A Clinical Decision Support System based on an Unobtrusive Mobile App

Ariella Richardson

, Avigail Perl

, Sapir Natan

and Gil Segev

Lev Academic Center, Jerusalem, Israel

BGSegev Ltd. (segevlabs.org), Jerusalem, Israel

Keywords:

Mobile Health, Digital Health, Digital Monitoring, Clinical Decision Support System (CDSS), Medical

Decision Support System (MDSS), Cardiovascular Disease, Silent Disease.

Abstract:

Clinical decision support systems typically rely on medical records and information collected in the doctor’s

ofﬁce. We propose a clinical decision support system that uses data collected from patients continuously and

in an unobtrusive manner. The system uses data collected from a mobile app installed on the patient’s device

(such as a mobile phone, smart-watch etc). The app collects data without user interference and combines it

with conventional medical records. Our system uses machine learning methods to extract meaningful insights

from the data. The output from the learning process is then presented to the doctor in a clear and meaningful

fashion on a web based platform. This system can be used to assist effective treatment selection, enable early

diagnosis, trigger alarms in case of an emergency and provide a tool for disease monitoring. We describe our

clinical decision support system and directions for future work.

1 INTRODUCTION

Often patients visit their doctor when they feel unwell

searching for a diagnosis. Doctors then try to deci-

pher the cause of the patients feeling. However, they

do not always have all the data necessary for precise

diagnosis, and the data that they have is often unclear

and hard to integrate. As the amount of data avail-

able for patients is vast, the need for clinical decision

support systems - CDSS (also called medical decision

support systems - MDSS) is great (El-Sappagh and

El-Masri, 2014). CDSS have been developed for var-

ious settings and conditions. A detailed architecture

for integrating electronic medical records from multi-

ple sources is presented in (El-Sappagh and El-Masri,

2014) and (Kawamoto et al., 2005; Shibl et al., 2013)

survey others. Developing a CDSS that makes a con-

tribution to the doctors practice is complex. Many

of them do not improve clinical practice, as shown

by (Kawamoto et al., 2005) who present a survey of

several systems and discuss parameters that correlate

with improving clinical practice. Some systems are

simply not accepted by practitioners, factors are de-

scribed in (Shibl et al., 2013).

Alongside CDSSs, health related applications for

mobile phones and smartwatches are capable of im-

proving health monitoring and detection. The number

of health related apps available is astounding, approx-

imating 40,000 apps in 2013 (Boulos et al., 2014) and

165,000 in 2015 (Terry, 2015) and growing continu-

ously. While CDSS are typically based on electronic

medical records and targeted at assisting the practi-

tioners, mobile apps often aim at providing feedback

to the patients themselves. We propose a CDSS that

collects data on a patients’ mobile phone, and then

uses it as input to the CDSS presented to the doctor.

Our proposed system is a CDSS that integrates tra-

ditionally documented information alongside sensor

information collected by the patient app on a mobile

device, such as a phone or smart-watch. The com-

bined data is analyzed using machine learning meth-

ods and presented in a clear and meaningful fashion

to the clinician. This enables assistance in making

medical diagnostic decisions. The system can also be

used to detect and alert the doctor or patient in case of

an emergency or deterioration. An important feature

of our system is that data is collected not only in the

doctor’s ofﬁce, but also between visits. This enables

performing a diagnosis based on a much more infor-

mation than is typically possible with other systems.

Our system overview appears in Figure 1. (patent re-

quest submitted (BGSEGEV, 2018)).

Among the surplus of medical apps, many require

the users to actively interact with the application in

order to achieve medical feedback, for example (Seo

et al., 2015; Zhang et al., 2015; Nam et al., 2014).

Richardson, A., Perl, A., Natan, S. and Segev, G.

A Clinical Decision Support System based on an Unobtrusive Mobile App.

DOI: 10.5220/0007587001670173

In Proceedings of the 5th International Conference on Information and Communication Technologies for Ageing Well and e-Health (ICT4AWE 2019), pages 167-173

ISBN: 978-989-758-368-1

167

Figure 1: Overview of system.

