Statistical analysis

Descriptive statistics were used to summarize baseline demographic and clinical characteristics of the patients. Continuous variables are presented as mean ± s.d. and categorical variables are presented as counts and percentages. Student’s t-test and Wilcoxon-Mann–Whitney U test were used to compare continuous variables. χ² test and Fisher’s exact test were used for categorical variables. A multivariate logistic regression model was used to identify independent predictors of severe GI disease. Sociodemographic and clinical variables associated with severe GI disease with P-values <0.1 in univariate analyses were included in the multivariate model.

Cumulative incidence of severe GI disease was calculated using a Kaplan–Meier plot. Mortality rates between subjects with and without severe GI disease were compared using a Cox proportional hazards analysis, with severe GI disease modelled as a time-dependent variable and adjusted for age and sex. Subjects were censored at death or last study visit if they were still alive and had not developed severe GI disease.

The impact of severe GI involvement on HRQoL was assessed using a longitudinal mixed effect model with SF-36 PCS and MCS as outcome variables and severe GI disease and time as exposure variables. The longitudinal dataset was only available for subjects enrolled in the CSRG registry and all consecutive outcome and exposure variables were included. In order to improve model fit, different models were compared using a likelihood ratio test (anova function in R) [23]. The final crude model was generated for the fixed effects of severe GI disease and both the fixed and the random effects of time. An adjusted multivariate model was also generated with additional covariates likely to impact HRQoL (i.e. fixed effects of age, sex, mRSS, FVC, DU, inflammatory myositis and inflammatory arthritis). All the mixed models were generated using the lmer and lme function under the lme4 and lmerTest package in R [24, 25].

In multivariate analyses, P-values ⩽0.05 were considered statistically significant. Longitudinal mixed effect model were generated with R version 3.2.0 for Windows (http://r-project.org) and all other statistical analysis were performed with SAS v.9.4 (SAS Institute, Cary, NC, USA).