Background/Purpose: A recent meta-analysis of cardiovascular diseases demonstrated that the odds of heart failure (HF) was more than 2.54-fold higher in primary Sjögren syndrome (pSS) patients than controls. Cardiac magnetic resonance imaging (CMR) is useful for the early assessment of myocardial abnormalities that precede the development of overt HF. Myocardial dysfunction may arise from a number of distinct processes, including myocardial fibrosis. Global longitudinal strain (GLS), using non-contrast feature tracking cardiac magnetic resonance (FT-CMR), has been reported to be significantly associated with the extent of myocardial fibrosis. Late gadolinium enhancement (LGE) is correspond to myocardial fibrosis. In the last years, artificial neural networks (ANNs) approaches have been shown to be an established method for analyzing large datasets. ANNs could be a useful prediction tool in medical scenarios. This study aimed to predict myocardial fibrosis in pSS assessed by FT-CMR and LGE, by using ANNs models.

Methods: This was a cross-sectional study of patients with pSS registered in our hospital between January 2014 and April 2018. Healthy volunteers were recruited as a control group to be frequency-matched to the age and sex distribution of the pSS patients. pSS patients and controls with no known heart disease or risk factors who underwent CMR. We used a random forest classifier to predict myocardial abnormality in two indices (LGE, GLS). This is an algorithm that uses multiple decision trees for classification. The number of trees in the forest were set as 100. The criteria used to measure the quality of a split was Gini impurity. The classification threshold was set to 0.5. Inputs for the classifier included many valuables including attributes and observed values. The number of variables was finally reduced to 10 (e.g., age, duration, Raynaud phenomenon, body mass index, Framingham score, ESR, rheumatoid factor, IgG, hemoglobin A1c, and N-terminal pro b-type natriuretic peptide) by feature selection based on the trained model.

Results: We evaluated 52 patients with pSS (100% women; mean age, 59.5 ± 9. 0 years) and 20 healthy controls (100% women; mean age, 55.7 ± 4.5 years). All 72 subjects underwent FT-CMR and 52 patients underwent LGE. The pSS patients had significantly lower GLS (p = 0.015) than controls. Abnormal LGE was seen in 10/52 subjects (19%). We created a mathematical model to be able to predict abnormal GLS and LGE with an area under the curve value of 0.72 and 0.79, respectively. The accuracy, sensitivity, specificity, positive predictive value, and negative predictive value for prediction of GLS and abnormal LGE value were 88%, 40%, 100%, 100%, 86%, and 67%, 30%, 88%, 56%, 70%, respectively.

Conclusion: We applied ANNs to identify a prediction model for myocardial fibrosis in pSS patients without cardiac symptoms assessed by CMR. The use of laboratory and clinical items enabled to construct a mathematical model, potentially identifying pSS patients with myocardial fibrosis. This prediction tool could be used in a clinical practice setting to stratify pSS patients according to myocardial fibrosis.

« Back to ACR Convergence 2021

ACR Meeting Abstracts - https://acrabstracts.org/abstract/artificial-neural-networks-approaches-to-predict-myocardial-fibrosis-in-primary-sjogren-syndrome-patients-without-cardiac-symptoms/