A five-factor risk-stratification model for cancer-associated myositis in idiopathic inflammatory myopathies and stage-specific survival analysis: a multicentre Japanese MYKO cohort study.

INTRODUCTION: Patients with idiopathic inflammatory myopathies (IIM) are at increased risk of malignancy, particularly around disease onset, but practical tools for risk stratification of cancer-associated myositis (CAM) remain limited. We aimed to characterise CAM in Japanese patients with IIM and develop a clinically applicable risk-stratification model. METHODS: We conducted a multicentre retrospective study using the Japanese MYKO cohort, including patients diagnosed with IIM between 2001 and 2024. CAM was defined as malignancy diagnosed within ±3 years of IIM onset. Baseline demographic, clinical, laboratory, and autoantibody variables were compared between patients with and without CAM. Standardised incidence ratios (SIRs) were calculated using age-, sex-, and calendar year-specific cancer incidence rates from the Japanese general population. Random forest analysis and logistic regression were used to derive a CAM risk-stratification model. Overall survival and stage-specific survival were assessed using Kaplan-Meier analysis. RESULTS: Among 364 patients with IIM, 42 (11.4%) had CAM. Most CAM cases occurred close to IIM onset, with 30 of 42 cases diagnosed within ±1 year. Compared with the general Japanese population, malignancy risk was increased within ±3 years of IIM onset (SIR 2.54, 95% CI 1.81-3.47) and was highest during the first year after onset (SIR 8.66, 95% CI 5.55-12.88). CAM was associated with older age at onset, male sex, smoking history, family history of malignancy, dermatomyositis subtype, dysphagia, elevated C-reactive protein (CRP) and anti-TIF1γ antibody positivity. Anti-TIF1γ positivity was the strongest predictor in random forest analysis. A five-factor model including anti-TIF1γ antibody positivity, elevated CRP, dermatomyositis subtype, older age at IIM onset, and family history of malignancy showed good discriminative performance (AUC 0.86; sensitivity 92.5%; specificity 65.1%). Five-year survival was lower in patients with CAM than in those without CAM (73.6% vs 94.6%) and was better in patients with early-stage than advanced-stage cancer (100% vs 44.0%). CONCLUSIONS: In this multicentre Japanese IIM cohort, malignancies in patients with CAM were diagnosed mainly around IIM onset. A machine learning-derived five-factor model identified patients at increased risk of CAM. An earlier-stage cancer at diagnosis was associated with better survival, highlighting the clinical importance of timely cancer detection.