Acceptance of a Digital Paper-based Diabetes Diary

Understanding Precursors of Acceptance of Digitally Assisted Diabetes Care

Andr

e Calero Valdez and Martina Zieﬂe

Human-Computer Interaction Center, RWTH Aachen University, Campus Boulevard 57, 52074 Aachen, Germany

Keywords:

Usability Testing, Novel Interaction Paradigms, Pen Computing, User Centered Design, Empirical Studies in

Interaction Design.

Abstract:

Diabetes prevalence has steadily been increasing over the last decades and is expected to continue in this

trajectory. Diary keeping is considered the central part of any successful therapy. Since paper-based diary

keeping is often poor and electronic diaries often challenging for elderly users, we developed Diabetto. Using

focus groups and interviews we analyzed the requirements for Diabetto. It uses a LiveScribe Echo pen as a

replacement for a regular pen in traditional paper-based diary keeping, while at the same time adds interactive

therapy support and access to a nutrition database through pen-input and text-to-speech. By enhancing an al-

ready familiar process we managed to achieve a high acceptance of the device. In an experimental evaluation

using an extended TAM model we analyzed the inﬂuence of user-diversity factors on the acceptance. Interest-

ingly typical predictors of acceptance, such as efﬁciency during use, did not inﬂuence ease of use, indicating

that the digital pen might not be perceived as computerized technology.

1 INTRODUCTION

With the rise of the “Internet of Things” (Atzori et

al., 2010) and omni-present computing (Davis, 2002)

almost every device will allow us to connect to the In-

ternet and exchange data anytime and anywhere. The

question is, whether this will allow overcoming some

of the barriers people perceive in regard to medical

technology, with their best interest in mind.

Although some may claim that usefulness

(Scheermesser et al., 2008) is the most determin-

ing factor for pervasive medical technology, technol-

ogy acceptance is neither static nor predictable using

standardized technology acceptance modeling (Zieﬂe

and Wilkowska, 2010). Illnesses inﬂuence the user

over time, as does aging. Salience of technology

may lead to increased perception of barriers in users,

who already feel disconnected from modern technol-

ogy, and reject the conscious use of modern technol-

ogy. Nonetheless almost all people use ICT everyday,

when they enter elevators, drive cars, or use their tele-

phone. There’s something about visibility of technol-

ogy that inﬂuences technology acceptance. Accep-

tance of ubiquitous computing and medical assistance

increases when technology is integrated into the fa-

miliar environment of the patient (Klack et al., 2011).

This article describes an approach to digital dia-

betes management system that tries to hide technol-

ogy as a computer from the user and tries to ﬁt com-

puting into their already existing behavioral patterns.

It puts the assistance into the pen that they use to ﬁll

out their diabetes diary.

1.1 Obstacles in Diabetes Therapy

Diabetes, if badly treated, causes a plethora of dam-

ages to the body (e.g. neural, cardial, etc.) resulting

in extremely high treatment costs in comparison to a

well-adjusted therapy (Hien and B

ohm, 1997).

The biggest problems in diabetes therapy con-

stitute therapy adherence and customization of the

therapy especially in regard to designing an assistant

(Hien and B

ohm, 1997; Chen, 2010). In order to ad-

dress the individual aspects of each patients illness,

patients are required to keep a diabetes diary. In this

diary vital parameters such as blood glucose levels are

recorded along with insulin dosages, activity data, and

food intake. The doctor must now understand the id-

iosyncratic reactions of the patients body to medica-

tion and lifestyle and adjust therapy accordingly.

Quality of therapy is highly dependent on the

quality of the diary keeping (Hader et al., 2004).

Users often refrain from keeping their diary in a pre-

cise fashion, but often ﬁll their diary at the end of the

day from memory (Stone et al., 2002). It is neces-

sary for a diabetic to calculate the insulin doses using

Calero Valdez, A. and Zieﬂe, M.

Acceptance of a Digital Paper-based Diabetes Diary.

DOI: 10.5220/0006734800740086

In Proceedings of the 4th International Conference on Information and Communication Technologies for Ageing Well and e-Health (ICT4AWE 2018), pages 74-86

ISBN: 978-989-758-299-8

cross-multiplication, as the dosage is both dependent

on current glucose levels and a correctional factor that

is individually established for different times of the

day. Also, a persisting diabetes can also lead to cog-

nitive deﬁcits (Yeung et al., 2009; Brands et al., 2005;

Brands et al., 2007), which in turn hampers therapy.

2 RELATED WORK

In the following we review the literature on digital

diabetes diaries and management applications. From

the literature we see different forms of application

and input methods. Generally electronic and paper

based diaries have been investigated with differing re-

sults. We then introduce the the LiveScribe Echo Pen,

which was used in an approach to combine the best of

both worlds—a digital pen paper-based diabetes di-

ary, that provides a multitude of the features that typ-

ical electronic diaries provide.

2.1 Diabetes Management

As early as 2003 Kerkenbush conducted a study on

the efﬁcacy of PDA-based diabetes diaries against

paper-based diabetes diaries. Kerkenbush and La-

some (2003) identiﬁed the importance of PDA based

solutions for diabetes care. They identiﬁed the impor-

tance of systematic and regular tracking in a diabetes

diary for patients health as well as the need to teach

potential users how to use a PDA based diary system.

The use of PDA based diabetes diaries is in par-

ticularly interesting as Stone et al. (2002) found a

high unwillingness to paper diaries in many diabetes

patients. This is not necessarily caused by a lack

of therapy adherence but might have been caused by

simple forgetting. An electronic device could have a

reminder function, leading to a possible increase in

diary keeping.

Broderick and Stone (2006) argue though, that

comparability in regard to the device type is not yet

possible. Reasons for compliance rates are not as-

sessable through the conducted studies and could be

conﬂicted by errors of self-reporting, motivation, and

lack of actual medical compliance analysis.

Nonetheless, Forjuoh et al. (2008) investigated the

medical impact of using PDA based diaries between

high and low usage of the PDA software “Diabetes Pi-

lot”. Over a course of six months high usage resulted

in an signiﬁcantly better increase in vital parameters

than in low usage users. Still both groups showed im-

proved of their vital parameters.

