(536b) Predictive Analytics for Irritable Bowel Syndrome Diagnosis and Management

Irritable Bowel Syndrome (IBS) is a common gastrointestinal disorder affecting the large intestine. It is estimated to affect 10-15% of the world population and 25 to 45 million people in the United States alone. IBS is classified as a chronic condition which is broadly defined by the Center for Disease Control and Prevention (CDC) as a condition that lasts one or more years, requires ongoing medical attention, and/or limits daily activities.¹ Symptoms of IBS include cramping, abdominal pain, bloating, constipation, diarrhea, or both.² These symptoms are often inconsistent and can become more severe as a response to certain stimuli such as stress, anxiety, or a poor diet. Dealing with such irregular symptoms decreases the quality of life of an individual. There are no definitive causes for IBS nor is there a definitive way to diagnose patients with IBS.

IBS can be modeled as a biopsychosocial disorder where biological (food allergies, dietary habits, etc.), psychological (anxiety, stress, etc.), and environmental and social factors (seasonal variations, work environment, etc.) act interdependently and trigger IBS symptoms.³ IBS is classified into four subtypes: IBS-C when patients predominantly experience constipation, IBS-D when patients predominantly experience diarrhea, IBS-M when patients experience both constipation and diarrhea equally, and IBS-U are unsubtyped patients that do not fit any of the previous three categories. Medical professionals may try to perform certain tests such as ordering a metabolic panel, testing for celiac disease, checking for lactose or glucose intolerance, etc., but there is no guarantee that these tests will indicate any issues within the individual.⁴ Criteria have been developed to try and aid in the identification and diagnosis of IBS. The most well-known of these criteria are the Manning criteria, the Kruis criteria, and the Rome criteria. These criteria consist of questionnaires and scoring to evaluate the possibility that an individual is experiencing IBS and assess what subtype of IBS an individual may have.⁵ Doctors base their diagnosis on patient-doctor discussions and the elimination of other possible conditions which can be a time-consuming and often tedious process. The lack of a clear biomarker to diagnose IBS is a key reason why extensive research has been performed in trying to find conclusive methods for diagnosis and what may trigger or worsen symptoms for individuals. Identifying conditions that may bring on or exacerbate symptoms will help medical professionals target root causes of these conditions in order to implement more comprehensive and effective treatment plans.

To this end, the focus of this work was to apply data analytics and machine learning methods to analyze the datasets collected from IBS patients and healthy controls. The data was provided by our clinical collaborators from the Cooper Medical School at Rowan University and consisted of qualitative surveys as well as the detailed stool sample analysis to identify the gut microbiomes and their concentration levels. Previous research has shown that the gut microflora of an individual can be linked to the symptoms experienced by an individual and hence we tried to link the IBS subtype identification to the stool sample data. The objective was to develop a definitive diagnosis and subtype identification method which can aid the clinicians in providing timely treatment and relief to IBS patients by recommending specific diets or probiotics to balance the gut microbiome and thus enhance their quality of life by minimizing or eliminating the exacerbated symptoms.

Clinical data consisted of 16 stool samples from different patients with additional details regarding patients' age and their High Fructose Corn Syrup (HFCS) consumption per year. Samples in quadruplicate were given from patients with IBS-C, IBS-D, IBS-M, and from the healthy control group. The stool samples included concentrations of 46 different bacteria. The stool samples were analyzed using bacterial ribosomal RNA (16S rRNA) gene sequencing. A random forest classification model was applied to evaluate the results of the stool sample bacterial concentrations and identify any correlations between certain bacteria and their relationship to specific symptoms and subtypes. The clinical data were limited for machine learning applications and therefore mock data was generated using a uniform sampling method to increase the data pool. 16 mock data samples were created for each subtype of IBS, totaling 64 data samples. To complete the random forest model the generated mock data was implemented into RStudio, the programming software of choice for this research.

The optimized random forest model determined that six bacteria were found to be most competent in determining the subtype of IBS for a patient: Megasphaera, Alistipes onderdonkii, Bacteroides, Ruminococcus, Akkermansia muciniphila, and Barnesiellaceae. This information was then used to successfully develop an online application (https://yenkiekm.com/computational-modules/) where the user (clinician or medical practitioner) could input values from new patient samples for these bacteria and get a prediction for the subtype of IBS. Thus, this work helped in exploring a more definitive method for diagnosing IBS and can serve as a preliminary framework for future studies. Further model improvements and a larger data pool can provide more accurate predictions for patients with IBS.

Acknowledgments: We thank the Inspira Health Network, New Jersey for the financial support and Dr. Sangita Phadtare from the Cooper Medical School at Rowan University (CMSRU) for assistance with the clinical data.

References:

1. International Foundation for Gastrointestinal Disorders, Inc. (2016, November 24). Facts About IBS. https://www.aboutibs.org/what-is-ibs/facts-about-ibs-2.html.
2. National Center for Chronic Disease Prevention and Health Promotion. (2020, November 17). About Chronic Diseases. Centers for Disease Control and Prevention. https://www.cdc.gov/chronicdisease/about/index.htm.
3. Pletikosić, S., & Tkalčić, M. (2016, April). The Role of Stress in IBS Symptom Severity. Retrieved April 8, 2021, from https://www.researchgate.net/publication/301695897_The_Role_of_Stress_in_IBS_Symptom_Severity#fullTextFileContent
4. Lacy, B. E., & Patel, N. K. (2017). Rome Criteria and a Diagnostic Approach to Irritable Bowel Syndrome. Journal of clinical medicine, 6(11), 99. https://doi.org/10.3390/jcm6110099
5. Irritable bowel syndrome. (2020, October 15). Retrieved November 19, 2020, from https://www.mayoclinic.org/diseases-conditions/irritable-bowel-syndrome/symptoms-causes/syc-20360016