TBM SYSTEMATIC REVIEWS

Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019

The impact of mobile applications on medication adherence:
a systematic review
Ricki Ng, Stephen R. Carter, Sarira El-Den

School of Pharmacy, Faculty Abstract contribute to disability and death, globally [5].
of Medicine and Health, The In 2008, Apple and Android launched their Application or Fortunately, the morbidity and mortality associated
University of Sydney, Australia
“App” stores. Since then, there has been a growing interest with many of these diseases can be managed using
in using mobile apps for improving medication adherence. medications; however, medication adherence is es-
However, research on the efficacy of apps, in terms of improved
sential for efficacy and therapeutic benefit.
medication adherence and clinical outcome and/or patient-
related outcome measures (PROMs) is scarce. The objective
Correspondence to: Stephen of this research was to systematically review the impact of Implications
R. Carter, stephen.carter@ apps on consumers’ medication adherence and to determine Practice: Mobile applications (apps) can be used
sydney.edu.au the effect on clinical outcome and/or PROM(s). A systematic to increase medication adherence in patients with
literature search was conducted to identify publications aimed chronic illnesses.
Cite this as: TBM 2019;XX:XX–XX at improving medication adherence published from January
doi: 10.1093/tbm/ibz125 2008 to April 2018. All studies were assessed for risk of bias
using either the Risk Of Bias In Non-randomized Studies-of
© Society of Behavioral Medicine
Interventions or the revised tool for Risk of Bias in randomized
Policy: Policies and regulations should be imple-
2019. All rights reserved. For permis- mented to ensure standardization of these apps
sions, please e-mail: journals.permis- trials tool, depending on study design. Eleven randomized
controlled trials (RCTs) and 10 non-RCTs were included. All
and provide patients with the assurance that apps
sions@oup.com.
11 RCTs showed improvements in adherence; however, only
are of high quality.
seven reported statistically significant improvements in at
least one adherence measure. Nine RCTs also demonstrated
improvements in clinical outcome/PROM(s), of which five were Research: For accurate comparisons within and
statistically significant, whereas two RCTs did not report on among studies, future research should design
clinical outcome/PROM(s). Only two studies using non-RCT high-quality studies involving blinding and con-
study designs showed statistically significant improvements in trols, using a variety of adherence measures, with
all measures of adherence and clinical outcome/PROM(s). The appropriate sample sizes and duration.
risk of bias was moderate or serious for all included studies.
Even though the use of an app may improve adherence, it is
difficult to draw conclusions regarding the impact of apps on
medication adherence due to the high degree of heterogeneity Adherence refers to how a person’s medication-
across studies, from the methodological design to the features taking behavior corresponds to the agreed recom-
of the app and the measure of adherence. mendations negotiated with the health care provider
[6]. The estimated adherence rate in high-income
Keywords countries is 50% for chronic medication therapy, re-
gardless of the disease [7]. As a result, the full benefits
Mobile applications, Apps, Medication adherence, of some medicines are not fully attained, and cer-
Medication therapy tain medical conditions continue to be a significant
health care problem. For example, nearly one third
INTRODUCTION of individuals taking antidepressants discontinue
Chronic diseases, such as cardiovascular diseases, treatment against medical advice after 1 month des-
cancer, respiratory diseases, and diabetes are a pite clear guidelines recommending that antidepres-
growing problem, worldwide [1]. Advances in med- sants be continued for 6 months after full remission
ical treatments and technologies have resulted in [8]. Furthermore, 45% of individuals with type 2 dia-
longer life expectancies, globally, and a rapidly aging betes fail to reach their target for glycemic control,
population, thereby also contributing to the burden partly due to poor medication adherence [9].
of chronic disease [2,3]. Chronic diseases account Failure to adhere to medication therapy may re-
for more than 14 million deaths each year and are sult in disease progression and ultimately an in-
expected to contribute to 65% of the global burden creased risk of morbidity and mortality [10]. There
of disease by 2020 [4]. Mental illnesses and human have been initiatives and medical advances aimed
immunodeficiency virus (HIV) and acquired im- at improving adherence, such as simplifying dosing
mune deficiency syndrome (AIDS) also significantly regimens. However, even with newer agents, such
TBM page 1 of 17
Published online: XX XXXX 2019
SYSTEMATIC REVIEWS

as oral anticoagulants that require less monitoring, A range of technological interventions including
medication adherence is still of utmost importance apps to promote medication adherence have been
due to the short half-life of these medicines [11]. In explored in the literature, but there is a lack of re-
2003, nonadherence was reported to be a signifi- search exploring their impact on adherence and clin-

Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019

cant, global health problem by the World Health ical outcomes. Previous reviews on adherence apps
Organization (WHO) [12], affecting not only the have focused on evaluating the features, properties,
individual but also the safety of the general popu- and qualities of currently marketed apps [23–25].
lation. For example, nonadherence to antiretroviral A recent review, which identified 272 medication
therapy (ART) among HIV-positive people can lead reminder apps, concluded that currently available
to the emergence of drug-resistant HIV, which not apps on the market are of low quality. It suggested
only increases the risk of developing AIDS but also that more than half the apps have not been updated
increases the risk of infecting others [13]. The WHO recently and do not contain important features, such
statement maintains that improving adherence may as medication tracking history, refill reminders, and
lead to better improvements in population health data security [24]. Another review evaluating apps
than any improvements generated from one specific that detect drug–drug interactions found that most
medical treatment [6,7]. currently available apps are delivering inaccurate
To improve adherence to medications, an in-depth and potentially unsafe information [25]. To the best
understanding of the underlying factors of inten- of our knowledge, studies that examine the effect-
tional and unintentional nonadherence is required iveness of apps for medication adherence have not
[14], and it has been suggested that a lack of under- been systematically evaluated. Therefore, given the
standing of these issues is a barrier to improving importance of medication adherence for a variety of
medication adherence [15]. Nonadherence is a medical conditions and the plethora of apps avail-
multifactorial and complex phenomenon that can able to support medication adherence, the aim of
be due to patient-, health care professional-, medica- this systematic review was to explore the current lit-
tion- and/or health care system-related factors [14]. erature relating to the impact of apps on consumers’
The most common patient-related factor associated medication adherence. Specifically, the objectives
with nonadherence is forgetting to take the medica- of this review are to:
tion [16]. Other patient-related factors may include
a lack of understanding of the nature of the disease • evaluate the impact of apps on medication adherence;
and the importance of adherence to treatment [17]. • determine whether demonstrated improvements in ad-
Such barriers may be overcome by effective com- herence through the use of apps leads to better clinical
munication between health care professionals and outcome(s) and/or patient-related outcome measures
patients or through the use of a mobile application (PROM(s)); and
or “app” [18]. For example, apps could be designed • make recommendations for future research evaluating
to support medication adherence through the inte- the impact of apps on medication adherence.
gration of reminders, among other features.
The management of chronic diseases has re-
cently evolved to include the use of technological METHODS
modalities. In 2008, Apple and Android launched
the Application or “App” store—Apple iTunes and
Selection of studies
Android Google Play, respectively, and by 2010
A systematic literature search was conducted in
there were approximately 150,000 apps available
the electronic databases MEDLINE, PubMed,
for download [19]. Mobile applications are down-
CINAHL, EMBASE, and PsycINFO from the period
loadable programs that run on the operating system
January 2008 to April 2018 to identify relevant pri-
of the mobile device, which may be stored locally
mary research publications exploring the use of ap-
or contain web-based features that require internet
plications to promote medication adherence. The
access for use [20]. This technological advance-
Preferred Reporting Items for Systematic Reviews
ment provided a new canvas for novel approaches
and Meta-Analyses (PRISMA) checklist guided this
to chronic disease management including medi-
systematic review. The search was limited to studies
cation adherence interventions and medication
conducted on a human sample and published in the
self-management. For the first time, it was possible
English language. Depending on the database, both
to reach patients, in real time, in any setting via an
key words and mapped subject headings of three
app. Since then, such apps have become a popular
different concepts were used to conduct the final
tool to address adherence and patient behavior, with
search strategy:
up to 325,000 health-related apps available from the
App store in 2017 [21]. Apps provide an innovative,
1. Telemedicine or telehealth or mobile application* or
practical, cost-effective approach to the promotion
app* or mobile phone* or smartphone* or reminder
of medication adherence and can be designed to fa-
system* or text messaging
cilitate behavior change [22].

page 2 of 17TBM
 SYSTEMATIC REVIEWS

2. Adherence or compliance or concordance or further ambiguities were discussed by all three mem-
persistence bers of the research team, until all authors agreed.
3. Drug therapy or medication therapy management or A second author (S.R.C.) reviewed all the included
pharmacological treatment* publications, and there was complete agreement be-

Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019

tween R.N. and S.R.C.
When screening potential publications for inclusion Owing to the significant clinical and methodo-
in this systematic review at the abstract stage, a publi- logical heterogeneity in the studies included, the as-
cation was excluded if the description of the interven- sessment of bias of individual studies was conducted
tion did not indicate that it was integrated within an using the revised tool for Risk of Bias in random-
app, as per the aforementioned definition of an app. ized trials (RoB 2) [26] and Risk Of Bias In Non-
randomized Studies-of Interventions (ROBINS-I)
Eligibility criteria tool for randomized control trials (RCTs) and
A structured inclusion and exclusion criteria check- quasi-randomized and nonrandomized studies, re-
list was developed and adapted to screen all eligible spectively [27]. The RoB 2 determines the risk of
studies, at the full-text stage, when considering pub- bias in individual studies across five domains: bias
lications for inclusion in this systematic review. For arising from the randomization process, bias due to
inclusion in this systematic review, all studies had to deviations from intended interventions, bias due to
meet the following inclusion criteria: missing outcome data, bias in measurement of the
outcome, and bias in selection of the reported results
1. The study evaluated the effectiveness of an app in pro- [26]. The ROBINS-I tool assesses the risk of bias in
moting medication adherence by comparing adher- individual studies across seven domains: bias due to
ence outcomes: confounding, bias in the selection of participants into
i. among an intervention and control group; and/or the study, bias classification of intervention, bias due
ii. by conducting pre/post intervention analyses. to deviation from intended interventions, bias due
2. The outcome of the study reported on adherence to a to missing data, bias in measurement of outcome,
medicine for any chronic medical condition or disease and bias in selection of reported results [27]
state (that requires adherence to a medication).
3. Participants of any age were included in the study: RESULTS
i. the app was used by an adult diagnosed with a
medical condition, or Study selection
ii. the app was used by a caregiver, parent, or guardian The literature search yielded a total of 2,809 poten-
of another person diagnosed with a medical condi- tially relevant publications. After the removal of all
tion (care recipient). duplicates, a total of 1,993 publications were con-
4. Participants (or care recipients) had a clinical diagnosis sidered for inclusion. Seven hundred and sixty-five
of a medical condition and were prescribed continuous articles were identified as non-primary research pub-
medication(s) for that specific medical condition. lications, and the remaining 1,228 publications were
5. The study was published as a primary research article. identified as potentially relevant. The remaining
1,228 citations were screened based on titles and,
Studies were excluded if: then, abstracts. One hundred thirteen full-text, pri-
mary research publications were screened, of which
1. the study only explored participants’ attitudes toward 21 were eligible for inclusion. A breakdown of this
an app; and process can be seen in the PRISMA flowchart in
2. the intervention used to support or improve medica- Figure 1. Full-text articles were excluded mainly
tion adherence was not integrated into an app. because the study explored technologies that were
not integrated within an app and/or web-based
platform and hence, did not meet the study aims.
Data extraction Another common reason for exclusion was that the
All the searches from different databases were ex- app in the study did not aim to improve adherence.
ported into EndNote and duplicates were removed
both automatically and manually. One author (R.N.) Study characteristics
screened through the titles, then the abstracts to Twenty-one studies were included in this systematic
determine whether potential publications were eli- review, with a total of 1,229 participants, ranging
gible for inclusion. The eligibility of the remaining from 8 to 90 years old. Of these 21 included studies,
full-text articles was determined using the inclu- 3 studies included children and adolescent partici-
sion and exclusion checklist. These articles were pants [28–30] and 17 studies included adult popu-
categorized as either included, excluded, or de- lations only [11,31–46] and 1 study included both
ferred due to ambiguity. All deferred articles were adult, children, and adolescent participants [47].
reviewed by a second author (S.R.C. or S.E.), and a The included publications fell into two categories,
decision was made once both authors agreed. Any in that they either reported only on the impact of the
TBM page 3 of 17
SYSTEMATIC REVIEWS

Records idenfied through

evidence) [29,32,33,46,47], two were interrupted
database searching time series studies (Level III.3 evidence) [31,43]
(n = 2809)
and one study was a prospective cohort study (Level
III.3 evidence) [28]. The duration of these interven-

Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019

Abstracts, le ers, thesis, reviews, study
tions was relatively short, ranging from three weeks
Records screened a‚er
duplicates removed protocol, case studies, books and book to 12 months.
(n = 1993) secons excluded
(n = 765) Regarding the nature of the comparator group,
three studies reported that participants received
a simplified version of the app [38,44,45] and two
Idenfied as potenally Records excluded based on
relevant abstracts and tles studies reported using a paper diary [33,41]. Six
(n = 1228) (n = 1094) studies reported that participants received usual
care [11,30,32,36,37,39] and four studies reported
61 arcles: intervenons were not
that the comparator group did not use an app
Full-text arcles assessed for
eligibility
integrated within an app [28,34,40,43]. One study used an electronic pillbox
(n = 134) 24 arcles: app did not aim to improve as the comparator group [35], and in another study,
adherence
participants in the comparator group received oral
14 arcles: only explored development,
feasibility and acceptability the app and written information on the safe use of medicines
14 arcles: text messaging/telephone calls [42]. In one study, participants in the comparator
intervenons group received extra online educational materials
[30] and three studies compared the effects of the
app by comparing the adherence before the inter-
Publicaons included in the
systemac review
vention, but the details of the intervention were not
(n = 21) specified [29,46,47].

Figure 1. PRISMA flow diagram of reviewed and included studies Characteristics of apps
Four broad categories of adherence promotion strat-
app on medication adherence or they reported on egies were identified: reminders, education, medica-
the impact of the app on both medication adherence tion e-dairy, and communication with a health care
and a clinical outcome/PROM(s). professional. Fourteen apps had a reminder feature,
Nine of the 21 apps identified supported adher- which was composed of any strategy that was aimed
ence to medicines for cardiovascular conditions, spe- at reminding and/or motivating an individual to take
cifically for hypertension (n = 2) [39,43], coronary their medication [11,30,32,34,35,37–39,41–43,45–
heart disease (n = 2) [31,41], heart failure (n = 2) 47]. Reminder strategies embedded into these apps
[35,46], atrial fibrillation (n = 1) [36], myocardial included simply sending text reminders, sending
infarction (n = 1) [38], and ischemic stroke (n = 1) daily motivational messages, as well as, customizable
[11]. A broad range of other chronic conditions were reminders based on the medicine’s unique time of
also identified and can be seen in Table A1, which administration. Educational features were embedded
presents the general characteristics of all included into 13 apps, to provide users with information to better
studies. The specific medical condition the app was understand the importance of, and thereby adhere
targeting was not reported in one study [42], and to their medications [28,29,31,32,34,36,38,40,42–
in another study, the app targeted multiple medical 44,46,47]. Medication and/or e-health features
conditions, including hypertension, dyslipidemia, were embedded into 12 apps and aimed to im-
heart failure, and HIV [32]. Studies were conducted prove adherence by creating a medication list and/
in seven middle- to high-income countries; however, or dosing schedule and also contained information
the majority of studies were conducted in the USA about side effects [28,30,33,35,36,38–41,43,45,46].
(n = 13; Table A1). The number of participants listed Communication features were embedded into seven
in Table A1 refers to the participants that completed apps and involved a range of two-way communica-
adherence measures in the study. tion strategies to provide motivation and social sup-
port [28,31,32,36,39,40,47]. Twenty included studies
Type of study, duration of intervention, and the nature of the reported on apps which used more than one adher-
comparator group ence promotion strategy [11,28,30–47]. One study
There was considerable variation among the reported on an app that supported medication adher-
studies regarding the methodological design. ence through education only, whereby educational
According to the National Health and Medical sessions were delivered through the Skype app [29].
Research Council (NHMRC) evidence hierarchy Other functionality features to support adherence in-
[48], 11 studies were RCTs; Level II evidence) cluded goal setting [31], providing a “vacation” fea-
[11,34,35,37–42,44,45], two were pseudo-RCTs ture to determine whether refills would be needed
(Level III.1 evidence) [30,36], five were case series before a certain date [30] and representing drug
with pretest and posttest outcomes (Level IV plasma concentrations to inform individuals [44].

page 4 of 17TBM
TBM
Table 1 | Outcomes of included publications using subjective measures of adherence only

