Uncovering Natural Clinical Data-Driven Clusters of Irritable Bowel Syndrome (IBS) Endophenotypes Using Machine Learning Techniques

Abstract

The physiopathology of Irritable Bowel Syndrome (IBS) is complex and multidimensional that has continued to baffle both researchers and clinicians. IBS has long been correlated with genetic factors, indicating a hereditary element and the efforts to identify these genetic risk factors for IBS have been restricted and inconclusive. In recent years, there has been considerable interest in the application of artificial intel ligence methodologies to the study of IBS. The aim of our study is to examine the UK BioBank, which is the largest database in the UK, in order to determine the endophe notypes of multidimensional patient data. This multi-dimensional clinical data contains information about a variety of factors like gastrointestinal indications, co-morbidities, demographics, physiology, and psychophysiology. Application of exploratory data analytics techniques on the data, and key variables such as age, BMI, and sex, reveals important insights into participant demographics and health patterns. Feature engineering techniques, including one-hot encoding and principal component analysis, are employed to handle high dimensional data and extract informative features for differentiating between IBS patients and the control group. This research focused on making use of unsupervised learning algorithms in order to find complex relations between different features in the data. By rigorously selecting ICD-10 codes as features and categorizing them into different groups, a concise subset of comorbidities strongly associated with IBS is identified. Cluster analysis and visualization further reveal hidden pattern and relationships among these comorbidities and results highlight the superior performance of the Kmeans clustering model. The findings of this research endeavour will enhance our comprehension of IBS and subsequently elevate the efficacy of treatment through customized medical interventions.