BioMed Wizard

An Approach for Gathering Personal Risk Factor Data

Mohammad Shafahi, Hamideh Afsarmanesh and Stefan Paap

Informatics Institute, Faculty of Science, University of Amsterdam, Science Park 904, Amsterdam, The Netherlands

Keywords:

BioMed Wizard, Personal Data, Privacy, Risk Factors, Survey, Data Collection, Socio-economical Aspects.

Abstract:

People can be at risk of developing some serious diseases without being aware of it. Such diseases either do

not present symptoms in early stages or have simple symptoms that are ignored or not properly identiﬁed by

patients, due to their lack of medical know-how. On the other hand, in order to provide patients with early

indications of their risk level on developing such diseases, specially for chronic diseases such as diabetes type

2, it is necessary to collect substantial amount of personal data about risk factors related to the disease. A

smart wizard software applying the approach developed in our study, which brings awareness about some

socio-economical concerns of patients, can increase patients’ engagement in providing their personal data.

The case study focuses on the diabetes type 2 and some socio-economical concerns of patients, including

privacy invasion, time, and cost. In this research, the willingness of a sample group of more than 100 people

is surveyed, in providing their personal data, for three different scenarios and related to nine main risk factors.

The results collected in this survey is then applied to develop four user-speciﬁc data collection ﬂow models,

to be implemented in a smart wizard software.

1 INTRODUCTION AND

RESEARCH APPROACH

For a number of diseases, people can be at risk of

developing them, without being aware. Such dis-

eases do not present symptoms in early stages or have

symptoms that are typically ignored or not properly

identiﬁed by patients due to their lack of medical

know-how in identifying symptoms. However, it is

very important to recognize diseases in their early

stages of development, in order to reduce the chance

of life threatening situations. An example of a disease

that can develop without alarming with early symp-

toms, is diabetes mellitus type 2. This is a chronic

disease that causes high levels of glucose in the blood.

Diabetes can result in serious complications such as

heart disease and stroke, obesity, high blood pressure,

blindness, kidney disease, nervous system related dis-

eases and more (Zhang and Zhao, 2013).

According to the World Health Organization, in

2012, 1.5 million deaths were directly caused by dia-

betes (World Health Organization, 2014) demonstrat-

ing the importance of detecting the risk of diabetes for

the general public in an early stage. Diabetes type 2

can be developed during any stage of ones life. Sev-

eral risk factors are known to be associated to this dis-

ease, which vary from someones lifestyle that can in-

ﬂuence the risk of developing diabetes type 2, but also

inheritance (i.e DNA) can be of inﬂuence. Therefore,

people are not always aware of being in risk of devel-

oping diabetes. Although lack of exercising and con-

suming unhealthy food might not have a direct (neg-

ative) inﬂuence on someone, but it increases his/her

chance of developing the disease. Early detection

can reduce the burden of complications of diabetes, if

treated on time (Knowler et al., 2002). As such bring-

ing awareness and insight of the possible risk level of

developing diabetes, based on information provided

about current personal habits and status (i.e risk fac-

tors), will deﬁnitely help with this illness. Although

many researchers have designed prediction models

for risk of developing type 2 diabetes(Abbasi et al.,

2012), the current process of assessing the risk of de-

veloping a disease entails setting appointments with

general practitioners or specialists as a starting point,

followed by conducting a blood test. This can be a

time consuming and costly process for diagnosing the

risks at the early stages. To reduce the time and cost

of such process, and in order to have a more extensive

coverage of the assessment process, one would like to

only perform such a process when necessary and after

an initial screening and diagnosis process.

298

Shafahi, M., Afsarmanesh, H. and Paap, S.

BioMed Wizard - An Approach for Gathering Personal Risk Factor Data.

DOI: 10.5220/0005701102980305

In Proceedings of the 9th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2016) - Volume 5: HEALTHINF, pages 298-305

ISBN: 978-989-758-170-0

There are online software wizards that provide

users with information about diabetes, but these are

often only based on a limited number of risk factors

(e.g. only asking about the gender). For example after

a gender selection option, the online wizard provides

all risk factors and symptoms speciﬁc for that speciﬁc

gender. Although this could be helpful, it still very

much leaves the assessment of the risk level to the pa-

tients, and depends on the proper identiﬁcation of the

symptoms by them. However, the indication of the

patients risk level can be further clariﬁed and can be

made more precise if more risk factors are measured

or identiﬁed from the user. For example tools such as

the Diabetes Risk Score (Lindstr

om and Tuomilehto,

2003) use information about certain risk factors such

as age, BMI and diet to calculate a score that indicates

the risk level of developing diabetes type 2.

