A new <scp>VISION</scp> to improve cardiac risk stratification in non‐cardiac surgery

In this issue of Anaesthesia, McAlister et al. examined the accuracy of the revised cardiac risk index (RCRI) and three atrial fibrillation (AF) thrombo-embolic risk models for predicting 30-day cardiovascular events after non-cardiac surgery in patients with a pre-operative history of AF 1. This was a follow-up study to a large 2014 international prospective study on myocardial injury by the vascular events in non-cardiac surgery patients cohort evaluation (VISION) writing group and a later 2015 VISION substudy looking at predictors of postoperative stroke or death in patients with pre-operative AF undergoing non-cardiac surgery 2, 3. Patients with a pre-operative history of AF (even if in sinus rhythm at the time of surgery) had a 30% higher risk of peri-operative cardiovascular events within 30 days after surgery compared with patients without a history of AF. The same authorship group have reported previously that patients with pre-operative AF had a 58% higher risk of postoperative stroke or mortality 3. A wider definition of adverse events in the current study probably accounts for the revised lower adjusted risk of adverse events of pre-operative AF. This new analysis expands the number of peri-operative cardiovascular events to include myocardial injury after non-cardiac surgery, heart failure, resuscitated cardiac arrest or cardiovascular death within 30 days, in addition to stroke and all-cause mortality. None of the three thrombo-embolic models or the revised cardiac risk index (RCRI) exhibited strong discrimination metrics at predicting peri-operative cardiovascular events. The RCRI was weakest with a disappointing c-index of 0.60. The R2CHADS2 was the pick of a modest bunch with a c-index of 0.65 4. The paper reports a 29% overall incidence of cardiovascular events with incidences of 43% and 24% in the highest (RCRI ≥ 3) and lowest (RCRI 0–1) risk groups, respectively. The relative risk ratio between the highest and lowest risk groups of approximately 1.8 highlights the limited clinical utility of RCRI in the prediction of cardiovascular events. This large multicentre paper by a very experienced and well-funded group undoubtedly has many strengths; methodology, execution and authorship are excellent. Unfortunately the paper also has limitations, many of which are recognised by the authors. Atrial fibrillation is a heterogeneous condition. Duration of AF, the presence of valvular disease and frequencies of paroxysms of AF (in those with a history of AF who were not in AF at the time of surgery) were all unknown and therefore will impact the applicability of the scoring models used. The study did not have access to biomarker results such as B-natriuretic peptide (BNP) or troponin at baseline, limiting the choice of thrombo-embolic scores for comparison, and preventing comparison with new AF thrombo-embolic risk scores such as the ATRIA or ABC scores 5, 6. This illustrates the vulnerability of studies of long duration to emerging new developments. In addition, it is conceivable that many of the control group could have had undiagnosed asymptomatic AF. The incidence of subclinical AF in patients with pacemakers was reported as ~10%, albeit in an older population ≥ 65 years 7. Risk assessment has multiple purposes that include: holistic patient information; informed consent; informing changes in management; and allowing outcome comparison between hospitals 8. Thorough assessment of risk is particularly important for patients undergoing high-risk non-cardiac surgery. Although many risk scoring systems are available, the most validated and widely used is the RCRI which consists of one procedural and five clinical risk factors. The six-parameter RCRI scoring system was derived from 4315 patients based in a single centre with data collected from 1989 to 1994. 9. The RCRI replaced Goldman's original nine parameter multifactorial index of cardiac risk, derived from just 1001 patients, which had been the previous gold standard for over two decades 10. Given that both scoring systems were developed from single centres using just 4315 and 1001 patients, respectively, it is remarkable that each survived over two decades as the gold standard for risk stratification. Although not designed as a peri-operative model, the R2CHADS2 score also consists of six parameters. Compared with RCRI, the main differences are: high-risk surgery and history of ischaemic heart disease are replaced by age and hypertension; and age and stroke/transient ischaemic attack/systemic embolism are given additional weighting. The R2CHADS2 score has some merit in that it recognises the importance of age as an independent risk-factor and the need to apply different weightings to risk factors. However, most readers should be reassured that, unlike Goldman's original scoring system or Lee's revised version, the R2CHADS2 score is unlikely to achieve gold standard status. It is inconceivable to these reviewers that high-risk surgery or ischaemic heart disease could be excluded from future peri-operative risk models. The most recent 2014 American College of Cardiology/American Heart Association, 2017 Canadian Cardiovascular Society (CCS) and 2018 European Society of Anaesthesiology guidelines all continue to endorse the RCRI 11-13. The American and European, but interestingly not the Canadian guidelines, also endorse the National Surgical Quality Improvement Program (NSQIP) score which includes over 20 universal risk factors 14. The 2017 CCS guidelines recommend RCRI over the NSQIP due to lack of external validation of NSQIP. The modest discrimination metrics demonstrated by the four prediction models in the paper by McAllister et al. predicate the need for a more accurate and reliable risk scores. Biomarkers are an emerging modality within predictive risk models. When combined with clinical risk factors, they often provide a superior predictive value for