Coeliac disease is an immune reaction in the gut to wheat, rye and barley, so the body absorbs fewer nutrients from food. Coeliac disease is common, affecting 1 in every 100 people in the UK, but fewer than a third of these are diagnosed. Early diagnosis allows dietary change to be recommended that avoids complications of growth and development in children and other detrimental health effects in adulthood, and reduces health care costs incurred by the NHS. Currently there is evidence of delay between people’s first presentation with symptoms and when they receive a diagnosis, with inequality in testing by socioeconomic deprivation.
To improve earlier diagnosis of coeliac disease in children and in adults, we will use routine medical records to look at how often they present to primary care and what symptoms or signs they present with. We will also look at whether they are admitted to hospital for other diagnoses, such as gastrointestinal or fertility problems.
Using this information we will develop and test a 'risk tool' using statistical methods called 'Bayesian learning methods' to identify predictive patterns of symptoms, signs, consultation behaviour, diagnoses, tests and prescriptions for people at risk of having coeliac disease. This will help us to identify children and adults at higher risk of the disease, thus enabling targeted testing to improve earlier diagnosis.
Background of need
Coeliac disease is common, affecting 1 in every 100 people in the UK, but fewer than a third of these are diagnosed. Early diagnosis allows dietary change to be recommended that avoids complications of growth and development in children and other detrimental health effects in adulthood, and reduces health care costs incurred by the NHS. Currently there is evidence of delay between people’s first presentation with symptoms and when they receive a diagnosis, with inequality in testing by socioeconomic deprivation.
Project aim
To improve earlier diagnosis of coeliac disease in children and in adults of those at higher risk of the disease using all their routine medical records, thus enabling targeted testing.
Proposed methods
1. Identifying risk factor patterns for coeliac disease: Using the information available for patients prior to Coeliac disease diagnosis or a pseudo index date in controls, we will use Bayesian supervised Latent Dirichlet Allocation topic modelling to learn predictive patterns of; symptoms, signs, consultation behaviour, diagnoses, tests and prescriptions; for both children and adults who are at risk of having coeliac disease, and compare these to their matched population controls.
2. Risk tool development: We will optimise an algorithm based on these risk factor patterns within the testing cohort to create a tool that flags people at risk of coeliac disease before their diagnosis whilst minimising false positives and false negatives. We will then perform internal validation and calibration on the final algorithm within a held-out validation cohort.
3. Comparison with traditional predictive model: We will compare the risk tool from 2. with more traditional predictive model methodology using logistic regression and selected candidate signs and symptoms based on current diagnostic guidelines, and assess the calibration and discrimination of this traditional predictive model compared to tool from 2. for identifying Coeliac disease early.
Coeliac disease
Colin Crooks - Chief Investigator - University of Nottingham
Laila Tata - Corresponding Applicant - University of Nottingham
Joe West - Collaborator - University of Nottingham
Timothy Card - Collaborator - University of Nottingham
HES Admitted Patient Care;ONS Death Registration Data;Patient Level Index of Multiple Deprivation;Practice Level Index of Multiple Deprivation;Practice Level Rural-Urban Classification