Intervention group
(IG) or comparator Method of adherence Adherence measure and change Clinical outcome/patient-related outcome measures Type of study, level of evidence as per
Study group (CG) measurement (significance) (significance) NHMRC hierarchy
Ammenwerth IG: used MyCor for Self-document daily meas- Mean adherence IG: 80% and QOL (MacNew questionnaire) improved from 5.5 to Interrupted time series without a
et al. [31] another 2 weeks urements and drug in- CG: 87% (SNR) 6.3 (p < .01). Reductions in blood pressure and parallel control group, Level III.3
after 12 weeks take through app heart rate were not observed (SNR)
without app sup-
port. CG: used
MyCor for 4 weeks
initially
Dietrich et al. IG: devise users. CG: Participants recorded their Mean number of times missed Admission rates for HMB IG: 0, CG: 19 (p = .01). Prospective cohort study, Level III.2
[28] did not use device current medications medication IG: 0.7 ± 1.1 and Mean episodes of breakthrough bleeding IG:
at all through app CG: 1.45 ± 1.18 (p = .03) 1.17 ± 1.27, CG: 2.32 ± 2.03 (p = .03). Mean
number of clinic visits IG: 2.95 ± 1.36 and CG:
3.2 ± 1.2 (NS). Mean number of medications used
IG: 2.04 ± 0.82 CG: 2.14 ± 1.25 (NS)
Ernst et al. [33] IG: participants Participants responded to 75% WWE tracked their medi- 42/66 women studied achieved pregnancy Case series with pretest/posttest out-
tracking medicine daily “pop up” reminders cation use on >80% of days. comes, Level IV
use on >80% of from the app to indicate Adherence rate IG: 97.71%
enrolled days whether they “took it,” and CG: 99.84% (SNR)
using WEPOD. “missed it,” or “took
CG: participants extra”
tracking <80% of
enrolled days
Goldstein et al. IG1: had medication IG: self-report through IG adhered 76% and CG adhered NR RCT, Level II
[35] reminders through app where the recorded 80% of the time (SNR)
app. IG2: had medication-taking
smartphones with events were compared
no reminders. CG1: with the scheduled
had medication times. CG: pillbox
reminder (alarm) openings
through electronic
pillbox, CG2: used
electronic pillbox
as a passive ad-
herence monitor
(Continued )
SYSTEMATIC REVIEWS

page 5 of 17
Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019
 SYSTEMATIC REVIEWS

Table 1 | Continued

Intervention group
(IG) or comparator Method of adherence Adherence measure and change Clinical outcome/patient-related outcome measures Type of study, level of evidence as per
Study group (CG) measurement (significance) (significance) NHMRC hierarchy
Guo et al. [36] IG: mAF. CG: usual 3-item adherence esti- Nonadherence score at 1-month QOL score baseline IG: 86.5 and CG: 71.3. 1-month p-RCT, Level III-1
care mator score IG: 0 (low risk) and CG: 4 IG: 87.6 and CG: 70.1. 3-month IG: 87.2 and CG:
(moderate risk; p < .001). At 69.9 (all ps < .05)
3-month IG: 2 (moderate risk)
and CG: 4 (moderate risk; p
< .001)
Johnson et al. IG: users of MMH. Self-reported through app. Change in adherence from base- Increased self-efficacy IG: 0.2826 and CG: 0.0291 p-RCT, Level III-1
[30] CG: receive usual Participants respond to line IG: 0.611 and CG: −1.345 (p = .02). Increased quality of life IG: 0.5301 and
care and online the reminder whether (p = .01) CG: 0.0957 (p = .04). Positive correlation for ACT
educational ma- they are taking, skipping, change IG: 1.74 and CG: 1.65 (NS)
terials about or holding the dose
asthma medication
management
Kim et al. [40] IG: smartphone Self-assessed adherence Improved adherence score in Median QLQAKA IG: from 67 to 70 (p = .03) and CG RCT, Level II
application user score from 0 to 100 IG. Median change adherence from 69 to 72 (NS). Median FEV1 IG: from 93% to
group. CG: nonuser of medication IG: 100–100 90% (NS) and CG: 91% to 100 % (NS). Median
group (p = .02) and CG: 100–100 ACT scores IG: 22–21 (NS) and CG: 22–23 (NS)
(NS)
Kim et al. [39] IG: wireless moni- MMAS-8 Mean MMAS-8 scores IG: 6.6– Reduced cigarette smoking per day IG: 16.5–2.6 (p RCT, Level II
toring program 6.7 (NS) and CG: 6.3–6.5 < .001) and CG: 17.1–0.3 (NS). Reduced alcohol
and disease (NS) drinking IG: 7.2–7.6 (NS) and CG: 6.2–5.8 (NS).
management. CG: Decreased systolic BP IG: 136.1–133.4 (NS) and
standard disease CG: 145.9–140.2 (NS). Decrease diastolic BP IG:
management 86.3–82.8 (NS) and CG: 93.1–85.3 (p = .001)
program
Mertens et al. IG1: use the app A14-scale questionnaire Significant improvement in both NR RCT, Level II
[41] system. IG2: use IG1 and IG2 when compared
the paper diary. to CG. IG1: 53.96, IG2: 52.60,
CG: before the and CG: 50.02 (p < .001
study without as- for both IG1 and IG2 when
sistive systems compared to CG). Stronger
adherence for IG1 than IG2 in
documenting medication in-
take (p < .001)
(Continued )