Duke et al. (2008) developed the Intelligent Di-

abetes Assistant (IDA). IDA uses machine learning

algorithms to improve therapy by measuring lifestyle,

nutrition and glucose levels. The measurements are

instantly shared with a physician to allow shorter

feedbacks loops on the diary evaluation. A similar

solution has been developed by (Tani et al., 2010),

which was also evaluated positively.

Burke et al. (2005) investigated the differences

in monitoring food intake on a PDA after training

hemodialysis and weight-loss patients in using a digi-

tal food diary. They found a higher loss of weight and

higher self-monitoring adherence in users of a PDA

based diary than users of paper-based diaries (Burke

et al., 2011). If daily feedback was given through the

PDA improvements were even larger.

When looking at PDAs (Arning and Zieﬂe ,2007)

and mobile phones (Zieﬂe, 2002) acceptance of these

devices might depend on prior experience and in par-

ticular be important when the users are part of the tar-

get demographic—-elderly users.

In a time when the technical challenges are no

longer the largest barriers, it becomes important the

regard the users emotive, hedonic (Alag

oz et al.,

2010) and cultural needs(Alag

oz et al., 2011), espe-

cially when developing medical technology. Users

might have a preference of using paper over PDAs

or mobile phones for sheer haptic reasons (Gregory et

al., 2008). When speciﬁcally looking at diabetes de-

vices age shows a particular negative inﬂuence on us-

age performance (Calero Valdez et al., 2009). Addi-

tionally the screen size of the device increases usabil-

ity in particular for older users (Calero Valdez et al.,

2010b; Calero Valdez et al., 2011). Since some users

are not familiar with the hierarchical menu structures

of mobile phones or PDAs, other forms of navigation

through a device should be considered (Calero Valdez

et al., 2010a). This is particularly the case in mobile

devices, where screen real-estate is scarce and can not

be using excessively for user guidance.

2.2 Technology Acceptance

Since actual usage of devices depends largely on user

acceptance it is necessary to incorporate acceptance

modeling in the development process of such a de-

vice. For our case we have chosen to rely on the Tech-

nology Acceptance Model (TAM) by (Davis, 1989).

TAM based the intention to use a system on two ex-

plaining variables. The perceived ease of use (PEU)

and perceived usefulness (PU) determine largely the

behavioral intention.

Other models such as UTAUT (Venkatesh et

al., 2003), TAM2 (Venkatesh and Davis, 2000) and

TAM3 (Venkatesh and Bala, 2008) were also con-

sidered but are not used due to the amount of fac-

Acceptance of a Digital Paper-based Diabetes Diary

tors that they introduce to the method. The TAM

model has successfully been used to model the accep-

tance of medical technology in a health care setting

(Venkatesh et al., 2011).

3 REQUIREMENT ANALYSIS

In order to develop a Livescribe-based diabetes-

management-assistant that is both usable and ac-

cepted the user must be taken into account. To get

insights into the life of a diabetic a focus group was

conducted, the results of which are presented next.

3.1 Focus Group Results

The focus group was conducted in order to get ﬁrst

insights into using a mobile device for diabetics. In

order to guarantee focus group success a list of core

questions was initially established, which were then

given to the participants. The questions were picked

with ﬁve topics in mind.

The ﬁrst topic is “diabetes in general” and ques-

tions were selected with special focus on therapy and

diabetes management. The second topic is “mobile

devices”. The questions regarding this topic were se-

lected to understand the impact of the level of technol-

ogy expertise in using medical technology (i.e. mo-

bile devices for diabetics). Special focus was put on

the features of the used devices and their acceptance.

The third topic is “diabetes diaries”. Questions for

this topic were chosen to improve understanding of

how potential users keep their diary. The last two top-

ics are usage motives and barriers. Questions for these

topics were picked to assess future hurdles to prevent

when implementing a device.

Five participants contributed to this setting and

shared the following demographic information:

1. Type 1 diabetic, male 26 years, student of Com-

puter Science, athlete

2. Type 1 diabetic, female, 14 years, pupil (accom-

panied by her mother)

3. Type 1 diabetic, female, 42 years, civil servant,

Insulin-pump

4. Type 2 diabetic, male, 46 years, university degree

in Mathematics

5. Type 2 diabetic, male, 63 years, retiree,

3.1.1 Results

In the focus group ﬁve participants were asked to

elaborate on the experience with diabetes, mobile

devices, diabetes management and possible motives

and barriers for using a diabetes management as-

sistant. The participants were selected from a pro-

active group of diabetics (i.e. visitors of a diabetes

congress), they were self-selected, and had no known

secondary disorders. They were mostly well-educated

and all on insulin therapy.

Important ﬁndings that would need consideration

from this focus group are (in order of occurrence):

1. Diabetes in general

(a) Diabetes strongly inﬂuences the daily routine.

(b) Physical activity needs special attention.

dent.

(d) Domain knowledge of diabetes is high, but

varies with severity of the disease.

(e) Informedness varies with understanding of dia-

betes.

(f) Different parameters are known differently well

depending on (perceived) domain knowledge.

(g) Important parameters are blood glucose level,

HbA

c, body-fat percentage & blood pressure.

2. Mobile devices

(a) Insulin pumps are perceived as very helpful, but

sometimes as foreign.

(b) General satisfaction with devices is high.

3. Diabetes diaries

(a) Paper-based diary keeping is usual but per-

ceived as cumbersome.

(b) Shortcuts are used to simplify logging.

4. Motives and barriers

(a) Therapy is cumbersome and device implemen-

tation a necessity.

(b) Operating expense (i.e. time) must be minimal.

digital ones.

(d) Data safety is important.

(e) Low level of perceived usefulness might be a

barrier.

(f) Beneﬁts for health are imaginable if the device

integrates well into the daily routine.

(g) Insulin-dosing calculation would be a helpful

feature.

(h) Deal-breakers are bad usability and public visi-

bility.

(i) Mobile phone integration is only attractive for

some users.

ICT4AWE 2018 - 4th International Conference on Information and Communication Technologies for Ageing Well and e-Health

4 AN INTERACTIVE

ANOTO-PAPER BASED

DIABETES DIARY

From the requirements gathered, we conceptualized a

diabetes diary using the Livescribe Echo digital pen.

The idea behind the software named Diabetto was

to let users keep their paper-based diaries but offer

interactive support using the Livescribe Echo. This

pen uses an infrared-camera to detect what is being

written with the pen and has a microphone and loud-

speaker included. Both can be used from software on

the pen. Usage feels like using a normal pen on regu-

lar paper, but with digital support.