In this research we aim to develop an approach

for user speciﬁc data collection that can be developed

within a smart wizard software to provide users with

an indication of their current risk level, with the con-

sideration of socio-economical aspects in the collec-

tion process. A smart wizard software automatically

adapts future data collection based on the data that

the user has provided up to that stage. By doing so,

the software creates a user speciﬁc ﬂow for entering

data, helping the user to traverse through the process,

making it easier, faster, and more efﬁcient in a user

centric manner. In the case of this research the fo-

cus is more on what, how, in which order, and how

frequently to collect risk factor data. Next to focus-

ing on these, we focus on factors such as cost, time,

and the needed effort, since these are also factors that

play important roles on the willingness of someone to

share data. Therefore, socio-economical aspects are

also taken into consideration. The goal here is to get

insight on what data can be realistically gathered and

from where, and how it can be gathered for different

types of scenarios when considering difﬁdent phases

of screening for diabetes as an example disease. As

such in this research we try to answer the following

research question:

How can personal data be collected from users

when considering socio-economic barriers?

In order to address this research question a ﬁve-

phase research approach has been designed, consist-

ing of the following phases: 1) Identiﬁcation of Main

Risk Factors and Relevant Bio-markers, 2) Design of

the Survey, 3) Evaluation of the Survey, 4) Conduct-

ing the Survey, and 5) Analysis of Results and Design

of User Speciﬁc Data Collection Flow Models.

2 RELATED WORK

With the introduction of electronic medical ﬁles, cre-

ated and stored in digital systems, it became more

easy to share medical information. Later with the in-

troduction of electronic health records (EHRs), elec-

tronically sharing medical information about patients

with third parties became feasible and trivial for hos-

pitals upon the request of patients. This provided pa-

tients with more control over the availability of their

medical information.

Although research in 2009 (Jha et al., 2009) shows

that US hospitals still struggle to implement the elec-

tronic forms of health records (9.1% make use of a

basic EHR system), a study in 2014 (Adler-Milstein

et al., 2014) showed that at least 50% of the US hospi-

tals make use of a basic EHR system. This adoption is

not limited to hospitals, but also other medical profes-

sions, such as physicians, show an increasing trend of

using EHR systems (Xierali et al., 2013). These stud-

ies show that EHR systems are becoming more pop-

ular and the use of them will continue to rise in the

coming years. Since these electronic health records

contain medical data about the patient, information

about risk factors are likely to be found in these ﬁles.

A study in 2009 (Hivert et al., 2009) used informa-

tion stored in electronic health records to ﬁnd pat-

terns that identify patients with metabolic syndrome

that would put them at larger risk of becoming obese

and develop diabetes. This study demonstrated that

electronic health records can play an import role as

data sources in identifying patients with a risk of de-

veloping diseases.

Although such sources of data look promising one

must investigate the level of willingness and the situ-

ation in which people are willing to provide access

to their electronic health records. Studies have inves-

tigated the willingness of users to share information

in general for a certain cause. A study (Teixeira et al.,

2011) has investigated the willingness of HIV patients

to share personal health information with others. The

founding of the study suggests that the vast major-

ity (84%) are willing to share this data electronically,

but the individuals who would receive the shared data

have to be involved in the direct care of the patient.

Fewer patients (39%) are willing to share it with non-

carers electronically. Another study (Beckjord et al.,

2011) has looked into the willingness of cancer pa-

tients to exchange personal related information elec-

tronically. They have compared participants who are

diagnosed with cancer against participants who are

not diagnosed, and have found that people who are di-

agnosed are much more willing to share this informa-

tion to help others. A research done in 2014 (Pickard

and Swan, 2014) has concluded that there is a strong

BioMed Wizard - An Approach for Gathering Personal Risk Factor Data

299

willingness of people to share personal data in order

to enable next generation health care services, ulti-

mately leading to improved health outcomes for all.

These ﬁndings suggest that people are willing to

share personal information electronically, which is

also an important requirement for a smart wizard soft-

ware, but these studies also suggest that the amount of

information shared differ depending on the amount of

engagement the patient has in a speciﬁc disease and

how the share of this information directly inﬂuences

their health conditions. In this research we further

investigate the willingness of people in sharing per-

sonal information in a more ﬁne grained manner (i.e

bio-markers’ level) while taking into account the in-

ﬂuence of their proﬁle (i.e demographics and engage-

ment with disease) and will investigate the possibility

of adjusting data collection based on the users proﬁle

and level of engagement.

