Using machine learning analysis to describe patterns in tissue Doppler and speckle tracking echocardiography in patients with transposition of the great arteries after arterial switch operation

Background: Advanced echocardiographic techniques such as Tissue Doppler imaging (TDI) and speckle tracking echocardiography (STE) can detect more subtle changes in ventricular performance. We aimed to study the ventricular performance in patients with transposition of the great arteries (TGA) at mid-term follow-up after the arterial switch operation (ASO) with advanced echocardiographic techniques. In addition, we sought to discover new clinical phenotypes using unsupervised machine learning. Methods: Conventional, TDI and STE echocardiographic parameters were prospectively obtained from 124 TGA patients (66.1 % male, age 10.8 ± 5.1 years, 24.2 % with ventricular septal defect) in this observational study. The data was analyzed with conventional statistics and new machine learning techniques. Results: TGA patients had reduced biventricular systolic (septal s’ Z-score −2.28 ± 1.26; RV s’ Z-score −2.16 ± 0.71; mean left ventricular longitudinal strain Z-score of the LV -2.49 ± 1.68) and RV diastolic performance (RV E/e’ Z-score 2.35 ± 1.70) mid-term after ASO. Unsupervised clustering within the TGA population revealed 3 clusters. Interestingly, cluster 3 defined a group of patients with older age at ASO, the most reduced ventricular performance as well as the highest rates of reoperations and interventions. Conclusions: Assessment of ventricular performance with TDI and STE 10 years after ASO showed that TGA patients have decreased biventricular systolic and diastolic function, especially at the septal regions. Novel analytical methods such as unsupervised clustering may help identify new clinical phenotypes from multiple variables and may contribute to improved risk stratification.