Clinical language analysis for rehabilitation in trauma (orthopaedic) in older years
Overview
Fragility fractures are a major public health concern, particularly among older adults. Despite clinical guidelines, access to rehabilitation services remains variable. Understanding the extent of this variation and how it may influence outcomes of usual care is challenging as rehabilitation is documented using ‘free text’ (sentences and paragraphs). It is difficult to combine free text across treatment sessions for one patient, or across treatment sessions for all patients, to gain an understanding of ‘usual care’.
However, there is potential to overcome these challenges by employing clinical natural language processing which could ‘unlock the black box’ of documentation by identifying established treatments (‘structured clinical concepts’) in a consistent way. These treatments may include mobilisation, activities of daily living, and exercise.
This study will see whether it is feasible to use clinical natural language processing in this way by applying the methods to data from Barts Health NHS Trust for rehabilitation provided to patients with fragility fractures over the past 10 years.
Why this study matters
If the proposed approach is feasible, it has the potential to enable employment of routinely collected data to better understand ‘usual’ rehabilitative care. This will allow us to see whether different access and delivery of care impact outcomes for patients, the NHS, and wider society. Further, it would also unlock the potential to provide more equitable and better tailoring of rehabilitation by understanding who responds to what treatments and when.
Given raw data for analysis will be free text, methods will be generalizable to other countries which use electronic free text data entry (standard practice for the UK, Ireland, Australia, Canada, US, Norway, among others).
What we’re doing
Participants are patients from Barts Health NHS Trust with fragility fractures, admitted for surgical repair of fractured hip, tibia, femur or ankle and treated at Barts Health NHS Trust over the past 10 years. Rehabilitation data will be selected retrospectively from anonymised health records.
Data will be extracted using clinical natural language processing tools to identify structured clinical concepts and rehabilitation dosage (frequency, intensity, time). For example, for each patient each occurrence of the concept ‘Promotion of walking using mobility aid’ will be extracted alongside the date (to enable estimation of frequency) and adjacent text e.g. 10 metres (to enable estimation of intensity) using a rules-based approach. We will seek to validate extraction against an annotated random sample of patient records.
We will use the structured concepts to assess variation in rehabilitation overall, by fracture type, and across years following surgical repair of lower limb fragility fractures, should the approach be deemed valid.
Who is involved
Lead investigator: Professor Katie Sheehan
Collaborators:
- Professor Xavier Griffin, Centre for Bone & Joint Health, QMUL
- Professor Catherine Sackley, University of Nottingham
- Professor Nadine Foster, University of Queensland
- Professor Finbarr Martin, Kings College London
- Professor Celia Gregson, University of Bristol
- Dr Charles Gutteridge, Chief Clinical Informatics Officer, Barts Health NHS Trust
- Dr Sophie Williams, Barts Life Sciences, Barts Health NHS Trust
- Dr Hiba Junaid, Barts Life Sciences, Barts Health NHS Trust
- Funded by: This was funded by a UK Research & Innovation Future Leaders Fellowship awarded to Professor Katie Sheehan.
Current status
Approvals are in place and we are preparing the data for analysis.
Get involved
Contact email: k.sheehan@qmul.ac.uk