3 IDENTIFICATION OF RISK

FACTORS AND RELEVANT

BIO-MARKERS

There are many studies on individual risk factors re-

lated to diabetes mellitus type 2, such as obesity

(Tirosh et al., 2011), diet (Montonen et al., 2004)

and ethnicity (Harris, 2001) but Mayo Clinic (Mayo

Clinic, 2015) and the Dutch National Institute for

Public Health and Environment (Rijksinstituut voor

Volksgezondheid en Milieu, 2013), have provided an

extensive collection of the main risk factors including:

age, obesity, ethnicity, family history, genetic factors,

diet, smoking behavior, pre-diabetes, gestational dia-

betes (pregnancy diabetes) and (in)activity. But in or-

der to assess the risk level of developing diabetes type

2, we are interested in collecting bio-markers related

to risk factors. Bio-markers are aspects of the body

that can be measured in order to understand the cur-

rent condition of the risk factors. As an example, for

the risk factor obesity, BMI (Body Mass Index) can

be considered as a bio-marker and can be calculated

based on two other measurable bio-markers being the

height and weight of the patient. The same applies

in the case of genome factors, where the genome se-

quence is considered, so that research can focus on the

speciﬁc parts needed to determine the genome factors

regarding diabetes.

Although most risk factors can easily be mapped

to a bio-marker and are easily perceived, there are oth-

ers that are perceived differently by different people

such as being active or not (Shephard, 2003). This

makes the measurement of these risk factors unreli-

able (Prince et al., 2008). Such risk factors have not

been incorporated in this research as covering all pos-

sible perspectives of such risk factors (e.g. cycling as

a perspective to measure activity) is not feasible and

would cause misinterpretation of the results of this re-

search.

Based on the above for the proof of concept, only

9 main risk factors/bio-markers have been selected in

order to be incorporated into the survey, including:

age, weight, height, ethnicity, family history, genome

sequence, diet, smoking behavior and pre-/pregnancy

diabetes. This is also due to the fact that in this re-

search we are not focused on identifying all possible

risk factors or providing the best possible risk pre-

diction model, rather investigating the feasibility of

a personalized data collection experience to reduce

socio-economical barriers.

4 DESIGN OF THE SURVEY

In order to measure the willingness of people to share

information about risk factors, a survey has been con-

ducted. The guidelines suggested by the literature

(Fink, 2012; D

ornyei and Taguchi, 2010) have been

followed to ensure the validity and quality of the re-

sults and the conclusions. The survey is designed

in a manner which: i) provides conﬁdence that the

questions are answered with appropriate awareness,

ii) the engagement of the participants in the survey

is maximized, and iii) false responses, due to privacy

concerns or boredom, is minimized. To achieve the

above a set of rules have been designed, that the sur-

vey should comply to: 1) The time needed to an-

swer all questions should be approximately 5 minutes;

2) The questionnaire should not contain more than 30

questions; 3) The questionnaire should be available

in the language of the majority of the sample popu-

lation (Dutch) and in English; 4) Only complete sub-

missions are acceptable, as such, partially ﬁlled ques-

tionnaires are not acceptable; 5) The survey should

be conducted in a completely anonymous manner;

6) The minimum age of the participants should be 18;

The questionnaire of the survey has been designed in

a manner that for each bio-marker, the participants

will be asked to respond to four types of questions:

Situation Questions. As we are interested in con-

sidering difﬁdent phases of screening, four different

scenarios with an increasing level of conﬁdentially

have been designed: 1) Social chatting 2) (General)

research 3) Family related 4) Diagnosed

Each scenario has been ﬁtted with an appropriate

image to visually attach the participant to the scenario

and a lines of text representing the situation. For each

HEALTHINF 2016 - 9th International Conference on Health Informatics

300

bio-marker, participant are asked to select in what sit-

uation(s)/scenario(s) they would provide information

regarding that speciﬁc bio-marker. Participants can

select none, one or more options. If one or more

options are selected, follow-up questions will be pre-

sented to them regarding that speciﬁc bio-marker oth-

erwise they can proceed to the question(s) about an-

other bio-marker.