page 6 of 17TBM
Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019
TBM
Table 1 | Continued

Intervention group
(IG) or comparator Method of adherence Adherence measure and change Clinical outcome/patient-related outcome measures Type of study, level of evidence as per
Study group (CG) measurement (significance) (significance) NHMRC hierarchy
Mira et al. [42] IG: tablet with MMAS-4 and number IG adherence increased by IG cholesterol level improved by 5%. Pre–post differ- RCT, Level II
the ALICE app of missed doses self- 28.3% and rate of missed ence in cholesterol IG: 5.7 and CG: −3.5 (p = .04).
installed and reported by patients dose fell by 27.3%. Pre–post Pre–post difference for glycated hemoglobin IG:
personalized ac- difference in MMAS-4 IG: 0.8 −0.4 and CG: 0.3 (NS). Pre–post difference for sys-
cording to their and CG: 0.1 (p < .001) tolic IG: 2.3 and CG: 3.2 (NS) and diastolic blood
prescribed medi- pressure IG: 1.7 and CG: 0.8 (NS). Pre–post differ-
cations. CG: oral ence for self-perceived health status IG: 3.3 and
and written infor- CG: 0.9 (NS)
mation regarding
the main risks
related to their
medication and
common errors of
taking medications
Walker et al. IG: post- MMAS-8 questionnaire MMAS-8 medication adherence NR Case series with pretest/posttest out-
[46] intervention. CG: was conducted during scale IG: 6.89 and CG: 6.44 comes, Level IV
pre-intervention initial enrollment visit (p = .10)
and over the phone after
3 months
ACT asthma control test; BP blood pressure; FEV1 forced expiratory volume in 1 s; HMB heavy menstrual bleeding; MMAS-4/8 4/8-item Morisky Medication Adherence Scale; MyCor/mAF/ALICE/MMH names of apps; NS nonsignificance; NR not reported; p-RCT pseudorandomized
controlled trial; QLQAKA quality of life questionnaire for adult Korean Asthmatics; QOL quality of life; RCT randomized controlled trial; SNR significance not reported; WWE women with epilepsy.
SYSTEMATIC REVIEWS

page 7 of 17
Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019
 SYSTEMATIC REVIEWS

Table 2. Outcomes of included publications using subjective and objective measures of adherence

Intervention group (IG) or Method of adherence Clinical outcome/patient-related outcome Type of study, level of evidence as per NHMRC
Study comparator group (CG) measurement Results (significance) measures (significance) hierarchy

