Page 24 - ANZCP Gazette May 2023

and the same increase for every 10 minutes duration of mean arterial pressure <50 mmHg.
The relationship between the duration of CPB and postoperative morbidity and mortality in patients undergoing cardiac surgery has been demonstrated (19). Despite the continuous improvement and development of new CPB techniques, the blood’s exposure to roller pumps, the air-blood interface, and its contact with artificial circuit components activates the coagulation and fibrinolytic system (20), platelets (21), complement system (22), and leukocytes with consequent degranulation and release of cytotoxic enzymes and inflammatory mediators (23). These sequalae increase the risk of postoperative multiorgan failure, infectious and bleeding complications (19). Factors such as operative planning and surgical team communication may modify risk of these outcomes if CPB time can be reduced. Since CPB duration is often related to the complexity of operation which itself may be related to postoperative outcome, we included both the procedure type and redo procedures in the variable selection process.
Not all preoperative variables in the model reported by Billah et al (1) were retained in our model. This reflects a change in the relative predictive value of these variables once the intraoperative variables were included in the bootstrap selection process.
In general, there is a paucity of knowledge regarding the relationships between CPB practice and outcome due to a lack of randomised studies. More specifically, the identification of modifiable CPB parameters associated with mortality are limited due to the large sample size required to reach adequate statistical power, and this study highlights the benefit of multicentre perfusion registries. The ANZCPR is unique in its integration of electronic perfusion data which have been collected in sites throughout Australia and New Zealand over a considerable period, and this study provided the opportunity to define and report which of these parameters play an important role in patient outcome.
Statistical modelling methodology can be described as either descriptive, explanatory or predictive. Predictive modelling can suggest improvements to existing explanatory models whilst also have an explanatory ability (24). The predictive modelling approach taken in this study enabled us to evaluate the relative importance of the CPB variables collected in our registry from an explanatory perspective through the bootstrap selection and ranking process, and by comparing the predictive ability of the model with or without the inclusion of the CPB variables.
An important perspective gained from undertaking this analysis concerning modelling of CPB related data was the inclusion of CPB time as quintiles in order to separate the impact of long bypass times. If CPB time was included in the model as a continuous variable, the effect of longer bypass times would be averaged across all patients, thereby overestimating the effect in patients undergoing shorter
CPB times. Given that CPB time was selected in 100% of the bootstrap models, this has a considerable influence on determining the influence of other variables in the final model. This may account for the inconsistency in findings reported by Turner et al who were unable to identify cardiac index <1.6 l/min/m2 or mean arterial pressure <50 mmHg as predictors of AKI (25); as CPB duration was treated as a continuous variable the effect of long CPB time may have overestimated the effect of CPB duration in patients with shorter bypass times, thereby influencing the independent effect of the pressure and flow variables.
Limitations of this study include its use of observational data and although the description of our patients was detailed and allowed us to extensively control for confounding, we cannot infer causality, nor exclude the possibility of residual confounding from factors not collected or included in the model. The metrics of model discrimination and calibration were consistent with those reported by Billah et al (1), albeit our base model of preoperative risk factors had a slightly lower ROC of 0.7833, compared with 0.8131 reported by Billah et al in single validation. Differences may be partially attributed to data collection over different time periods and a difference in the number of contributing sites in each registry. ANZCPR does not collect data on preoperative inotropic, nitrate use or anticoagulant medication, rather collects critical preoperative state (ventricular tachycardia / ventricular fibrillation or aborted sudden death, cardiac massage, ventilation before anaesthetic room, inotropes or IABP, acute renal failure) and this was included in development of the model. Given that the final model relies upon inclusion of intraoperative parameters, it cannot be used preoperatively.
Despite these limitations, the strengths of our study include the use of a large and well described multicentre CPB registry of prospectively collected data. Furthermore, randomly splitting the dataset into two cohorts provided the opportunity to both cross validate the performance of the model and to compare metrics of model performance with a similar model from an Australian dataset with comparable data definitions. Since the patient sample was large, selection of patients was randomly performed, patient characteristics between cohorts were similar and discrimination, calibration, and goodness-of-fit metrics were similar for the training and validation sets, we have provided evidence that the model is robust and did not suffer from overfitting.
Conclusion
The inclusion of CPB parameters augments the prediction of 30-day mortality following cardiac surgery. Modifiable CPB parameters including CPB time, red blood cell transfusion, mean arterial pressure <50 mmHg, minimum oxygen delivery and cardiac index <1.6 l/min/m2 were identified. Randomised trials designed to evaluate modifiable CPB parameters will determine their impact on mortality.
21 MAY 2023 | www.anzcp.org