Means of Providing Information. If the partici-

pant selects one or more scenarios, he/she is asked

about the means of which he/she prefers providing

information about the speciﬁc bio-marker in the spe-

ciﬁc scenario(s). All bio-markers are provided with

the option to manually provide information as well

as through an electronic health record (EHR). We

have assumed that if EHR is chosen as a means of

providing bio-marker information, then this informa-

tion in deﬁnitely available in the EHR of the partici-

pant. Besides these two options, several bio-markers

have other option. Age can also be shared through

social networks. The same accounts for family his-

tory. Although it is very unlikely to ﬁnd informa-

tion about diabetes of other family members on their

social network, relationships between them can be

extracted from these networks. By questioning the

users about diabetes in their families, a more de-

tailed proﬁle can be created of the user: a parent is

of larger inﬂuence than a far relative. Finally, in-

formation about pregnancy diabetes can be shared

through social networks. If, for example, informa-

tion about pregnancy is found on the user’s network,

by analyzing their posts, a more speciﬁc question

about their pregnancy can be asked. As for weight,

another mean of providing information is presented:

a smart scaling device. These devices, such as the

Fitbit

https://www.ﬁtbit.com/aria Aria and Withings

smart body analyser

, are able to measure someones

weight and wirelessly transmit it to an external ap-

plication. Over all, depending on the scenario and

bio-marker a sub-set of the following means of pro-

viding information, that are applicable, is presented:

1) Manually providing the data through an online

form 2) Through Social Media 3) Through your Elec-

tronic Health Record (EHR) 4) Through smart de-

vices In this type of question multiple options could

be selected and a minimum of one option is required.

Frequency of Providing Information. For certain

questions another follow-up question is the frequency

https://www.ﬁtbit.com/aria

http://www.withings.com/eu/en/products/smart-body-

analyzer

of providing information. Certain risk factors, such

as weight, diet & smoking behavior can frequently

change and therefore have an important inﬂuence on

the current risk of developing diabetes. So the fre-

quency of sharing the information by the participants

is also an important factor that should be measured

in order to get a more detailed insight. For each mean

of providing information question, a separate question

for the frequency of sharing information might also be

asked. So for example if both manually ﬁlling in an

online form and through a smart scaling device are se-

lected, two extra questions appeared: one question for

each. This is done due to the fact that there could be

a large difference between ﬁlling in an online form

or transmit data automatically with a smart scaling

device with regards to effort as well as cost. Man-

ually providing the information requires more time

than stepping on a scale that automatically transmits

the data to the application, but the purchasing price

of such a scaling device is also a factor to consider.

Therefore, the participants are given the ﬂexibility to

accurately indicate this. Not every means of provid-

ing information question have a follow-up question

for frequency. For example, length (as an adult), eth-

nicity and genome sequence are factors that do not

change over time. Over all, depending on the means

of providing information, scenario and bio-marker a

sub-set of the following frequencies of proving in-

formation, that are applicable, is presented: 1) Only

once, 2) Monthly, 3) Weekly, 4) Daily, 5) Whenever

I feel like, and 6) Permanent Access. In this type of

question multiple options could be selected.

Demographics. Questions related to each of the

nine risk factors of diabetes used for this research and

the associated questions are located in separate pages

of the questionnaire. Besides these pages, two other

pages are placed before these questions. The ﬁrst

page contained demographic questions, such as what

is your sex, what is your age group and do you know

someone with diabetes in order to investigate possi-

ble differences between demographic groups. For ex-

ample the willingness to share information for dia-

betic research in people that know someone close to

them diagnosed with diabetes compared to others is

higher as they know how it is like to live with diabetes

(Beckjord et al., 2011). The ages groups are divided

into three groups: youth (18-24 years), adult (25-64

years) and senior (65+ years), based on the Canadian

age classiﬁcation standard. No option has been pro-

vided for ages less than 18 as the minimum age of 18

is set as a rule for the questionnaire, in order for the

participant him/her-self be able to decide on sharing

personal information.

BioMed Wizard - An Approach for Gathering Personal Risk Factor Data

301

5 EVALUATION OF THE SURVEY

In order to develop a solid and sound questionnaire

and in order to evaluate the compliance of the sur-

vey with the set rules discussed in section 4 an up-

date cycle is used to improve the questionnaire. After

each round of the cycle, the updated questionnaire’s

compliance to the rules are re-veriﬁed. In the ﬁrst

step the questionnaire is tested by three experts. The

purpose is to test if the setup is correct (e.g. to ver-

ify the questions, the question formulation, availabil-

ity of answers ) and to test the initial ﬁtness to the

set rules. Then a small pilot group of 10 participants

are invited to participate in the questionnaire.The pur-

poses of this phase is to verify the understandability of

the questionnaire for a broader audience and to mea-

sure the required time to answer the questionnaires.

The later is needed in order to verify if the question-

naire satisﬁed the set rules. The results indicate that

in average the time needed to answer the question-

naire is 6 minutes, 19 seconds and the median is 5

minutes, 21 seconds. After further investigation it be-

comes clear that two participants took almost twice

as long as the others due to external interruptions (e.g

phone call), making the average time longer than pro-

jected. After removing the outliers, the time needed

to complete the questionnaire is satisfying (5 minutes,

42 seconds). Finally, every member of the pilot group

is contacted and asked for feedback. Each question

is discussed as well as the setup of the questionnaire.

