Management of Multiple Data Streams in Sensor Networks for Medicine

Jan Sliwa and Emmanuel Benoist

Bern University of Applied Sciences, Quellgasse 21, CH-2501 Biel, Switzerland

Keywords:

Sensor Networks, Data Management, Quality Assurance, Medical Research.

Abstract:

In this paper we discuss the privacy aware management of data streams generated by wireless sensor networks

(WSN) in medical applications. Such data are primarily used to ensure the direct health support for a speciﬁc

patient. They can also be used for other purposes, like quality assurance of applied devices and comparative

evaluation research of medical treatments. Considering the novelty of the WSN-based medical application,

every information helping to improve the service is extremely useful. As privacy in medicine is an important

issue, care should be taken to harmonize the legal requirements with the medical necessities. Data used for

different purposes have different characteristics, different users and different access rules. Therefore the raw

data stream has to be split and processed (condensed, anonymized, etc.) in various ways dependent on the use.

The access to data has to deﬁned differently for different output data streams.

1 INTRODUCTION

This paper shows the problems regarding data man-

agement in wireless sensor networks (WSN). We will

concentrate our attention on the applications in the

medicine. For brevity, we use the term ’sensor net-

works’ but we should not forget that such networks

often also contain actors (or actuators) - devices act-

ing directly on the patient, like deﬁbrillators or in-

sulin pumps. In this paper however we focus on

the incoming data streams. In our previous papers

we have discussed the ethical aspects of enhancing

human organisms with technical instruments (Sliwa

and Benoist, 2011b) and presented this overwhelm-

ing development in a historical perspective (Sliwa and

Benoist, 2011a). We present here a basic introduction

to this subject, a broader overview (with further refer-

ences) is given in the aforementioned papers.

The development in the area of medical WSN

based systems is driven by some enabling technolo-

gies. New devices permit measuring physical and

chemical properties that were until now difﬁcult to

treat. They can also act on the human body in new

ways, like stimulating the heart or dosing a medicine.

Local networks permit the devices to cooperate and

wide area networking allows them to exchange data

with a hospital or a data center. Reducing power

consumption, energy harvesting and remote powering

permit a long term functioning.

The basic conﬁgurations are: Body Area Net-

works (BAN) and Ambient Assisted Living (AAL).

A BAN consists of a set of wearable and im-

plantable devices carried by the patient. They can

measure the health parameters (sensors) or act on the

body (actors or actuators). They exchange data with

a data aggregator (typically a smartphone) that sends

periodical reports to the hospital or alerts it in the case

of an emergency.

In the AAL scenario the patient lives in a home

equipped with stationary medical devices that provide

a supporting environment. In this way an elderly per-

son may live longer in his/her familiar environment

and still have an adequate level of security.

In this paper, ﬁrst we set the problem of the data

management in a productive deployment of medical

WSN systems. Then we explore the various uses of

the data generated by them: direct health support,

quality assurance and medical research. For each of

these uses we discuss the characteristics of the re-

lated data stream and analyze to whom, when and how

should they be available. Finally, we outline some re-

search challenges related to the data management fac-

ing the specialists who develop, deploy and operate

such systems.

2 SETTING THE PROBLEM

In this paper we want to discuss the aspects of the

managements of the data generated by the sensor net-

works in medicine. Presently, medical applications of

399

Sliwa J. and Benoist E..

Management of Multiple Data Streams in Sensor Networks for Medicine.

DOI: 10.5220/0004325403990402

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2013), pages 399-402

ISBN: 978-989-8565-37-2

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

the WSN are in the experimental phase. The technical

challenges are immense, therefore most efforts con-

centrate on solving speciﬁc problems, ranging from

the design of the sensors resistant to a hostile chemi-

cal environment via studying the propagation of the

electromagnetic waves in the human tissue, energy

harvesting for a long-term unattended operation to the

development of efﬁcient signal processing algorithms

- and many more. In the experimental phase, the sys-

tems are installed and monitored by skilled and moti-

vated technical and medical specialists, including the

designers themselves. The data sets are small and are

accessed by a small number of trusted scientists.

After the large scale deployment the situation

changes radically. The systems are installed and oper-

ated by less specialized personnel. Their operation is

not supervised. Data sets are large and the number of

possible users increases. It may be tempting to inte-

grate the systems using the bottom-up approach, just

combining the available elements in a working sys-

tem. It seems however to be now the right time for

some more general reﬂections, setting the goals and

analyzing the requirements.

In the area of data management, we argue that fol-

lowing uses of the collected data should be consid-

ered:

• direct health support

• quality assurance

• medical research

A reference scenario, used throughout this paper,

is a Body Area Network for treating heart problems.

A hospital diagnoses the patient, deﬁnes a treatment

based on a WSN system and delivers a set of de-

vices to him/her. It may be just a wearable device

handed out with a short instruction, or an implantable

device that requires an operation. The data aggrega-

tor (a smartphone) is conﬁgured by the hospital staff.

