Patient-centric Handling of Diverse Signals in the mHealth Environment

Jan Sliwa

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

Keywords:

Mobile Health, Signals, Cyber-Physical Systems, Real Time.

Abstract:

In the context of the Mobile Health (or Telemedicine) many signals (data items) are exchanged between the

medical devices, data storage units and involved persons. They are often treated as a uniform mass of “medical

data”, especially in the Big Data community. At a closer look, they unveil various characteristics, like type

(single / continuous), required quality and tolerance for missing values. As in medical care time is crucial, real-

time characteristics are important, like the sampling rate and the overall reaction time of the emergency system.

The handling of data depends also on the severity of the medical case. Data are transferred and processed by

external systems, therefore the overall function depends on the environment and the persons involved: from

the user/patient to a dedicated medical emergency team. The collected data can be anonymously reused to

gain or verify knowledge, what calls for a fair trade-off between the interests of the individual patient and

the community. This paper discusses the semantics of mHealth data, like medical requirements, physical

constraints and human aspects. This analysis leads to a more precise mathematical deﬁnition of the required

data handling that helps to construct mHealth systems that better fulﬁll the health support function.

1 INTRODUCTION

A mobile health system, serving to treat a serious

medical case has to meet many demands. For a de-

signer, it is easy to be biased by his/her own experi-

ence. A developer of smartphone applications, a Big

Data analyzer, a sensorics specialist - each of them

has a different viewpoint and puts stress on differ-

ent aspects of the problem. As the technical issues

are truly challenging and crucial for the success, they

tend to play the dominant role, especially in small

and young enterprises. We will try to connect various

technical viewpoints with the medical perspective in

order to build a uniﬁed picture.

Mobile health systems generate and process large

amounts of data. It is important to see their medical

signiﬁcance which is not equal for all of them. Some

of them are required in a predeﬁned frequency, some

are helpful but optional. Some need a certain preci-

sion, some serve only as orientation values. Some

have to ensure guaranteed reaction times, for some

timing constraints are irrelevant. Some, when an-

alyzed statistically, need a well balanced, unbiased

sample, for some sheer quantity makes them valuable.

This paper takes a closer look at the above

mentioned aspects of mHealth data and brings to

a technically-oriented reader a better view on their

meaning. This semantic view allows in the next step

to formalize the requirements regarding time con-

straints and data extraction, conversion, transmission

and storage. The ﬁnal goal is to help the reader to

construct better patient-centric medical systems.

Remark: We use here interchangeably the terms

signals or data (items). The word signal stresses the

capture from sensors, whereas data emphasizes the

information content. We also use the term patient,

although in the case of ﬁtness tracking user would be

more correct. As the borders between the application

areas are not sharp, we stay with the ﬁrst term.

2 RELATED WORK

There are several good overviews of today’s mobile

health technology and applications (Baig et al., 2015),

(Islam et al., 2015), (Silva et al., 2015), (Soh et al.,

2015). A detailed analysis of remote blood glucose

monitoring is given in (Lanzola et al., 2016). Car-

diovascular health and disease is discussed in (Eapen

et al., 2016). The authors also propose a roadmap for

mobile health in this area. An Android-based system

for ECG arrhythmia classiﬁcation and heart rate vari-

ability analysis is presented in (Xue et al., 2015).

It is useful to look at the mHealth from the

medical perspective. (Itrat et al., 2016) discuss the

Sliwa J.

Patient-centric Handling of Diverse Signals in the mHealth Environment.

DOI: 10.5220/0006298505610568

telemedicine in prehospital stroke evaluation. (Ag-

boola et al., 2016) presents a bitter reality check for

smartphone apps. A top-selling app for managing and

diagnosing skin cancer was only able to accurately

classify 10 of 93 biopsy-proven melanomas. As for

insulin dose calculation, 67% of the apps gave inap-

propriate output dose recommendations that put users

at risk (Huckvale et al., 2015).

