Document resource
Introduction Cancer may precede, follow, or coincide with SSc onset. Risk factor identification has yielded heterogeneous results.Material and Methods EUSTAR-CP154 is a 1:1 case-control study including SSc patients with/without cancer (age and SSc duration-matched). Cancers were defined as synchronous (diagnosed within 3 years before or after SSc onset), subsequent (>3 years after SSc), or previous. Risk factors (vs no-cancer) and disease trajectories were explored with multivariable logistic, mixed effect, Cox and survival analyses. Four machine learning models (Logit, RandomForest, XGBoost, SVM) were 5x-cross-validated to intercept synchronous/subsequent cancers.Results We included 908 patients: 89% female; mean age 55±13, disease duration 4±5years; 32% anti-topoisomerase-ATA and 45% anticentromere-ACA; dcSSc 28%, digital ulcers 20%, ILD 31%, pulmonary hypertension 4.4%, esophageal involvement 55%.Cancers were synchronous in 134 patients, previous in 90 (median 10years before SSc), subsequent in 230 (12years after). Breast (32%) and lung (16%) were the most frequent malignancies.Risk factors for synchronous cancers included smoking (OR 1.6; 95% CI 1.1-2.4), anti-POLR3 (OR 2.1; CI 1.1-3.7), U1RNP (OR 3.6; CI 1.1-12); digital ulcers were protective (OR 0.5; CI 0.3-0.9). Patients with synchronous cancers were more frequently negative for both ATA and ACA (36% vs 23%, p=0.01), especially in dcSSc (18% vs 5.5%, p=0.001). Patients with previous cancer showed no difference vs. cancer-free patients, but more anti-POLR3 (19% vs 8%; p=0.03).Subsequent cancers were associated with cyclophosphamide (OR 2.6; CI 1.5-4.7), negatively with mycophenolate (OR 0.4; CI 0.2-0.6) and methotrexate (OR 0.5; CI 0.3-0.8).Breast cancer was associated with anti-POLR3 (OR 2.9; CI 1.6-5.3) and negatively with digital ulcers (OR 0.4; CI 0.2-0.7). Lung cancer was associated with disease duration (OR 1.04; CI 1-1.08), smoke (OR 3.3; CI 1.9-5.9), ILD (OR 2; CI 1.1-3.5), ATA (OR 2.6, CI 1.4-5.0).At the end of follow-up, 802 (88%) patients were alive. Cancer reduced SSc survival, especially subsequent (p=0.01) malignancies. The longitudinal incidence of SSc manifestations was similar in patients with/without cancer. Risk factors for incident cancers included ILD (HR 1.7, CI 1.3-2.3), telangiectasias (HR 1.6, CI 1.2-2.1), anti-Th/To (HR 3.2, CI 1.3-7.6); presence of multiple autoantibodies (HR 0.8, CI 0.7-0.9) and intestinal involvement (HR 0.7, CI 0.6-0.96) were protective.Machine learning models showed fair sensitivity but poor specificity in intercepting cancers (figure 1).Conclusions Emerging risk factors for cancer in SSc include selected autoantibodies and clinical features; associations with therapies warrant exploration. Cancers impacts SSc survival but poorly influences clinical trajectories. Machine learning could aid cancer screening.Abstract P.357 Figure 1