fetching data ... 
Background: Disease course in Systemic Sclerosis (SSc) ranges from mild, to severe with progressive organ involvement within months. Guidelines for follow-up are mainly based on expert consensus, and advocate annual assessment. So far, no data driven guidelines exist that describe tailormade systematic assessments for individual patients in line with individual disease course.
Objectives: To develop a prediction model to guide annual assessment of SSc patients tailored in accordance to disease activity.
Methods: A machine learning approach was used to develop a model that can identify patients without disease progression. SSc patients included in the prospective Leiden SSc cohort and fulfilling the ACR/EULAR 2013 criteria were included. The primary endpoint in the prediction model was disease progression which was defined as progression in ≥1 organ system, and/or start of immunosuppression or death between the two most recent visits. Using elastic-net-regularization, and including 90 independent clinical variables (100% complete), we trained the model on 75% and validated it on 25% of the patients in order to perform internal validation of the final model. We optimized the model on negative predictive value (NPV) to minimize the likelihood of missing progression. By expert assessment of the test characteristics, including swarm plots of the probability scores, cut-offs were identified for low, intermediate and high risk for disease progression.
Results: Of the 492 SSc patients (range of follow-up: 2-10yrs), disease progression during follow-up was observed in 52% (median time 4.9yrs), including myocardial progression in 29%, lung progression in 23%, skin progression in 16%, and death in 12%. Performance of the model in the test set showed an AUC-ROC of 0.66. Probability score cutoffs were defined: low risk for disease progression (<0.197, NPV:1.0; 29% of patients), intermediate risk (0.197-0.223, NPV:0.82; 27%) and high risk (>0.223, NPV:0.78; 44%). The predictive variables included in the model were: previous use of cyclophosphamide or corticosteroids, start with immunosuppressive drugs, previous gastrointestinal progression, previous cardiovascular event, pulmonary arterial hypertension, modified Rodnan Skin Score, creatinine kinase, and diffusing capacity for carbon monoxide.
Conclusion: Our machine-learning-assisted model for disease progression enabled us to classify 29% of SSc patients as ‘low risk’. In this group annual assessment programs could be less extensive than indicated by international guidelines.
| Baseline characteristics | Total
| Non-Progressors
| Progressors
|
| Demographics | |||
| Female, n (%) | 389 (79) | 193 (82) | 196 (76) |
| Age, mean (SD) | 55 (14) | 55 (15) | 55 (13) |
| Disease duration nonRP, median (IQR) | 3.2 (0.9-10.3) | 3.5 (0.8-10.5) | 3.6 (1.1-9.3) |
| Organ involvement | |||
| DcSSc, n (%) | 118 (24) | 34 (15 ) | 84 (33 ) |
| DLCO% of pred, mean (SD) | 66 (18) | 69 (18) | 64 (17) |
| FVC% of pred, mean (SD) | 98 (23) | 96 (24) | 97 (21) |
| ILD on HRCT, n (%) | 183 (37) | 66 (28 ) | 117 (46 ) |
| PAH, n (%) | 26 (5) | 10 (4) | 16 (6) |
| GAVE, n (%) | 9 (2) | 4 (2) | 5 (2) |
| Cardiac involvement, n (%) | 28 (6) | 14 (6) | 14 (5) |
| Myositis, n (%) | 8 (2) | 6 (3) | 2 (1) |
| Renal crisis, n (%) | 14 (3) | 6 (3) | 8 (3) |
| Autoantibodies | |||
| Anti-centromere, n (%) | 194 (39) | 118 (50 ) | 76 (30 ) |
| Anti-topoisomerase, n (%) | 116 (24) | 42 (18 ) | 74 (29 ) |
RP=raynaud phenomenon, dcSSc= diffuse cutaneous systemic sclerosis, mRSS=modified rodnan skin score, DU=digital ulcera, DLCO= single-breath diffusing capacity for carbon monoxide, FVC= forced vital capacity, ILD=interstitial lung disease, HRCT= high resolution computed tomography, PAH= pulmonary arterial hypertension, GAVE= gastric antral vascular extasia.
Disclosure of Interests: None declared