If we analyze the function of the systems not in

the lab, but in real life, practical experience is of

use. (Espinoza et al., 2016) present the design of a

telemedicine management system in Ecuador where

speciﬁc local challenges of a rural environment in a

less developed country had to be addressed.

3 SIGNAL TYPES

3.1 Basic Signals

We can divide the basic signals in following cate-

gories:

• Single measurements: (t, value)

• Point events: (t, type)

• Continuous waveforms: (y = f(t))

If we take a closer look, the divisions between

them are blurred. In the case of a chronic dis-

ease, single measurements form a sequence of values.

Point events may be entered as such, like an epileptic

seizure or tumbling. If they are detected by sensors,

they are derived from other measurements, as shown

below. Continuous signals are created by sampling

a property, so formally they are sequences of values,

at a high sampling rate. When we speak of such sig-

nals like ECG (electrocardiogram) or EEG (electroen-

cephalogram), they may come from a single sensor (in

a very basic version) or from a set of sensors placed on

the chest or on the head in well deﬁned locations. In

the latter case we obtain a set of synchronized wave-

forms.

3.2 Derived Signals

From the basic signals we can derive more condensed

information. It is especially useful in the case of con-

tinuous signals the volume of which is too large to

store or transfer. Equally, in the raw form it is not

yet very useful. For the electrocardiogram, a well

known condition is the ST Segment Elevation My-

ocardial Infarction (STEMI). The cardiac cycle has

characteristic points P-Q-R-S-T, and the elevation of

the S-T segment is a signal warning about the risk of

a myocardial infarction, i.e. heart attack. Similarly,

irregularities of the cycle frequency can be identiﬁed

as arrhythmia, in several variants, like tachycardia or

bradycardia (heartbeat too fast or too slow). For ar-

rhythmia to be detected, the basic signal has to be an-

alyzed over many cycles. The severity of the case de-

pends on the intensity and duration of the abnormal

condition.

As mentioned above, a point event can be detected

by sensors. For example, tumbling of the patient

can be detected by accelerometers, e.g. in a smart-

phone. In this case, an algorithm extracts a character-

istic waveform from a continuous signal.

When we monitor vital signals, we have to treat

adequately missing values. The fact that the measure-

ment is missing may be an information itself. For ex-

ample, missing a required value for a prolonged time

may indicate that the device is not working (e.g. bat-

tery empty), that something has happened to the pa-

tient or that the patient is not using the device because

it is obtrusive, he/she went on travel and: has forgot-

ten it at home / has left it because of its weight / has

no plug adapter for the charger. In a similar way, sys-

tematic outlier values may mean a health problem or a

wrong placement or poor body contact of the device.

We list those cases in order to stress that the same

observed situation may have very different causes.

Some of them may require intervention of the system

operator (hospital).

3.3 Complex Signals

Until now we have discussed signals coming from

single sources. From the medical point of view, it

is often useful to combine information from many

sources. One of the main reasons is making sharper

distinctions between the cases and eliminating false

alarms. There are many papers presenting methods to

detect patients’ falls with the use of the accelerome-

ters. Typically the accelerometers built in the smart-

phones are considered as described in (Sannino et al.,

2015). They are reasonably ubiquitous, however the

assumption that they are worn all the time seems not

to hold. In any case, if such a device detects the pa-

tient tumbling on the ﬂoor, it is useful to verify it with

more data. Especially if the detected condition is se-

vere, requires an action and this action is costly - like

sending an ambulance - it is crucial to detect only real

cases. False alarms, even not frequent, will erode the

conﬁdence in the service.

If we want to design a novel architecture that con-

nects various devices from unrelated producers, we

face the problem of the interoperability. Such devices

- microphone, ECG sensor, EEG headset - typically

are delivered with a connection to the smartphone or

the cloud and visualization and/or analysis applica-

tion. If we want to connect them into a combined

device, we have to go on the level of the internal inter-

faces (rarely disclosed) and to write our own applica-

tion. The system described in (Sannino and De Pietro,

2014) detects fainting and falling of patients by con-

necting the heart rate variability in the ECG with the

information from accelerometers and other body and

environmental sensors. In this way the decisions gen-

erated by the system take the context into account

what increases their reliability.

