing the best with an AUC of 0.84 (0.08). Epistemic uncertainty was improved using the PN approach, compared to GVI. UAN is able to use an interpretable and flexible deep learning approach to provide estimates of model uncertainty alongside state-of-the-art predictions. The model has been made freely available as an easy-to-use web application demonstrating that by designing uncertainty-aware models with innately explainable predictions deep learning may become more suitable for routine clinical use.

Tree-based survival analysis improves mortality prediction in cardiac surgery.

Objectives: Machine learning (ML) classification tools are known to accurately predict many cardiac surgical outcomes. A novel approach, ML-based survival analysis, remains unstudied for predicting mortality after cardiac surgery. We aimed to benchmark performance, as measured by the concordance index (C-index), of tree-based survival models against Cox proportional hazards (CPH) modeling and explore risk factors using the best-performing model. Methods: 144,536 patients with 147,301 surgery events from the Australian and New Zealand Society of Cardiac and Thoracic Surgeons (ANZSCTS) national database were used to train and validate models. Univariate analysis was performed using Student's T-test for continuous variables, Chi-squared test for categorical variables, and stratified Kaplan-Meier estimation of the survival function. Three ML models were tested, a decision tree (DT), random forest (RF), and gradient boosting machine (GBM). Hyperparameter tuning was performed using a Bayesian search strategy. Performance was assessed using 2-fold cross-validation repeated 5 times. Results: The highest performing model was the GBM with a C-index of 0.803 (0.002), followed by RF with 0.791 (0.003), DT with 0.729 (0.014), and finally CPH with 0.596 (0.042). The 5 most predictive features were age, type of procedure, length of hospital stay, drain output in the first 4 h (ml), and inotrope use greater than 4 h postoperatively. Conclusion: Tree-based learning for survival analysis is a non-parametric and performant alternative to CPH modeling. GBMs offer interpretable modeling of non-linear relationships, promising to expose the most relevant risk factors and uncover new questions to guide future research.

Association of Perioperative Cryoprecipitate Transfusion and Mortality After Cardiac Surgery.

BACKGROUND: Cryoprecipitate is often transfused in patients undergoing cardiac surgery. However, its safety and effectiveness remain uncertain. METHODS: This study was a propensity score-matched analysis of data from the Australian and New Zealand Society of Cardiac and Thoracic Surgeons National Cardiac Surgery Database. The study included adults undergoing cardiac surgery between 2005 and 2018 across 38 sites. The association between perioperative cryoprecipitate transfusion and clinical outcomes was estimated, with a primary outcome of operative mortality. RESULTS: Of 119,132 eligible patients, 11,239 (9.43%) patients received cryoprecipitate. The median cumulative dose was 8 U (interquartile range, 5-10 U). After propensity score matching, we matched 9055 cryoprecipitate recipients to 9055 control subjects. Postoperative cryoprecipitate transfusion was associated with reduced operative mortality (odds ratio [OR], 0.82; 99% CI, 0.69-0.97; P = .002) and long-term mortality (hazard ratio, 0.92; 99% CI, 0.87-0.97; P = .0042). It was also associated with a reduction in acute kidney injury (OR, 0.85; 99% CI, 0.73-0.98; P = .0037) and all-cause infection (OR, 0.77; 99% CI, 0.67-0.88; P < .0001). These findings were observed despite increased rates of return to the operating room (OR, 1.36; 99% CI, 1.22-1.51; P < .0001) and cumulative 4-hour postoperative chest tube output (adjusted mean difference in mL, 97.69; 99% CI, 81.65;113.74; P < .0001). CONCLUSIONS: In a large, multicenter cohort study and after propensity score matching, perioperative transfusion of cryoprecipitate was associated with reduced operative and long-term mortality.

Clinical Outcomes in Surgical and Transcatheter Aortic Valve Replacement: An ANZSCTS Database Review 2001-2019.

BACKGROUND: Since the last formal publication reporting on the findings of the Australian and New Zealand Society of Cardiac and Thoracic Surgeons (ANZSCTS) database on surgical aortic valve replacement (SAVR) and transcatheter aortic valve replacement (TAVR) in 2016, transcatheter approaches have become common practice. There has been an increase in use of TAVR following large, randomised control trials that only report on short-term outcomes in a selective cohort. This study aims to report on primary outcome measures and identify complications associated with SAVR and TAVR from a large national database. METHODS: From the ANZSCTS database (2001-19), 14,097 SAVR and 1,194 TAVR patients were identified with clinical details and 30-day follow-up available. The primary endpoint was the composite of all-cause mortality and/or permanent stroke at 30 days. Secondary endpoints were post-procedure complications requiring treatment. Logistical regression followed by propensity score matching was performed. RESULTS: Using logistical regression when all patient factors considered for all patients who had SAVR and TAVR, the only preoperative factors that had an impact on 30-day mortality was cerebrovascular disease, respiratory disease, preoperative dialysis, angina, and hypertension. Primary outcome 30-day mortality rate was 1.83% in the SAVR group, and 1.68% in patients in the TAVR group, p=0.7001, and permanent stroke was seen in 1.07% patients in the SAVR group, and 1.26% patients in the TAVR group. Acute limb ischaemia, aortic dissection, ventricular tachycardia, bradyarrhythmia and heart block were more common following TAVR (p<0.001), while reintubation and atrial arrhythmia were more common following SAVR (p<0.001). CONCLUSIONS: In the real world SAVR and TAVR have been used in very different patient groups and it is difficult to compare as different baseline characteristics and complications. The two patient groups maintain similarities in primary and secondary endpoints, but differences in life threatening and life altering morbidity remains significant. Collection of SAVR and TAVR data in a combined database may help to better capture and compare these complications and institute strategies to prevent them.

