Related Experiment Video
Updated: Jul 17, 2026
08:08
Cardiopulmonary Bypass in a Mouse Model: A Novel Approach
Published on: September 22, 2017
11.1K
Unsupervised machine learning to explore inflammation following cardiopulmonary bypass
Enrico Squiccimarro1,2, Roberto Lorusso2,3, Paolo Vetuschi4
1Division of Cardiac Surgery, Department of Medical and Surgical Sciences, University of Foggia, Foggia, Italy.
Perfusion
|August 27, 2025
Summary
Unsupervised machine learning identified two types of systemic inflammatory reaction syndrome (SIRS) after cardiac surgery: maladaptive SIRS linked to worse outcomes and adaptive SIRS with favorable results, enabling personalized care.
Area of Science:
- Cardiovascular Surgery
- Immunology
- Data Science
Background:
- Cardiac surgery with cardiopulmonary bypass (CPB) frequently triggers systemic inflammatory reaction syndrome (SIRS).
- The impact of SIRS on postoperative outcomes remains complex, with potential for both beneficial and detrimental inflammatory responses.
- Understanding these distinct inflammatory patterns is crucial for optimizing patient care.
Purpose of the Study:
- To utilize unsupervised machine learning to differentiate between adaptive and maladaptive SIRS following cardiac surgery.
- To investigate the association between identified SIRS subtypes and major postoperative adverse outcomes.
- To establish a foundation for personalized treatment strategies targeting inflammation in cardiac surgery patients.
Main Methods:
- A post hoc analysis of 1908 adult patients undergoing elective cardiac surgery with CPB.
- Application of the partitioning around medoids (PAM) algorithm using Gower distance for patient clustering based on SIRS criteria 12 hours post-surgery.
- Multivariable logistic regression analysis to assess the relationship between SIRS subtypes and a composite outcome (death, stroke/TIA, renal replacement therapy, reoperation, mechanical support, prolonged ICU stay).
Main Results:
- Systemic inflammatory reaction syndrome (SIRS) was observed in 28.7% of patients.
- Machine learning identified two distinct SIRS clusters: maladaptive SIRS (52.9%) with higher preoperative risk and significantly worse outcomes, including 30-day mortality (21.7% vs 1.6%).
- Adaptive SIRS (47.1%) was associated with favorable outcomes, comparable to SIRS-negative controls. In specific clusters, SIRS was linked to a reduced risk of the composite outcome (OR 0.44).
Conclusions:
- Unsupervised machine learning effectively distinguishes between adaptive and maladaptive SIRS in cardiac surgery.
- This classification provides a basis for personalized postoperative management strategies.
- Identifying modifiable factors associated with maladaptive SIRS may pave the way for precision medicine approaches to mitigate harmful inflammation.