Not only data formats may pose a problem, also

communication protocols may be different (periodic

sending / on demand), time and data granularity or

measurement precision. When detecting complex

events, we need synchronized data. If they come

marked by internal timestamps, we have to compen-

sate possible differences or to force the synchroniza-

tion of the clocks.

4 CASE SEVERITY

The treatment of the signal depends on the severity of

the medical case. We can list here following classes:

• ﬁtness, general health, behavior modiﬁcation

• chronic disease

– mild

– severe

• life saving

The inﬂuenced factors are:

• required quality

• necessity, sampling rate

• reaction time

If the devices are used for ﬁtness improvement,

the measurements are performed or checked accord-

ing to the interest of the user. The value is rather

used for general information, and there are no tim-

ing requirements. The user often loses interest for

the measurements after a certain time. If in mean-

time he/she changed to a healthier lifestyle, the basic

goal has been achieved.

There is however a risk that the user gets obsessed

with the ﬁtness goals. Some people measure their

weight on bathroom scales many times per day, not

taking account of normal daily variation and measure-

ment errors. In the same way, trying to constantly in-

crease the daily step count, especially when compar-

ing to the group and obtaining (verbal) rewards from

the device, may pose a health risk. There is a certain

optimum and not always more is better. On the other

hand, the device typically just counts steps, so climb-

ing a mountain is like walking, or even of smaller

value, as the covered distance is smaller. Equally

swimming, when the device is left in a locker, does

not count at all. Therefore the user has to be aware

that his/her virtual health is only an imperfect model

of the real one.

This becomes a problem, if the user has an agree-

ment with an insurance company to have a healthy

lifestyle in exchange for lower primes. A simple de-

vice, like an accelerometer in a smartphone, registers

only certain types of activities. On the other hand, the

user may be tempted to cheat the device, by simulat-

ing realistic oscillations.

In a case of a chronic disease, the device is typ-

ically used to monitor the state of the patient, detect

the abnormalities and inform the doctor or hospital

that handles his/her case. For a mild condition, the

measurement can be done occasionally, more or less

periodically or if the patient does not feel well. There-

fore missing values are not problematic, possibly the

patient feels no necessity to act. He/she should not

be burdened too much by reminders. He/she can be

called for a periodic check, as it is for apnea patients

using a ventilator. The measurement should have a

reasonable precision, determined on the basis of the

medical science. It is necessary to eliminate the vari-

ability caused by imprecise placement of the sensor,

too high (or too low) humidity of skin, wrong oper-

ating mode of the device, or similar. If - as we as-

sume - the measurement is performed by the patient

at home, the handling should be entirely simple and

clear. The device has to be approved by the doctor

what eliminates most cheap sensors and easily down-

loadable smartphone applications.

For a severe chronic disease, the requirements for

quality and regularity of the measurements are more

stringent. Regular measurements may show the in-

creased risk before the actual event occurs and the pa-

tient may come for an extensive check or a preven-

tive hospital stay. As the deterioration of the health

state may be rapid and serious, time plays an impor-

tant role. Ergonomics has to be carefully designed,

as in an emergency the patient’s capabilities are im-

paired. For example, during an insulin shock the pa-

tient is dizzy and nervous and his/her vision is blurred.

If the patient is not able to act on his/her own and has

no assistance, an automatic communication with the

hospital is necessary.

