324 language (v.3.8.5, Python Software Foundation, https://www.python.org), using the pandas (v.1.2.3) and numpy (v.1.20.0) packages and SPSS Statistics software package (v.25.0) (SPSS Inc., Chicago, IL, USA). Data visualization was performed using seaborn (v.0.11.1) and matplotlib (v.3.4.1) packages in Python and GraphPad Prism (v.9.1) (La Jolla, CA, USA). P-values ≤ 0.05 were considered statistically significant. Discrimination analyses of Montreal Behavior subclasses Univariable logistic regression analyses (method: enter) were performed to assess the discriminative ability of the biomarkers regarding Montreal disease classifications. Receiver operating characteristics (ROC) statistics with the area under the curve (AUC) as an overall measure of fit and corresponding 95% confidence intervals (CI) were used to assess the discriminative ability of biomarkers with regard to the outcomes. ROC curves and AUCs were computed using the non-parametric, tie-corrected trapezoidal approximation method. Significant results (preselection threshold: nominal P-value ≤ 0.05) from univariable analyses were incorporated into multivariable logistic regression analyses. First, biomarker levels were adjusted for demographic or clinical characteristics by performing multivariable backward logistic regression analyses, taking into account variables that were significantly associated with the outcome of interest (derived from univariable logistic regression analyses). Second, biomarker levels were adjusted for their unstandardized residual values by performing multivariable logistic regression analysis in order to determine the predictive value of solely the biomarker. Unstandardized residual values were derived from linear regression using the same confounding factors from the first multivariate analysis in relation to the specific biomarker. The discriminative performance of adjusted models was determined by ROC estimation of combined predicted probabilities from the models with bootstrap inference (n=500 iterations). Additionally, multivariable models were internally validated using k-fold cross-validation (k=10). In this procedure, the dataset was randomly divided into k equally sized folds, where each fold was then left out (10% of cases) while the model was fitted to the remaining k-1 folds (90% of cases, ‘training set’) and predictions were obtained for the left-out part (‘test set’). This procedure was repeated 10 times where AUCs from each fold were averaged and bootstrapped to achieve statistical inference, resulting in a cross-validated AUC (cv-AUC). Prospective (follow-up) analyses Biomarkers were analyzed for associations with the risk of future progression or recurrence of stricturing or penetrating disease or the risk of surgical interventions using Kaplan-Meier survival analysis. Survival distributions were assessed for tertiles of biomarker levels, which were pairwise compared using log-rank tests. Survival time was defined from the time of sampling (baseline) until the first date that progression was evident (by endoscopy and/or radiography) or CD-related surgical intervention was performed, or until the last contact date with their treating gastroenterologist (end of follow-up). Cox proportional hazards regression analyses were performed to assess the prospective associations between biomarker levels and the risk of progression or occurrence of surgical interventions. Results were expressed as hazard ratios Chapter 10
RkJQdWJsaXNoZXIy MjY0ODMw