The most important result of this exploratory analysis was its demonstration that the predictive ability of simple cardiac risk scores, such as the RCRI and AUB-HAS2, could be improved by adding the preoperative NT-proBNP cut-off.

Our data also showed that the score proposed by Andersson et al. is insufficient to assess postoperative morbidity, including all individual components of the CME in this study, after non-cardiac surgery in cohorts containing both patients with and without HF. The Andersson score has been validated for 30-day mortality risk in a large non-cardiac surgery cohort in which all patients had a known history of HF (Andersson et al. 2014). Therefore, the impact of the isolated Andersson score criteria appears to differ substantially from what is needed when assessing perioperative morbidity risk in mixed cohorts. This finding might be particularly aggravated by age and body mass index being weighted sevenfold and fourfold in the original score, respectively. When the Andersson score is calculated, the data from our cohort suggest that these two criteria are overemphasised. It is plausible that age and weight alone do not determine postoperative morbidity without other comorbidities in general populations. Therefore, its predictive ability to identify patients at risk for postoperative complications was non-existent in our analysis.

The correlation between preoperative NT-proBNP, which was predictive of postoperative complications in our cohort, and the Andersson score was very weak. The correlation was also weak between preoperative NT-proBNP and the RCRI, which also failed to significantly predict the CME in our small cohort. However, its predictive ability for AKI was sufficient and comparable to preoperative NT-proBNP. Therefore, we generated some preliminary evidence that the RCRI may also be used to predict postoperative AKI after major non-cardiac surgery. The RCRI has generally been shown to discriminate moderately well in mixed non-cardiac surgery cohorts but did not perform well in vascular surgery cohorts (Ford et al. 2010). However, it was shown that its predictivity for major adverse cardiovascular events could be improved by preoperative biomarkers, such as high-sensitivity cardiac troponin T/I and NT-proBNP (Kyeong et al. 2008; Vernooij et al. 2021; Rodseth et al. 2011; Park et al. 2011). Unfortunately, there is still a lack of clinically applicable strategies combining the RCRI with preoperative biomarker screening.

In our analysis, the modified RCRI, including the weighted preoperative NT-proBNP cut-off and original RCRI components, showed improved predictivity compared to the original RCRI. However, the multivariable logistic regression model showed that preoperative NT-proBNP had an almost fivefold greater adjusted predictive ability for the CME than the RCRI components. A meta-analysis by Rodseth et al. showed that reclassifying the RCRI using a preoperative NT-proBNP cut-off improved risk prediction for cardiovascular deaths and nonfatal myocardial infarction (Rodseth et al. 2011). Therefore, our data are consistent with previously published data, expanding the improvability of the RCRI to the field of postoperative morbidity. However, it must be noted that the original RCRI criteria had only a small impact on the modified score because of the strongly weighted preoperative NT-proBNP cut-off. Therefore, this finding once again underscores the superiority of preoperative NT-proBNP alone over clinical scores, such as the RCRI, in predicting the CME. In addition, its overall predictivity for the CME could not be improved compared to preoperative NT-proBNP alone.

The AUB-HAS2 showed a good correlation with preoperative NT-proBNP in our analysis. It also showed better risk prediction than the RCRI for the CME but for ADHF and postoperative infections. The criteria used in the AUB-HAS2 seem to be of higher relevance for the perioperative outcome than the parameters of the other scores, leading to the improved predictive value for the CME. For example, symptoms of heart disease are a criterion that can be positive, even if the patient has not yet diagnosed with HF or chronic coronary syndrome. Furthermore, anaemia is not considered in the RCRI and Andersson score; however, preoperative anaemia is strongly associated with postoperative AKI and adverse outcomes (Katayama et al. 2021). The AUB-HAS2 was initially validated for risk prediction of 30-day death, myocardial infarction, and stroke in a general non-cardiac surgery cohort (Dakik et al. 2019b). It was subsequently shown to be predictive in various surgical subpopulations, including vascular surgery (Dakik et al. 2019a; Msheik et al. 2021). Therefore, we expanded the evidence that AUB-HAS2 can also predict postoperative morbidity, showing better discriminatory power than the RCRI in our cohort. Our data is consistent with previous analyses showing that AUB-HAS2 was superior to the RCRI (Dakik et al. 2019a, 2019b). AUB-HAS2 predicted risk for the CME was comparable to that of preoperative NT-proBNP alone. alone. Therefore, both tools can be suitably used to assess perioperative morbidity risk in patients undergoing non-cardiac surgery. While the individual strength of the AUB-HAS2 depends on accurate information in the patient’s history, preoperative NT-proBNP measurement offers an additional independent parameter.

