AI model predicts heart surgery complications from routine blood tests
| Source: BIOENGINEER | Original article
An interpretable machine learning model using routine blood tests can predict major adverse cardiac and cerebrovascular complications after cardiac surgery.
A new interpretable machine‑learning model that relies only on routine blood‑test results has been shown to predict major adverse cardiac and cerebrovascular events (MACCE) after cardiac surgery. The model was trained and validated on a multicentre retrospective cohort of patients who underwent cardiopulmonary‑bypass surgery at four tertiary cardiac centres in China between 2019 and 2022. By focusing on biomarkers that reflect the inflammation‑coagulation axis, the algorithm can flag patients at high risk of postoperative complications before they leave the operating theatre.
The development matters because current risk‑stratification tools in cardiac surgery depend on clinical scores that often overlook subtle biochemical signals. An approach that extracts predictive information from inexpensive, routinely collected laboratory data could improve pre‑operative counselling, tailor postoperative monitoring, and potentially reduce mortality and stroke rates. Moreover, the model’s interpretability allows clinicians to understand which blood‑test components drive the risk assessment, addressing a common barrier to AI adoption in high‑stakes medical settings.
The research builds on earlier work that demonstrated the value of multi‑output machine‑learning frameworks for predicting several postoperative outcomes simultaneously, integrating patient blood‑management biomarkers with other clinical variables. Going forward, the authors plan prospective validation in additional hospitals and exploration of how the model can be embedded into electronic health‑record systems to provide real‑time decision support. Watch for subsequent trials that test whether the tool can meaningfully alter clinical pathways and improve patient outcomes across diverse health‑care settings.
Sources
Back to AIPULSEN