The feedback of the participants are then applied to

the questionnaire. For example, it has been noted that

the term genome sequence is not very clear, as well

as pre-diabetes and pregnancy diabetes. In order to

clarify the risk factors/bio-markers additional links to

external web-pages are added with more information

about genome sequences, pre-diabetes and pregnancy

diabetes. Also the images of the scenarios are up-

graded to better quality images. After these and other

changes are applied, the questionnaire is then tested

again.

6 CONDUCTING THE SURVEY

The survey is implemented using LimeSurvey

, an

open source software for implementing complex sur-

veys. LimeSurvey provides several features such as

exportable paper-based version of the online ques-

tionnaire, follow up questions, multilingual support,

and time-tracking. The PHP web application is in-

stalled on a private web server and made available. To

https://www.limesurvey.org

ensure the privacy of the participant, secure connec-

tions are established using the SSL encryption pro-

tocol to prevent unwanted interception between the

private server and the participants web browser. The

start page provides the participant with information

about the context of the questionnaire as well as an es-

timated completion time and information on how their

information will be processed and handled, anony-

mous. Participants can also choose their preferred

language: English or Dutch with the default set to

English. When the participants are ﬁnished, they are

presented with a thank you screen with a link to a

page with information on diabetes type 2. This page

is provided in order to create more awareness as well

as inform the participant about diabetes mellitus type

2 and its risk factors. The survey is also advertised

by means of email, social media and physical posters.

The survey is conducted for two consecutive weeks

and then closed. In total 116 responders have partic-

ipated in the survey, 11 of which are incomplete. As

such a total of 105 completed submissions are gath-

ered resulting in a completion rate of 90.5%. The av-

erage answering time is 6 minutes, 46 seconds. This

average response time is longer than projected (5 min-

utes), but further investigation shows some high re-

sponse times (>25 minutes) that could indicate exter-

nal interruptions in the survey process for some par-

ticipants. The median however is 5 minutes and 22

seconds, which is satisfying regarding the rules set.

Four participants out of the 105 complete results have

only selected one scenario for all risk factors , while

the intention is to select all applicable options. Since

these results could act as outliers, they have been ex-

cluded from the analyse, as such reducing the num-

ber of valid responses to 101. Out of the valid re-

sponses, 39 participants are youth (18-24 years), 58

adults (25-64 years) and only 4 seniors (65+ years).

Although the senior population is signiﬁcantly lower

in the sample group, this will not inﬂuence a smart

wizard software for screening diabetes as the seniors

are not in the target group for such a software. On

the other hand 46 of the participants are male and

55 are female, suggesting an acceptable split between

both genders. Although the participant are from 17

different nationalities, 74% of them have a dutch na-

tionality. This is due to the fact the the research is

conducted in the Netherlands. Finally our popula-

tion is quite well distributed between different dis-

ciplines (23 from Computer & Information sciences,

24 from Biological & Health Sciences, 22 from Engi-

neering, Physical & Social sciences and Psychology,

and 32 from other educational backgrounds). While

the level of education has shown not to inﬂuence the

understanding of privacy, the ﬁeld of study has been

HEALTHINF 2016 - 9th International Conference on Health Informatics

302

Table 1: Overview of selection rates for frequency of pro-

viding information.

Frequency Selection Rate

Whenever I feel like 38.5%

Only once 19.6%

Monthly 17.6%

Weekly 10.6%

Permanent Access 3.2%

Daily 2.9%

suggested to have such an inﬂuence (Ferreira et al.,

2011). Other than this, the understanding of the im-

portance of sharing this data for medical research

might inﬂuence participants choice on sharing their

information. The distributed between different disci-

plines in our study prevents biased results due to these

factors.

7 STATISTICAL ANALYSIS AND

USER SPECIFIC DATA

COLLECTION FLOW

The statistical analysis shows that, as expected, an

increasing amount of positive responses can be seen

when the scenarios become more conﬁdential. The

diagnosed scenario is the most selected, with a 94.4%

selection rate, followed by family related situation,

(general) research and social chat scenarios with se-

lection rates of 90.7%, 69.9%, and 53%, respectively.

When looking at the means of providing information,

manual submission of information is by far the most

selected option with a selection rate of 28.9% fol-

lowed by EHR, Smart Devices, Social Network with

selection rates of 10.9%, 7.1%, and 1.9% , respec-

tively. However, when only looking at means of pro-

viding the genome sequence, we can see that the

selection rate for providing through EHR is 30.7%

while manual submission is 29.8%. For all other

risk factors/bio-markers, manual submission has been

chosen substantially more. Finally, the frequency of

providing information when applicable (weight, diet

and smoking) are shown in Table 1. As visible, the

option Whenever I feel like is by far the most selected

frequency. Although for the risk factors weight and

diet the second most selected frequency is Monthly,

and for smoking, the second most selected frequency

is Only once.