For a life saving condition, the precision and re-

action times are even more important. Let us ﬁx our

attention on a patient with a cardiac disease, having

one or more wearable / implantable devices connected

via a wireless Body Area Network (BAN) to a smart-

phone that can alarm the hospital in case of emer-

gency. The devices are active, i.e. can induce a life

saving action locally. As the patient is at risk, he/she

has to adapt his/her life habits accordingly. As the

emergency call system depends on functioning com-

munication, the loss of the phone signal or of data

roaming is a problem. This may occur in a rural area,

in a location not properly covered by the patient’s

provider or in a restaurant restroom in the basement

under a thick concrete ﬂoor. Therefore the patient

should avoid such locations, limit the stay and take

care. The current risk level may be evaluated by the

devices and communicated to the patient. In a better

phase less precautions have to be taken. The user in-

teraction design is delicate, as the patient has to know

about the increased risk but on the other hand should

not be overwhelmed by messages. Too many warn-

ings will themselves increase his/her anxiety or with

time will be ignored. An intervention is costly, pos-

sibly includes sending an ambulance, therefore mak-

ing a clear distinction between real and false alarms is

extremely important. When in doubt, the emergency

team may try to contact the patient. However, if the

risk condition is generated by a complex algorithm

combining many factors, the hospital emergency team

may know about the upcoming event earlier than the

patient him-/herself. This evidently is true if the mea-

surement system is reliable and the algorithm is cor-

rect. We see that taking quickly correct, resolute de-

cisions is not easy.

5 REAL TIME ANALYSIS

In the systems that handle serious medical conditions

where emergencies may occur, reaction time is essen-

tial. At the lowest level we treat basic communica-

tions issues, like network architecture, data quantity,

channel capacity and similar. Several papers discuss

these problems and propose various feasible architec-

tures (Thelen et al., 2015), (Castellano et al., 2015),

(Hossain and Muhammad, 2016), (Kang et al., 2015).

Let us analyze the sequence of events (Figure 1)

that if not handled on time, can lead to a catastrophic

outcome, like death or a durable health damage.

We monitor the health state of the patient with pe-

riodic measurements. The time from the occurrence

of the emergency state to a catastrophic outcome is

a medical fact, as well as the possible outcome itself

(risk level). The time from noticing the emergency to

an intervention depends on technology. The period of

the measurements has to be adapted adequately. For

fast events, like a heart attack, where the time limit

Figure 1: Timing of a medical intervention.

for an intervention is around one hour, the monitoring

should be continuous (measurement period reduced to

almost zero) and the intervention delay reduced to a

minimum. For slow events, like breast cancer, where

the disease develops in months, the measurement pe-

riod is decisive. Not observing this limits makes the

monitoring system virtually useless.

Under warning, as indicated in the ﬁgure 1, we un-

derstand some early signals that suggest an increased

risk (yellow alarm). Early warning permits to extend

our time reserve for action. As the indication in this

stage is less decisive, full scale intervention is not ap-

propriate. The patient may however reduce the risk by

taking a rest, performing additional tests or visiting a

doctor.

Breast cancer screening with mammography is a

well studied example of risk analysis (Wegwarth and

Gigerenzer, 2013). The authors show we should be

wary of overdiagnosis and overtreatment.

When deciding what and how to measure and

what actions to take, following factors have to be con-

sidered:

• disease

– development time

– medical risk when not treated on time

– cost of intervention

• measurements (possible overdiagnosis)

– cost (ﬁnancial and organizational)

– medical side effects

• false positives (overtreatment)

– probability

– cost of unnecessary treatment

– medical side effects

The ﬁgure 2 depicts the basic trade-offs in this

process. Mobile health technology mainly permits to

reduce the cost of the measurement and in this way to

make more frequent measurements without visiting a

doctor - at home and in travel. It also permits to send

Figure 2: Trade-offs between various factors inﬂuencing a

successful intervention.

quickly the data and emergency alarms, allowing the

patient to maintain the same risk level even having an

active life.

In the case of a disease with high risk and short

required reaction times (like heart disease), it is im-

portant to design carefully the emergency service, en-

suring high reliability and respecting the time limits.

As this problem goes beyond the desine of the device

itself, it will be treated in the following section, re-

garding the Cyber-Physical-Social Systems.

6 CYBER-PHYSICAL-SOCIAL

SYSTEMS

It is important to see that mHealth systems consist of

more than the purely technical elements. They inter-

act intensely with the physical world. The sensors

themselves are material - they need electrical power,

their probes can break, sensing surfaces can have poor

contact to the skin, output nozzles can be clogged. If

we speak of Body Area Network, we have to remem-

ber that tissue and clothing damp the wireless signal.

