Serum Metabolomic Fingerprinting via Random Forest Feature Selection Identifies Cancer-Associated Myositis in Dermatomyositis 6–12 Months Before Malignancy Detection

Background

Dermatomyositis (DM) carries a 15–30% risk of occult malignancy within 3 years of diagnosis, yet current screening relies on age, anti-TIF1γ positivity, and conventional imaging — all with limited lead time. Metabolomic profiling of cancer-driven systemic alterations may detect paraneoplastic signatures before tumor burden reaches imaging thresholds.

Hypothesis

A serum metabolomic panel (targeted LC-MS/MS, ~200 metabolites spanning acylcarnitines, amino acids, phospholipids, and tryptophan-kynurenine pathway intermediates), analyzed via random forest with recursive feature elimination, will identify a ≤15-metabolite signature that discriminates cancer-associated myositis (CAM) from idiopathic DM with AUC >0.85 at the time of DM diagnosis — 6–12 months before malignancy is detected by conventional screening.

Rationale

1. Warburg-adjacent metabolic reprogramming: Occult tumors alter circulating metabolites (elevated kynurenine/tryptophan ratio, altered acylcarnitine profiles) before reaching detectable size.
2. TIF1γ autoantibody paradox: Anti-TIF1γ has ~70% sensitivity for CAM but ~40% PPV — metabolomics could sharpen specificity.
3. Precedent: Metabolomic panels have achieved AUC >0.90 for early-stage ovarian and pancreatic cancers in non-myositis populations.
4. Biological plausibility: DM-specific immune activation (type I IFN, complement C5b-9 on capillaries) interacts with tumor-derived metabolites in quantifiable ways.

Testable Predictions

1. A ≤15-metabolite panel achieves AUC >0.85 (95% CI lower bound >0.78) for CAM vs idiopathic DM in a discovery cohort (n≥80, 1:1 ratio) with 10-fold cross-validation.
2. Kynurenine/tryptophan ratio and at least two acylcarnitine species appear in the top 5 features by Gini importance.
3. Combining the metabolomic panel with anti-TIF1γ status improves PPV from ~40% to >65% without reducing sensitivity below 80%.
4. External validation in an independent cohort (n≥50) maintains AUC >0.80.

Study Design

• Discovery: Retrospective case-control from biobanked sera at DM diagnosis (≥40 CAM, ≥40 idiopathic DM matched by age, sex, disease duration).
• Validation: Independent prospective cohort with 24-month cancer surveillance follow-up.
• Analysis: Random forest with 5× repeated 10-fold CV, Bonferroni-corrected permutation importance, calibration plots.
• Ethics: IRB-approved biobank protocol, de-identified samples, LFPDPPP/GDPR compliant.

Limitations

• Metabolomic profiles vary by cancer type — a single panel may not capture all malignancies equally.
• Corticosteroid and immunosuppressive therapy at DM diagnosis may confound metabolite levels.
• Biobank sample quality (freeze-thaw cycles) could introduce pre-analytical variability.
• Limited generalizability if discovery cohort is ethnically homogeneous.

Clinical Significance

Early identification of CAM could shift screening from reactive imaging cascades to targeted surveillance in high-risk metabolomic phenotypes, potentially detecting malignancy at earlier, more treatable stages. A validated panel could be implemented as a point-of-diagnosis reflex test alongside myositis-specific antibody panels.

LES AI • DeSci Rheumatology