Using the above statistical analysis and by ﬁnding

signiﬁcant differences between different demographic

groups, a user speciﬁc data collection ﬂow model

is designed using BPMN 2.0 for each scenario, that

represents a smart wizard software for each screen-

ing phase. The signiﬁcant differences are found used

Table 2: Overview of selection rates for risk factors for the

social chat scenario.

Bio-marker/Risk factor Selection Rate

Age 80%

Smoking 79%

Ethnicity 75%

Height 74%

Diet 58%

Weight 49%

Family history 29%

Pre-/Pregnancy diabetes 24%

Genome sequence 14%

the binary logistic regression method. And if a sig-

niﬁcant difference is found, the Crosstabs method is

used to discover the cause of the difference. A sig-

niﬁcance level of p <0,05 is used to indicate signiﬁ-

cant differences for all methods. In order to demon-

strate how these signiﬁcant values are used to imple-

ment the model, one of the developed models (i.e so-

cial chat representing the initial screening phase) will

be discussed in detail. Each model consists of three

lanes, two of which are the most important: the soft-

ware lane and the user lane. The software lane repre-

sents the action preformed by the smart wizard soft-

ware, such as asking a question. The user lane on the

other hand represents the action by a participant in the

screening process, such as providing an answer. Fi-

nally, the third lane represent external services that are

invoked to retrieve information. The order in which

the questions are asked in the model is based on the

willingness of people to share their data related to a

bio-marker. Table 2 demonstrates the distribution in

which the respondents are willing to share informa-

tion in social chat context. As visible most of the par-

ticipants are willing to provide their age, while the

least are willing to provide their genome sequence.

This order is used to prevent offending users of the

software, by asking for data that are not likely to be

share by them, and to keep the user engagement with

the wizard to its maximum.

Now that the order in which questions are asked

by the software has been set, the focus lies on what

can and cannot be asked to the users regarding a

higher chance of offending them. To achieve this

as stated in the method, a binary logistic regression

has been conducted. If a signiﬁcant result was found,

a crosstabs method was used to get insight on what

caused the signiﬁcant difference. Our ﬁndings sug-

gest that women are signiﬁcantly less likely to pro-

vide information about their age, height and weight

compared to men. On the other hand our ﬁnding

also suggest that non-Dutch (minorities) are signif-

icantly less likely to provide information about their

age and height. These signiﬁcant differences are mod-

elled into the smart wizard software model. There-

BioMed Wizard - An Approach for Gathering Personal Risk Factor Data

303

fore, the software asks the user for his/her sex as well

as nationality ﬁrst in order to decide what questions

should be presented further. The complete model for

the social chat scenario is illustrated in the BPMN no-

tation

. The ﬁrst two questions ask about the sex and

nationality. After which the age is asked about. How-

ever, due to the identiﬁed signiﬁcances the model is

adjusted. The same approach is followed for height,

weight and pre-/pregnancy diabetes.

Regarding the means of providing information, it

is important to notice that all information have to be

provided manually with the exception of genome se-

quence. Since there is almost no signiﬁcance differ-

ence between manual submission and EHR the deci-

sion has been made to provide the user with both op-

tions. A XOR statement enables the user to select one

of the options. If the user selects the option to share it

through his/her electronic health record, a request to

an external service is made, i.e. the service that holds

the data of the electronic health record.

Finally, the frequency by which data can be gath-

ered is also implemented in the model. As discussed

before and visible in Table 1, the option to share in-

formation whenever the user feels like is the most se-

lected option. For both diet and weight the second

most selected option is monthly and therefore a re-

minder is modelled for these factors. For these cases

the user is send a monthly reminder via a notiﬁcation

(e.g phone notiﬁcation or email), stating that the user

can update its information if he or she feels like doing

so. It is important to state that at each point the user

has the ability to skip every question.

