In the contemporary landscape of cardiac surgery, minimizing postoperative complications remains a central goal. One such complication, systemic inflammatory response syndrome (SIRS), emerges as a critical concern due to its substantial influence on postoperative outcomes. This research, published in the Journal of Cardiothoracic and Vascular Anesthesia, rigorously examines the impact of SIRS on 30-day mortality after cardiac surgery and explores the potential of machine learning in predicting SIRS incidence.
The study retrospectively analyzed 1,908 patients who underwent elective or urgent cardiac surgeries involving cardiopulmonary bypass (CPB) between 2016 and 2020 at a single tertiary hospital. A key finding was the significant incidence of SIRS—present in 28.7% of patients—and its strong correlation with elevated 30-day mortality (12.2% in SIRS-positive patients versus 1.5% in SIRS-negative patients).
Researchers utilized established criteria from the American College of Chest Physicians/Society of Critical Care Medicine to diagnose SIRS within the first 12 postoperative hours. A comprehensive set of preoperative, intraoperative, and postoperative variables were meticulously collected and analyzed to uncover potential predictors of SIRS and mortality. Multivariable regression analysis identified several strong predictors of SIRS: female sex, preoperative anemia, leukocytosis, lymphopenia, and thrombocytosis. Additionally, intraoperative factors such as lower hemoglobin nadir, higher peak lactate levels, and the need for vasopressors were associated with increased SIRS risk.
To robustly assess the relationship between SIRS and mortality, propensity score matching created balanced cohorts, adjusting for major confounders. Even in these matched groups, SIRS was independently associated with a nearly threefold increase in 30-day mortality (odds ratio 2.77). Structural equation modeling (SEM) provided further insight by quantifying how intraoperative factors impacted mortality both directly and via SIRS mediation. For example, 24.3% of the effect of hemoglobin nadir on mortality was mediated through SIRS.
Importantly, the study incorporated machine learning to build predictive models for SIRS using clinical data. The baseline risk model (BRM), which relied solely on preoperative variables, achieved an AUC of 0.77 in cross-validation and 0.73 on a test set. A more advanced procedure-adjusted risk model (PARM), which included intraoperative variables, outperformed the BRM with an AUC of 0.81 in cross-validation and 0.82 on the test set. These models used random forest algorithms and were evaluated using standard metrics including precision, recall, and the Hosmer-Lemeshow goodness-of-fit test.
The integration of machine learning not only provided a powerful tool for risk stratification but also highlighted actionable variables that may be modifiable to improve patient outcomes. Notably, intraoperative anemia and elevated lactate levels—both modifiable through targeted interventions—emerged as top predictors. Shapley additive explanation (SHAP) values further clarified the contribution of each variable to the prediction, offering interpretability to the model outputs.
Despite its strengths, the study acknowledges limitations including the single-center, retrospective design, and lack of external validation for the machine learning models. The reliance on SIRS definitions originally developed for sepsis, rather than surgery-specific inflammatory responses, may also limit generalizability. Furthermore, while the study emphasizes SIRS as a mediator of poor outcomes, the inherent complexity of inflammation and potential confounding variables suggest that causal relationships should be interpreted cautiously.
Nevertheless, this work contributes meaningful evidence to the literature by quantifying the risk that postoperative inflammation poses after cardiac surgery and demonstrating how predictive analytics can support clinical decision-making. The authors suggest that future randomized controlled trials might benefit from incorporating machine learning-based risk stratification to enrich study populations likely to develop postoperative inflammation, potentially leading to more effective therapeutic evaluations.
In conclusion, SIRS is not merely a marker of inflammation but a potent predictor of short-term mortality following cardiac surgery. Early identification and proactive management—guided by machine learning models—could be instrumental in improving outcomes. This pioneering approach offers a template for integrating artificial intelligence into perioperative care, enabling more personalized and effective interventions in cardiac surgery.
