Frontiers in Immunology (Sep 2022)

Prediction model for the pretreatment evaluation of mortality risk in anti-melanoma differentiation-associated gene 5 antibody-positive dermatomyositis with interstitial lung disease

  • Xianhua Gui,
  • Wangzhong Li,
  • Wangzhong Li,
  • Wangzhong Li,
  • Yanzhe Yu,
  • Tingting Zhao,
  • Ziyi Jin,
  • Kaifang Meng,
  • Rujia Wang,
  • Shenyun Shi,
  • Min Yu,
  • Miao Ma,
  • Lulu Chen,
  • Wei Luan,
  • Xiaoyan Xin,
  • Yuying Qiu,
  • Xiaohua Qiu,
  • Yingwei Zhang,
  • Min Cao,
  • Mengshu Cao,
  • Jinghong Dai,
  • Hourong Cai,
  • Mei Huang,
  • Yonglong Xiao

DOI
https://doi.org/10.3389/fimmu.2022.978708
Journal volume & issue
Vol. 13

Abstract

Read online

BackgroundAnti-melanoma differentiation-associated gene 5 antibody-positive dermatomyositis with interstitial lung disease (anti-MDA5 DM-ILD) is a disease with high mortality. We sought to develop an effective and convenient prediction tool to estimate mortality risk in patients with anti-MDA5 DM-ILD and inform clinical decision-making early.MethodsThis prognostic study included Asian patients with anti-MDA5 DM-ILD hospitalized at the Nanjing Drum Hospital from December 2016 to December 2020. Candidate laboratory indicators were retrospectively collected. Patients hospitalized from 2016 to 2018 were used as the discovery cohort and applied to identify the optimal predictive features using a least absolute shrinkage and selection operator (LASSO) logistic regression model. A risk score was determined based on these features and used to construct the mortality risk prediction model in combination with clinical characteristics. Results were verified in a temporal validation comprising patients treated between 2019 and 2020. The primary outcome was mortality risk within one year. The secondary outcome was overall survival. The prediction model’s performance was assessed in terms of discrimination, calibration, and clinical usefulness.ResultsThis study included 127 patients, (72 men [56.7%]; median age, 54 years [interquartile range, 48-63 years], split into discovery (n = 87, 70%) and temporal validation (n=37, 30%) cohorts. Five optimal features were selected by LASSO logistic regression in the discovery cohort (n = 87) and used to construct a risk score, including lymphocyte counts, CD3+CD4+ T-cell counts, cytokeratin 19 fragment (CYFRA21-1), oxygenation index, and anti-Ro52 antibody. The retained predictive variables in the final prediction model were age, Heliotrope, fever, and risk score, and the most predictive factor was the risk score. The prediction model showed good discrimination (AUC: 0.915, 95% CI: 0.846–0.957), good calibration (Hosmer–Lemeshow test, P = 0.506; Brier score, 0.12), and fair clinical usefulness in the discovery cohort. The results were verified among patients in the temporal validation cohort (n = 38). We successfully divided patients into three risk groups with very different mortality rates according to the predictive score in both the discovery and validation cohorts (Cochran-Armitage test for trend, P < 0.001).ConclusionsWe developed and validated a mortality risk prediction tool with good discrimination and calibration for Asian patients with anti-MDA5 DM-ILD. This tool can offer individualized mortality risk estimation and inform clinical decision-making.

Keywords