Figure - available from: Scientific Reports
This content is subject to copyright. Terms and conditions apply.
Learning curves of Model 1 (mRSS and WHO-FC ≥ II), (a) Training and validation loss, (b) the model accuracy in training and validation. Title = (a) Model loss (b) Model accuracy. X-axis = (a) Loss (b) Accuracy. Y-axis = (a) number of epochs (b) number of epochs.
Source publication
Clinical predictors of mortality in systemic sclerosis (SSc) are diversely reported due to different healthcare conditions and populations. A simple predictive model for early mortality among patients with SSc is needed as a precise referral tool for general practitioners. We aimed to develop and validate a simple predictive model for predicting mo...
Citations
The clinical manifestations of systemic sclerosis (SSc) are highly variable, resulting in varied outcomes and complications. Diverse fibrosis of the skin and internal organs, vasculopathy, and dysregulated immune system lead to poor and varied prognoses in patients with SSc subtypes. Therefore, this study aimed to develop a personalized tool for predicting the prognosis of patients with SSc.
A cohort of 517 patients with SSc were recruited between January 2009 and November 2021 at Xijing Hospital in China, and 266 patients completed the follow-up and performed in the survival analysis. Risk factors for death were identified using Cox survival analysis and random survival forest-based machine-learning methods separately. The consistency index, area under the curve (AUC), and integrated Brier scores were used to compare the predictive performance of the different prognostic models.
The results of Cox-based multivariate regression analysis suggested that pulmonary arterial hypertension, digital ulcer, and Modified Rodnan Skin Score (mRSS) were independent risk factors for poor prognosis in patients with SSc and significant risk factors in random survival forest (RSF) surveys. A nomogram was plotted to evaluate the prognostic risk to facilitate clinical assessment; the RSF model had better predictive performance than the Cox model, with 3- and 5-year AUCs of 0.74 and 0.78, respectively.
Machine-learning models can help us better understand the prognosis of patients with SSc and comprehensively evaluate the clinical characteristics of each individual. The early identification of the characteristics of high-risk patients can improve the prognosis of those with SSc.
Key Points
• Regarding predictive performance, the random survival forest model was more effective than the Cox model and had unique advantages in analyzing nonlinear effects and variable importance.
• Machine learning using the simple clinical features of patients with systemic sclerosis (SSc) to predict mortality can guide attending physicians, and the early identification of high-risk patients with SSc and referral to experts will assist rheumatologists in monitoring and management planning.
