Operational machine learning and mechanistic modelling for supporting patient flow at Great Ormond Street Hospital

Lead Research Organisation: University College London
Department Name: Institute of Health Informatics


Great Ormond Street Hospital (GOSH) has recently developed a cloud-based data and analytics platform, the digital research environment (DRE), and an associated digital research and informatics unit (DRIVE). Through them, the hospital has established the underlying infrastructure and governance (including existing HRA/REC approval for use of routinely collected de-identified hospital data) to support a wide range of research projects. These projects apply contemporary machine learning and AI methods to routinely collected child health data. These data are deployed within a rapidly evolving data lake which contains >100 million unique mapped records to date. Data ingest is continuing into 2020 in addition to ongoing continually updated Epic Electronic Patient Records (EPR) feeds.

Both DRIVE's clinical research and informatics program (CIRP), and its associated MRes/PhD students are expected to utilise data-intensive methods to produce high-quality research. The defining characteristic of these projects is their ambition to generate output which can be rapidly translated into clinical or operational practice to ultimately improve child health. A key area of operational delivery where data science methods successfully intersect with traditional hospital planning is modelling patient flow pathways. Several NHS trusts have approached this key problem from a statistical perspective. This model-driven approach has led to increased efficiency in areas such as emergency and elective list planning, bed capacity planning, and pathway optimisation (1-7). Although demonstrably successful, the generalisability of these solutions to additional healthcare providers remains questionable. This is particularly true for GOSH because it lacks an accident and emergency (A&E) department and routinely provides highly specialised care. The range and severity of rare conditions treated at GOSH result in nuanced and complex care pathways, and this directly impacts on the trust's ability to estimate key markers of patient flow.

These specific circumstances mean that key flow-related activities at GOSH, such as list planning, have historically been dealt with in a siloed way, for example, by relying on departmental data. The DRE's unification of data sources now presents an exciting and unique opportunity for a PhD student to model these data at a trust-wide level to inform future flow planning and strategy. Our academic partnerships and integration with trust systems will enable the student to address both the academic side of model building, using rich data, and the operationalisation of the developed models, thus closing the translational research loop.

Overall Research Aims (MRes & PhD)
1. Utilise data mining and unsupervised learning methods to algorithmically characterise, visualise, and label trust-wide flow pathways from the DRE's patient-level data lake.
2. Compare and contrast statistical, machine learning, and deep learning methodologies based on their ability to deliver accurate, interpretable, and measurable patient flow predictions from standardised model input pipelines.
3. Collaborate with academic modelling and trust operations experts to build departmentally-focused, sustainable and data-driven patient flow models. These models will enhance existing trust data sources, particularly for problem areas such as intensive care units.
4. Integrate all previous work into a trust-wide predictive model of patient flow.
5. Examine and deliver computational approaches to operationalise patient flow predictions within GOSH and evaluate, both quantitatively and qualitatively, barriers to their uptake.



Abigail East (Student)


10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/S021612/1 01/04/2019 30/09/2027
2245620 Studentship EP/S021612/1 23/09/2019 30/09/2023 Abigail East