Presence of Fistulas As Predictor of Stenoses in Crohn S Diseas

2

Jürgens M et al. Fistulas as predictor of stricturing Crohn’s disease

Supplementary Data

Details on the statistical analysis

All statistical analysis were performed using the software packages STATA 10.0 (StataCorp. 2007) and R 2.6.2 (R Development Core Team, 2008). All continuous variables were expressed by their mean and corresponding standard deviations. Categorical variables were evaluated by frequencies and proportions. For presence of concomitant intestinal stenoses based on fistulas as diagnostic predictor, sensitivity, specitivity, positive predictive value (PPV), negative predictive value (NPV), and corresponding binomial 95% confidence intervals (95% CI) according to Wilson were calculated. In addition, we used (bivariate) logistic regression analysis to measure the strength of the association between stenosis and fistula presented as odds ratio (OR) and corresponding 95% CI according to Wald. In addition, we built a multivariate logistic regression model to predict the presence or absence of stenosis considering the following potential clinical and health-related predictors: fistula, gender, age (in years), disease duration (in years), smoking behaviour, body mass index (BMI), and therapy with infliximab and immunosuppressive agents. To account for the possibility of non-linear relationships between stenosis and continuous covariates such as age, age at disease onset, disease duration, and BMI, we applied fractional polynomials to build multivariate logistic regression models. We used the multivariable fractional polynomials (MFP) algorithm developed by Sauerbrei and Royston 18-20, which combines the selection of the functional forms of each continuous covariate using fractional polynomials of first and second degree with the selection of all (continuous and non-continuous) covariates via backward elimination. We set the selection level for the fractional polynomials to 0.05 using closed test procedure and set the nominal p-value for the potential predictors to 0.01 to prevent the risk of overfitting. In order to judge the ability of the multivariate logistic regression model to discriminate between patients with and without stenosis, we converted the linear predictor of the logistic regression model to the estimated probability of having stenosis. Then, each patient can be classified as having stenosis if their estimated probability exceeds a chosen value (“cutting point”) or as not having stenosis. The lower the chosen cutting point, the higher the sensitivity and the lower the specificity will be. The sensitivity and specificity of all possible cutting points can be graphically represented as the receiver operating characteristic (ROC) curve by plotting the sensitivity on 1 minus specificity. In addition, we calculated the area under the ROC curve (AUC) as a measure of overall discriminatory power of the multivariate logistic regression model. Further, the fit of the final multivariate logistic regression model was judged by an intense graphically based residual analysis 21.

To quantify the impact of the genotypes, we created contingency tables to describe the association between the presence or absence of stenosis and the different NOD2/CARD15 and IL23R genotypes. We used Fisher’s exact test to analyze the association between the presence of stenosis and each genotype. Overall, we tested 14 genotypes: 10 IL23R gene variants, the three main NOD2 mutations, and the presence or absence of any NOD2 mutation. To adjust for multiple testing, we used the method proposed by Benjamini and Hochberg to control the false discovery rate 22. To assess the additional genotype influence on the ability to discriminate between patients with and without stenosis, we used only the statistically significant compound genotypes variables and entered them in the developed multivariate logistic regression model. The results were plotted as an ROC curve and the AUC was calculated.

Supplemental Table 1. Demographic characteristics of the patient population of the retrospective and prospective study.

Retrospective data analysis

Prospective data analysis

Age at diagnosis

Disease duration

Supplemental Table 2. Fistula distribution and concomitant stenosis stratified for the anatomic region of fistula formation for both the retrospective and the prospective study. Considering that some patients had concomitantly fistulas in several anatomic regions, the total number of fistulas was higher than the number of patients.

*Analyzed were 161 fistulas in 145 different patients in the retrospective study and 44 fistulas in 42 patients in the prospective study.

Supplemental Table 3. In patients with fistulas and stenosis, fistulas were stratified for “symptomatic” and “asymptomatic” fistulas and the time of the diagnosis of fistula compared to the date of diagnosis of stenosis. This table contains n=120 patients (n=48 with “symptomatic” fistulas, n=72 with “asymptomatic” fistulas) out of the group of 125 patients with fistulas and stenosis for which exact time information on the diagnosis of fistulas and stenosis was available.

Supplemental Table 4. Disease localization 7 of the patients of the retrospective study. This analysis includes all patients of the retrospective study with complete information on their disease localization based on endoscopic and radiological methods (n=86 patients without fistulas and stenosis; n=19 with fistulas and without stenosis; n=125 with fistulas and stenosis).

Supplemental Table 5. Immunosuppressive and immunomodulatory treatment in the study population. The data are presented for n=231 CD patients of the retrospective study which had complete information on their medical treatment available. *In this subgroup, information on immunosuppressive therapy was available only in 18 patients, data on immunomodulatory agents were available in 19 patients.

Supplemental Table 6. Distribution of CD-associated NOD2/CARD15 genotypes in stricturing and non-stricturing CD

*This number represents 1 patient homozygous for one of the 3 CD-associated NOD2 variants and 2 compound heterozygous patients.

** This number represents 24 patients homozygous for one of the 3 CD-associated NOD2 variants and 13 compound heterozygous patients. The distribution of each genotype regarding the presence of stenosis (+: stenosis present; -: stenosis absent) is given in brackets.

-/-: wild-type; +/-: heterozygous; +/+: homozygous for the minor allele.

Supplemental Table 7. Distribution of CD-associated IL23R genotypes dependent on the presence of stenosis in the patients analyzed in the retrospective study. The distribution of each genotype regarding the presence of stenosis (+: stenosis present; -: stenosis absent) is given in brackets. -/-: wild-type; +/-: heterozygous; +/+: homozygous for the minor allele.