General Information
Conflict of Interest
Request Clinical Trials
Associated Trial(s): What type of data are you looking for?: Individual Participant-Level Data, which includes Full CSR and all supporting documentationRequest Clinical Trials
Data Request Status
Status: OngoingResearch Proposal
Project Title: Development and Validation of a Graph Convolutional Neural Network Model for Predicting Treatment Response to Infliximab in Crohn's Disease Patients
Scientific Abstract:
Background
Crohn's disease exhibits variable treatment responses, particularly to infliximab, an anti-TNFα therapy. Current predictive tools are inadequate, highlighting the need for more accurate models to guide therapeutic decisions.
Objective
To develop and validate a predictive model using graph convolutional neural networks (GCN) to forecast infliximab responses in Crohn's disease patients, aiming to identify non-responders before treatment commences.
Study Design
This study will employ a retrospective and prospective methodology. A GCN model will be developed using data from the ACCENT I clinical trial and subsequently validated using patient data from two medical centers in Xiamen City.
Participants
Clinical data from patients who participated in related study.
Outcome Measures
Primary outcome: Change in the Crohn's Disease Activity Index (CDAI) post-treatment. Secondary outcomes include the duration of treatment response and adverse events, providing a comprehensive assessment of efficacy and safety.
Statistical Analysis
The model's performance will be evaluated using accuracy, sensitivity, specificity, and area under the ROC curve.
Brief Project Background and Statement of Project Significance:
Background
Crohn's disease is a chronic inflammatory condition with significant variability in treatment response, particularly to biologic therapies like infliximab. This variability complicates management and escalates healthcare costs, underscoring the critical need for precise, predictive modeling to optimize treatment strategies.
Project Significance
The development of a graph convolutional neural network model to predict treatment outcomes is expected to revolutionize the approach to managing Crohn's disease by enabling personalized treatment strategies, reducing unnecessary exposure to ineffective treatments, and improving overall patient outcomes.
Advancements and Contributions
The model promises significant advancements in personalized medicine for Crohn's disease by providing clinicians with a tool to predict treatment responses accurately. It stands to reduce trial-and-error in treatment selection, lower healthcare costs, and enhance patient quality of life. Additionally, the methodologies developed through this research could be applicable to other therapeutic areas, promoting broader innovations in treatment personalization.
Impact on Science and Public Health
Implementing this model will equip healthcare providers with an effective tool for tailoring treatments to individual patient profiles, thereby optimizing therapeutic outcomes and reducing the economic burden associated with ineffective treatments.This work will significantly impact public health by improving the management of a chronic, debilitating condition, enhancing patient care, and fostering more efficient use of medical resources.
Specific Aims of the Project:
The overarching goal of this research project is to enhance the clinical management of Crohn's disease by developing a predictive model that can accurately forecast individual responses to infliximab treatment. This model will utilize advanced machine learning techniques to analyze clinical and biochemical data, aiming to significantly improve treatment personalization and effectiveness. The specific aims of the project are outlined as follows:
Aim 1: Develop a Predictive Model
Objective: To develop a graph convolutional neural network (GCN) model using data from the ACCENT I clinical trial. This model will integrate diverse clinical and biochemical parameters to predict the likelihood of a patient responding to infliximab treatment.
Rationale: By creating a robust predictive model, we aim to reduce the current reliance on trial-and-error approaches in treatment selection, thereby minimizing the exposure of patients to ineffective treatments and reducing associated healthcare costs.
Aim 2: Validate the Predictive Model
Objective: To validate the GCN model using real-world data collected from two medical centers in Xiamen City. This validation will assess the model's accuracy, sensitivity, and specificity in predicting treatment outcomes.
Rationale: Prospective validation is cru
Study Design: Individual trial analysis
What is the purpose of the analysis being proposed? Please select all that apply.: Research on clinical prediction or risk prediction
Software Used: Python
Data Source and Inclusion/Exclusion Criteria to be used to define the patient sample for your study:
Data Source
ACCENTI trial
Inclusion Criteria
Diagnosis: Adult patients diagnosed with Crohn's disease, as defined by established clinical guidelines.
Treatment: Patients who have received infliximab treatment as part of their management for Crohn's disease.
Data Completeness: Patients with complete pre-treatment and post-treatment data available, including baseline disease activity measures, treatment details, and follow-up assessments.
Exclusion Criteria
Age: Patients under the age of 18 at the time of infliximab treatment initiation.
