P249 Development and validation of a novel model to predict disease behavior progression in Crohn's disease patients with intestinal stenosis

Y Wang,H Chen,R Mao

DOI: https://doi.org/10.1093/ecco-jcc/jjad212.0379

2024-01-01

Journal of Crohn's and Colitis

Abstract:Abstract Background Approximately more than half of patients with Crohn's disease (CD) develop a stenosis phenotype and are at high risk of requiring endoscopic or surgical intervention. Surgical rates vary widely among Crohn's disease patients with intestinal strictures present. The aim of this study was to stratify Crohn's disease patients with intestinal stenosis using multiple algorithms to differentiate their poor prognostic characteristics using currently available clinical indicators. Methods This is a single-center retrospective cohort study that retrospectively collected CD patients with intestinal stenosis admitted from October 2018 to October 2022 at the First Hospital of Sun Yat-sen University. Demographic, symptomatic, and clinical test indicators at the time of first diagnosis of intestinal stenosis at baseline in our hospital were collected. Study endpoints were penetrating phenotype, or bowel resection. Imaging or surgical data were collected from patients at every 6 months, 1 year, 2 years, and 3 years from the baseline of stenosis, thus determining whether the patients had undergone disease behavioral progression. Patients who experienced progression within 3 months were excluded. Logistics regression model, random forest model, and xgboost model were constructed based on the above information. Results Twenty-seven baseline variables were retrospectively collected from 288 patients for analysis (Table 1). 202 cases were in the training set and 86 cases were in the test set, in which a five-fold cross-validation was performed. According to different algorithms, we constructed logistic regression model, random forest model and xgboost model, among which the xgboost model had the best efficacy in the test set, with an AUC of 0.734 for predicting the progression of disease behaviors after 3 years of stenosis, compared to 0.687 and 0.709 for logistic regression model and random forest model, respectively (Figure 1). in terms of terms of model interpretability, the first 4 important variables for exploring xgboos, were albumin, globulin, abdominal cramps, and abdominal distension. Finally, we attempted to predict disease progression at 6 months, 1 year, and 2 years using the model with AUCs of 0.749, 0.760, and 0.784, respectively. Conclusion We constructed a model based on simple metrics that can accurately predict the progression of disease behaviour in patients with intestinal stenosis Crohn's disease.

gastroenterology & hepatology

What problem does this paper attempt to address?

This paper aims to solve the problem of predicting the progression of disease behavior in Crohn's Disease (CD) patients with intestinal stenosis. Approximately more than half of CD patients will develop a stenotic phenotype, and these patients are at a relatively high risk of requiring endoscopic or surgical intervention. However, the surgical rates vary considerably among different CD patients. Therefore, the objective of the study is to stratify CD patients with intestinal stenosis using multiple algorithms based on existing clinical indicators, in order to distinguish the characteristics of their poor prognosis.

Specifically, the study retrospectively collected data of CD patients with intestinal stenosis who visited the First Affiliated Hospital of Sun Yat - sen University from October 2018 to October 2022, analyzed 27 baseline variables, and constructed a logistic regression model, a random forest model and an XGBoost model to predict the progression of disease behavior in these patients. Among them, the XGBoost model performed best in the test set, with an AUC value of 0.734 for predicting the progression of disease behavior after 3 years, while the AUC values of the logistic regression model and the random forest model were 0.687 and 0.709 respectively. In addition, the model was also used to predict disease progression at 6 months, 1 year and 2 years, with AUC values of 0.749, 0.760 and 0.784 respectively.

In conclusion, this study constructed a model based on simple indicators that can accurately predict the progression of disease behavior in Crohn's disease patients with intestinal stenosis. This achievement helps clinicians better assess the development of patients' conditions and provides a scientific basis for treatment decisions.