Researchers studied 227 patients with inflammatory bowel disease (IBD) to figure out which patients would respond well to two specific medications: vedolizumab and ustekinumab. Using artificial intelligence and machine learning, they analyzed blood tests and patient information to predict who would get better. They found that certain blood markers—like white blood cell count, inflammation levels, and vitamin B12—were the best predictors of whether a treatment would work. This research suggests doctors could use these simple blood tests to help choose the right medication for each patient, rather than guessing which treatment to try first.

The Quick Take

  • What they studied: Can doctors predict which IBD patients will respond well to specific biologic medications by looking at their blood tests and medical history?
  • Who participated: 227 patients with inflammatory bowel disease (either Crohn’s disease or ulcerative colitis) who received treatment at a Spanish hospital between 2015 and 2022
  • Key finding: Machine learning models using blood test results and patient information could predict treatment success with 65-87% accuracy. The most important predictors were white blood cell count, fecal calprotectin (a gut inflammation marker), C-reactive protein (a blood inflammation marker), and vitamin B12 levels. Patients with higher inflammation markers were less likely to respond well to treatment.
  • What it means for you: In the future, your doctor might use a simple blood test to predict whether a specific IBD medication will work for you before starting treatment. This could help avoid wasting time on medications that won’t help and get you on the right treatment faster. However, this is still early research and needs testing in larger groups before it becomes standard practice.

The Research Details

Researchers collected medical information from 227 IBD patients treated at one hospital in Spain over 7 years (2015-2022). They gathered three types of data: basic patient information (age, sex), clinical details (disease type, symptoms), and blood test results (white blood cells, inflammation markers, vitamin levels). They then used a computer program called XGBoost—a type of artificial intelligence—to learn patterns from this data. The AI was trained to predict whether patients would go into remission (no symptoms) or respond well to treatment at two time points: 26 weeks and 52 weeks after starting medication. The researchers tested this on two different medications: vedolizumab and ustekinumab. They measured how accurate the predictions were using several methods, including checking if the predictions worked equally well for men and women, and for younger and older patients.

This approach is important because IBD patients currently have to try medications somewhat randomly—doctors prescribe a treatment and wait weeks to see if it works. If a medication doesn’t help, patients suffer longer while waiting to try something else. By identifying which blood tests predict success, doctors could make smarter choices upfront. The study also used ‘real-world evidence’ from actual patient records rather than controlled experiments, which means the results reflect what actually happens in regular medical practice.

This is an early-stage exploratory study, which means it’s a first step rather than definitive proof. The researchers were honest about limitations: the study was done at only one hospital with a relatively small group of patients, which means results might not apply everywhere. The AI predictions worked better for some outcomes (87% accuracy) than others (65% accuracy). The study also noted that predictions were less reliable for smaller groups of patients (like very young or very old patients), probably because there wasn’t enough data to learn from. The researchers recommend larger studies across multiple hospitals before doctors start using this in regular practice.

What the Results Show

The machine learning models successfully predicted treatment outcomes with varying accuracy levels. For predicting remission at 52 weeks, the model achieved an F1 score of 0.842 (a measure of accuracy combining precision and recall). For predicting response at 26 weeks, it achieved 0.869. For predicting response at 52 weeks, it achieved 0.649. In simpler terms, the model was quite good at predicting short-term success but less reliable at predicting long-term outcomes. The most important blood test predictors were: white blood cell count, fecal calprotectin (a marker of gut inflammation), C-reactive protein (a marker of general inflammation), and vitamin B12 levels. Patients with higher levels of these inflammation markers were significantly less likely to respond well to treatment, suggesting that inflammation severity is a key factor in treatment success.

When researchers looked at whether the predictions worked equally well for different groups, they found some differences. The models performed differently for men versus women, and for younger versus older patients, though these differences were partly due to having fewer patients in some groups. This suggests that treatment response might vary somewhat between different demographic groups, though more research is needed to understand why. The study also confirmed that both vedolizumab and ustekinumab showed variable responses across the patient population, supporting the idea that personalized prediction tools could be valuable for both medications.

This research builds on existing knowledge that inflammation markers predict IBD treatment response, but it’s novel in using machine learning to combine multiple factors together. Previous studies looked at individual markers one at a time; this study shows that combining several blood tests and patient information creates better predictions. The finding that white blood cells, calprotectin, CRP, and B12 are important matches what doctors have observed clinically, but using AI to weight these factors together appears to improve prediction accuracy compared to traditional methods.

The study has several important limitations. First, it included only 227 patients from one hospital in Spain, so results might not apply to different populations or geographic regions. Second, the study was retrospective, meaning researchers looked back at past records rather than following patients forward, which can introduce bias. Third, the predictions were less accurate for long-term outcomes (52 weeks) than short-term (26 weeks), suggesting the models might be less reliable for predicting lasting remission. Fourth, the study had unequal numbers of men and women and different age groups, making it harder to know if predictions work equally well for everyone. Finally, the researchers acknowledge this is an exploratory study meant to generate ideas for future research, not provide definitive clinical guidance.

The Bottom Line

Based on this early research, we cannot yet recommend that doctors routinely use these predictions in clinical practice. However, this research suggests promise for future use. The findings support the importance of measuring blood inflammation markers (especially fecal calprotectin and CRP) in IBD patients before starting biologic therapy. Doctors should continue using current methods to choose treatments, but this research points toward a future where blood tests might help personalize medication selection. Confidence level: Low to Moderate—this is preliminary evidence that needs validation in larger studies.

This research is most relevant to: (1) IBD patients considering vedolizumab or ustekinumab therapy who want to understand what factors predict success, (2) gastroenterologists treating IBD who are interested in emerging personalized medicine approaches, (3) researchers developing better tools for IBD management. This research should NOT be used by patients to self-diagnose or change their treatment without consulting their doctor. The findings are too preliminary for routine clinical use.

If this research leads to clinical tools, benefits would likely appear within 26-52 weeks of starting treatment, as that’s when the models make predictions. However, it will take several years of additional research before these tools are ready for regular use in doctors’ offices. Patients should not expect their doctor to use these predictions immediately.

Want to Apply This Research?

  • Track your blood test results monthly: white blood cell count, fecal calprotectin level, C-reactive protein (CRP), and vitamin B12. Record these alongside your symptom severity score (0-10 scale) to see if patterns emerge in your personal response to treatment.
  • Work with your doctor to ensure you get regular blood tests before and after starting IBD medication. Use the app to record these results and share them with your healthcare provider. This creates a personalized record that could be valuable as new prediction tools become available.
  • Create a dashboard showing your inflammation markers over time alongside your symptom improvements. This helps you and your doctor see patterns in how your body responds to treatment and could support future conversations about medication adjustments or changes.

This research is preliminary and exploratory in nature. The machine learning models described have not been validated for clinical use and should not be used to make treatment decisions without consulting your gastroenterologist. If you have IBD, continue following your doctor’s recommendations for treatment selection. This article is for educational purposes only and does not constitute medical advice. Always discuss any changes to your IBD treatment plan with your healthcare provider.