Project Summary/Abstract
Distal radius fractures (DRF) are the most common upper extremity fracture. They disproportionately affect
older women, and the overall incidence is increasing worldwide. Deciding on appropriate DRF treatment
remains a challenge, with multiple high-quality studies reporting conflicting results due in part to inadequate
patient-specific granularity in treatment decision-making. A better understanding of the recovery process,
especially early in treatment, is needed. Although several outcome measures are available to evaluate early
recovery following DRF open reduction and internal fixation (ORIF), including strength/range of motion testing,
patient-reported outcomes (PROs), and motor testing, they all fall short in accurately describing functional
use of the upper extremity (UE). Standard accelerometry tracking is also inadequate since it cannot reliably
differentiate between functional and nonfunctional UE use. The objective of our proposed study is to evaluate
the application of refined machine learning (ML) algorithms to parse out the accelerometer data and gain unique
insights as to how the arm is being used as an indication of functional recovery after DRF ORIF.
We will train these algorithms on patients’ data from controlled environments and then test on accelerometry
data from at-home use. This training is based on videos of patients performing tasks in our lab with
accelerometers on their wrists. The data obtained from the accelerometers will be analyzed through a series of
ML algorithms designed to categorize functional versus non-functional UE movement, then compared to
determinations of UE use from video review. Once the analysis algorithms are refined and reliably indicate
functional versus non-functional use from accelerometry, we will use them to analyze accelerometry data
obtained at home during tasks that will also be video recorded, scored, and compared to accelerometry results.
This will show how accurately we can use accelerometer data to track UE functional use in a natural setting.
Throughout these phases we will obtain PRO data and compare these data to video results. With this we aim
to understand the value of PROs in describing UE function, as well as use the ML algorithm analyses of early
accelerometry data to predict long-term recovery and outcomes.
The shortcomings of available instruments for evaluating early functional recovery following DRF ORIF
remain. Additionally, full-time patient observation is impractical and tedious, while accelerometry data alone are
inaccurate in differentiating functional versus nonfunctional UE use. Our approach is based on prior successful
work but is novel and innovative for the evaluation and management of surgically-treated DRF. By
accomplishing these aims, we will introduce and operationalize a suite of ML algorithms that can be
readily applied to accelerometry data and accurately capture functional UE use to help optimize patient-
centered DRF post-operative recovery with patient-specific granularity.