In our system, we propose an app that will be able

to provide meaningful information, without requiring

user actions for data collection and input. Our CDSS

works in conjunction with an application that accom-

panies the patient and collects data on-line using a

mobile phone application. We collect data during reg-

ular phone usage, without the need for direct patient

involvement. This characteristic is what makes our

system different from many other health apps.

We currently focus on vascular diseases such as

stroke, heart and peripheral vascular disease but have

also been expanding our work to other conditions

where sensor data has predictive capabilities such as

cancer and osteoporosis. We collect data from many

sources, for example: typing patterns, voice record-

ings, pulse measurements, walking patterns and in-

jury related data.

Generally, during a doctor’s visit, the doctor re-

lies on descriptions of the patient’s subjective feeling,

and the patient’s description of his health difﬁculties.

These descriptions are often inadequate, and affected

by recent subjective feelings, rather than unbiased,

long-term and continuous health monitoring. Visits to

the doctor are sporadic and often far and between due

to the constraints of the health system. The physician

has no way of detecting deterioration in the patient’s

condition between visits. Patients often visit the doc-

tor only when the pain is unusual and sometimes this

is a sign that the disease has already progressed. Our

system has the beneﬁt of being able to collect data

over time and between doctor visits. The use of such

data, may present a clearer picture of the patients con-

dition than data collected periodically during doctor

visits, or inexact patient descriptions of symptoms.

Moreover, the patient may not feel any symptoms

of a disease. It is known, for example, that many peo-

ple with PAD (Peripheral Artery disease) report a lack

of symptoms, including those with a relatively serious

illness (Criqui and Aboyans, 2015). With our CDSS,

diseases may be detected even before the patient be-

comes symptomatic. Identifying diseases in an early

stage is critical to treating them and may even prevent

their onset, hence the great contribution of our frame-

work. Obviously security measures must be consid-

ered when handling sensitive medical data, these are

out of the scope of this study.

There is need for a system that will provide a

maximal response to all of its needs without requir-

ing active involvement from the patient during data

collection. Our CDSS offers state-of-the-art manage-

ment, documentation and user interface platform that

receives constant feedback from the patient’s applica-

tion about dynamic data collected in the background

and processed using data mining. Our proposed sys-

tem is autonomous, and therefor reduces the number

of required doctor visits, reduces human error, and

enriches the medical diagnosis and monitoring pro-

cess with information from new sources such as the

smartphone sensors. The data mining performed in

the system will provide rich output to the doctor and

patient, and may also be used for real-time events and

for alerting caregivers.

ICT4AWE 2019 - 5th International Conference on Information and Communication Technologies for Ageing Well and e-Health

168

2 RELATED WORK

We consider two types of studies to be relevant to our

work. The ﬁrst are studies on clinical decision support

systems. The second set of studies are mobile appli-

cations for healthcare, as our CDSS is heavily based

on a mobile application for data collection.

CDSSs cover a broad variety of topics. Some

systems are speciﬁc to a condition such as diabetes

(Weymann et al., 2016) or retinal disease (Bourouis

et al., 2014). Many others are aimed at assisting di-

agnosis based on the vast amount of medical infor-

mation available. (El-Sappagh and El-Masri, 2014)

present an architecture for combining electronic med-

ical records into a single DSS to assist clinicians. A

survey of CDSSs along with a detailed discussion ex-

plaining the need for such systems can be found in

(Castaneda et al., 2015). Although CDSSs are plen-

tiful, they are often slow to be accepted and do not

always improve medical practice. (Shibl et al., 2013)

survey several CDSS and discuss factors that impact

system acceptance. One of the main parameters found

to affect acceptance was whether the system inter-

fered with the regular work-ﬂow. Systems that re-

quired a break in the normal work-ﬂow were not usu-

ally accepted. Surprisingly, ease of use was consid-

ered less important. Clinicians claimed that if a sys-

tem was helpful they would be prepared to make an

effort to use it. Other parameters are described in

the paper. (Kawamoto et al., 2005) perform a simi-

lar survey to identify features that impact the success

of CDSSs. They claim that clinicians were found to

adopt systems, where the information was provided

automatically. Clinicians were less likely to use sys-

tems that required an active search for information.

