Subsets in systemic sclerosis: One size does not fit all

Purpose: Systemic sclerosis (SSc) is a heterogeneous disease that is often divided into subsets to stratify patients and predict prognosis. We hypothesized that individual methods of subsetting would not prognosticate equally well for different outcomes or in patients at different stages of disease. Methods: We subsetted subjects with SSc using three approaches: limited versus diffuse cutaneous SSc (lcSSc, dcSSc); grouped by SSc-specific antibodies; and, grouped using unsupervised clustering. We studied patients with <2 years or between 2–4 years of disease duration, separately. Outcomes were time to death and time to development of (a) SF-36 Physical Component Score <40, (b) forced vital capacity <70% predicted, (c) echocar-diographic pulmonary hypertension, and (d) interstitial lung disease. We used Cox proportional hazards models to determine the ability of the subsets to predict the outcomes of interest, and Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) to compare the performance of the models. Results: In this international, multicentered cohort of over 500 SSc subjects with less than four years of disease duration, none of the three methods of subsetting studied was able to predict all of the outcomes of interest. Different subsetting methods predicted different outcomes within and between each disease duration group. In general, subsetting by skin performed somewhat better than the two other methods, but this was not consistent and there was considerable variability in the models. Conclusions: Subsetting SSc to consistently predict morbidity and mortality in subjects at different stages of disease remains an important challenge.