If the patient moves in the environment, the wire-

less signal connecting him/her with the server may be

blocked by obstacles (e.g. when visiting the restroom

in a restaurant’s basement). He/she can also move into

the area where his provider has a poor signal or data

roaming is impossible.

The emergency systems (Figure 3) are operated

Figure 3: Emergency service.

by humans (Human in the Loop) and depend on their

cognitive skills and information state. For example,

if a heart monitoring system issues an alarm and the

ambulance scheduler has no experience with auto-

mated systems, his/her thinking time adds directly to

the delay of the rescue. The hospital that sends the

ambulance is located near to the current position of

the patient is not necessarily the same that handles

his/her case and owns his/her health record. This re-

quires to arrange the cooperation in advance, includ-

ing a smooth interoperable data transfer and ﬁnancial

agreements. In the case of a heart attack the whole

process has to be concluded during the golden hour

and leaves no reserve for doubts and clariﬁcations.

This shows that we should ensure that our assump-

tions about the system reliability and availability are

not simplistic. We have to remember that in real life

even a trivial cause (empty battery of a sensor, dry

skin under the sensor patch, loosened contact of a ca-

ble) may have dramatic consequences for the patient’s

health. Therefore when analyzing the system with

formal methods we should be aware that the model

and the real object are different and imagination and

common sense will be helpful.

7 VERIFICATION OF

HYPOTHESES

Data generated in an mHealth supported therapy can

be reused to verify our assumptions and obtain new

knowledge. Let us consider a heart monitoring sys-

tem issuing alarms and warnings in emergency cases.

If the system is working continuously, it can detect

signiﬁcant events also if the patient is not aware of

anything. They may consist of changed ECG wave-

forms, slower / faster / less periodic frequency or else.

The intensity and duration of such events may play a

role, as well as their sequence. Combining them with

other vital signals will make the detection more spe-

ciﬁc.

Figure 4: Using knowledge / hypothesis.

Normally, as shown in the ﬁgure 4, we assume that

we know the rules and just detect the characteristic

signals that indicate that an emergency event is immi-

nent.

Figure 5: Searching for a new hypothesis.

The area of continuous ECG analysis, especially

combined with other signals, is however fairly new.

Therefore such event detection rules have to be ver-

iﬁed, possibly on many patients, in many situations.

Mobile health systems provide us such opportunity.

We can analyze the actually occurring events and look

back at the signals (Figure 5) searching for telltale

data events, determine their characteristics and ver-

ify if the correlation is signiﬁcant. In this way, we

can create new hypotheses based on data. Evidently,

quantifying variability of continuous signals is not

easy, especially if we do not know beforehand what

are we looking for. This shows that storing raw data

(or extensive excerpts of them) for later analysis is

useful, even if the beneﬁt is not immediately evident.

8 COMMUNICATION AND

STORAGE - SECURITY

Data collected from a mHealth device are sent to a

receiver. Locally, it may be a smartphone or another

data aggregator. A device can also have a direct con-

nection to a server in the cloud. It is interesting what

happens to those data later: where are they processed

and stored, who owns them, for which purposes can

they be used.

8.1 Architectures

Following factors are important in the analysis of the

transmission and processing of data:

• data quantity

• network independence

• local use of data

– user feedback

– local action

• data reuse at a server

• cost of local computation

• cost of transmission

Especially in handling continuous signals there is

a trade-off between storage, computation and com-

munications. For taking decisions, we are interested

in global parameters (pulse frequency, oxygen satura-

tion) or special states or events (arrhythmia). If the

rules to extract such condensed information are well

deﬁned, it is better to do the processing locally on the

strongest device that can handle this task. This could

mean that the implanted sensor device sends raw data

to the smartphone, and the smartphone executes the

complex calculation. This reasoning is only an ex-

ample, in a speciﬁc case a precise analysis of compu-

tational power, transmission channel capacity, energy

costs, etc. has to be performed.