Similarly, systems that used a computer to generate

decision support were more effective than those that

required manual intervention. The studies provide

an indication, that our proposed system that requires

minimal intervention on the side of the patient, and

highly automated and clearly presented integrated in-

formation to the clinician, will have a high chance of

acceptance.

Similar to our framework is the work by (Artikis

et al., 2012). They propose a CDSS that integrates

data from sensors collected on the patient, alongside

other medical data, just as we do. The focus of their

work is on dementia and depression, whereas we are

looking at cardiovascular disease and other conditions

with physical deterioration. To the best of our under-

standing their work was not expanded.

Another system that must be discussed in close re-

lation to our work is that of the MobiGuide Project

(Peleg et al., 2014; Peleg et al., 2017). MobiGuide fo-

cuses on arterial ﬁbrillation, and gestational diabetes.

The proposed system is targeted strongly at supply-

ing the patient with a DSS system to help manage

the disease. Although doctors also use the system,

it seems that focus of this large impressive study is

different to ours. We are less concerned with provid-

ing a DSS system for the patient. Our concept uses

the data collected from the patient as input to the clin-

icians decision. Our system collects as much data as

possible without inconveniencing the patient, in order

to supply the doctor with as much information as pos-

sible. The information is beneﬁcial, as it is collected

between the ofﬁce visits, and we hope it will assist

in making a decision that is supported by long term

conditions, as opposed to solely relying on recent in-

formation reported by the patient in the ofﬁce.

Aside from CDSSs, it is important to consider mo-

bile health applications that are relevant to our frame-

work. The number of mobile applications for health-

care is vast and constantly expanding (Boulos et al.,

2014; Terry, 2015). Most mobile health apps are tar-

geted at monitoring and rehabilitation such as (Seo

et al., 2015) who tested the feasibility of a mobile app

for patients who had suffered a stroke. The app was

aimed at managing risk factors such as blood pressure

and diabetes management. Other applications (Zhang

et al., 2015; Micallef et al., 2016), accompany the pa-

tient by encouraging exercises, following up on taking

pills, and logging mood reports.

Some of the apps are aimed at bridging the dis-

tance between the patients to medical assistance.

(Nam et al., 2014) provide a stroke screening appli-

cation. The application shows a set of cartoons repre-

senting stroke symptoms. Potential patients can fol-

low the cartoons and try to determine whether they

may be suffering from a stroke. (Mitchell et al., 2011)

bridge the gap by providing a teleradiology system

that enables a doctor to interpret a CT scan. (Demaer-

schalk et al., 2012) provide high-quality video tele-

conferencing.

Several surveys on the use of smartphones in

medicine (Ozdalga et al., 2012; Boulos et al., 2014;

Dobkin and Dorsch, 2011) show how smartphones

can be used for patient care, monitoring and rehabil-

itation alongside accessing clinical data such as elec-

tronic medical records CT scans etc. Most of these

systems require active and sometimes heavy user in-

volvement. Patients are expected to use the app fre-

quently and enter the relevant information for data

collection. We did not encounter a system that com-

bines both of an unobtrusive application on the pa-

tient side with a CDSS for the doctor, in the structure

of the system that we propose in this paper.

A Clinical Decision Support System based on an Unobtrusive Mobile App

169

3 DESCRIPTION OF OUR CDSS

The main focus of this paper is the CDSS for the doc-

tor. As shown in Figure 1 data from phone sensors

is aggregated with medical history. After this data is

analyzed using machine learning, a clear presentation

of the output including diagnostic suggestions is dis-

played to the doctor. We developed a prototype web

based application that presents this information to the

doctor, and describe it in section 3.1. We also devel-

oped a data collection system to collect labeled data

in order to train our system, described in Section 3.2.

3.1 CDSS

In designing the DSS for the doctor, we kept in mind

that the system should be designed in a convenient

and intuitive manner. Physician activity while us-

ing the system should not deviate much from regular

practice, as claimed by (Shibl et al., 2013). Improve-

ments will aim at saving time and improving diagno-

sis accuracy. We take User interface and experience

into account in our design. The CDSS has been de-

veloped as a prototype (in Hebrew), with the actual

machine learning engine left for future work.

