PREDICTIONS OF BLOOD PRESSURE AND PULSE RATE
USING VECTOR AUTOREGRESSIVE MODEL
Xuesi Liu
1,3
, Yanyan Guo
2
1
Beijing University of Posts and Telecommunications, Beijing, P. R. China
2
Shanxi university, Taiyuan, P. R. China
3
Science and Technology on Information Transmission and Dissemination in Communication Networks Laboratory
Email: liuxuesi@bupt.edu.cn, guoyanyan@sxu.edu.cn
Abstract
Recently, the evolving of IoT (Internet of Things) and
WeHealth(Wireless e-Health) has resulted in the promotion of
intelligent medical services. Many devices and sensors are
producing lots of time series data, which requires an effective
method to process and analysis. Time series analysis is a
common way to build models and make predictions for data.
Furthermore, it can also remind patients to be more cautious
once the trend for data is found to be unusual. In this paper, we
collect time series data from a WeHealth platform in Beijing.
The data contains systolic, diastolic blood pressure and pulse
rate. We build a vector autoregressive model with these data,
determine the parameters and order of the model. Then we
predict the patients' systolic, diastolic blood pressure and pulse
rate for next 10 days, and compare it with the real data. As the
result, VAR model has good performance, it obtained higher
accuracy and less mean absolute prediction error than AR
model.
Keywords: Healthcare system, vector autoregressive model,
time-series biological data, WeHealth
1 Introduction
With the growing aging population in China, people are getting
more concerned about the health care and disease prevention
of elders. Hypertension is a chronic medical symptom in which
the blood pressure in the arteries is always very high, and it is
also the most important factor in causing cardiovascular
disease. Currently the average hypertension rate in China is
about 24 percent, In some communities, the hypertension rate
has already climbed up to 60 percent [1]. If the incidence of
hypertension has not been effectively controlled, the
cardiovascular rate may also grow. Therefore, supervising
patients’ blood pressure in real time, predicting the trend for
blood pressure, giving warnings once the blood pressure is out
of the boundary, all make contribution in keeping the patients
healthy.
In recent years, as the Internet of Things (IOT) and smart health
has been used, Wireless E-health (WeHealth) has been
proposed and applied, and relevant products called portable
health equipment are popular among people [2]. To obtain
health data, we build several WeHealth platform in Beijing.
WeHealth platform is a community medical treatment
information system, the system uses a variety of check-up
devices to obtain biological data of users, then transmit to the
server via Bluetooth, 3G and Wi-Fi. We analysis user data at
server, then give health advice to users. In this proposal, we
select several series of data from the WeHealth biological
database.
Time series analysis mainly focuses on analyzing time series
data, which is widely used in medicine, economics and other
fields [3]. Blood pressure is an important data for judging the
seriousness of hypertension [4], but people are more concerned
about the current state of blood pressure, ignoring the trend of
historical data. Actually, continuous measuring blood pressure
data can be used as a time series data, which after analyzing
will tell us about the future state of people's health. To be more
specific, once we master the trend for a patient’s blood pressure,
we can take some actions before his blood pressure elevates,
which means that the patient’s blood pressure can be controlled
through taking medicine or adjusting the diet, thus reducing the
risk of getting hypertension.
Besides, pulse rate also has an impact on blood pressure, so we
build a three-dimension time series data by systolic, diastolic
blood pressure and pulse. Here, the data can be seen as a vector
[5]. In this article, we will analysis the multi-dimension time
series data and build a vector autoregressive model to predict
future blood pressure status, and compare the predictions
between the autoregressive model and vector autoregressive
model.
This paper is organized as follows. Section 2 describes how we
choose model for predicting biological data. Section 3
describes the structure, parameters and order of VAR model.
Section 4 provides the experimental results. Both VAR model
and AR model are used to make prediction, and we compare
the performance between them. Conclusions and discussions
of some future works are given in section 5.
2 Related Work
Generally, time series data have linear association between
lagged observations, this feature suggests past observations
might predict the future observations. Autoregressive(AR)
process is such a model that makes predictions with past
observations. AR model was proposed in the 1920s, and it is
widely applied in many fields. An AR model use past data
points to fit a linear function, and estimate current data point
with the function. It is popular because it is simple in structure
and computationally efficient. In 2012, Tanii,H collecting a