Editorial Comment: Re: Evaluation of the Aristotle complexity models in adult patients with congenital heart disease

Abstract

With an increasing number of adult congenital heart (GUCH) surgery patients, a suitable risk stratification model should be available for outcome evaluation, but it would tend to be incomplete due to the very heterogeneous and often complex patient groups with multiple measured as well as unmeasured risk factors. The paper by Horer et al .[ 1] contained valuable data concerning the procedural factors associated with early death and the complications following GUCH surgery. In addition to identifying the associations between early death and procedural factors, the authors looked for associations between items contained within the Aristotle comprehensive complexity (ACC) score. They commented that some of the items in the ACC ‘did not achieve statistical significance’ in terms of an association with early death, putting this down to the low prevalence of these factors within the dataset or to the small number of deaths. However, the lack of inclusion of complex patients who were not offered high-risk surgical options might skew the analysis. A third possibility is the lack of an association between all items in the ACC (developed primarily for paediatric cardiac patients) and early death in adult patients. Having found that the ACC score had a fairly good discrimination in the GUCH patient population, the authors discussed the appropriateness of the ACC for predicting outcome. Some care is required here. The authors talked of predictive accuracy, but no assessment of the accuracy of ACC was presented. To be very clear, discrimination (the extent to which a model can discern between low-risk and high-risk cases) is not a measure of predictive accuracy for individual patients or patient sub-groups. As we may remember, the Aristotle score was not created to predict mortality [2]. The score would give an average value of mortality for a given paediatric cardiac procedure which every centre or surgeon could use for performance evaluation. The authors also commented on the potential utility of adding diagnostic information such as ‘univentricular status’ to augment procedural information in risk stratification schemes such as the ACC. We certainly see the value in this and have recently completed an analysis of the UK Central Cardiac Audit Database (CCAD) to develop a risk model for paediatric cardiac surgery that incorporates such information on diagnosis [3]. In risk modelling for paediatric cardiac surgery, the trend is towards systems based on the analysis of data rather than expert opinion [4]. Nevertheless, the authors’ vision of their work stimulating the collection and publication of larger series of data is one we fully endorse.