Does ML model prediction for IBD disease exist?

Inflammatory bowel disease (IBD) is one of the top five most expensive gastrointestinal conditions, with annual costs exceeding $25 billion. However, recent advances in IBD understanding, diagnosis, and treatment have led to improved remission rates and reduced surgical needs. 

In this post, I try to answer the question asked in the title. To do that, I will introduce you to one of the findings described in the publication titled: “Machine Learning-based Prediction Models for Diagnosis and Prognosis in Inflammatory Bowel Diseases: A Systematic Review”. This publication seeks to test whether computer-based prediction methods, specifically ML model predictions of IBD disease, are better than traditional methods. As AI becomes increasingly prevalent in medical research, it’s crucial to assess the reliability and validity of these custom medical algorithms. 

The analysis included cohort studies in which machine learning models based on collected clinical data in patients with IBD were developed or validated. The purpose of these models was to predict the risk of onset or progression of adverse clinical events. Then they compared the predictive performance of these models with traditional statistical models that predicted the same events.  

The study excluded ML models that: 

• utilized variables not commonly found in standard medical records, 
• employed AI to analyse endoscopic, radiological, or histopathological images, 
• included the same patient population across multiple analyses to prevent result duplication. 

 A detailed assessment of the risk of bias was carried out in each of the included studies. To achieve this, researchers used the PROBAST tool, which is specifically designed to evaluate predictive model studies. This tool allows you to assess whether the test was conducted in a reliable way and the results are reliable. The assessment included aspects such as: selection of study participants, definition of factors affecting the result, method of data analysis and the possibility of applying the results in other situations. 

Key findings on the features and outcomes associated with the model and the research were summarized. The characteristics of the patient’s examination and treatment are also presented in detail. Based on this data, the authors divided and evaluated the studies into individual criteria: 

1. Predicting response to treatment with biologics, 
2. Predicting response to thiopurines, 
3. Longitudinal disease activity and complications, 
4. Outcomes in patients with acute severe ulcerative colitis. 

Below, we’ll discuss two case studies that show the most promising results. 

The researchers wanted to predict whether treatment with a new drug for bowel disease would be effective in a particular patient. They used special computer models called “random forests”. These models analysed various information about patients, such as blood test results, to predict whether the patient’s disease would improve. 

All three studies showed that the models created were useful in predicting the effectiveness of treatments. The best results were obtained when the model analysed data collected both at the start of treatment and after several weeks. Researchers also created simplified models that were easier to use but still produced good results. 

Could the researchers predict how patients with inflammatory bowel disease would respond to thiopurine drug treatment? To find out, they created special computer models that analysed various data about patients, such as blood test results. All studies showed that the computer models were very effective in predicting different outcomes of thiopurine treatment.  

These models were able to predict: 

• whether a patient would respond well to the drug, 
• whether they would follow the doctor’s instructions,  
• and whether there would be certain problems with the drug. 

Analysts investigated whether computer learning models can better predict how the disease will progress in patients with enteritis (IBD) than traditional statistical methods. Researchers have analysed many different studies that have used ML models to predict various things related to IBD, such as: 

• Will the treatment be effective? 
• Will the disease get worse? 
• What will be the long-term effects of the disease? 

The authors of the described publication also investigated whether ML model predictions of IBD disease, can better predict the course of the disease in patients with enteritis than traditional methods. It turned out that ML models are more effective in predicting various aspects of the disease, such as the effectiveness of treatment or the development of complications. However, to fully exploit the potential of this technology, further research is needed to create even more precise models. 

If this topic is interesting to you I recommend our previous blog posts where we described other predictive algorithms in the medical field: Algorithm to predict pancreatic cancer risk based on disease trajectory or Predictive AI kidney disease models in forecasting end-stage. 

References: https://doi.org/10.1093/ecco-jcc/jjab155