# Predictive accuracy of diagnostic tests for excessive bleeding in cardiac surgery: The COPTIC‐C study

## TL;DR

This study tested if adding biomarkers of aging and multimorbidity improves predictions of bleeding after heart surgery, but found only small improvements over standard tests.

## Contribution

The study evaluates the added value of aging and multimorbidity biomarkers in predicting post-surgical bleeding beyond traditional coagulation tests.

## Key findings

- Adding biomarkers of multimorbidity and aging improved predictive accuracy only slightly over coagulation tests.
- The best model had an AUROC of 0.701 using lab tests and biomarkers, but clinical utility was not better than lab tests alone.
- Discrimination was higher for specific bleeding outcomes like red cell transfusion and procoagulant transfusion.

## Abstract

We tested the hypothesis that the addition of biomarkers of multimorbidity and biological aging would improve the predictive accuracy of point‐of‐care viscoelastometry or laboratory tests of coagulation for clinically important bleeding following cardiac surgery.

This predictive accuracy study included 2437 participants in the coagulation and platelet laboratory testing in cardiac surgery (COPTIC study) with complete clinical, TEG 5000 thromboelastography, ROTEM, multiplate aggregometry, full blood count, laboratory reference tests of coagulopathy, and biomarkers of biological aging and multimorbidity. Models with different biomarkers to predict the composite primary outcome, clinically important bleeding, were developed using logistic regression and internally validated using 10‐fold cross‐validation. Discrimination, calibration, and clinical utility of the models were assessed comprehensively.

For the composite primary outcome, the AUROC for the best predictive model using TEG or ROTEM plus other biomarkers was 0.694 (0.612–0.775). The best predictive model overall included laboratory reference tests of coagulation, full blood count results, and biomarkers of multimorbidity and aging, AUROC = 0.701 (0.620–0.781), although clinical utility was not superior to using laboratory reference tests alone. Discrimination was higher for individual components of the primary outcome: large volume (≥4 units) red cell transfusion 0.754 (0.602–0.903) and large volume procoagulant transfusion 0.723 (0.590–0.857), but not for excess loss in drains/re‐sternotomy 0.701 (0.613–0.788). Calibration was generally good among the models.

The addition of biomarkers of multimorbidity and biological aging yielded only small improvements in model predictive accuracy for bleeding over tests of coagulation. Existing clinical definitions of bleeding likely represent heterogeneous phenotypes and disease mechanisms.