Skip to main content
ACR Convergence 2023 Main banner
Final Program


Schedule at a Glance
Legend
Session Format
In Person
In Person
Livestreamed
Livestreamed
On Demand
On Demand
Back

Poster Session B

Myopathic rheumatic diseases (polymyositis, dermatomyositis, inclusion body myositis)

Session: (1155–1182) Muscle Biology, Myositis & Myopathies – Basic & Clinical Science Poster II

1180: Machine Learning Approach for the Prediction of anti-MDA5 Positive Dermatomyositis Mortality

Monday, November 13, 2023
9:00 AM – 11:00 AM PT
Location: Poster Hall
On Demand In Person

    Abstract Poster Presenter(s)

  • XL

    Xinlei Liu', MMed

    Department of Rheumatology and Immunology, Tianjin Medical University General Hospital
    Tianjin, Tianjin, China

    Disclosure information not submitted.

Xinlei Liu1, Fumin Qi2, Wei Wei1 and Yin Zhao1, 1Tianjin Medical University General Hospital, Tianjin, China, 2Tianjin Medical University, Tianjin, China

Background/Purpose: We describe a single-center clinical cohort of MDA5 positive dermatomyositis(DM) in China and apply a machine learning approach to predict the risk of mortality in patients.

Methods: We conducted a retrospective collection of clinical characteristics from 70 Asian patients diagnosed with MDA5+ DM between June 2017 and March 2023. The collected data encompassed various aspects, including basic information, clinical symptoms and signs, prognosis, treatment methods, imaging information, pulmonary functions, and laboratory examinations. The cohorts were divided into training and validation sets in a 4:1 ratio. Machine learning approaches (Ridge, LASSO, Elastic Net) were employed to identify the optimal variables among the 152 variables, which were used to construct a COX regression model. A Nomogram was subsequently developed to provide guidance for clinical practice.

Results: Demographic and disease characteristics of the 70 MDA5 positive DM patients can be found in Table 1. The median follow-up time was 10 months (IQR 2-40, range 0.5-71), and 29 patients (41.4%) died in our study. The deceased patients were older in age, had a shorter duration of disease, exhibited fever, respiratory symptoms, and poorer lung condition, as well as higher levels of inflammatory markers. After comparing three machine learning methods, it was found that the Partial Likelihood Deviance of the Lasso-COX regression model was 5.001, which was not the lowest. However, it was comparable to the other models in terms of deviation and showed good performance with a c-index of 0.7421. This indicates that the model has strong discriminatory power. Therefore, the Lasso-COX model was ultimately chosen to establish the risk prediction model in this study (Fig).

Conclusion: A machine learning approach was applied to construct a mortality prediction model for patients with MDA5 positive DM by variables such as age, smoke, HRCT semiquantitative visual scores, and ALB, with good discrimination and calibration.


Comparison of clinical and laboratory characteristics between surviving and deceased patients with anti-MDA5-positive


Nomogram for predicting 30-, 90-, and 180-day survival (A). ROC curves for 30-day, 90-day 180-day mortality models fitted by Lasso-COX models (B). Calibration curve of nomogram predictions of 30-day (C), 90-day (E) and 180-day (G) survival. Decision curve analysis plots of nomogram predictions of 30-day (D), 90-day (F) and 180-day (H) survival.


X. Liu: None; F. Qi: None; W. Wei: None; Y. Zhao: None.