Summary
Background: Many older adults in the US prefer to live independently for as long as they are
able, despite the onset of conditions such as frailty and dementia. Solutions are
needed to enable independent living, while enhancing safety and peace of mind for
their families. Elderly patients are particularly at-risk for late assessment of cognitive
changes.
Objectives: We predict early signs of illness in older adults by using the data generated by
a continuous, unobtrusive nursing home monitoring system.
Methods: We describe the possibility of employing a multiple instance learning (MIL) framework
for early illness detection. The MIL framework is suitable for training classifiers
when the available data presents temporal or location uncertainties.
Results: We provide experiments on three datasets that prove the utility of the MIL framework.
We first tuned our algorithms on a set of 200 normal/abnormal behavior patterns produced
by a dedicated simulator. We then conducted two retrospective studies on residents
from the Tiger Place aging in place facility, aged over 70, which have been monitored
with motion and bed sensors for over two years. The presence or absence of the illness
was manually assessed based on the nursing visit reports.
Conclusions: The use of simulated sensor data proved to be very useful for algorithm development
and testing. The results obtained using MIL for six Tiger Place residents, an average
area under the receiver operator characteristic curve (AROC) of 0.7, are promising.
However, more sophisticated MIL classifiers are needed to improve the performance.
Keywords
Automated pattern recognition - multiple instance learning - eldercare