Measurements can be used for local action with

strong requirements for precision and reaction time,

like in glucose management system. It is essential to

ensure this function locally, also in the absence of the

connection with the remote system. In this case we

often speak of Fog Computing - like in Cloud Com-

puting we have storage and processing nodes but not

high in the sky, but rather locally, near to the ground

(Gia et al., 2015), (Chakraborty, 2016).

8.2 Security

As mobile health systems handle very personal data,

they have to be properly secured. This concerns both

data transmission and storage. It is easier said than

done. Sensor and actuator nodes are low power de-

vices and handling strong security is costly. Basically

all exposed networks should be properly managed -

this includes applying security patches if necessary.

However distributing software updates via network is

itself an attack vector.

We have a set of heterogeneous devices com-

ing from different sources, with different lifetimes.

Establishing secure communications with the Public

Key Infrastructure (PKI) is difﬁcult. We have also to

be aware of the trade-off between security and func-

tion. On one hand, the devices should be sure they

talk to trusted partners. On the other hand, if the se-

curity certiﬁcate expires on a node and the communi-

cation is blocked, this would stop the proper function

of the device that may be critical for the health or life

of the patient.

Various architectural options, also in the context

of the Internet of Things (IoT) and cloud computing,

with the stress on the security aspects are presented

in (AlTawy and Youssef, 2016), (Gejibo et al., 2015),

(Samie et al., 2016), (Suciu et al., 2015), (Larburu

et al., 2016) and (Sukor et al., 2015).

9 INDIVIDUAL PATIENT AND

POPULATION - PRIVACY

Data regarding individual patients can be collected

and reused for many purposes. Fitness tracking ap-

plications typically permit to send own data to the

pool and to compare personal results with the com-

munity. This community has some basic stratiﬁca-

tion, e.g. with respect to gender and age. The quality

of input data is unproven, no sources of bias are con-

sidered. The power of the solution lies in quantity of

data. Such comparison is used mostly for personal

satisfaction and motivation for further effort. It has

to be mentioned that more effort is not always good,

especially for elderly with osteoporosis and worn-

out knees. The competition may cause an addiction

where gaining points and virtual rewards count more

than the actual health.

In order to participate, the patient has to agree, i.e.

to express consent to share data. However, the exact

conditions of this consent, if published, are never con-

sulted. All detailed data reside on the servers of the

provider and the patient somehow assumes that exact

times and locations of his/her walks will not be dis-

closed to third persons or sold.

In the treatment of more serious medical cases,

data can be aggregated and analyzed in order to ob-

tain and verify knowledge. This can help to identify

risk factors or to issue recommendations for healthy

behavior.

If statistics based on collected data are used for

generating decisions (actionable knowledge), we have

to ensure adequate quality and statistical validity. Var-

ious aspect of medical data reuse are discussed in

depth in (Sliwa, 2016b). If the decisions apply to

the entire population, we should eliminate bias. If

the population shows strong variations, it should be

properly stratiﬁed, and the statistics for each category

have to satisfy the quality criteria. This is not easy,

as it is more practical to observe the population than

to execute a formal clinical trial. To put it simply: if

mostly young, technically oriented patients are will-

ing to share their data, those data should not be used

as a benchmark for the entire population, including

the elderly.

Evidently, the system provider has access to all

data and can use them at least to improve the ser-

vice. Due to the fast pace of the technical develop-

ment, much faster than the legislation, from formal

legal point of view the area of reusing mobile health

data is a gray zone. The question of data ownership

in a multi-party Internet of Things (IOT) system, with

smart medical devices as one of the examples, is dis-

cussed in (Sliwa, 2016a).

10 CONCLUSIONS AND FUTURE

WORK

The basic goal of this analysis is to raise awareness for

the semantic aspects of data processing in mHealth

systems. It is important to understand the properties

of the signals ﬂowing from the sensors and their rel-

evance to the overall health support function of the

system. Their properties are determined by the med-

ical factors, like the severity of the case, the possible

outcome. the necessary intervention and its time con-

straints.