AntiPORT: adaptation of a transfusion prediction score to an Australian cardiac surgery population.

Introduction: Risk scoring systems exist to predict perioperative blood transfusion risk in cardiac surgery, but none have been validated in the Australian or New Zealand population. The ACTA-PORT score was developed in the United Kingdom for this purpose. In this study, we validate and recalibrate the ACTA-PORT score in a large national database. Methods: We performed a retrospective validation study using data from the Australian and New Zealand Society of Cardiac and Thoracic Surgeons Database between 1 September 2016 and 31 December 2018. The ACTA-PORT score was calculated using an equivalent of EuroSCORE I. Discrimination and calibration was assessed using area under the receiver operating characteristic (AUROC) curves, Brier scores, and calibration plots. ACTA-PORT was then recalibrated in a development set using logistic regression and the outcome of transfusion to develop new predicted transfusion rates, termed "AntiPORT", using AusSCORE "all procedures" as the regional equivalent of EuroSCORE I. The accuracy of these new predictions was assessed as for ACTA-PORT. Results: 30 388 patients were included in the study at 37 Australian centres. The rate of red blood cell transfusion was 33%. Discrimination of ACTA-PORT was good but calibration was poor, with overprediction of transfusion (AUROC curve, 0.76; 95% CI, 0.75-0.76; Brier score, 0.19). The recalibrated AntiPORT showed significantly improved calibration in both development and validation sets without compromising discrimination (AUROC curve, 0.76; 95% CI, 0.75-0.76; Brier score, 0.18). Conclusions: The AntiPORT is the first red cell transfusion risk scoring system for cardiac surgery patients to be validated using Australian data. It is accurate and simple to calculate. The demonstrated accuracy of AntiPORT may help facilitate benchmarking and future research in patient blood management, as well as providing a useful tool to help clinicians target these resource-saving strategies.

Linkage of Australian national registry data using a statistical linkage key.

BACKGROUND: Data from clinical registries may be linked to gain additional insights into disease processes, risk factors and outcomes. Identifying information varies from full names, addresses and unique identification codes to statistical linkage keys to no direct identifying information at all. A number of databases in Australia contain the statistical linkage key 581 (SLK-581). Our aim was to investigate the ability to link data using SLK-581 between two national databases, and to compare this linkage to that achieved with direct identifiers or other non-identifying variables. METHODS: The Australian and New Zealand Society of Cardiothoracic Surgeons database (ANZSCTS-CSD) contains fully identified data. The Australian and New Zealand Intensive Care Society database (ANZICS-APD) contains non-identified data together with SLK-581. Identifying data is removed at participating hospitals prior to central collation and storage. We used the local hospital ANZICS-APD data at a large single tertiary centre prior to deidentification and linked this to ANZSCTS-CSD data. We compared linkage using SLK-581 to linkage using non-identifying variables (dates of admission and discharge, age and sex) and linkage using a complete set of unique identifiers. We compared the rate of match, rate of mismatch and clinical characteristics between unmatched patients using the different methods. RESULTS: There were 1283 patients eligible for matching in the ANZSCTS-CSD. 1242 were matched using unique identifiers. Using non-identifying variables 1151/1242 (92.6%) patients were matched. Using SLK-581, 1202/1242 (96.7%) patients were matched. The addition of non-identifying data to SLK-581 provided few additional patients (1211/1242, 97.5%). Patients who did not match were younger, had a higher mortality risk and more non-standard procedures vs matched patients. The differences between unmatched patients using different matching strategies were small. CONCLUSION: All strategies provided an acceptable linkage. SLK-581 improved the linkage compared to non-identifying variables, but was not as successful as direct identifiers. SLK-581 may be used to improve linkage between national registries where identifying information is not available or cannot be released.

Machine Learning Algorithms for Predicting and Risk Profiling of Cardiac Surgery-Associated Acute Kidney Injury.