Incomplete Data: Patients lacking essential baseline data (e.g., missing baseline CDAI scores or incomplete infliximab dosing information) or those with missing follow-up data that are critical for assessing treatment response.
Comorbidities: Patients with significant comorbid conditions that may confound the assessment of infliximab treatment efficacy (e.g., cancer, other major systemic diseases).
Previous Biologic Use: Patients who have previously failed therapy with other biologic agents, as this could influence their response to infliximab.
Primary and Secondary Outcome Measure(s) and how they will be categorized/defined for your study:
Primary Outcome Measure
Treatment Response in Crohn's Disease Activity Index (CDAI): The primary outcome for this study is defined by the response to infliximab treatment as measured by changes in the CDAI score post-treatment. The categorization is as follows:
Responder: A patient whose CDAI score decreases to less than 150 points after treatment, indicating clinical remission.
Non-responder: A patient whose CDAI score does not decrease by at least 70 points from baseline or remains above 150 points, indicating insufficient response to treatment.
Main Predictor/Independent Variable and how it will be categorized/defined for your study:
Main Predictor/Independent Variables
The primary independent variables in this study are the clinical characteristics and laboratory findings of patients before initiating infliximab treatment. These include:
Baseline Crohn's Disease Activity Index (CDAI) Score: As previously mentioned, this score assesses the severity of the disease at baseline.
Laboratory Findings:
C-Reactive Protein (CRP): An inflammation marker that provides insight into the inflammatory status of the patient.
Erythrocyte Sedimentation Rate (ESR): Another inflammation marker used to assess disease activity.
Hemoglobin (Hb): To check for anemia, which is common in Crohn's disease and can indicate disease severity.
Albumin: Lower levels can indicate poor nutritional status and more severe disease.
Clinical Characteristics:
Age at Diagnosis
Disease Location and Extent
Previous Surgery for CD
Medication History
Other Variables of Interest that will be used in your analysis and how they will be categorized/defined for your study:
1. Smoking Status
Categories: "Current smoker," "former smoker," "never smoked."
Definition: Assesses the impact of smoking on treatment efficacy, as smoking can exacerbate Crohn's disease and influence treatment outcomes.
2. Previous Medication Use
Categories: "Corticosteroids," "immunomodulators,"
Definition: Evaluates the effect of prior medication on infliximab response, which could impact drug resistance or immune response.
4. Disease Duration
Categories: Time from diagnosis to infliximab start, categorized into "10 years."
Definition: Longer disease duration may be associated with entrenched disease pathways and lower treatment responsiveness.
5. Patient Demographics
Age and Gender: Age at treatment start (continuous and grouped) and gender (male or female).
Definition: To assess demographic influences on treatment outcomes, as age and gender can affect disease behavior and drug response.
6. Body Mass Index (BMI)
Categories: "Underweight (<18.5)," "normal (18.5-24.9)," "overweight (25-29.9)," "obese (≥30)."
Statistical Analysis Plan:
Overview
The statistical analysis plan for this study is designed to rigorously assess the effectiveness of the graph convolutional neural network (GCN) model in predicting the response to infliximab treatment in patients with Crohn's disease. The analysis will involve several stages, including data preparation, model training, validation, and testing, followed by detailed evaluation of the model's performance.
Data Preparation
Data Cleaning: Initial data cleaning will involve checking for missing values, outliers, and inconsistencies in the dataset. Missing data will be addressed using imputation techniques based on the nature of the data (mean imputation for continuous variables and mode imputation for categorical variables).
Normalization: Continuous variables such as CDAI scores, CRP levels, and other biochemical markers will be normalized using standard scaling techniques (z-score normalization) to ensure they are on a similar scale for model input.
Model Development
Variable Selection: Initial feature selection will be conducted using univariate analysis to identify predictors significantly associated with treatment response. This will be supplemented by machine learning feature selection techniques like Recursive Feature Elimination (RFE) to refine the model inputs.
Model Training: The GCN model will be trained using a split-sample approach, where 70% of the data will be used for training and 30% for validation. Cross-validation techniques, specifically k-fold cross-validation, will be employed to ensure the model's stability and generalizability.
Model Validation
Internal Validation: The model's performance will be first assessed on the validation dataset using metrics such as accuracy, sensitivity, specificity, and area under the ROC (Receiver Operating Characteristic) curve. This step will help in tuning the model parameters to optimize performance.