After the semantic analysis the elementary prop-

erties of the signals can be extracted, which permit to

build a formal model:

• data type

• conversion algorithm

• timing requirements

• data quantity

• duration and location of storage

• ownership and protection

Nevertheless, it has to be stressed that for complex

Cyber-Physical-Social-Systems, as in Mobile Health,

the formal model is only an approximation and has

to be constantly veriﬁed. The environmental condi-

tions are diverse, the technology changes rapidly and

human behavior is difﬁcult to predict, therefore the

usage of such systems has to observed and the design

assumptions have to be periodically reviewed.

REFERENCES

Agboola, S. O., Bates, D. W., and Kvedar, J. C. (2016). Dig-

ital health and patient safety. JAMA, 315(16):1697–

1698.

AlTawy, R. and Youssef, A. M. (2016). Security tradeoffs

in cyber physical systems: A case study survey on im-

plantable medical devices. IEEE Access, 4:959–979.

Baig, M. M., GholamHosseini, H., and Connolly, M. J.

(2015). Mobile healthcare applications: sys-

tem design review, critical issues and challenges.

Australasian Physical & Engineering Sciences in

Medicine, 38(1):23–38.

Castellano, N. N., Gazquez, J. A., Salvador, R.

M. G., Gracia-Escudero, A., Fernandez-Ros, M., and

Manzano-Agugliaro, F. (2015). Design of a real-time

emergency telemedicine system for remote medical

diagnosis. Biosystems Engineering, 138:23–32.

Chakraborty, S. (2016). Data Aggregation in Healthcare

Applications and BIGDATA set in a FOG based Cloud

System. PhD thesis, University of Cincinnati.

Eapen, Z. J., Turakhia, M. P., McConnell, M. V., Graham,

G., Dunn, P., Tiner, C., Rich, C., Harrington, R. A.,

Peterson, E. D., and Wayte, P. (2016). Deﬁning a

mobile health roadmap for cardiovascular health and

disease. Journal of the American Heart Association,

5(7).

Espinoza, J., Chandy, D., Ochoa, S., Jim

enez, J., Huerta,

M., Soto, A., Sagbay, G., and Avila, R. (2016). Design

of telemedicine management system in ecuador. In

Ecuador Technical Chapters Meeting (ETCM), IEEE,

volume 1, pages 1–6. IEEE.

Gejibo, S., Mancini, F., and Mughal, K. A. (2015). Mo-

bile data collection: A security perspective. In Mobile

Health, pages 1015–1042. Springer.

Gia, T. N., Jiang, M., Rahmani, A.-M., Westerlund, T.,

Liljeberg, P., and Tenhunen, H. (2015). Fog com-

puting in healthcare internet of things: A case study

on ecg feature extraction. In Computer and In-

formation Technology; Ubiquitous Computing and

Communications; Dependable, Autonomic and Se-

cure Computing; Pervasive Intelligence and Comput-

ing (CIT/IUCC/DASC/PICOM), 2015 IEEE Interna-

tional Conference on, pages 356–363. IEEE.

Hossain, M. S. and Muhammad, G. (2016). Cloud-assisted

industrial internet of things (iiot)–enabled framework

for health monitoring. Computer Networks, 101:192–

202.

Huckvale, K., Adomaviciute, S., Prieto, J. T., Leow, M. K.-

S., and Car, J. (2015). Smartphone apps for calcu-

lating insulin dose: a systematic assessment. BMC

Medicine, 13:106.

Islam, S. R., Kwak, D., Kabir, M. H., Hossain, M., and

Kwak, K.-S. (2015). The internet of things for health

care: a comprehensive survey. IEEE Access, 3:678–

708.

Itrat, A., Taqui, A., Cerejo, R., Briggs, F., Cho, S.-M., Or-

ganek, N., Reimer, A. P., Winners, S., Rasmussen, P.,

Hussain, M. S., et al. (2016). Telemedicine in pre-

hospital stroke evaluation and thrombolysis: taking

stroke treatment to the doorstep. JAMA neurology,

73(2):162–168.

