Using Electronic Health Records to Detect and Predict Inpatient Falls

Lead Supervisor
Dr Julie Whitney
Lecturer in Physiotherapy
School of Population Health and Environmental Sciences, King’s College London

Dr James Teo (King’s College Hospital) / Dr Fiona Gaughran (King’s College London)

Project Details

Inpatient falls are among the most commonly reported patient safety incidences in hospitals. Falls result in injury and a reduction in confidence and function. They are also costly due to the impact on length of stay, subsequent care needs and litigation. To date, there is neither an effective way to identify risk of falling, nor effective interventions to prevent falls in acute inpatient settings.  This is compounded by poor recording of fall events making it difficult to fully appreciate the scale and the nature of the problem.

The candidate will:

  1. Work with KCH and SLAM electronic health records research systems like Cogstack to develop NLP models that can detect inpatient fall episodes, that is superior to formal reporting mechanisms;
  2. Expand the model to identify a faller digital phenotype from structured and unstructured health data of inpatients that are at high risk of falls;
  3. Develop machine learning for a fall risk prediction model and validate these processes in a second hospital site. 

Terms associated with a fall event will form the unstructured data that informs the fall detection model. The team at the CTI have already begun work on classifying terms associated with risk of falling into SNOMED, which along with ICD10 codes will provide structured data to be used. Models for fall detection will be tested against the number of fallers identified on hospital reporting systems (i.e. Datix). The hypothesis is that conventional incident reporting misses significant amount of falls and that a model trained to handle structured and unstructured data would complement conventional systems or even be superior.

Characteristics in EHRs associated with inpatient falls will then be used to build a phenotype for an inpatient at high risk of falls. Classical statistical analysis will be used to identify factors that are independently associated with falls and models developed and compared for predictive accuracy. 

If an accurate prediction model is established, machine learning options will be explored and models developed, investigating supervised and unsupervised techniques. This will be tested on a second hospital site. Cross validation methods will be used to test these processes. The best model for detecting fall risk will be used to develop a decision support tool.

The student will be exposed to and expected to learn:

  • Core Data science skills including querying of both structured and unstructured data; and natural language processing (NLP);
  • Machine learning and classical statistical techniques;
  • Systems thinking for understanding healthcare processes
  • General understanding of interoperable Healthcare ontology

The supervision for this project will include a supervisor with a background in clinical research into fall prediction and prevention, who currently leads on the National Audit of Inpatient Falls. The second supervisor will bring expertise in data science, natural language processing and machine learning to this project and experience of working on similar projects. 


The principle ethical and information governance issues affecting this project are that it will use the electronic health records of patients. The models will be developed at King’s College Hospital (KCH), which is at the forefront of the development of this technology. To that effect, KCH has recently set up the KERAI group which in conjunction with the Health Research Authority (HRA) and local research and innovation departments, have developed a protocol for reviewing all applications to use EHR data. This project will have to go through this committee as part of the approvals process. 

The project will be undertaken in compliance with the code of conduct for data driven health technology ( and NHSX (


Falls, Inpatients, risk factors, prediction