All versions of the Livescribe Echo are based on

an ARM9 processor. The pen is larger (158x19mm)

and heavier (36g) than an average pen. Data transfer

happens using a micro-USB connector. The pen also

has a 96x18 pixel OLED-Display, an integrated mi-

crophone and and integrated loudspeaker. It also has

a headphone jack, which further allows recording of

stereo-data.

The software Diabetto does not focus on the eval-

uation of generic diabetes diary keeping, but on as-

sisting the user by verbal feedback on diary entries.

Furthermore it should allow self-deﬁned shortcuts the

simplify diary keeping in order to reduce barriers.

The following section introduces how the the Diabetto

software was developed, which is then evaluated in

the next section.

4.1 Diabetto – A Digital Pen based

Diabetes Diary

The ﬁrst essential part of Diabetto is the paper-based

diary. The diary consists of three parts. The ﬁrst part

is the actual diary pages that the users uses to log his

diary entries. The second part is a set of pages that

contain a database of food items that the user can

modify as a shortcut to regularly consumed meals.

The last part is a set of general abbreviations that al-

lows the user to record and input with a self-deﬁned

abbreviation.

4.1.1 The Diary Pages

The ﬁrst part is the actual diary (see Figure 1). The di-

ary itself consists of lined paged that are empty. Each

page refers to a certain day. Users can write on these

lines indicating what they have done on a particular

day.

When a line starts with a time-information (e.g.

9:00) the diary assumes this to be the time of the en-

try (used when logging forgotten information). When

no time is given the current system time is used as

the time of entry. Time-information is recognized by

a regular expression detecting any four or ﬁve set of

numerical characters with a colon in between.

2IU basal.

9:30 one roll 2bu.

11:00 one hour walking.

pear?

3:00pm forest.

120 bs.

Figure 1: A sample diary page. The diary page contains a

list of entries, one per line. Entries that are striked-through

are deleted entries. Entries that have no time are recorded

using the system time.

Blood glucose readings are detected, when a line

ends with the characters “bs”. The pen then tries to

recognize the number before these characters and in-

terprets them and gives advice accordingly. When the

numbers are on the low end of health blood glucose

levels an auditory warning is issued: “Warning, your

glucose levels are low. Please verify your measure-

ment or react to this state.” When levels were too high

an analogous warning was issued. This message also

included a suggestion of how much insulin to admin-

ister in order to reduce the amount of blood glucose

level to a healthy level. In the case of normal mea-

surements a conﬁrmatory message containing the nu-

merical value written in the diary was played.

Whenever the two characters “BU” are found after

numerical characters a food consumption entry is as-

sumed. The amount of BU-units is played as auditory

feedback and then recorded.

Insulin dosages were recognized when the charac-

ters “IU” were found after numerical characters. Au-

ditory feedback on the amount of administered inter-

national units of insulin was played.

Acceptance of a Digital Paper-based Diabetes Diary

When a word is written that ends with a question

mark, the pen looks into his nutrition facts or abbrevi-

ation database. If an entry in the nutrition facts exists,

the amount of BU-units is reported to the user using

audio feedback. If the same word is written with a dot

at the end, the BU-units are recorded as a diary entry.

Found abbreviations are “unfolded” and the reinter-

preted.

The rest of the line, which is not interpreted, is

used as a commentary to the respective diary entry.

If no interpretable data is contained in a single

row, the entry is recorded as an activity (e.g. “one

hour walking”).

Entries can be deleted either by striking them

through or by clicking the X-button on the bottom of

the page immediately after input.

Furthermore every page has a help button (show-

ing a “?”) that starts a short audio-tutorial to the pen,

as well as three buttons to adjust the volume of the

pen (i.e. volume up/down/mute).

The Nutrition Facts Pages

The second part of the diary contains a table with

columns (see Figure 2). The left column is used to

enter the name of a meal or a food item that should

be stored in the diary. The right column (in the same

row) takes the amount of BU-units that are associated

with this entry. These entries can be used in the di-

ary as a shortcut. Furthermore deletion of entries is

possible by striking them through.

The Abbreviations Pages

The third and last part of the diary is very similar to

the second part. It also has two columns. The left col-

umn takes the abbreviation, while the right column

takes the expanded full text. Whenever a user writes a

word from the left column of the abbreviations pages

the text of right column of the same row is virtually

inserted and interpreted. This allows for complex ab-

breviations like “9:00 Breakfast, 2 cheese rolls, 6BU.”

that may occur regularly in a users life.

4.2 The Applications

Two applications are part of Diabetto. The ﬁrst appli-

cation is the penlet that runs within the pen to provide

the functionality of the diary. Since any of the fea-

tures that were described in the previous section are

actually a functionality of the pen and not the paper,

no detailed further information about the penlet ap-

plication is necessary. This already indicates that the

pen itself is not considered as the interface, but the pa-

per is, although technically the opposite applies. The

Banana

Foodstuff

Dietary Chart

Figure 2: Example page of the Diabetto diary showing a

nutrition facts sheet. A user may write food items he con-

sumes often in the left column and its BU-units in the right

column.

application automatically starts, when the pen is used

on Diabetto-diary paper. The second application is

desktop-application to synchronize data. This appli-

cation is not part of our evaluation here.

5 QUESTIONS ADDRESSED

Classical technology acceptance using a TAM model

requires to assess the usefulness and ease of use of

a technology to estimate the behavioral intention to

use the technology. In our case we wanted further

information to be available for analysis.

We were interested in ﬁnding out whether the

auditory-feedback and the writing-based input were

usable and helpful to the patients. We assumed that

the integration into the already known process of

keeping a paper-based diary would be easier than

teaching a PDA or mobile phone application. Addi-

tionally we were interested in seeing how age, exper-

tise, domain knowledge, and computer self-efﬁcacy

inﬂuenced the evaluation of the device, as previous

studies had shown a strong inﬂuence of these fac-

tors (Calero Valdez et al., 2009; Calero Valdez et al.,

2011) in PDA based devices. In particular the ef-

fectiveness of use has shown to be good predictor of

perceived ease of use in previous exploratory studies

ICT4AWE 2018 - 4th International Conference on Information and Communication Technologies for Ageing Well and e-Health

(Calero Valdez et al., 2009). Under the assumption

that a pen might be less likely seen as a technological

device, inﬂuences off effectiveness during use could

differ from previous studies.