adverse cardiovascular events, although supporting statistical values are not strong. Interestingly, based on moderate quality evidence, the 2017 CCS guidelines also made a strong recommendation for the use of pre-operative biomarkers in three distinct groups undergoing elective surgery: patient age ≥ 65 years; RCRI ≥ 1 and patient age 45–64 years with significant cardiovascular disease. There are many emerging biomarkers with clinical utility, however, two of note, namely troponin and B-type natriuretic peptide (NT-proBNP), merit brief discussion. Cardiac troponin has been an important cardiac biomarker for decades. Recently, the development of increasingly sensitive troponin assays has enabled more accurate detection of myocardial injury. The troponin substudy by Hijazi et al. demonstrated that troponin elevation is positively related to both the risk of stroke and cardiac death independently of patient characteristics 15. NT-proBNP, released due to myocyte stress, is an established powerful prognostic biomarker in patients with cardiovascular disease 16. Its prognostic role in AF patients was initially demonstrated by Hijazi et al., who showed that higher NT-proBNP levels correlated with up to five-fold higher risk of cardiovascular mortality as compared with subjects with normal levels 17. When added to either CHADS2 or CHA2-DS2-VASc, discriminatory performance also increased. Interestingly, the same trial clearly demonstrates a further risk prediction increase when both troponin and NT-proBNP levels are combined with CHADS2 score 17. In a prospective single-centre study, Choi et al. reported that the addition of biomarkers to RCRI increased the relative risk of RCRI for clinical events three-fold 18. In addition, Ruff et al. demonstrated higher biomarker levels were associated with a 15-fold gradient of increased cardiovascular risk, providing more weight to the inclusion of biomarkers to risk scoring models 19. Emerging biomarker research has informed newer scoring models, such as the ABC score, which uses both troponin and BNP for predicting stroke in patients with AF to produce a calibrated score with higher c-indices than conventional models, despite increases being modest 6, 20. The CCS guidelines recommend communication of peri-operative risk on the expected event rate among 100 patients. It is currently difficult to convert any given RCRI score to percentage risk. Lee's original quoted complication rates of 0.4%, 1%, 7% and 11% in patients with RCRI points of between 0 and ≥ 3 was based on single-centre data collected over two decades ago and underestimates the actual percentage risk. The new RCRI revised estimates put the comparable risks closer to 4%, 6%, 10% and 15%, respectively. McAlister et al.'s study highlights the limitations of basing risk classification decision making on a binary response six-parameter model conceived over two decades ago and sub-division of patient age into just three risk categories 12. Future members of guideline committees seeking a solution to this problem could consider either the established risk stratification model used for cardiac surgery, the logistic EuroSCORE, which is used to communicate the risk of death to patients before cardiac surgery 20. Alternatively, a similar methodology has also been used to produce the non-cardiac risk scoring system known as the Surgical Outcome Risk Tool (SORT) 21. Both predictive models produce the percentage chance of death based on a number of validated variables. The 2018 European Society of Anaesthesiology guidelines do not endorse SORT due to limited external validation 13. We have two additional concerns with SORT. Firstly, it still categorises age into three groups. In an electronic era, we believe that future scoring systems should utilise absolute patient age to improve discriminatory power. Secondly, SORT utilises the subjective assessment of ASA physical status, despite its known inter-rater variation. We believe it is inevitable that future scoring systems will be used to facilitate comparisons of non-cardiac surgical outcomes between centres. For this reason, we judge it is extremely desirable that only objective risk markers are included in future scoring systems. The VISION group has now collected detailed information on multiple risk factors and outcomes in over 40,000 patients undergoing non-cardiac surgery. It should be relatively straightforward to derive and validate a logistic scoring system for non-cardiac surgical patients from this sized database. The majority of the original five original clinical factors and high-risk surgery category could be subdivided to increase the discriminatory ability of the model. Consideration should also be given to factoring the additional risks of emergency and urgent surgery into the model. The impact of adding additional risk factors such as, sex, peripheral vascular disease, chronic obstructive pulmonary disease, frailty, body mass index and AF could also be tested. The development of an application similar to either logistic EuroSCORE or SORT would enable rapid estimation of risk in both in the pre-operative assessment clinic and at the bed-side. It would also potentially simplify pre-operative assessment in the elective surgical limb of the 2017 CCS guidelines regarding indications for additional biomarkers. The newly derived percentage risk score may eventually replace the current combination of patient age, RCRI score and the presence of significant cardiovascular disease to guide which patients need additional biomarkers. In conclusion, we believe the case of need for a one-off peri-operative prediction model exclusively for the sub-group of patients with AF undergoing non-cardiac surgery is unproven. However, we propose there is a strong case for using the large and expanding VISION database to derive and validate a peri-operative cardiac risk model for use in all patients undergoing non-cardiac surgery 22.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.