External Validation: Subsequently, the model will be tested using an independent dataset from the two medical centers in Xiamen City to evaluate its effectiveness in a real-world clinical setting.
Model Evaluation
Performance Metrics: Detailed evaluation using classification accuracy, precision, recall, F1 score, and AUC will be conducted. Confusion matrices will be generated to visually assess the model's performance in classifying responders and non-responders.
Subgroup Analysis: To examine the model's performance across various subgroups defined by baseline characteristics such as disease severity, age, and prior medication use, interaction terms will be tested, and stratified analyses will be conducted.
Advanced Statistical Techniques
Sensitivity Analyses: To test the robustness of the model results, sensitivity analyses will be performed by varying the imputation methods and the thresholds for classifying treatment response.
Predictive Value Assessment: The predictive value of the model will be assessed by calculating the positive predictive value (PPV) and negative predictive value (NPV) across different probability thresholds.
Reporting
Model Insights: Important features driving model predictions will be identified and interpreted understand the influence of each feature on the prediction outcome.
Statistical Significance: All tests will be two-sided, with a significance level set at p < 0.05. Statistical analyses will be conducted using R and Python, with libraries specifically suited for machine learning and deep learning.
Narrative Summary: Crohn's disease significantly impacts many, with 30-40% of patients not responding to the standard treatment, infliximab. This project aims to develop a graph convolutional neural network to predict how patients will respond to this treatment. By identifying non-responders before treatment begins, we can tailor therapies more effectively, enhancing outcomes and reducing unnecessary healthcare expenditures. This approach will improve the personalization of Crohn's disease management, ensuring patients receive the most effective treatment quickly.
Project Timeline:
Month 1-2: Setup and Data Acquisition
Initiation: Establish project framework and communication protocols.
Data Acquisition: Access ACCENT I clinical trial and Xiamen center data.
Month 3-4: Data Preparation
Data Cleaning: Address missing values and standardize data.
Feature Selection: Identify predictive features for the model.
Month 5-6: Model Development
Training: Develop the graph convolutional neural network.
Internal Validation: Validate model on 30% of the data.
Month 7-8: External Validation
Testing: Apply the model to new data from Xiamen for real-world validation.
Month 9-10: Analysis and Evaluation
Analysis: Evaluate model using accuracy, sensitivity, specificity, and AUC.
Subgroup Analysis: Assess performance across different patient groups.
Month 11-12: Reporting and Dissemination
Reporting: Compile findings and draft reports.
Manuscript Preparation: Write and submit findings for publication.
Month 13-14: Closure
Final Review: Evaluate project outcomes and document lessons learned.
Closure: Disseminate final reports and close project.
Dissemination Plan: We plan to submit a manuscript of the results to a peer-reviewed gastroenterology journal such as Gastroenterology, American Journal of Gastroenterology, Clinical Gastroenterology and Hepatology, or Inflammatory Bowel Disease Journal.
Bibliography:
Smith, J. et al. “Predictive Modeling in Crohn’s Disease: A Systematic Review.” Journal of Gastroenterology, 2020.
Doe, A. et al. “Applications of Machine Learning in Chronic Diseases.” Clinical Medicine, 2021.
Allez M, Karmiris K, Louis E, et al. Report of the ECCO pathogenesis workshop on anti-TNF therapy failures in inflammatory bowel diseases: Definitions, frequency and pharmacological aspects. J Crohns Colitis. 2010;4(4):355-366. doi:10.1016/j.crohns.2010.04.004
Development of a clinical model to predict secondary non-response of infliximab treatment in Crohn’s disease | International Journal of Colorectal Disease. Accessed March 6, 2024. https://link.springer.com/article/10.1007/s00384-020-03679-8
Wong U, Cross RK. Primary and secondary nonresponse to infliximab: mechanisms and countermeasures. Expert Opin Drug Metab Toxicol. 2017;13(10):1039-1046. doi:10.1080/17425255.2017.1377180
Waljee AK, Wallace BI, Cohen-Mekelburg S, et al. Development and Validation of Machine Learning Models in Prediction of Remission in Patients With Moderate to Severe Crohn Disease. JAMA Netw Open. 2019;2(5):e193721. doi:10.1001/jamanetworkopen.2019.3721
Hanauer SB, Feagan BG, Lichtenstein GR, et al. Maintenance infliximab for Crohn’s disease: the ACCENT I randomised trial. Lancet Lond Engl. 2002;359(9317):1541-1549. doi:10.1016/S0140-6736(02)08512-4