# Developing a practical machine learning model to predict post implantation syndrome after endovascular aneurysm repair

## TL;DR

This study develops a machine learning model to predict post-implantation syndrome after a common aneurysm repair procedure, using patient data to improve clinical decision-making.

## Contribution

A novel machine learning model using LDA is developed to predict post-implantation syndrome after EVAR, incorporating 11 preoperative and intraoperative risk factors.

## Key findings

- 11 risk factors were identified for predicting post-implantation syndrome after EVAR.
- The LDA model achieved an AUC of 0.794 and accuracy of 0.697 in predicting PIS.
- The model may assist clinicians in identifying patients at risk for PIS and its complications.

## Abstract

Post-implantation syndrome (PIS) is recognized as a systemic inflammatory response following endovascular aneurysm repair (EVAR), characterized by a high frequency of occurrence and the capacity to provoke cardiovascular complications and extend the duration of hospitalization. The objective of our study is to construct a predictive algorithm through the application of machine learning (ML) techniques to forecast the onset of PIS subsequent to EVAR procedures.

The data of 618 patients were retrospectively retrieved from the Electronic Health Record (EHR) system of Foshan First People’s Hospital, covering the period from January 2018 to December 2022. Least absolute shrinkage and selection operator (LASSO) regression is used for data preprocessing and variable selection. Eight ML models are developed to predictive PIS after EVAR. The area under the receiver operating curve (AUC), F1-score, accuracy, sensitivity, and specificity were evaluated as the model performances.

According to the exclusion criteria of 618 patients, 594 patients were finally included in the statistical analysis, and the incidence rate of PIS was 16.8%. Our research results show that there are 11 features that predict risk factors for PIS, including intraoperative use of etomidate, muscle relaxants, polyester endograf (knitted process), polyester endograf (woven process), glucocorticoids, phenylephrine, platelet count, age, absolute neutrophil count, surgical duration, and creatinine. The linear discriminant analysis (LDA) model performs the best among prediction models, with an AUC of 0.794, F1 score of 0.438, sensitivity of 0.7, specificity of 0.697, and accuracy of 0.697.

Our study selected 11 preoperative and intraoperative variables to develop a ML model based on LDA for predicting PIS after EVAR and the model may help assist clinical decision-making.

The online version contains supplementary material available at 10.1186/s42155-026-00668-w.