EndoSysScore (ESS) v3.0 — Phenotype-aware synthetic data-augmented risk prediction for 30-day mortality after cardiac surgery for infective endocarditis

EndoSysScore (ESS) v3.0 is a phenotype-aware stacking ensemble for predicting 30-day mortality after cardiac surgery for infective endocarditis, developed on the GIROC Italian multicenter registry (24 centres, 2010–2023; derivation cohort n=4,334) and evaluated in an independent external validation cohort of 881 patients from three centres.

Performance (independent external validation, n=881, 3 centres):
- AUC 0.789 (95% CI 0.751–0.827)
- 6/6 DeLong superiority over all comparators (all p<0.05)
- O/E ratio 0.994
- Mean absolute deviation on rare phenotypes: 3.8 pp vs 9.4–38 pp for comparators

Six high-risk phenotypes (A–F) identified in derivation data and validated externally. TIMA-3 blind realism test: 10 cardiac surgeons, 1,000 judgements, pooled accuracy 49.5% (p=0.776 vs chance).

This is the updated version of the original deposit (doi:10.5281/zenodo.20327975, SYS Score v2.0). The model architecture has been updated from bagged XGBoost to a LightGBM meta-learner with phenotype-aware sample weighting. The validation cohort has been expanded from a single centre (n=597) to three independent centres (n=881).

Live calculator: https://endosysscore.org
GitHub: https://github.com/sandro2462/sys-score