6 EXPERIMENTAL EVALUATION

In order to understand, whether possible users would

accept a digital paper-based diary and how well they

could use such a device, a user study was conducted.

Focus of the study was to let users use the pen in a

scenario-based task, assess their performance and fur-

thermore assess their acceptance of the device.

6.1 Method

The experiments were conducted at the RWTH

Aachen in a laboratory environment. Additional ex-

periments were conducted at the ofﬁce of Dr. L

atzsch

in Aachen in order to remove some effect of the self-

selection that occurs when participants have to travel

to the experimental destination. Participants were di-

rectly contacted in the physician’s ofﬁce or for the

control group taken from the immediate social net-

work of the author.

6.2 Experimental Procedure

The experiment was conducted in several steps. First

participants were informed about the experiment and

that their data was being recorded. In order to mea-

sure performance, videos of the experiments were

also recorded. The participant was informed that the

interaction would be recorded on video, not exposing

the face (only the diary and the hand were in frame).

Then participants had to answer a questionnaire.

In order to get accustomed to the digital pen, ﬁrst

four tasks were instructed, in which the participant

would also receive assistance from the experimenter.

These tasks could also be repeated as often as de-

sired. Once a participant stated that he understood

the concept behind the digital pen, participants were

instructed to perform three performance tasks. Af-

ter ﬁnishing the task sets, a post-experimental ques-

tionnaire had to be answered. After this last ques-

tionnaire participants were instructed about how they

could contact the experimenter in case they wanted

any of the data deleted at any time afterwards.

6.3 The Questionnaire

The questionnaire was separated in two parts. Partici-

pants were asked demographic data such as age, gen-

der and work. Furthermore handedness was assessed

because handedness could inﬂuence accuracy of the

text recognition.

Additionally health status was assessed. This

means type of diabetes, diabetes duration and type of

therapy were assessed. Technical expertise was as-

sessed with both household and expertise in comput-

erized technology (ECT) as well as expertise in health

technology (EHT). For descriptive purposes coping

scales and domain knowledge scales were assessed

(i.e. DP, DCI, TDC, IFG, IFO).

Lastly the computer self-efﬁcacy (CSE) (Beier,

2004) and behavioral intention to use the diary were

assessed. Computer self-efﬁcacy was measured using

eight items, while behavioral intention was measured

using three items.

6.4 The Task Set

In order to assure equal conditions between partici-

pants a ﬁxed task set was designed, which included

all features of the Diabetto diary. Instructions were

available in paper-based format (i.e. printed between

the pre- and post-questionnaire). Questions to the ex-

perimenter were allowed during the ﬁrst four tasks.

The following four tasks were used as introductory

tasks:

1. Opening the Help Function. Users were asked to

activate the help function. This task required them

to understand that they could “click” on buttons

on the paper, which activates behavior of the pen.

Furthermore participants were encouraged to set

the audio level output to a comfortable level.

2. Write a Simple Sentence. Users were asked

to write in a line of the diary “I am taking part

in an experiment”. To this entry the pen would

give audio feedback: “An activity for <date of

experiment> was logged”. This introduced the

participant to the idea that the diary recognizes in-

put as a diary and stores the information.

3. Log a Glucose Measurement. Users were asked

to add an entry to the diary saying that they had

a blood glucose reading of 100mg/dl. The pen

would give audio feedback and comment on the

healthy value of 100. This introduced participants

to the idea that the diary could be used to record

glucose measurements, and that the pen would

comment on the value.

4. Log Insulin Dosing. Users were asked to log two

entries of insulin dosing. This introduced the par-

ticipants to how insulin dosing can be logged in

Diabetto.

Acceptance of a Digital Paper-based Diabetes Diary

After completing the introductory tasks, the par-

ticipant was informed that the following tasks were

the actual experiment. Then the performance task sets

were presented. The task set consisted of the follow-

ing complex tasks:

5. Log a Whole Day. The user was asked to record

typical measurements and doses and activities of

a whole day in the diary.

6. Usage of Nutrition Facts. The users were asked

to retrieve the BU-unit value for an already stored

food item from the database. Furthermore they

were asked to create their own entry and retrieve

that value as well.

7. Usage of Abbreviations. Users were asked to

create an abbreviation and use it in the diary.

After the completion of these tasks, the pen was

connected to a Laptop and the Desktop application

was shown, as well as a printout of the diary entries

was handed to the participants.

In some cases participants strayed from doing the

actual task, because the data given in the task did

not coincide with their personal life. In those cases,

participants were gently reminded and asked whether

they could, after entering the own personal data, com-

plete the actual task. This behavior deemed successful

because participant’s curiosity, how the pen behaved

in their own lives was satisﬁed, but nonetheless all

tasks were completed.

In cases were the pen did not recognize readings,

and when it was obvious that recognition failure was

caused by the limited dictionary provided by the pro-

totype, participants were reminded of the prototypic

nature of the experiment, and asked to retry the task

using the words given in the task description. The

amount of additional help was noted for additional

evaluation.

6.5 Perfomance Measurment

In order to analyze performance video data is ana-

lyzed using a form sheet. The sheet has predeﬁned

actions that when recognized in the video recording

are counted for evaluation. Approx. 14 hours of video

data was analyzed for the data.

Items that are counted are the participant’s secu-

rity in using the device (i.e. “asks for help” and “cor-

rects mistakes on his own”) and user satisfaction (i.e.

“positive mentions of the device”, “negative mentions

of the device”, “laughter”). Additionally it is recorded

whether the participant ignores visual and auditory

feedback of the device.

Furthermore it is recorded when technical prob-

lems (e.g. OCR fails) occur, what participants say

about the device and a general impression of the ex-

perimenter of the participant.

Timing is measured by entering the time-codec of

the frame when a user clearly starts reading a task,

when he starts writing, and when he ﬁnishes writing

(i.e. the pen is lifted). Furthermore it is noted whether

the task was effectively completed.

Time on task was calculated as the difference of

ﬁnishing time codec from start time codec.