We describe and display some of the more inter-

esting features of our CDSS. The ﬁrst feature is the

disease simulator. The simulator automatically ticks

boxes that the system has detected from the app data.

The doctor can then discuss these choices with the

patient and change the selection of symptoms. Once

ticked symptoms are agreed on, the system outputs

suggestions for possible diagnosis. The diagnosis out-

put by the system takes into account both the data en-

tered by the doctor, but also the input collected auto-

matically by the app on the patients device. The in-

tegration of these two sets of data, both that inserted

by the doctor, and that automatically collected from

the patient app are what make this simulator unique.

An example of what this screen looks like is shown in

Figure 2. The simulator provides the doctor with a list

of possible symptoms, and the doctor inputs data in a

simple manner by ticking boxes. Examples to symp-

toms (that are ticked) are limping, slow ulcer heal-

ing, ulcers on legs and weight loss. After the aggre-

gation with the patient app data the diagnostic sug-

gestions are presented on the screen, and can be seen

in red. For example on the this screen the proposed

diagnoses are Peripheral Vascular disease (PVD) and

Osteoporosis.

One of the special features in our system, is plots

of the data collected by the patient app that can be

viewed in real time as the data accumulates or at a

later time. The doctor may choose to enter the CDSS,

Figure 2: Diagnosis simulator. Symptoms in tic boxes, Di-

agnosis in red.

Figure 3: Plots of patient symptoms, alongside proposed

diagnosis with conﬁdence rating.

between ofﬁce visits, to follow how a patient is per-

forming. The screen displays the name of the patient,

and a set of plots for graphs of ulceration information,

walking patterns, pulse measurements and sleep pat-

terns. These are all shown in Figure 3. The graphs

show the patient’s condition in the various proﬁles.

The graph is accompanied by a box with the sug-

gested diagnosis along with a degree of conﬁdence

that this diagnosis is correct. Plots can also be be

printed or saved.

Alerting the doctor to suspicious events that re-

quire attention are an important part of our system.

Figure 4 is the description of the proﬁle belonging to

a patient as shown to the doctor. The data includes the

patient ID, age, name, phone number, address, gender

etc. The data also includes medical information such

as smoking habits, alcohol consumption, medications,

sport, diet etc. In case of a suspicious event, an alert

is displayed on the top of the screen, as shown in the

ﬁgure. Then details of the patients condition can then

be seen in Figure 5, that shows an example of a re-

ICT4AWE 2019 - 5th International Conference on Information and Communication Technologies for Ageing Well and e-Health

170

Figure 4: Patient proﬁle, with an alarm indicating a suspi-

cious event.

port that summarizes a patients data collection. Each

row has the date, type of activity performed, state de-

tected and the suspected condition. Events that have

been detected as suspicious are marked in red. Ulcers

are accompanied by a picture of the ulcer.

Our system also includes standard medical system

functionality such as the display of patient informa-

tion in a table form, as shown in Figure 6 that de-

scribes the patients information (name, ID, visit num-

ber) Then the diagnosis, the reason for the visit, extra

referrals and recommendations. Aside from the pa-

tient information, the doctor is offered links to extra

reading material as shown in Figure 7, enabling ac-

cess to the latest medical articles on relevant topics.

3.2 Labeled Data Collection

As we have stated the app used to collect patient data

is unobtrusive. Data is collected without the need for

user intervention. However, in order to build machine

learning models we need to collect labeled data for

the conditions we wish to diagnose. In order to for-

ward our research we developed a data collection app

for research. This application (that is different to the

application for data collection for the CDSS) requires

the user to describe the type of information being col-

lected, and then collect the information. Examples to

types of data that can be collected for research with

this app are various walking activities such as run-

ning, walking and even falling, or dropping the phone.

The are examples to activities that might play an im-

portant part in stroke detection. We can collect the

typing patterns that are created as patients type mes-

sages or other text into the phone. We also collect

voice recordings that are obtained during phone calls.

