Predictive models for non-response to conventional treatment in polymyositis and dermatomyositis.

INTRODUCTION

Polymyositis (PM) and dermatomyositis (DM) are the major subtypes of idiopathic inflammatory myopathies with heterogeneous treatment responses. This study aimed to develop a predictive model for identifying PM/DM patients who are non-responsive to conventional therapy by assessing a range of diagnostic tools to guide individualized treatment.

METHODS

Patients with PM/DM from two medical centers (January 2010-December 2024) were included. Baseline and 12-week post-treatment clinical data were collected. Logistic regression was employed to develop both comprehensive and non-invasive predictive models. Model performance was validated using the area under the receiver operating characteristic curve (AUC) in internal and external test sets.

RESULTS

The Qingdao cohort comprised 84 patients (age 57.7 ± 1.5 years; 54 females [62.5%]; DM = 53, PM = 31) and the Liaocheng cohort included 22 patients (age 56.4 ± 3.8 years; 15 females [68.2%]; DM = 13, PM = 9). Gender, mSUVmax, and muscle bundle atrophy were identified as independent predictive factors. The comprehensive model achieved an AUC of 0.900 (95% confidence intervals [CI]: 0.813-1.009) in the training set, demonstrating superior diagnostic performance compared to the non-invasive model. The non-invasive model yielded AUCs of 0.857 (95% CI: 0.766-0.972), 0.742 (95% CI: 0.599-0.984), and 0.765 (95% CI: 0.632-0.987) in the training, internal test, and external test sets, respectively, indicating broader applicability across different cohorts. Both models showed good discrimination and calibration, and decision curve analysis further confirmed their clinical value.

DISCUSSION: These findings suggest that predictive models based on clinical, pathological, and imaging features can effectively identify PM/DM patients who are non-responsive to conventional therapy, potentially providing a tool for personalized treatment.