Optimization and Validation of a Classification Algorithm for Assessment of Physical Activity in Hospitalized...

Area of Science:

• Gerontology
• Rehabilitation Medicine
• Biomedical Engineering

Background:

• Hospitalized patients often exhibit low physical activity (PA) and prolonged sedentary behavior.
• Inactivity during hospitalization can lead to negative health consequences.
• Objective PA monitoring is crucial, but lacks validated algorithms for this population.

Purpose of the Study:

• To optimize and validate a classification algorithm for monitoring physical activity in hospitalized patients.
• To differentiate between sedentary, standing, and dynamic activities using accelerometer data.
• To record postural transitions in free-living conditions.

Main Methods:

• An accelerometer-based algorithm was developed and optimized using data from orthopedic and elderly acute care patients.
• Parameters like window size (WS), PA threshold (PA Th), and sensor orientation threshold (SO Th) were optimized.
• Validation was performed against video analysis, assessing sensitivity, specificity, accuracy, and percentage error.

Main Results:

• Optimized parameters (WS 4s, PA Th 4.3 cps, SO Th 0.8g) yielded the best performance.
• The algorithm achieved >80% sensitivity, specificity, and accuracy for most activities.
• Classification of standing activity was less accurate than other activities.

Conclusions:

• The validated algorithm is suitable for classifying physical activity in hospitalized patients.
• This tool can support interventions to increase PA and reduce sedentary time.
• Further refinement may improve standing activity classification.