Kang, K., Wang, Q., Hur, J., Park, K.-J., and Sha, L. (2015).

Medical-grade quality of service for real-time mobile

healthcare. IEEE Computer, 48(2):41–49.

Lanzola, G., Losiouk, E., Del Favero, S., Facchinetti, A.,

Galderisi, A., Quaglini, S., Magni, L., and Cobelli, C.

(2016). Remote blood glucose monitoring in mhealth

scenarios: A review. Sensors, 16(12):1983.

Larburu, N., Bults, R., van Sinderen, M., Widya, I.,

and Hermens, H. (2016). Quality-of-data broker for

quality-of-data-aware telemedicine systems. IRBM,

37(4):210–218.

Samie, F., Bauer, L., and Henkel, J. (2016). Iot tech-

nologies for embedded computing: A survey. In

Hardware/Software Codesign and System Synthesis

(CODES+ ISSS), 2016 International Conference on,

pages 1–10. IEEE.

Sannino, G., De Falco, I., and De Pietro, G. (2015). A

supervised approach to automatically extract a set of

rules to support fall detection in an mhealth system.

Applied Soft Computing, 34:205–216.

Sannino, G. and De Pietro, G. (2014). A mobile system for

real-time context-aware monitoring of patients health

and fainting. International journal of data mining and

bioinformatics, 10(4):407–423.

Silva, B. M., Rodrigues, J. J., de la Torre D

ıez, I., L

opez-

Coronado, M., and Saleem, K. (2015). Mobile-health:

a review of current state in 2015. Journal of biomedi-

cal informatics, 56:265–272.

Sliwa, J. (2016a). A generalized framework for multi-party

data exchange for iot systems. In 2016 30th Interna-

tional Conference on Advanced Information Network-

ing and Applications Workshops (WAINA), pages 193–

198. IEEE.

Sliwa, J. (2016b). Reuse of data from smart medical devices

for quality control and evidence based medicine. In

Dobre, C. and Xhafa, F., editors, Pervasive Comput-

ing: Next Generation Platforms for Intelligent Data

Collection, chapter 9, pages 255–297. Morgan Kauf-

mann.

Soh, P. J., Vandenbosch, G. A., Mercuri, M., and Schreurs,

D. M.-P. (2015). Wearable wireless health monitoring:

Current developments, challenges, and future trends.

IEEE Microwave Magazine, 16(4):55–70.

Suciu, G., Suciu, V., Martian, A., Craciunescu, R., Vulpe,

A., Marcu, I., Halunga, S., and Fratu, O. (2015). Big

data, internet of things and cloud convergence–an ar-

chitecture for secure e-health applications. Journal of

medical systems, 39(11):1–8.

Sukor, J. A., Mohktar, M. S., Redmond, S. J., and Lovell,

N. H. (2015). Signal quality measures on pulse oxime-

try and blood pressure signals acquired from self-

measurement in a home environment. IEEE journal

of biomedical and health informatics, 19(1):102–108.

Thelen, S., Czaplik, M., Meisen, P., Schilberg, D.,

and Jeschke, S. (2015). Using off-the-shelf med-

ical devices for biomedical signal monitoring in a

telemedicine system for emergency medical services.

IEEE journal of biomedical and health informatics,

19(1):117–123.

Wegwarth, O. and Gigerenzer, G. (2013). Overdiagnosis

and overtreatment: evaluation of what physicians tell

their patients about screening harms. JAMA internal

medicine, 173(22):2086–2087.

Xue, S., Chen, X., Fang, Z., and Xia, S. (2015). An

ecg arrhythmia classiﬁcation and heart rate variabil-

ity analysis system based on android platform. In

Future Information and Communication Technologies

for Ubiquitous HealthCare (Ubi-HealthTech), 2015

2nd International Symposium on, pages 1–5. IEEE.