The used devices are well deﬁned, the patients and

their devices are registered. A patient is related to

a main hospital handling his/her case, called here a

home hospital, where the patient’s records are stored

and which is called in case of an emergency. Thanks

to a cooperation with other hospitals, the patient may

be also treated in a remote location, if an emergency

occurs there. The hospital, or a group of hospitals, op-

erate a registry that helps to evaluate statistically the

quality of the implemented devices and the correct-

ness of the applied treatments.

Our experience in managing medical data used for

multiple purposesby various user’s groups is based on

the platform for medical registries that we have de-

veloped and operated for about 10 years (R¨oder et al.,

2006). Its goal is to store the patients’ records and to

make the anonymized extracts available for research.

The architecture based on the physical separation of

personal and clinical data enhances the privacy pro-

tection (Sliwa and Benoist, 2012). Currently we in-

put data using fairly static online forms. Nevertheless,

this experience is useful for solving similar problems

arising in the WSN ﬁeld, regarding storing and pro-

tecting data, controlling the access and presenting in

various ways for various user groups.

3 DATA FOR HEALTH SUPPORT

3.1 Data Characteristics

The basic goal of medical WSN systems is provid-

ing health support for speciﬁc patients. The delivered

information can vary with respect to its quantity and

temporal characteristics. It may consist of single val-

ues, signal waveforms or images. A local unit can

process the continuous waveforms and extract fea-

tures from them - either calculate general properties

or detect events. Data may come from one or from

many sensors, it may be also a combination of them

or a sequence of events. For example, when a drug

is automatically delivered into blood, the timing of its

propagation and of other chemical reactions to it may

be important.

Sensor data may just periodically report the health

parameters with no special action required. They may

also detect emergencies - in this case an action has to

be taken. This action may be executed by actuators

on the patient remotely controlled from his/her home

hospital, or it may involve real people and equipment.

In the latter case it is necessary to know the actual lo-

cation of the patient, and the task has to be assigned to

the nearest cooperating hospital. This example shows

us that real-life deployment of a health support sys-

tem requires important organizational measures and

merely technical solutions are not sufﬁcient.

3.2 Data Access

For regular treatments the patient’s case can be han-

dled by his/her home hospital. The difference in com-

parison to the traditional way is data transmission via

a wireless channel and data storage that both occur

automatically. Long distance data transmission from

the data aggregator has to be protected like any se-

cure wireless transmission. The problem is the trans-

mission between the sensors/actors and the data ag-

gregator. These miniaturized, low power devices can-

not support the same level of security measures (like

strong encryption) due to their limited capabilities.

HEALTHINF2013-InternationalConferenceonHealthInformatics

400

When data reach the hospital, they are treated in

a similar way as other data in a clinical information

system and haveto be protected accordingly. They are

stored on disks accessible to the system operators and

are visible to the personnel that normally is allowed

to see them.

A different class of problems appear when an

emergency service is needed in a remote location as

in the example mentioned above. First, a network

of cooperating institutions that agree for a mutual as-

sistance has to be deﬁned. Then, procedures of the

transfer of the patient’s history are necessary. The in-

tervening team has to know all relevant information,

and have it available fast. When the case is closed, the

home hospital needs the update of the patient’srecord,

data at the remote hospital have to be removed as no

more necessary. Similar problems are being solved in

the European e-Health Project epSOS

4 DATA FOR QUALITY

ASSURANCE

4.1 Data Characteristics

Data collected from the medical WSN can be used to

evaluate the quality of the deployed equipment. Al-

though they are thoroughly tested and formally ap-

proved, only the actual operation can give us infor-

mation about long term results, unforeseen adverse

reactions and rare incidents. If an unexpected event

happens it may have various causes, just to name a

few: mechanical problem - ﬁxture loosened, part bro-

ken; ﬂuid sensor dirty, nozzle clogged; poor usability

of the user interface - display unreadable, small keys,

dialog unclear; battery depleted too fast; no phone

signal available; external attack.

Some faults can be diagnosed on the basis of the

analysis of the sensor messages. In an optimal case,

an intelligent device performs regular self tests and

informs about the possible and actual problems, but

often a problem analysis by a human is necessary.

Industrial networks, like train control systems, peri-

odically test their integrity. They have also well de-

ﬁned real-time properties. In a similar way, a health

supporting system has certain temporal requirements,

depending on the severity of the treated disease.

The named problems can be ﬁxed in very differ-

ent ways, like upload of a corrected software version,

device replacement, device redesign or organizational

changes. Therefore in order to keep track of the qual-

ity problems and solutions a registry is necessary.

http://www.epsos.eu/ (visited on 2012-10-26)

Medical software based devices pose challenging

problems to the statistical analysis. One of the prob-

lems is assuring a reasonable size of a statistical sam-

ple of comparable, uniform enough data. The prod-

ucts are often upgraded, therefore there are not so

many identical devices deployed. Even when only a

software bug on a device is corrected, the device be-

haves differently, so formally speaking, it is not the

same device as before.

Moreover, a patient has not just one device, but

a set of cooperating medical devices together with a

speciﬁc model of a smartphone with speciﬁc phone

applications loaded. Because of the interdependen-

cies of the elements, any change of any element

makes a different system, a different case to be evalu-

ated. Because of this confused, dynamically evolving

situation, the deﬁnition of the analysis processes re-

quires an involvement of humans that understand the

underlying problems.

