Who can the service see? Observability, health equality and AI in inflammatory bowel disease

Background

Health services can only act on what they can see. Inflammatory bowel disease (IBD) is lifelong and affects nearly one in a hundred people in Lothian [1], yet most of a patient’s life happens between appointments, and routine records capture the care people received rather than the care they needed. We call this observability: how well a service can infer a patient’s current disease state from the signals it holds, in time to act. In Lothian, 56% of people with IBD had no IBD clinic attendance over a five-year period. Some were well; others had disease nobody was seeing.

Observability is unlikely to be shared equally. People who are poorer, older, frailer, more rural, from minority ethnic groups or living with mental illness may be seen less often, tested less often and recorded less completely. Prediction models trained on routine data then learn from the best-observed patients and risk widening the inequalities they are meant to reduce [2]. This project makes that gap measurable and tests how to close it.

Aims

1. Map observability across a whole population. Using the Lothian IBD Registry linked through DataLoch (around 10,000 people over 20 years, including primary care, laboratory, prescribing and hospital data) [3], describe for each person, over time, how recently and how reliably the service could infer their disease state.
   1. Measure inequality in observability and its consequences. Test whether observability differs by deprivation, age, sex, ethnicity, rurality, frailty and mental illness, and whether low observability leads to emergency admission, steroid exposure, surgery and other avoidable harm.
   2. Develop and test methods that close the gap: dynamic prediction that treats missing observations as informative rather than random, fairness audits of risk models across observability groups and, where clinical text is available, language models that recover disease state.
2. Replicate in a second health system. With Karolinska Institutet, test whether the patterns hold in Swedish national register data, using what the registers capture well: specialist contacts and admissions, dispensed drugs, the Swedish IBD quality register (SWIBREG) [4], socioeconomic data and mental illness.
3. Translate the findings into equitable monitoring and recall rules for SENTINEL, the proactive IBD platform being deployed in NHS Lothian, so the student sees the work used in practice.

Training Outcomes

The student will train in population epidemiology and causal inference; longitudinal and survival modelling, including dynamic prediction by landmarking; methods for informative missingness; machine learning, fairness evaluation and clinical natural language processing; record linkage and reproducible analysis in trusted research environments; and replication across two national data systems, with time at Karolinska Institutet. They will also learn to take analytical work into a live NHS implementation setting, with patients and clinicians as partners: the foundation for a future leader in equitable, data-driven care.

References:

1. Jones GR, et al. IBD prevalence in Lothian, Scotland, derived by capture–recapture methodology. Gut 2019;68:1953–60.

2. Sisk R, et al. Informative presence and observation in routine health data: a review of methodology for clinical risk prediction. J Am Med Inform Assoc 2021;28:155–66.

3. Constantine-Cooke N, et al. Large-scale clustering of longitudinal faecal calprotectin and C-reactive protein profiles in inflammatory bowel disease. Clin Gastroenterol Hepatol 2025.

4. Ludvigsson JF, et al. Swedish Inflammatory Bowel Disease Register (SWIBREG): a nationwide quality register. Scand J Gastroenterol 2019;54:1089–101.