SERUM BASED BIOMARKER ALGORITHM TO PREDICT RAPID PROGRESSION OF FVC IN SSC ACROSS DISEASE SUBSETS.

Vishal, K.; Kim, S.; Di Battista, M.; Ross, R.; Bi, Y.; Sornasse, T.; Del Galdo, F.

SERUM BASED BIOMARKER ALGORITHM TO PREDICT RAPID PROGRESSION OF FVC IN SSC ACROSS DISEASE SUBSETS.

All Authors

Vishal, K.

Kim, S.

Di Battista, M.

Ross, R.

Bi, Y.

Sornasse, T.

Del Galdo, F.

LTHT Author

Del Galdo, Francesco
Ross, Rebecca

LTHT Department

NIHR Leeds Biomedical Research Centre
Rheumatology

Publication Date

2024

Item Type

Conference Abstract

Abstract

Introduction: Systemic sclerosis (SSc) is a multisystemic connective tissue disease, characterised by widespread organ fibrosis, vasculopathy and immune dysregulation, occurring at different stages of the condition. ILD remains the leading cause of mortality in SSc. In this study, we implemented a retrospective longitudinal serum proteomic analysis to identify prognostic biomarkers (pBMs) providing the likelihood of FVC decline in SSc. Material(s) and Method(s): We identified baseline sera with matched FVC measurements (+-3 months from the date the sera were obtained) from 100 patients (33 Very Early Diagnosis of SSc [VEDOSS], 21 Limited Cutaneous SSc [LcSSc] and 46 Diffuse Cutaneous SSc [DcSSc]) from a large multicentre observational study. FVC rapid progressors (rP) were defined as patients with a monthly rate of FVC% predicted deterioration of over 0.5%. Olink proteomics was employed to determine the relative levels of 1536 pBMs (95.8% passed quality control) in the patients' sera. Machine Learning (Random Forest) analysis was employed to identify and rank pBMs that best separated the two groups (rP vs non rP). Logistic regression and standard ROC analysis were built to determine the predictive value of pBMs. Backward elimination was then performed on the regression analysis to select the pBMs with the highest statistical significance. Result(s): 20 patients (20%) met the rP criteria in the cohort (3 VEDOSS, 2 LcSSc and 15 DcSSc). Random Forest analysis prioritised 18 pBMs as the best predictors of rapid progression event independent of disease subset or duration. The discovery statistical model including all 18 pBMs, showed an 80% positive predictive value (PPV) and an 87% negative predictive value (NPV) in identifying rP patients. ROC curve analysis showed an AUC of 0.8725 (Figure 1). Backward elimination identifying a model using 4 specific pBMs had the greatest statistical significance. The PPV was 92.3% and NPV of 90.8%. ROC curve analysis showed an AUC of 0.848 (Figure 1). Conclusion(s): In this study we have performed initial exploratory analysis to identify a group of pBMs potentially associated with the rapid decline in FVC across SSc subsets. Although our results are encouraging, validation through standard proteomic quantification and further studies in a large independent cohort would be necessary to confirm the reliability and robustness of this predictive model. This would also inform on the clinical value of a combined biomarker blood test in aiding the stratification for risk of ILD progression in SSc.