4.2 Data Access

The analysis of the data for quality assurance needs

no personal information about the patients, therefore

should be performed on anonymized data. This is en-

tirely true if we are interested just in the number of

faults for a device model. If we want to ﬁnd the cause

of a problem, the procedure is more similar to an anal-

ysis of an airplane crash. Still, the name of the patient

is not relevant, but we may need supplementary infor-

mation, for example what is his/her diet, does he/she

any physical exercises, what is his/her age, gender,

education level. If in case of an emergency no help

came, or it came too late, it may have been caused by

a software bug on the device, a loss of phone signal,

inefﬁcient information ﬂow at the hospital or no free

ambulance. Therefore it may be necessary to perform

the analysis of the case at a very detailed level, also

with the access to the raw sensor data.

General statistical analysis is best performed by a

team of independent researchers, trained in medicine

and statistics, understanding information and commu-

nication technology. For a precise analysis of a tech-

nical fault, representatives of the producers may have

to be involved. The scenario outlined above shows

that an interdisciplinary problem needs an interdisci-

plinary solution.

We also see how important, and how difﬁcult it is

to deﬁne the rules what data should be available, to

whom and under what conditions.

ManagementofMultipleDataStreamsinSensorNetworksforMedicine

401

5 DATA FOR MEDICAL

RESEARCH

5.1 Data Characteristics

Data collected by the WSN-based systems can be ﬁ-

nally used for medical research. As the approval pro-

cess for intelligent software-based devices is not as

strict as for drugs (chemicals), it is important to eval-

uate the merits of various devices in their actual ap-

plication. Apart of comparing speciﬁc models, it is

interesting to determine if a treatment method pro-

vides expected results. In the case of drugs we ex-

pect healing or at least slowing down the progress of

the disease. In the case of medical devices the effect

may consist of enhancing the quality of life, reduc-

ing the costs and effort for routine measurements and

consultations, improving the security or delaying the

transfer to a nursing home.

The information about the effectiveness of such

systems is important for the hospitals applying them

and having to choose the best one. It is also impor-

tant for approval institution in order to reevaluate the

deployed systems (postmarket vigilance) and to as-

sess the costs, especially if they are covered by public

funds. Furthermore, this informationcan serve the de-

vice developers and scientists in choosing the viable

directions for further research.

5.2 Data Access

In the medical research the identity of the patient has

no relevance, therefore anonymized data should be

used. It is much more important to accumulate data

from many medical institutions, related to systems of

many producers. Also combining data from various

sources is useful. For example, correlating medical

data with data about lifestyle or social environment

may give clues about the causes of good or poor ef-

fectiveness of the systems. Of course, such corre-

lated data permit to re-identify the patients if enough

is known about their cases. It is however counterpro-

ductive to obfuscate data used for scientiﬁc research.

Bad data may lead to bad, harmful conclusions. It is

therefore much more important to control the group

of people allowed to access data, recording suspi-

cious queries and limiting the size of detailed, non-

aggregated data sets accessible to any single person.

It may be also useful - under certain conditions -

to retrieve speciﬁc patients. Normally statistics is in-

terested in the typical cases, not in the outliers. It may

however happen, that some patients or some groups

of them respond differently to a treatment. In the

case of drugs, it could be a patient having an unex-

pected resistance to the disease. In such a case, a

more profound analysis of previously not considered

factors could help to understand unknown aspects of

the treatments.

6 RESEARCH CHALLENGES

Managing data - collecting, storing and distributing -

will be a major issue in the real-life deployment of

the medical applications based on the wireless sensor

networks. We can name here some of the problems to

be solved:

• compressing raw data streams for different recip-

ients and for different purposes

• designing a hardware and software architecture

for storing and querying large amounts of data

• protecting the data stream on its way end-to-end

from the sensors to the ﬁnal destinations

• removing the patient’s identity from the data

while still permitting to retrieve the detailed case

if necessary

• deﬁning a reasonable and enforceable privacy

protection scheme permitting quality assurance

and medical research

• developing secure and portable calculation and

communication intensive applications for the

smartphones

• organizing the cooperation between the data pro-

ducers and data users

In this paper admittedly more questions are posed

than resolved. We expect to take an active part in pro-

viding the answers.

REFERENCES

R¨oder, C., M¨uller, U., and Aebi, M. (2006). The rationale

for a spine registry. European Spine Journal, 15:S52–

S56.

Sliwa, J. and Benoist, E. (2011a). Pervasive computing -

the next technical revolution. In Developments in E-

systems Engineering (DeSE), 2011, pages 621 –626.

Sliwa, J. and Benoist, E. (2011b). Wireless sensor and actor

networks: e-health, e-science, e-decisions. In Mobile

and Wireless Networking (iCOST), 2011 International

Conference on Selected Topics in, pages 1 –6.

Sliwa, J. and Benoist, E. (2012). A web architecture based

on physical data separation supporting privacy protec-

tion in medical research. International Journal of Re-

liable and Quality E-Healthcare, 1(4).

HEALTHINF2013-InternationalConferenceonHealthInformatics

402