Machine learning models for predicting postoperative paraplegia in acute type A aortic dissection patients

Comprehensive Summary

Acute Type A aortic dissection (ATAAD) is a life-threatening cardiovascular condition, with 50% of those who do not undergo surgery dying from aortic rupture. Of those who do undergo surgery, 2-4% will experience paraplegia as a post-operative complication. Occurrence of paraplegia confers a worse prognosis for patients and increased hospitalization costs. In this study, Zhang et al. identified risk factors for post-operative paraplegia in patients with ATAAD and validated machine learning (ML) models for risk prediction. Using data from 572 surgical patients, the authors used LASSO regression to identify seven key risk factors: elevated pancreatic amylase and lipase, need for a secondary surgery, left subclavian artery involvement, Sun’s procedure, low hemoglobin, and older age. Seven ML predictive models were trained and compared, with the Neural Networks model emerging as the top performer (AUC 0.829). Using SHAP analysis, the model’s interpretability was improved by ranking each variable's contribution to the predicted risk. The authors conclude that the Neural Networks-based ML model can aid physicians in stratifying paraplegia risk before an operation begins. This can guide individualized surgical planning, and facilitate earlier recognition and treatment of this serious complication.

Outcomes and Implications

This study generated a practical, data-driven prediction tool to help physicians determine paraplegia risk in patients with ATAAD before any surgery has begun. The identification of several risk factors may prompt clinicians to broaden routine screening for variables such as pancreatic enzyme levels. While this model is an exciting step forward, it is limited to its single-center retrospective dataset with a small sample size of paraplegia cases. External, multi-center validation will be necessary before this model can be incorporated into routine practice.