Using a large national database of cardiac surgical procedures, we applied machine learning (ML) to risk stratification and profiling for cardiac surgery-associated acute kidney injury. We compared performance of ML to established scoring tools. Four ML algorithms were used, including logistic regression (LR), gradient boosted machine (GBM), K-nearest neighbor, and neural networks (NN). These were compared to the Cleveland Clinic score, and a risk score developed on the same database. Five-fold cross-validation repeated 20 times was used to measure the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. Risk profiles from GBM and NN were generated using Shapley additive values. A total of 97,964 surgery events in 96,653 patients were included. For predicting postoperative renal replacement therapy using pre- and intraoperative data, LR, GBM, and NN achieved an AUC (standard deviation) of 0.84 (0.01), 0.85 (0.01), 0.84 (0.01) respectively outperforming the highest performing scoring tool with 0.81 (0.004). For predicting cardiac surgery-associated acute kidney injury, LR, GBM, and NN each achieved 0.77 (0.01), 0.78 (0.01), 0.77 (0.01) respectively outperforming the scoring tool with 0.75 (0.004). Compared to scores and LR, shapely additive values analysis of black box model predictions was able to generate patient-level explanations for each prediction. ML algorithms provide state-of-the-art approaches to risk stratification. Explanatory modeling can exploit complex decision boundaries to aid the clinician in understanding the risks specific to individual patients.

Predicting Acute Kidney Injury After Cardiac Surgery Using a Simpler Model.

OBJECTIVE: To develop a simple model for the prediction of acute kidney injury (AKI) and renal replacement therapy (RRT) that could be used in clinical or research risk stratification. DESIGN: Retrospective analysis. SETTING: Multi-institutional. PARTICIPANTS: All cardiac surgery patients from September 2016 to December 2018. INTERVENTIONS: Observational. MEASUREMENTS AND MAIN RESULTS: The study cohort was divided into a development set (75%) and validation set (25%). The following 2 data epochs were used: preoperative data and immediate postoperative data (within 4 h of intensive care unit admission). Univariate statistics were used to identify variables associated with AKI or RRT. Stepwise logistic regression was used to develop a parsimonious model. Model discrimination and calibration were evaluated in the test set. Models were compared with previously published models and with a more comprehensive model developed using the least absolute shrinkage and selection operator. The study included 22,731 patients at 33 hospitals. The incidences of AKI (any stage) and RRT for the present analysis were 5,829 patients (25.6%) and 488 patients (2.1%), respectively. Models were developed for AKI, with an area under the receiver operating curve (AU-ROC) of 0.67 and 0.69 preoperatively and postoperatively, respectively. Models for RRT had an AU-ROC of 0.77 and 0.80 preoperatively and postoperatively, respectively. These models contained between 3 and 5 variables. Comparatively, comprehensive least absolute shrinkage and selection operator models contained between 21 and 26 variables, with an AU-ROC of 0.71 and 0.72 for AKI and 0.84 and 0.87 for RRT respectively. CONCLUSION: In the present study, simple, clinically applicable models for predicting AKI and RRT preoperatively and immediate postoperatively were developed. Even though AKI prediction remained poor, RRT prediction was good with a parsimonious model.

Post Cardiotomy Extra Corporeal Membrane Oxygenation: Australian Cohort Review.

BACKGROUND: Over the last two decades, technological advancements in the delivery of extra corporeal membrane oxygenation (ECMO) have seen its use broaden and results improve. However, in the post cardiotomy ECMO patient group, survival remains very poor without significant improvements over the last two decades. Our study aims to report on the Australian experience, with the intention of providing background data for the formation of guidelines in the future. METHODS: Retrospective analysis of prospectively collected data from the Australian and New Zealand Society of Cardiothoracic Surgeons (ANZSCTS) Database was performed. The ANZSCTS database captures at least 60% of cardiac surgical data in Australia, annually. Data was collected on adult patients who received ECMO post cardiotomy from September 2016 to November 2017 inclusive. Transplant and primary cardiomyopathy patients were excluded. RESULTS: Of the 16,605 adult patients undergoing cardiac surgery in the 15-month period of the study, 87 patients required post cardiotomy ECMO (0.52%). The average age of the entire cohort was 56 years. Overall survival to discharge was 43.7% (n=38). Multivariable logistic regression analysis demonstrated that multiorgan failure (MOF), increasing age and longer cardiopulmonary bypass time were significant predictors of in hospital mortality. CONCLUSIONS: Post cardiotomy ECMO support is an uncommon condition. Survival in this study appears to be better than historical reports. Identification of poor prognostic indicators in this study may help inform the development of guidelines for the most appropriate use of this support modality.

The Australian and New Zealand Society of Cardiac and Thoracic Surgeons Database Program - Two Decades of Quality Assurance Data.

Over two decades, the Australian and New Zealand Society of Cardiac and Thoracic Surgeons (ANZSCTS) cardiac surgery database program has evolved from a single state-based database to a national clinical quality registry program and is now the most comprehensive cardiac surgical registry in Australia. We report the current structure and governance of the program and its key activities.