However, it must be noted that NT-proBNP can be negative in up to 20% of patients with HF with preserved ejection fraction and might be influenced by other factors, such as age, weight, sex, and renal function (McDonagh et al. 2021). Therefore, combining preoperative NT-proBNP and clinical risk scores appears attractive. When preoperative NT-proBNP was combined with the AUB-HAS2 after the multivariable analysis in our study, the AUCROC was numerically improved compared to AUB-HAS2 or preoperative NT-proBNP alone. However, the absolute difference in the AUCROC was small, showing statistical significance to the original score but not NT-proBNP alone, potentially reflecting our study’s small sample size. Furthermore, preoperative NT-proBNP was weighted stronger than all other AUB-HAS2 parameters together. Therefore, patients with preoperative NT-proBNP below the cut-off but with clinically relevant chronic HF would be at risk of false-negative classification with the modified score calculated in our study. These findings suggest that risk stratification should still be performed by combining biomarkers and risk scores derived from the patient’s history because no existing clinical risk scores sufficiently include preoperative biomarkers.

Nevertheless, our data showed that combining preoperative NT-proBNP with established risk scores might improve outcome prediction for postoperative morbidity. Furthermore, while the AUB-HAS2 was validated for death, myocardial infarction, and stroke at 30 days, it could also predict the morbidity measures analysed in our study. These morbidity events occur more frequently than the fatal outcomes initially assessed. For example, the AUB-HAS2 was validated in a cohort with a 1.2% rate of major adverse cardiovascular events (Dakik et al. 2019b). However, there are good reasons to consider morbidity measures, such as those analysed in our study, when assessing perioperative risk. Postoperative morbidity after non-cardiac surgery is not only individually relevant but is also economically important because the morbidity events analysed in our study have already caused longer hospital and intensive care unit stays (Schmidt et al. 2024). Therefore, when risk scores and biomarker screening jointly indicate high perioperative risk, distinct therapeutic strategies are indicated and should be evaluated in further studies.

Our study had several limitations that must be acknowledged. First, its sample size was small, and the initial sample size was not powered to assess differences in perioperative risk scores. Therefore, the results of this pilot analysis should be interpreted as hypothesis-generating because its small sample size limited its statistical analysis. For example, advanced statistical measures, such as machine learning models, could not be implemented in our study to improve predictivity for the CME. Second, we did not validate the modified scores in an independent validation cohort. Further studies with larger cohorts are necessary to create more accurate predictive models considering preoperative NT-proBNP testing after major non-cardiac surgery. Third, our study only considered elective non-cardiac and non-vascular surgeries. Therefore, the predictive abilities of the analysed scores were impaired because not all of their criteria were covered by this analysis. For example, the AUB-HAS2 considers both vascular and emergency surgery for risk prediction; however, it still had the best predictivity of all analysed scores in our cohort. Fourth, only patients undergoing surgery with intermediate or high surgical risk were analysed. Therefore, our study did not examine risk prediction in patients with low surgical risk but potentially high patient-related risk. The predictivity of the analysed scores may be reduced in patients undergoing low-risk surgery, which could be classified as at high risk for the CME.