7 HYPOTHESES

As independent variables demography and health val-

ues were used. Age, gender and diabetes related fac-

tors were assessed. Furthermore expertise with tech-

nology and domain knowledge, as well as computer

self-efﬁcacy were assessed. As intermediary variables

task performance was measures in regard to effective-

ness and efﬁciency. Lastly as dependent variables per-

ceived ease of use and perceived usefulness were as-

sessed in order to assess inﬂuence on behavioral in-

tention (see Figure 3).

We assume that age inﬂuences computer self-

efﬁcacy negatively (H

), similar as age should in-

ﬂuence expertise with technology negatively (H

Both these factors, as shown in previous experiments,

should inﬂuence the efﬁciency during the experiment

). The effectiveness of using a diabetes diary

in general should be positively inﬂuence (H

) by ei-

ther knowing more about diabetes or being more ex-

perienced in using diabetes diaries (i.e. longer dia-

betes duration). Efﬁciency and Effectiveness should

both inﬂuence the perception of ease using the diary

), while perceived usefulness should also depend

on experience, our previous experiments have not

conﬁrmed this ﬁnding (Calero Valdez et al., 2009).

Therefore this is not a formulated hypothesis. Fur-

thermore we expect to see a positive inﬂuence on be-

havioral intention using both PEU and PU (H

8 DESCRIPTION OF THE

SAMPLE

A set of 27 participants took part in the study. The

average age was 38 years (SD = 15.8, 22-73 years).

Eleven participants were male (41%), and 16 were fe-

male. Eight participants were diabetics. Three per-

sons were left-handed. Healthy participants were re-

cruited from announcements in a local newspaper,

and diabetics via the social network of diabetes pa-

tients.

ICT4AWE 2018 - 4th International Conference on Information and Communication Technologies for Ageing Well and e-Health

Figure 3: Visual representation of hypotheses.

The participants were split in three groups (tertial-

split) according to their age. The young age group

showed a mean age of 25 (SD = 2.7, n = 10), the

medium age group showed a mean age of 34 years

(SD = 5.7, n = 10), and the old age group showed a

mean age of 62 (SD = 5.7, n = 7).

Diabetics showed a mean age of 49 years (SD =

17.7), while non-diabetics only showed a mean age

of 33 years (SD = 12.7). Mann-Whitney-U testing

reveals that diabetics were indeed older than non-

diabetics (U = 35.5, z = −2.157, p < .05, r = −.42).

9 RESULTS

In the following sections results are presented. Re-

sults are analyzed using bi-variate correlations, uni-

variate analysis of variance (ANOVA) and multiple

linear regression analysis.

9.1 Descriptive Results

In order to get a broad overview into the results, ﬁrst

descriptive statistics for the three age groups are pre-

sented. When a inﬂuence seem plausible, correlation

data is also presented. First we look into perception

of aging and technical expertise. Then diabetes scales

are presented. Lastly the various performance met-

rics, as well as a pen evaluation is presented.

9.1.1 Technical Expertise

Technical expertise seems to be highly and equally

well distributed for the ECT and EHT scale (see Fig-

ure 4). Only expertise with medical technology is

very low for all age groups. Computer self-efﬁcacy

seems to decrease with age. Overall the sample seems

to be highly tech-savvy with the typical decrease in

self-efﬁcacy with age.

<= 28 29-41 41+

Level of Agreement

Age Groups

Technology Expertise and Computer

Self-Efficacy

EHT

ECT

CSE

Figure 4: Technical expertise and efﬁcacy scale marginal

means over age. EHT=Expertise with household tech-

nology, ECT= Expertise with computerized technology,

CSE=Computer self-efﬁcacy. Error bars denote standard er-

ror.

Correlational analysis conﬁrms that age correlates

signiﬁcantly with computer self-efﬁcacy (CSE). (r =

. − 457, p < .05). Interestingly the seemingly equal

technology expertise with computerized technology

(ECT) correlates highly signiﬁcantly with the CSE

(r = .577, p < .01).

9.1.2 Diabetes: Coping Scales and Domain

Knowledge

When looking at coping scales no differentiation be-

tween age groups can be made, because of the small

sample of diabetics (n = 8). Nonetheless diabetics

seem to perceive their diabetes to be as only lightly

pervasive (DP: M = 2.4, SD = 1.3). The diabetics

seem to have a high desire for information and con-

Acceptance of a Digital Paper-based Diabetes Diary

trol (DCI: M = 4.7, SD = 0.7) as well as trust in their

doctor’s competence (TDC: M = 4.1, SD = 0.5).

Diabetics were very well informed about diabetes

and rather fairly informed about their obesity (IFG:

M = 3.9, SD = 0.3, IFO: M = 2.5, SD = 0.2). This is

not unusual because all diabetics were type-1 diabet-

ics.

Correlational analysis showed that both IFO and

IFG did not correlate with any other measures, except

with themselves (r = .824, p < .05). The duration

of diabetes only interacted also interacted with the

informedness of obesity related factors (r = −.793,

p < .05). The longer a participant has had diabetes

(early onset) the less he was informed about obe-

sity, which is externally valid, since obesity related

information is of more interest to late-onset diabetes.

Furthermore we found that diabetes pervasiveness did

correlate with perceived cognitive deﬁcits (r = .812,

p < .05). All other measures did not correlate with

coping style scales.

Further analyses of these variables are not per-

formed.

9.1.3 Task Performance

Task performance was determined by both task suc-

cess rate for effectiveness and time on task as efﬁ-

ciency. Effectiveness seems to be very high in the

younger age group over all tasks (see Figure 5). The

older age group seems in contrast to have more prob-

lems with the later tasks. The only tasks that seems

to make no difference between age groups are task

one and task ﬁve. Correlational analysis shows that

tasks three (r = −.473, p < .05) and four (r = −.450,

p < .05) do correlate signiﬁcantly with age. The older

a user is the less effective he is at completing these

tasks.

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

<= 28 29-41 41+

Success Rate (%)

Age Groups

Task Effectiveness

Task 1

Task 2

Task 3

Task 4

Task 5

Task 6

Task 7

Figure 5: Marginal means of effectiveness for all tasks in

percent over age. Error bars denote standard error.

When looking at efﬁciency for individual task (see

Figure 6) individual task length becomes visible. Task

ﬁve seems to take the most time from all participants,

while task two seem to take the least.

