Background:  To deliver high-quality, resource-efficient health care, the next outpatient appointment needs to be scheduled just in time. Too soon wastes resources and inconveniences patients. Too late risks adverse health outcomes, including irreversible sight loss. Clinicians make a complex judgement about timing of next follow-up, implicitly weighing a vast array of factors. Recent advances in machine learning offer promising potential for medical diagnosis and risk prediction.1,2 

Novelty & Importance:  Ophthalmology is the busiest outpatient specialty, with >7.5 million appointments, and 10% of the NHS waiting list. Much service capacity is consumed by necessary follow-up of chronic sight-threatening conditions. This PhD will develop multimodal deep learning that integrates medical images, electronic health records (EHR), unstructured text data extracted with CogStack,3 lab results, and -omics data in patients with inflammatory eye disease. The model will learn from patients' visit histories to predict treatment response, vision, and quality of life outcomes, helping to address a stakeholder research priority area,4 and to optimise follow-up care, just in time.

The project will use two deeply phenotyped datasets: Guy’s and St Thomas’ Electronic Records Research Interface (GERRI); and Birdshot ABC, a UK multi-centre prospective cohort study, including risk/outcome variables not available in routine record data. 

Aims & Objectives

1. To develop a multimodal deep learning model that integrates medical images, EHR, lab results, and -omics data, learning from patients' visit histories to predict treatment response, vision, and quality of life outcomes

2. To predict the optimal timing of just in time clinical review

References

1. Liu F, Zhu T, Wu X, et al. A medical multimodal large language model for future pandemics. NPJ Digit Med 2023;6:226.

2. Zhou RL, L.; Xiang, T.; Liang, Z.; Clifton, D.A.; Dong, Y.; Zhang, Y.T. Semi-supervised Lerning for Multi-Label Cardiovascular Diseases Prediction: A Multi-Dataset Study. IEEE transactions on pattern analysis and machine intelligence 2024;46:3305-20.

3. Bean DM, Kraljevic Z, Shek A, Teo J, Dobson RJB. Hospital-wide natural language processing summarising the health data of 1 million patients. PLOS Digit Health 2023;2:e0000218.

4. James Lind Alliance. Sight loss and vision priority setting partnership. London, UK: The College of Optometrists, Fight for Sight and the James Lind Alliance; 2013.