Figure 5: Example of visit report.

Figure 6: Patient data in tabular form, with suspicious

events marked in red.

Some of the application screens are shown in Figures

8 and 9. Figure 8 shows the screen where user infor-

mation is inserted. The user ﬁrst deﬁnes a proﬁle, by

entering a name, gender, and age. Then the user may

select a condition from a lost of conditions, such as

PVD, and continues by pressing the ”continue” button

on the bottom of the screen. This brings the patient to

the next screen, for recording the activity. This data

is used to identify the subject, as the app may be used

to collect data from many subjects. Figure 9 is the

screen used for recording an activity. The activity will

be recorded using all sensors, and labeled according

the the information provided in the ﬁrst screen. First

the type of activity is selected (i.e. walking, climbing

stairs), then the phone position is selected, and then

recording is performed by pressing the red round but-

ton, and stopped by pressing the black square button.

A Clinical Decision Support System based on an Unobtrusive Mobile App

171

Figure 7: Extra reading material.

Figure 8: App for collecting patient information for re-

search - subject data.

Data were collected for patients with various car-

diovascular diseases: PVD, stroke, heart disease and

also for osteoporosis. Early detection is a common

necessity for all these diseases, and is important for

the effective treatment of the disease. The ability to

use the information obtained from the various sensors

of the application contributes greatly to carry out the

Figure 9: App for collecting patient information for re-

search - activity description.

disease detection. This app can easily be extended to

perform data collection for research for many more

conditions, and used for other research applications

as well.

4 DISCUSSION AND FUTURE

WORK

This study presents the development of an CDSS sys-

tem for a doctor that works together with the patient’s

app. The patient app collects data from the patient,

during regular phone usage, without requiring that

user actively assist the data collection (but obviously

with patient permission). The data is collected on the

cloud and used by the doctor CDSS system to mon-

itor and detect various diseases. It seems that the

structure of the system, which combines all the in-

formation from the patient, constitutes a signiﬁcant

milestone in the management of the patient’s medi-

cal ﬁle, to the point of identifying diseases at an early

stage and even before their outbreak. The system re-

ICT4AWE 2019 - 5th International Conference on Information and Communication Technologies for Ageing Well and e-Health

172

lies on having successful data mining models, and a

description of these is left for future work, as it is a

separate module of the system. Using data mining to

analyze the data collected in the app brings the sys-

tem even further towards better disease management.

In the future, we will offer research and analysis of

data from other diseases, such as cancer that our team

has begun to study, and adapt the system accordingly.

We have begun studying various data mining meth-

ods (Richardson et al., 2019) in order to select the

most appropriate models from our CDSS and will re-

port on progress in future work. Future studies will

also involve testing the system with both patient and

caregiver subjects.

REFERENCES

Artikis, A., Bamidis, P. D., Billis, A., Bratsas, C.,

Frantzidis, C., Karkaletsis, V., Klados, M., Konstan-

tinidis, E., Konstantopoulos, S., Kosmopoulos, D.,

et al. (2012). Supporting tele-health and ai-based

clinical decision making with sensor data fusion and

semantic interpretation: The useﬁl case study. In

International workshop on artiﬁcial intelligence and

NetMedicine, page 21.

BGSEGEV (2018). BGSegev STROKETOR - stroke detec-

tor. https://stroketor.segevlabs.org/.

Boulos, M. N. K., Brewer, A. C., Karimkhani, C., Buller,

D. B., and Dellavalle, R. P. (2014). Mobile medical

and health apps: state of the art, concerns, regula-

tory control and certiﬁcation. Online journal of public

health informatics, 5(3):229.

Bourouis, A., Feham, M., Hossain, M. A., and Zhang, L.

(2014). An intelligent mobile based decision support

system for retinal disease diagnosis. Decision Support

Systems, 59:341–350.

Castaneda, C., Nalley, K., Mannion, C., Bhattacharyya, P.,

Blake, P., Pecora, A., Goy, A., and Suh, K. S. (2015).

Clinical decision support systems for improving di-

agnostic accuracy and achieving precision medicine.