As only the last three tasks are measured as a per-

formance experiment, reliability of these three task as

a scale was assess with a Cronbach’s α = .734. This

indicates that the last three tasks can be used as one

scale. Because these three individual measurements

(ToT

,ToT

) have different maxima and min-

ima, the geometric mean is chosen as a function to

unite these variables to a total time on task (ToT) as

follow:

ToT =

ToT

∗ ToT

The average ToT a user takes is M = 10 seconds

(SD = 3.71s). The minimal ToT is 4.4 seconds and

the maximal ToT is 20.8 seconds.

100

150

200

250

300

350

400

450

<= 28 29-41 41+

Time on Task (s)

Age Groups

Task Efficiency

Task 1

Task 2

Task 3

Task 4

Task 5

Task 6

Task 7

Figure 6: Efﬁciency measure marginal means for all tasks

over age. Error bars denote standard error.

It seems necessary to incorporate the success rate

into the efﬁciency measurement, because users could

be faster at a task, when they simply skipped certain

elements of a task. Therefore a corrected time on task

(CTOT) is calculated using the effectiveness measure

(SR=success rate) of the last three tasks.

CToT =

ToT

/SR

∗ ToT

/SR

∗ ToT

/SR

This measure shows a mean of 14.4 seconds

(SD = 14) with a minimum of 4.4 seconds and a max-

imum of 62.4 seconds.

Univariate ANOVA analysis indicates that age

groups do not differ in regard to efﬁciency

(F(2,25) = 2.245, n.s.). Correlation analysis con-

ﬁrms this (r = .306, p = .14).

9.2 Effects on Performance

When trying to understand how performance is de-

termined multiple linear regression analyses are per-

formed.

ICT4AWE 2018 - 4th International Conference on Information and Communication Technologies for Ageing Well and e-Health

First a model for success rate is generated using

age, diabetes duration, technical expertise and com-

puter self-efﬁcacy as possible predictors. Interest-

ingly only a model with a single predictor remained

signiﬁcant. The model used expertise in household

technology as a predictor and was able to explain

27% (adj. r

= .27) more variance than the scale

mean (F(1, 17) = 7.714, p < .05). The constant term

showed a coefﬁcient of B = 2.185 (SEB = 0.473)

and the predictor EHT showed a coefﬁcient of B =

−0.265 (SEB = 0.09) and a standardized slope of

β = −.559. This means that household technology

actually inﬂuences effective negatively.

Secondly a model of corrected time on task was

generated using the same predictors as in the last re-

gression. A model using EHT and CSE as predictors

was shown to explain 47% (adj. r

= .47) more vari-

ance than the scale mean (F(2, 15) = 8.52, p < .01).

Expertise with household technology again showed

negative inﬂuence on performance while computer

self-efﬁcacy showed positive effects (see Table 1).

Variance inﬂation was negligible (V IF = 1.036). The

only downside of this model is the relatively high

standard error of the constant term. Removing the

weakest predictor from the model, did

Table 1: Linear regression table for corrected time on

task. All predictors increased the explained variance sig-

niﬁcantly.

Predictor Unstand. coeff. Standardized slope

B SEB β

(Constant) -32.476 37.140

EHT 16.952 5.732 .531

CSE -8.508 3.711 -.412

The only downside of this model is the relatively

high standard error of the constant term. Removing

the weakest predictor (i.e. CSE) from the model, did

not decrease the standard error to a drastically more

acceptable size (SEB = 32.7), while at the same time

the explained variance decreases to 33% (adj. r

.33, F(1,16) = 9.358, p < .01).

9.3 Pen Evaluation

After the experiment was ﬁnished it was of high inter-

est to ﬁnd out, how much participants liked the digital

pen-based diabetes diary. We based the assessment on

a TAM model using both perceived ease of use and

perceived usefulness as predictors for the behavioral

intention. Furthermore we wanted to investigate how

technical expertise inﬂuenced behavioral intention.

We measured perceived ease of use and usefulness

for nine features and combined them into two scales.

Reliability was analyzed using Cronbach’s α. Per-

ceived ease of use showed excellent reliability with

a Cronbach’s α = .914, similarly to perceived use-

fulness, which showed also an excellent reliability of

α = .909. The scales are calculated as the mean of the

items.

The three behavioral intention items showed an

acceptable reliability of α = .782. The scale BI was

calculated as the mean of these items.

When looking at how the age group inﬂuences the

three scales PEU, PU, and BI (see Figure 7), differ-

ences are hard to make out. Only perceived ease of

use seems to show a difference between young and

medium aged users. Though Univariate ANOVA re-

jects this hypothesis (F(2,19) = 2.373, n.s.). In gen-

eral perceived ease of use is high (M = 5.01, SD =

0.75), perceived usefulness is very high (M = 5.44,

SD = 0.58) and behavioral intention is also very high

(M = 5.38, SD = 0.58).

<= 28 29-41 41+

Level of Agreement

Age Groups

Pen Evaluation

PEU

Figure 7: Marginal means of PEU, PU, and BI over age

groups. Error bars denote standard error.

In order to determine what factors predict the be-

havioral intention a TAM based model using both

PEU and PU as predictors was used. Additionally

diversity factors like age, gender (dummy coded), di-

abetes duration, technology-expertise, and computer

self-efﬁcacy were used as predictors. Only the pure

TAM based model using a two predictor model with

PEU and PU as predictors was able to signiﬁcantly

explain more variance than the scale mean. This

model was able to explain 86% (adj. r

= .859) more

variance than the scale mean (F(2,15) = 52.747, p <

.01). This is a very large effect. The perceived use-

fulness was about three times stronger in predicting

the behavioral intention than the perceived ease of use

(see Table 2). Both predictors had a positive inﬂuence

on the behavioral intention.

Acceptance of a Digital Paper-based Diabetes Diary

Table 2: Linear regression results for behavioral intention.

Predictor Unstand. coeff. Stand. slope

B SEB β

(Constant) -0.04 0.54

PEU 0.210 0.123 .244

PU 0.798 0.155 .734

10 QUALITATIVE INSIGHTS

During the experiments positive feedback about the

pen in general was extensive. Participants were sur-

prised that a pen was able to “understand” what they

were writing. One participant (female, 55 years) was

very reluctant to take part in the experiment because,

she “hate[s] all technology.” She said:

“Stay away from me with this nonsense. All you

engineers should try to learn that normal people like

me cannot use these complicated devices.”

