Milea Timbergen

191 R0 resection was assumed in case the surgical margin was provided as “0”, “R0” or “clear”. R1 resection was assumed in case the surgical margin was provided as a “1”, or “involvement”. In cases with “2”, R2 resection was assumed and these patients were excluded from the analysis. Cases with unknown resection margin, stated as “not applicable” (NA), “unknown”, “Rx” or “999” were marked as unknown. RFS was calculated using the date of surgery and the date of recurrence. In case the date of surgery was not provided, the date of diagnosis and the date of recurrence were used to calculate this variable as we assumed that most patients underwent surgery within two months after diagnosis. End of follow-up was considered as “last date of follow-up”, or “date life or death”. IPD integrity Data of individual patients were not subjected to data checking since cohorts were from various countries and often included patients from various hospitals due to the rarity of DTF. Data integrity was checked by comparing published articles with shared databases containing data of individual patients. Statistical analysis An one-stage approach was used for this meta-analysis with IPD. All variables were collected in a single database. Categorical variables are summarized as frequencies and corresponding percentages. Continuous variables are summarized as mean values with SD or as median with IQR. Analysis of variance and the chi-square test were used to evaluate the associations between the variables tumour size, tumour site, sex, mutation type and cohort. The Kaplan-Meier method was used to calculate RFS, which was defined as the time between the date of surgery (or the date of diagnosis in case the date of surgery was not available) and the date of recurrence (or the end of follow-up). Univariable and multivariable Cox proportional hazards analyses were performed to assess the association between the outcome (recurrence) and the independent variables (age, sex, mutation type, tumour site, tumour size [log-transformed], resection margin status and cohort). The proportional hazards assumption was tested for each independent variable by including an interaction effect of the independent variable with time since surgery in a Cox regression with time-dependent covariates. In case of significance of the interaction effect, the proportional hazards assumption was considered not to be met and the variable was included as a stratification 7