Statistical Analysis

Patients who completed a PSC were grouped by the number of relapses in the last 12 months (0, 1, 2, and ≥3 relapses). Demographic and clinical characteristics were described for each group using summary statistics. Numeric variables were described using count, mean, standard deviation (SD), median, interquartile range (IQR), min, and max; categorical variables were described using count and percentage. Analysis of variance (ANOVA) and Chi-squared tests were used to compare outcomes across groups for numeric and categorical variables, respectively.

Multiple regression analyses were conducted with selected outcomes as the dependent variable and relapse groups as the main independent variable of interest. Other regression covariates adjusted for were age, sex, body mass index (BMI) and Charlson comorbidity index. The type of regression model varied according to the dependent/outcome variable, with linear regression for numerical outcomes (reported as β coefficients), logistic regression for binary outcomes (reported as odds ratios [ORs]), and multinomial logistic regression where the outcome was categorical with more than 2 categories (reported as relative risk ratios [RRRs]). Wald tests were used to generate p-values for each relapse group, with 0 relapse as the reference. Likelihood ratio tests were also used to generate p-values comparing 0 relapses against 1, 2, and ≥3 relapses combined (i.e., 0 relapses compared to any number of relapses).

Assumptions for models were checked by examining the residuals for linearity and normality (where appropriate) and examining the variance inflation factors and condition number for multicollinearity. In the majority of models, no issues were observed with either. However, some issues were observed in regressions conducted on a small sample size so these results should be interpreted more cautiously.

All analyses were conducted in Stata v16.1 (StataCorp, 2019).