8 DISCUSSION

One should have in mind that people are not always

completely honest when asked about privacy related

matters and in real-life might act differently. This

might jeopardize the results of this research, as such

on one hand, we have tried our best not to reveal

the intention of the questionnaire to reduce any pre-

judgement and perception that might cause this phe-

nomenon and on the other hand, we observe the par-

ticipants tendency to being truthful, by their response

to situation questions addressed in section 7. An-

other possible challenge with our approach rises when

the user is not asked to provide bio-marker informa-

tion for some of risk factors. For example BMI for

obesity in women and non-Dutch users in the model

of the social chat cannot be calculated. However,

BMI is a very important factor to know, since over-

Available online http://goo.gl/AUDJ1a

weight is the primary risk factor for diabetes type 2

for both adults (Hartemink et al., 2006) and children

(Wei et al., 2003). This becomes further clear when

looking into a study (Baan et al., 1999) that has in-

vestigated what data is essentially required to identify

patients at a greater risk of developing diabetes type 2.

This study created three risk prediction models. The

ﬁrst model is based on data gathered in ﬁles of a gen-

eral practitioner, such as age, sex and absence of obe-

sity. The second model used additional information

by asking questions about family history and smok-

ing. The third model incorporated medical data, such

as diastolic blood pressure and systolic blood pres-

sure. The study has concluded that while the ﬁrst

and second models generate similar results, adding

the risk factors in the third model is not useful, and

the extra medical data used in the third model has no

additive predictive value. At the end researchers have

recommend the ﬁrst model, since it uses less infor-

mation that is usually available in the patient speciﬁc

notes of a general practitioner. Even when looking

at the minimal model still obesity, as a risk factor,

plays a critical role and as such the bio-marker BMI

is critical in properly assessing the risk of developing

diabetes type 2. A possible solution can be to esti-

mate critical bio-markers such as BMI for patients us-

ing knowledge-based and learning-based approaches

(Khan et al., 2012) to ﬁll in gaps where information is

missing.

9 CONCLUSION AND FUTURE

WORK

The aim of this research is to achieve a model for user

speciﬁc data collection, using a smart wizard software

that would ultimately provide users with an indica-

tion of their current risk level, when also considering

socio-economical aspects. As a ﬁrst step in this pro-

cess, we have created four models (for different types

of possible software scenarios) that can be used to

create a user speciﬁc software wizard to gather data

related to diabetes type 2. A survey is conducted to

measure the willingness of users to share data about

different risk factor. Four scenarios are used to distin-

guish possible differences in the willingness to share

information in different phases of screening for dia-

betes type 2. The users are also asked about the pre-

ferred means of sharing the information and, if appli-

cable, the frequency in which they are willing to share

this information. Future work can use the method

proposed in this research to develop similar data col-

lection models for other diseases. Another further

research can be conducted with larger and more di-

HEALTHINF 2016 - 9th International Conference on Health Informatics

304

verse populations to investigate possible differences

between regions and countries. Also, once the wizard

is developed, evaluation of the approach can be pre-

formed using A/B testing and measuring parameters

such as the dropout rate.

REFERENCES

Abbasi, A., Peelen, L. M., Corpeleijn, E., van der Schouw,

Y. T., Stolk, R. P., Spijkerman, A. M., Moons, K. G.,

Navis, G., Bakker, S. J., Beulens, J. W., et al. (2012).

Prediction models for risk of developing type 2 di-

abetes: systematic literature search and independent

external validation study. Bmj, 345:e5900.

Adler-Milstein, J., DesRoches, C. M., Furukawa, M. F.,

Worzala, C., Charles, D., Kralovec, P., Stalley, S., and

Jha, A. K. (2014). More than half of us hospitals have

at least a basic ehr, but stage 2 criteria remain chal-

lenging for most. Health Affairs, pages 10–1377.

Baan, C. A., Ruige, J. B., Stolk, R. P., Witteman, J., Dekker,

J. M., Heine, R. J., and Feskens, E. (1999). Per-

formance of a predictive model to identify undiag-

nosed diabetes in a health care setting. Diabetes care,

22(2):213–219.

Beckjord, E. B., Rechis, R., Nutt, S., Shulman, L., and

Hesse, B. W. (2011). What do people affected by

cancer think about electronic health information ex-

change? results from the 2010 livestrong electronic

health information exchange survey and the 2008

health information national trends survey. Journal of

Oncology Practice, 7(4):237–241.

ornyei, Z. and Taguchi, T. (2010). Questionnaires in sec-

ond language research: Construction, administration,

and processing. Routledge.

Ferreira, A. M., Tziortzios, C., and Shafahi, M. (2011).

Passwords awareness in the academic world. Tech-

nical report, University of Amsterdam.

Fink, A. (2012). How to conduct surveys: A step-by-step

guide. Sage Publications.

Harris, M. I. (2001). Racial and ethnic differences in health

care access and health outcomes for adults with type

2 diabetes. Diabetes care, 24(3):454–459.

