Deploying machine learning algorithms to stratify patients with IBD using routinely collected EHR

Title: Deploying machine learning algorithms to stratify patients with inflammatory bowel disease using routinely collected electronic health records

Abstract: Background: Personalised medicine approaches are eagerly awaited to facilitate individualisation of medical care for patients with inflammatory bowel disease (IBD). Multiple approaches have already been explored in attempts to stratify patients into different prognostic trajectories. In this study we aimed to use unsupervised machine learning algorithms to cluster patients based on their routinely collected electronic health records in an unsupervised approach.

Methods: we interrogated the rich electronic health records (EHR) of 700 patients with a diagnosis of IBD requiring an advanced therapy and attending the IBD service at King’s College Hospital (KCH), London, UK. Data for a seven-year period (June 2016-May 2023) were included with variables that captured diagnosis, treatments, service utilisation and investigations. Ethical approval was provided following application to the KERRI committee (KCH committee for accessing anonymised HER). In a preliminary analysis, we employed a multi-step approach to analyse patient data, identify significant features through k-prototypes clustering and a combination of parametric and non-parametric statistical tests. We examined different cluster solutions up to six clusters.

Results: we included 38 quantitative and 32 qualitative variables for 350 patients with a diagnosis of CD and 269 with a diagnosis of UC. Missing data were addressed by imputation by k-NN, mode value and zero. In the context of the unsupervised learning approach, when k=3, key differentiating features included faecal calprotectin levels, body mass index, mean platelet volume, marital status, taking advanced therapy for less than 3 years, and service access (telephone or clinic appointments).

Conclusion: this first pass analysis suggests that patients who are relatively early in their journey with at least moderately active IBD take up more recourses of the IBD service. This associates also with their social circumstances. Further work will explore these interactions in more detail.

Biography: Haider Mian is a Research Assistant at the Fair Machine Learning and Topological Data Analysis Lab supervised by Dr Raquel Iniesta. He studied a BSc and MSc in Statistics in Pakistan. As a NIHR pre-doctoral funded fellow Haider completed an MSc in Applied Statistical Modelling and Health Informatics from the Department of Biostatistics and Health Informatics at King's College London. He is interested in using Machine learning methods to identify interesting groups of patients of Inflammatory Bowel Disease that can help the identification of treatment markers and ultimately be used for the personalisation of therapies.