After persuading her to take part in the experiment

and using the pen she stated that she loved the friendly

interaction with the pen, and the pen actually felt good

in her hand. She would like to have pens be as heavy

as this one, as normal pens were to tiny for her to use.

Most participants showed similar reactions. Even

the non-diabetic participants were delighted with the

pen usage and in particular with the feeling that tech-

nology could “understand” them. Some even asked

whether they could buy a pen for non-diabetic related

tasks.

11 DISCUSSION

The results (see Figure 8) from the Diabetto user stud-

ies are interesting because some of the results are

rather unexpected, while others are very much ex-

pected. We could conﬁrm that age does have a neg-

ative effect on computer self-efﬁcacy and expertise

with household technology (supporting H

and H

while at the same time it has no effect on computer-

ized technology (disagreeing with H

). This could be

a sample effect. The size of the sample is rather small

and the sample is very tech savvy.

Much more interesting though is that the exper-

tise with household technology exerts a negative ef-

fect on the efﬁciency with the pen (disagreeing with

), while computer self-efﬁcacy exerts a positive ef-

fect (supporting H

). A possible explanation might

be that a pen that is used by hand with no buttons

Translated from german.

or other interaction features that are typical for com-

puterized technology is more like a non-technology

object. People that tend to type more and write less

might be slower at writing with a digital pen as well.

The only troublesome ﬁnding is that computer self-

efﬁcacy in turn does inﬂuence efﬁciency positively.

This could be explained in a way that people that are

more conﬁdent in using technology feel less intimi-

dated by the computerized interaction with the pen.

Inﬂuences of diabetes expertise through duration

or domain knowledge were not found in regard to ef-

fectiveness (rejecting H

). This could mean that the

diary was as easy to use for diabetics as it was for non-

diabetics. This effect could be explained by the exten-

sive help-function that explained many of the features

anytime during the experiment.

Any effect of efﬁciency or effectiveness on the

pen evaluation was not found (rejecting H

). These

were expected as similar ﬁndings existed in the ex-

periments described in (Calero Valdez et al., 2009).

The inﬂuence of perceived ease of use and per-

ceived usefulness on behavioral intention was simi-

larly strong as predicted by theory (supporting H

Both showed a strong inﬂuence. The weakness of

prediction of perceived ease of use could be explained

with the general high rating of perceived ease of use

for the whole sample. Looking at perceived useful-

ness though reveals a similar distribution. Usefulness

still seems to be more important than ease of use,

when ease of use is high. Ease of use could there-

fore be seen as a necessary condition for acceptance,

while usefulness poses as a sufﬁcient condition.

12 LIMITATIONS AND FUTURE

WORK

This study tried to use both, qualitative and quantita-

tive methods in order to develop a deeper understand-

ing of the requirements of diabetes patients and to im-

plement the user requirements into the development

of the prototype. However, future studies will have

to validate the device performance data with a larger

sample. On the base of the present data, extrapolation

is not yet feasible.

When reviewing how effectiveness is measured

and what other effectiveness measures exist, we can

see that only effectiveness of introductory tasks corre-

lated with age. Further research could therefore look

into other tasks and learnability of these tasks as well.

Strangely trust in doctors competence played a

large role in perceived ease of use and perceived use-

fulness. This exploratory ﬁnding might be a ﬂuke

or could indicate high agreeableness in a participant.

ICT4AWE 2018 - 4th International Conference on Information and Communication Technologies for Ageing Well and e-Health

Figure 8: Visual representation of results, showing both correlational and linear regression data. The minus sign in front of a

value indicates the direction of the correlation, and is not result of an complex number squared. It was added after squaring.

Examining this and further personality types might be

of interest in further research.

ACKNOWLEDGEMENTS

We would like to thank Hennadiy Verkh for the de-

velopment of the software prototype. Furthermore we

would like to thank the participants for taking part in

the study and sharing their insights. We would also

like to thank the anonymous reviewers for their input

on a previous version of this article.

REFERENCES

Alag

oz, F., Calero Valdez, A., Wilkowska, W., Zieﬂe, M.,

Dorner, S., & Holzinger, A. (2010). From cloud com-

puting to mobile internet, from user focus to culture

and hedonism: The crucible of mobile health care

and wellness applications. In 2010 5th International

Conference on Pervasive Computing and Applications

(ICPCA), pages 38–45. IEEE.

Alag

oz, F., Zieﬂe, M., Wilkowska, W., & Calero Valdez,

A. (2011). Openness to accept medical technology –

a cultural view. In Information Quality in e-Health,

pages 151–170. Springer.

Arning, K. and Zieﬂe, M. (2007). Understanding age differ-

ences in PDA acceptance and performance. Comput-

ers in Human Behavior, 23(6):2904–2927.

Atzori, L., Iera, A., & Morabito, G. (2010). The internet of

things: A survey. Computer Networks, 54(15), 2787–

2805.

Beier, Guido. (2004). Kontroll

uberzeugungen im Umgang

mit Technik: ein Pers

onlichkeitsmerkmal mit Rele-

vanz f

ur die Gestaltung technischer Systeme. (Doc-

toral Thesis) dissertation.de.

Brands, A. M., Biessels, G. J., Haan, E. H. d., Kappelle, L.

J., & Kessels, R. P. (2005). The effects of type 1 dia-

betes on cognitive performance a meta-analysis. Dia-

betes Care, 28(3):726–735.

Brands, A., Van Den Berg, E., Manschot, S. M., Biessels,

G. J., Kapelle, L., De Haan, E. H., & Kessels, R. P.

(2007). A detailed proﬁle of cognitive dysfunction and

its relation to psychological distress in patients with

type 2 diabetes mellitus. Journal of the International

Neuropsychological Society, 13(02):288–297.

Broderick, J. E., & Stone, A. A. (2006). Paper and elec-

tronic diaries: Too early for conclusions on compli-

ance rates and their effects – Comment on Green,

Rafaeli, Bolger, Shrout, & Reis (2006).

Burke, L. E., Warziski, M., Starrett, T., Choo, J., Music,

E., Sereika, S., Sevick, M. A., et al., (2005). Self-

monitoring dietary intake: current and future prac-

tices. Journal of Renal Nutrition, 15(3), 281–290.

Burke, L. E., Conroy, M. B., Sereika, S. M., Elci, O. U.,

Styn, M. A., Acharya, S. D., ... and Glanz, K. (2011).

