Biomedical Engineering Reference
In-Depth Information
The influence of the resting HR at the beginning of a training in S3 leads to a good
precision of the model during the training itself. This predictor is probably influenced
by many other, hard to measure variables like medical treatment, stress, dehydration
and coffee consumption, which might have a strong impact on the metabolic system.
This leads to the unexpected observation that the given blood pressure values show
only a very small effect on the HR. Blood pressure kinetics are in close relationship to
HR, but not to absolute values, due to antihypertensive treatment in most patient's.
The prediction can be used to estimate the patient's physical state on the day of
testing and thereby help to define an appropriate training intensity before the training
starts.
The phase-wise prediction in S4 during the runtime of the training shows a relative
error below 5%. This should be precise enough to robustly detect abnormal HR de-
velopments and calculate the optimal load for the next phase. In future we will focus
on the analysis of other time dynamic predictors that might increase the model accu-
racy and also facilitate high refresh rates without the abstract distinction between
training phases.
Integration of the first two HR models into the PHR showed to be a promising way
to bring them into application. Key challenges for the adoption have been addressed
by the implementation of standards and by localizing the system in the user's own
home to avoid acceptance problems. Whether this concept is successful also depends
on future backup solutions of the PHR's health data and the access models that have
to be implemented, when such a system will be integrated into existing health infra-
structures.
5
Conclusions
We created a statistical model to predict HR as an important vital parameter for the
rehabilitation training of cardiopulmonary patients and integrated it into a PHR that is
localized in the user's home to overcome acceptance and interoperability problems.
We considered demographic data, training plan information, vital parameters and
weather information as potential predictors and classified them into five aim-specific
scenarios where they can be used as individualized initial or reference values to para-
meterize alert or training control algorithms. The implementation of the first two ap-
plication scenarios into the training plan creation of the PHR was presented as a proof
of concept for the integration of the model. The validation of the model revealed that
weather and the measured blood pressure have nearly no direct influence on HR. Age
and previously measured HR based variables like the resting HR strongly influence
the responding HR.
The models prediction results in an overall low error of ≈11 bpm in median, when
used for the creation of a training schedule (scenario 1). The error is reduced by about
50%, when the model is used for prediction at the beginning of a training session. The
error decreases to less thanthe significance level when the model is used during a
training to predict HR at the beginning of each of the four training phases. This makes
it potentially suitable to detect critical situations before they appear.
The precision of the prediction might be improved by additionally including expert
knowledge and further statistical methods, but it already serves as a good basis for the
Search WWH ::
Custom Search