Machine Learning Identifies New Predictors on Restenosis Risk after Coronary Artery Stenting in 10,004 Patients with Surveillance Angiography

Abstract

Objective: Machine learning (ML) approaches have the potential to uncover regular patterns in multi-layered data. Here we applied self-organizing maps (SOMs) to detect such patterns with the aim to better predict in-stent restenosis (ISR) at surveillance angiography 6 to 8 months after percutaneous coronary intervention with stenting.

Methods: In prospectively collected data from 10,004 patients receiving percutaneous coronary intervention (PCI) for 15,004 lesions, we applied SOMs to predict ISR angiographically 6-8 months after index procedure. SOM findings were compared with results of conventional uni- and multivariate analyses. The predictive value of both approaches was assessed after random splitting of patients into training and test sets (50:50).

Results: Conventional multivariate analyses revealed 10, mostly known, predictors for restenosis after coronary stenting: balloon-to-vessel ratio, complex lesion morphology, diabetes mellitus, left main stenting, stent type (bare metal vs. first vs. second generation drug eluting stent), stent length, stenosis severity, vessel size reduction, and prior bypass surgery. The SOM approach identified all these and nine further predictors, including chronic vessel occlusion, lesion length, and prior PCI. Moreover, the SOM-based model performed well in predicting ISR (AUC under ROC: 0.728); however, there was no meaningful advantage in predicting ISR at surveillance angiography in comparison with the conventional multivariable model (0.726, p = 0.3).

Conclusions: The agnostic SOM-based approach identified-without clinical knowledge-even more contributors to restenosis risk. In fact, SOMs applied to a large prospectively sampled cohort identified several novel predictors of restenosis after PCI. However, as compared with established covariates, ML technologies did not improve identification of patients at high risk for restenosis after PCI in a clinically relevant fashion.

Keywords: artificial intelligence; coronary artery disease; machine learning; percutaneous coronary intervention; prediction; restenosis.

Figure 1

Workflow used in this study. The left branch (in light red) shows the Viscovery workflow to identify predictors of restenosis: In the pre-processing steps, the attributes (measured parameters) are thoroughly checked; feature selection and extraction takes place as well as outlier removal and data transformation to have an optimal input for the SOM algorithm. In the clustering steps, the attributes, which are used for SOM calculation, are defined and prioritized, the SOM is calculated, and clusters on top of the map are identified by a SOM–Ward algorithm. By these steps, 19 factors of restenosis were found. These 19 potential predictors were used as input to compare the predictive power of the SOM-based model vs. the conventional model (right branch). The right branch (blue and white) shows the comparison of predictions by the conventional model (white) and the SOM model (light blue): Data had been randomly split into a training set, for identification of predicting variables, and a test set. The models, based on predictors from the training set, were applied to the test set and compared by using ROC curves.

Figure 2

Nine different clusters were retrieved after generation of a SOM ordered by restenosis parameters with subsequent SOM–Ward Clustering. Color coding from low (blue) to high (red) shows the distribution of values for parameters used for patient ordering (see Table 3). The ‘high restenosis’ cluster is indicated by labeling (top left panel).

Figure 3

Parameters compared between the ’high restenosis’ cluster against the remaining clusters as a group. The plotted ‘Profile values’ indicate the magnitude of difference in terms of numbers of standard deviations.