JMIR Publications

ABSTRACT

Background:

Physical activity is a highly important aspect in an older patients’ rehabilitation process after hip fracture surgery. To gain more insights into an older patients’ physical activity progress during rehabilitation after hip fracture surgery, we recently assessed the patterns of overall physical activity over time. Results showed four common patterns of overall physical activity over time during rehabilitation, which were significantly associated with the duration of rehabilitation stay. Knowing the expected pattern of overall physical activity early in the rehabilitation phase could provide an early indication of the rehabilitation stay. Furthermore, it could provide healthcare professionals and patients valuable information about what to expect during rehabilitation.

Objective:

To explore the early prediction of patterns of overall physical activity in older patients rehabilitating after hip fracture surgery at a skilled nursing home.

Methods:

Physical activity of surgically treated hip fracture patients (≥70 year) was continuously monitored during rehabilitation at a skilled nursing home using an accelerometer. Physical activity patterns used in this study were described in our previous study: the upward linear pattern (n=15) and the S-shape pattern (n=23). Features from the intensity of physical activity and clinical features were calculated for multiple time windows to assess at which early rehabilitation moment the patterns could be predicted most accurately: window of the first 5 days, the first 6 days, the first 7 days, and the first 8 days. Multiple classifiers were used: decision trees, discriminant analysis, logistic regression, support vector machines, nearest neighbors, and ensemble classifiers. The performance was assessed by calculating precision, recall, F1-score, and micro F1 (micro-averaged F1-score).

Results:

The overall intensity of physical activity at the first day of rehabilitation and morphological features describing the shape of the patterns were the best selected features for all time windows. Features extracted from the first 7 days resulted in a cosine k-nearest neighbor model with the highest overall prediction performance (micro F1 = 1) and a 100% correct classification of the physical activity patterns.

Conclusions:

Results showed that the physical activity patterns could be 100% correctly classified by a cosine k-nearest neighbor model using features extracted from the first 7 days. Continuous physical activity monitoring is important to assess the recovery progress of older patients during hip fracture rehabilitation and could benefit health care organizations, healthcare professionals and patients themselves.