https://doi.org/10.1007/s00392-024-02526-y
1Universitätsklinikum Frankfurt Med. Klinik III - Kardiologie, Angiologie Frankfurt am Main, Deutschland; 2Klinikverbund Südwest - Klinikum Sindelfingen-Böblingen GmbH Medizinische Klinik II, Kardiologie Sindelfingen, Deutschland; 3Goethe Universität Frankfurt am Main Zentrum für Molekulare Medizin, Institut für Kardiovaskuläre Regeneration Frankfurt am Main, Deutschland
Background:
Although myocardial fibrosis (MF) is common in patients with severe aortic stenosis (AS) and independently predicts adverse outcome, examination by cardiac MRI is not routinely available in the majority of patients undergoing transcatheter aortic valve replacement (TAVR).
Aim & Methods:
In this study, we examined novel circulating fibrosis marker and established myocardial injury proteins as a platform to improve risk prediction in 393 consecutive patients undergoing TAVR. Implementation of myocardial injury and fibrosis marker into TAVR risk stratification was tested by an artificial intelligence (AI)-assisted machine-learning (ML) algorithm followed by two-year survival analyses.
Results:
Among a bundle of circulating fibrosis marker associated with maladaptive fibrotic remodeling, elevated TIMP1 levels were found to independently predict reduced survival in univariate (hazard ratio, 4.8; 95% CI, 7.0–8.5 [P<0.001]) and multivariate (adjusted hazard ratio, 2.0; 95% CI, 1.0–4.0 [P<0.05]) Cox regression analyses. Combining elevated TIMP1 levels with the STS score significantly resulted in a higher overall prediction of reduced survival compared to STS score alone (AUC 0.764 [95% CI: 0.705-0.823] vs. AUC 0.682 [95% CI: 0.599-0.743], P< 0.01), thereby improving risk stratification into subgroups (P for trend <0.001). Strikingly, AI-driven algorithms discovered a simple combination of two biomarker (TIMP1 and hs-cTnT) with superior prognostic value compared to routinely used STS score alone (AUC 0.756 [95% CI: 0.690-0.821] vs. AUC 0.682 [95% CI: 0.599-0.743], P< 0.05). Notably, elevation of both biomarker allowed reclassification into subgroups ranging with reduced two‐year survival rates of 88.0 % (low risk subgroup) to 26.0 % (high risk subgroup).
Conclusion:
Artificial intelligence for predictive biomarker discovery and implementation enables improved prediction of adverse outcome in patients undergoing TAVR.