Hartemink, N., Boshuizen, H. C., Nagelkerke, N. J., Jacobs,

M. A., and van Houwelingen, H. C. (2006). Com-

bining risk estimates from observational studies with

different exposure cutpoints: a meta-analysis on body

mass index and diabetes type 2. American journal of

epidemiology, 163(11):1042–1052.

Hivert, M.-F., Grant, R. W., Shrader, P., and Meigs, J. B.

(2009). Identifying primary care patients at risk for

future diabetes and cardiovascular disease using elec-

tronic health records. BMC health services research,

9(1):170.

Jha, A. K., DesRoches, C. M., Campbell, E. G., Donelan,

K., Rao, S. R., Ferris, T. G., Shields, A., Rosenbaum,

S., and Blumenthal, D. (2009). Use of electronic

health records in us hospitals. New England Journal

of Medicine, 360(16):1628–1638.

Khan, A., Doucette, J. A., Cohen, R., and Lizotte, D.

(2012). A hybrid design for medical decision support

using data mining to impute missing data.

Knowler, W. C., Barrett-Connor, E., Fowler, S. E., Ham-

man, R. F., Lachin, J. M., Walker, E. A., and Nathan,

D. M. (2002). Reduction in the incidence of type 2

diabetes with lifestyle intervention or metformin. The

New England journal of medicine, 346(6):393.

Lindstr

om, J. and Tuomilehto, J. (2003). The diabetes risk

score a practical tool to predict type 2 diabetes risk.

Diabetes care, 26(3):725–731.

Mayo Clinic (2015). Risk factors type 2 diabetes

risk factors. http://www.mayoclinic.org/diseases-

conditions/type-2-diabetes/basics/risk-factors/con-

20031902. Accessed: 2015, May 21.

Montonen, J., Knekt, P., J

arvinen, R., and Reunanen, A.

(2004). Dietary antioxidant intake and risk of type 2

diabetes. Diabetes Care, 27(2):362–366.

Pickard, K. T. and Swan, M. (2014). Big desire to share big

health data: A shift in consumer attitudes toward per-

sonal health information. In 2014 AAAI Spring Sym-

posium Series.

Prince, S. A., Adamo, K. B., Hamel, M. E., Hardt, J., Gor-

ber, S. C., and Tremblay, M. (2008). A comparison of

direct versus self-report measures for assessing phys-

ical activity in adults: a systematic review. Interna-

tional Journal of Behavioral Nutrition and Physical

Activity, 5(1):56.

Rijksinstituut voor Volksgezondheid en Milieu (2013).

Welke factoren beinvloeden de kans op diabetes mel-

litus? welke factoren benvloeden de kans op diabetes

mellitus? - nationaal kompas volksgezondheid. http://

www.nationaalkompas.nl/gezondheid-en-ziekte/

ziekten-en-aandoeningen/endocriene-voedings-en-

stofwisselingsziekten-en-immuniteitsstoornissen/

diabetes-mellitus/welke-factoren-beinvloeden-de-

kans-op-diabetes-mellitus. Accessed: 2015, May 21.

Shephard, R. J. (2003). Limits to the measurement of habit-

ual physical activity by questionnaires. British journal

of sports medicine, 37(3):197–206.

Teixeira, P. A., Gordon, P., Camhi, E., and Bakken, S.

(2011). Hiv patients willingness to share personal

health information electronically. Patient education

and counseling, 84(2):e9–e12.

Tirosh, A., Shai, I., Afek, A., Dubnov-Raz, G., Ayalon, N.,

Gordon, B., Derazne, E., Tzur, D., Shamis, A., Vinker,

S., et al. (2011). Adolescent bmi trajectory and risk of

diabetes versus coronary disease. New England Jour-

nal of Medicine, 364(14):1315–1325.

Wei, J.-N., Sung, F.-C., Lin, C.-C., Lin, R.-S., Chiang,

C.-C., and Chuang, L.-M. (2003). National surveil-

lance for type 2 diabetes mellitus in taiwanese chil-

dren. Jama, 290(10):1345–1350.

World Health Organization (2014). Global health estimates:

Deaths by cause, age, sex and country, 2000-2012.

Geneva, WHO.

Xierali, I. M., Hsiao, C.-J., Puffer, J. C., Green, L. A.,

Rinaldo, J. C., Bazemore, A. W., Burke, M. T., and

Phillips, R. L. (2013). The rise of electronic health

record adoption among family physicians. The Annals

of Family Medicine, 11(1):14–19.

Zhang, J. and Zhao, Y. (2013). A user term visualization

analysis based on a social question and answer log.

Information Processing & Management, 49(5):1019–

1048.

BioMed Wizard - An Approach for Gathering Personal Risk Factor Data

305