Journal of clinical bioinformatics, 5(1):4.

Criqui, M. H. and Aboyans, V. (2015). Epidemiology

of peripheral artery disease. Circulation research,

116(9):1509–1526.

Demaerschalk, B. M., Vegunta, S., Vargas, B. B., Wu, Q.,

Channer, D. D., and Hentz, J. G. (2012). Reliability

of real-time video smartphone for assessing national

institutes of health stroke scale scores in acute stroke

patients. Stroke, 43(12):3271–3277.

Dobkin, B. H. and Dorsch, A. (2011). The promise of

mhealth: daily activity monitoring and outcome as-

sessments by wearable sensors. Neurorehabilitation

and neural repair, 25(9):788–798.

El-Sappagh, S. H. and El-Masri, S. (2014). A distributed

clinical decision support system architecture. Journal

of King Saud University-Computer and Information

Sciences, 26(1):69–78.

Kawamoto, K., Houlihan, C. A., Balas, E. A., and Lobach,

D. F. (2005). Improving clinical practice using clin-

ical decision support systems: a systematic review

of trials to identify features critical to success. Bmj,

330(7494):765.

Micallef, N., Baillie, L., and Uzor, S. (2016). Time to ex-

ercise!: an aide-memoire stroke app for post-stroke

arm rehabilitation. In Proceedings of the 18th Inter-

national Conference on Human-Computer Interaction

with Mobile Devices and Services, pages 112–123.

ACM.

Mitchell, J. R., Sharma, P., Modi, J., Simpson, M., Thomas,

M., Hill, M. D., and Goyal, M. (2011). A smartphone

client-server teleradiology system for primary diagno-

sis of acute stroke. Journal of medical Internet re-

search, 13(2).

Nam, H. S., Heo, J., Kim, J., Kim, Y. D., Song, T. J., Park,

E., and Heo, J. H. (2014). Development of smartphone

application that aids stroke screening and identifying

nearby acute stroke care hospitals. Yonsei medical

journal, 55(1):25–29.

Ozdalga, E., Ozdalga, A., and Ahuja, N. (2012). The smart-

phone in medicine: a review of current and potential

use among physicians and students. Journal of medi-

cal Internet research, 14(5).

Peleg, M., Shahar, Y., and Quaglini, S. (2014). Making

healthcare more accessible, better, faster, and cheaper:

the mobiguide project. European Journal of ePrac-

tice: Issue on Mobile eHealth, 20:5–20.

Peleg, M., Shahar, Y., Quaglini, S., Broens, T., Budasu,

R., Fung, N., Fux, A., Garc

ıa-S

aez, G., Goldstein,

A., Gonz

alez-Ferrer, A., et al. (2017). Assessment of

a personalized and distributed patient guidance sys-

tem. International journal of medical informatics,

101:108–130.

Richardson, A., Shani Ben Ari, and Sinai, M., Atsmon, A.,

Conley, E. S., Gat, Y., and Segev, G. (2019). Mobile

applications for stroke: A survey and a speech classi-

ﬁcation approach. In ICT4AWE, page to appear.

Seo, W.-K., Kang, J., Jeon, M., Lee, K., Lee, S., Kim, J. H.,

Oh, K., and Koh, S.-B. (2015). Feasibility of using

a mobile application for the monitoring and manage-

ment of stroke-associated risk factors. Journal of Clin-

ical Neurology, 11(2):142–148.

Shibl, R., Lawley, M., and Debuse, J. (2013). Factors inﬂu-

encing decision support system acceptance. Decision

Support Systems, 54(2):953–961.

Terry, K. (2015). Number of health apps soars but use does

not always follow. Medscape Medical News.

Weymann, N., H

arter, M., and Dirmaier, J. (2016). Informa-

tion and decision support needs in patients with type

2 diabetes. Health informatics journal, 22(1):46–59.

Zhang, M. W., Yeo, L. L., and Ho, R. C. (2015). Harness-

ing smartphone technologies for stroke care, rehabili-

tation and beyond. BMJ innovations, 1(4):145–150.

A Clinical Decision Support System based on an Unobtrusive Mobile App

173