Enhancing predictions of subclinical cardiac dysfunction in SLE patients through integrative machine learning...

To investigate the two-dimensional speckle-tracking echocardiography (2D-STE) parameters associated with early impaired left ventricular systolic function in SLE patients and to estimate the potential clinical factors that may trigger and influence left ventricular systolic dysfunction.

This study collected a total of 36 patients admitted to the rheumatology and immunology department of Sun Yat-sen University between January 2020 and December 2021, who were newly diagnosed with SLE and had a Systemic Lupus Erythematosus Disease Activity Index 2000 Score≥4 points. An equal number of healthy controls matched for gender and age were included. All participants underwent routine echocardiography and two-dimensional speckle-tracking echocardiography (2D-STE) examinations. Various clinical data were also collected. Machine learning and regressions were used to estimate potential risk factors for left ventricular systolic dysfunction in SLE patients.

Significant differences in 2D-STE parameters were found, including global longitudinal peak systolic strain (GLPS) (p-adjust<0.001), GLPS strain obtained from the apical two-chamber view and GLPS strain obtained from the apical four-chamber view (GLPS-A4C) (p-adjust=0.005), and GLPS strain obtained from the apical long-axis view (GLPS-APLAX) (p-adjust=0.003) between SLE patients and controls. Machine learning models, particularly GLPS-APLAX, showed excellent discrimination ability with an AUC of 0.93 (95% CI: 0.89 to 0.96) and an area under the precision-recall curve of 0.96. Multivariate regression further highlighted the inverse relationship between anti-U1 small nuclear ribonucleoprotein (U1RNP) antibodies and four GLPS-related continuous variable measures, with GLPS, GLPS-A4C and GLPS-APLAX measures having statistically significant effects (eg, GLPS coefficient=-3.71, 95% CI: -5.91 to -1.51, p=0.002).

This case-control study revealed that 2D-STE parameters can be used to predict subclinical cardiac dysfunction in SLE patients, and anti-U1RNP antibodies may be an essential predictive clinical factor. Machine learning may further assist in preliminary screening and quantifying left ventricular systolic dysfunction reasons in SLE patients.