Atypical femoral fractures - a personalised risk assessment tool powered by machine learning combining clinical, radiographic, and genetic risk factors
Description: Objectives
To develop an atypical femoral fracture (AFF) risk assessment tool combining clinical, radiographic, and genetic risk factors using machine learning methods, for an ageing population with prevalent osteoporosis and increased use of anti-resorptive drugs.
Hypothesis to be tested
Patients at risk of atypical femoral fractures could be identified from clinical, radiographic, and genetic risk factors using a machine learning AFF prediction model.
Design and subjects
This territory-wide study involves the Hong Kong Osteoporosis study cohort and two major regional hospitals. Patients aged 65 or above with osteoporosis who received anti-resorptive drugs are screened for inclusion. Sample size is calculated to be adequate having recruited 50 AFF patients and 500 controls with genotyping data available.
Instruments
Clinical risk factors include demographics, anthropometry, comorbidities, serological, and pharmacological parameters. Radiographic risk factors involve radiographic femur morphology. Genetic risk factors are identified through genome-wide association study.
Interventions
Machine learning models will be constructed using gradient boosting machine and neural networks to predict AFF.
Main outcome measures
AFF prediction model accuracy will be evaluated in terms of area under curve, sensitivity, specificity, positive and negative predictive values.
Data analysis
AFF prediction model will be developed using gradient boosting machine with XGBoost and neural networks involving deep learning with long short-term memory and gated recurrent units, with subsequent validation.
Expected results
An accurate AFF prediction model is expected, as our pilot study of osteoporotic hip fracture machine learning prediction model achieved 84.9% sensitivity, 77.3% specificity, and area under curve of 0.889.
Funder: Health and Medical Research Fund
Amount: HKD 95,900
Funding type: grant
Funder scheme: HMRF - Health and Health Services (including HHSRF)
Start date: 2020-01-01
Investigator: Siu Him Janus WONG (lead)
Institutions: University of Hong Kong, Department of Orthopaedics and Traumatology, The University of Hong Kong
Subfield: Orthopedics and Sports Medicine
Field: Medicine
Domain: Health Sciences
Funded outputs count: 0
Funder award ID: 19201491
OpenAlex ID: G7600873609