The Effect of Electronic Self-Monitoring on Weight

Loss and Dietary Intake: A Randomized Behavioral

Weight Loss Trial. Obesity, 19(2), 338–344.

Calero Valdez, A., Zieﬂe, M., Horstmann, A., Herding, D.,

& Schroeder, U. (2009). Effects of aging and domain

knowledge on usability in small screen devices for di-

abetes patients. In HCI and Usability for e-Inclusion,

pages 366–386. Springer.

Calero Valdez, A., Zieﬂe, M., Alag

oz, F., & Holzinger, A.

(2010a). Mental models of menu structures in diabetes

assistants. In Computers Helping People with Special

Needs, pages 584–591. Springer.

Calero Valdez, A., Zieﬂe, M., Horstmann, A., Herding, D.,

& Schroeder, U. (2010b). Task performance in mobile

and ambient interfaces. does size matter for usability

of electronic diabetes assistants? In Information Soci-

ety (i-Society), 2010 International Conference on the

Information Society, pages 514–521. IEEE.

Calero Valdez, A., Zieﬂe, M., Horstmann, A., Herding, D.,

& Schroeder, U. (2011). Mobile devices used for med-

ical applications: Insights won from a usability study

Acceptance of a Digital Paper-based Diabetes Diary

with diabetes patients. In The International Journal of

Digital Society (IJDS), 2(1):337–346.

Chen, Y. (2010). Take it personally: accounting for individ-

ual difference in designing diabetes management sys-

tems. In Proceedings of the 8th ACM Conference on

Designing Interactive Systems, DIS ’10, pages 252–

261, New York, NY, USA. ACM.

Davis, F. D. (1989). Perceived usefulness, perceived ease

of use, & user acceptance of information technology.

MIS Quarterly, 13(3):319–340.

Davis, G. B. (2002). Anytime/anyplace computing and the

future of knowledge work. Communications of the

ACM, 45(12), 67–73.

Duke, D. L., Thorpe, C., Mahmoud, M., & Zirie, M.

(2008, March). Intelligent Diabetes Assistant: Us-

ing machine learning to help manage diabetes. In

Computer Systems and Applications, 2008. AICCSA

2008. IEEE/ACS International Conference on (pp.

913–914). IEEE.

Forjuoh, S. N., Reis, M. D., Couchman, G. R., & Ory, M.

G. (2008). Improving diabetes self-care with a PDA

in ambulatory care. Telemedicine and e-Health, 14(3),

273–279.

Gregory, C. L. (2008). But I Want a Real Book. Reference

& User Services Quarterly, 47(3), 266–273.

Hader, C., Beischer, W., Braun, A., Dreyer, M., Friedl, A.,

usgen, I., Gastes, U., Gr

uneklee, D., Hauner, H.,

obberling, J., et al., (2004). Diagnostik, Therapie

und Verlaufskontrolle des Diabetes mellitus im Alter.

Diabetes und Stoffwechsel, 13:31–56.

Hien, P., & B

ohm, B. (1997). Diabetes-Handbuch: eine An-

leitung f

ur Praxis und Klinik. Springer.

Kerkenbush, N. L. (2003). A comparison of self-

documentation in diabetics: electronic versus pa-

per diaries. In AMIA Annual Symposium Proceedings

(Vol. 2003, p. 887). American Medical Informatics

Association.

Kerkenbush, N. L., & Lasome, C. E. (2003). The emerging

role of electronic diaries in the management of dia-

betes mellitus. AACN Advanced Critical Care, 14(3),

371-378.

Klack, L., Schmitz-Rode, T., Wilkowska, W., Kasugai, K.,

Heidrich, F., & Zieﬂe, M. (2011). Integrated Home

Monitoring and Compliance Optimization for Patients

with Mechanical Circulatory Support Devices. Annals

of biomedical engineering, 39(12), 2911–2921.

Krishna, S., Boren, S. A., & Balas, E. A. (2009). Healthcare

via cell phones: a systematic review. Telemedicine and

e-Health, 15(3), 231–240.

Scheermesser, M., Kosow, H., Rashid, A., & Holtmann, C.

(2008, January). User acceptance of pervasive com-

puting in healthcare: Main ﬁndings of two case stud-

ies. In Pervasive Computing Technologies for Health-

care, 2008. PervasiveHealth 2008. Second Interna-

tional Conference on (pp. 205–213). IEEE.

Stone, A. A., Shiffman, S., Schwartz, J. E., Broderick, J.

E., & Hufford, M. R. (2002). Patient non-compliance

with paper diaries. BMJ, 324(7347):1193–1194.

Tani, S., Marukami, T., Matsuda, A., Shindo, A., Take-

moto, K., & Inada, H. (2010). Development of a health

management support system for patients with diabetes

mellitus at home. Journal of medical systems, 34(3),

223–228.

Venkatesh, V. and Davis, F. D. (2000). A theoretical exten-

sion of the technology acceptance model: Four longi-

tudinal ﬁeld studies. Management science, pages 186–

204.

Venkatesh, V., Morris, M. G., Davis, G. B., & Davis, F. D.

(2003). User acceptance of information technology:

Toward a uniﬁed view. MIS Quarterly, 27(3):425–

478.

Venkatesh, V. and Bala, H. (2008). Technology acceptance

model 3 and a research agenda on interventions. De-

cision Sciences, 39(2):273–315.

Venkatesh, V., Zhang, X., & Sykes, T. A. (2011). Doctors

do too little technology: A longitudinal ﬁeld study of

an electronic healthcare system implementation. In-

formation Systems Research, 22(3):523–546.

Yeung, S. E., Fischer, A. L., & Dixon, R. A. (2009). Explor-

ing effects of type 2 diabetes on cognitive functioning

in older adults. Neuropsychology, 23(1):1–9.

Zieﬂe, M. (2002). The inﬂuence of user expertise and phone

complexity on performance, ease of use and learnabil-

ity of different mobile phones. Behaviour & Informa-

tion Technology, 21(5), 303-311.

Zieﬂe, M., & Wilkowska, W. (2010, March). Technol-

ogy acceptability for medical assistance. In Pervasive

Computing Technologies for Healthcare (Pervasive-

Health), 2010 4th International Conference on (pp. 1–

9). IEEE.

ICT4AWE 2018 - 4th International Conference on Information and Communication Technologies for Ageing Well and e-Health