Anglada- IG: used Medplan after CG SMAQ and PDC from phar- SMAQ: improved adherence, IG: 55.9%, CG: 36.5% (p Systolic BP IG: 131.3 and CG: 130.2 Case series with pretest/posttest outcomes,
Martinez for 3 months. CG: usual macy refills < .01). Decreased mean no. of days missed doses IG: (NS). Diastolic BP IG: 75.4 and CG: Level IV
et al. [32] care for 3 months 0.4 and CG: 3.5 (p = .02). PDC: no improvement in 79.9 (NS). Cholesterol IG: 147.2 and
adherence IG: 83.4 and CG: 85.8 (p = .25). Mean ad- CG: 207 (NS). Triglyceride IG: 185 and
herence rate measured using app: 58.4% CG: 263.5 (NS). EQ-5D questionnaire
viral load IG: <37 and CG: <37 (SNR)
Johnston et al. IG: interactive patient sup- MARS-5, pill count and No difference in MARS-5 mean score IG: 24.4 and CG: Increased number of quitters among ac- RCT, Level II
[38] port tool containing an self-reported drug 24.5 (NS). Pill count: no difference in adherence tive smokers IG: 16 and CG: 5 (NS).
extended drug adherence nonadherence score in (SNR) Self-reported mean nonadherence score was Increased median change in exercise
e-diary and secondary the app significantly lower in IG. Mean nonadherence score IG: minutes/week IG: +90 and CG: +65
prevention education 16.6 and CG: 22.8 (p = .03) (NS). Increased QOL measured by
modules. CG: simplified EQ-5D VAS IG: 14.7 and CG: 8.4 (NS)
e-diary only
Leonard et al. IG: used ITP for 6 months. MPR based on pharmacy MPR IG: 0.72 and CG: 0.65 (p = .28). Compliance based Serum ferritin level at 6 months Case series with pretest/posttest outcomes,
[47] CG: baseline without app refill rate and self-record on self-reported log at 6-month follow-up: 85% follow-up decreased by 434.1 ng/ Level IV
videos of daily adminis- mL (NS)
tration using the app
Patel et al. IG1: 3 months using medi- Pharmacy refill rates to cal- Significant PDC difference between IG1: 0.58, IG2: 0.46 Significantly higher baseline systolic BP. Interrupted time series without a parallel con-
[43] cation reminder appli- culate PDC and Morisky and CG: 0.54 (p = .003). Significant difference be- Baseline systolic BP 144/89: sig- trol group, Level III.3
cation. IG2: 3 months self-reported medication tween IG1 and IG2 (p < .001). Increased adherence nificantly higher than IG1 136/84
after withdrawal of the scale score between CG and IG2 (p = .06). MMAS-4 mean and (p = .04), IG2 135/85 (p = .01), and
medication reminder ap- median score baseline 2.4 and 2.0 increased to 3.2 CG 137/85 (p = .31)
plication. CG: 3 months and 3.0 at study completion, respectively (p < .001)
prior to study entry
Perera et al. IG: augmented version Self-report MARS-9, phar- IG significantly higher self-reported adherence (MARS-9 Detectable HIV viral load (>20 copies RCT, Level II
[44] containing components macy dispensing data score) to ART IG: 48.93 [95% CI = 48.36–49.50], of HIV RNA/mL) IG decreased from
that illustrate partici- and HIV viral load CG: 47.09 [95% CI = 44.60–49.58] (p = .03). baseline 26% to 7% (SNR); CG: from
pants’ current estimated Pharmacy dispensing data: no differences between baseline 18% to 37% (SNR)
plasma concentration IG and CG, IG: 100.00 [95% CI = 100.00–100.00],
of antiretroviral drugs CG: 93.21 [95% CI = 80.34–107.48] (p = .18). IG
and immune protec- HIV viral load significantly lower at 3-month follow-up
tion provided by ART. IG: 1.30 [95% CI = 1.29–1.31], CG: 1.70 [95%
CG: standard version of CI = 1.67–1.72] (p = .02)
smartphone application
ART antiretroviral treatment; BP blood pressure; CI confidence interval HIV human immunodeficiency virus; MARS medication adherence rating scale; Medplan/ITP name of the app; MMAS Morisky Medication Adherence Scale; MPR mean procession ratio; NS not significant; PDC proportion of days covered; QOL quality of life; RCT
randomized controlled trial; SMAQ Simplified Medication Adherence Questionnaire; SNR significance not reported; VAS European quality of life–5 dimensions visual analogue scale.

page 8 of 17TBM
Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019
TBM
Table 3. Outcomes of included publications using objective measures of adherence only

Intervention group (IG) or Method of adherence Clinical outcome/patient-related outcome Type of study, level of evidence as per
Study comparator group (CG) measurement Adherence results (significance) measures (significance) NHMRC hierarchy
Fenget al. [34] IG: receive WeChat Dose count based on the Absolute difference in mean ad- Endoscopic findings: only granulation score RCT, Level II
services. CG: did not re- returned medication herence rate between IG and in CG increased more than IG (p < .001);
ceive WeChat service bottle CG: 17.3% (p < .001) other changes were similar (NS); SNOT-
20 score: no difference between IG and
CG (NS)
Hammonds IG: app users. CG: app Manual pill count at the IG: 3.5 times more likely to be Depressive symptoms reduced, but the RCT, Level II
et al. [37] nonusers beginning of the study adherent. Rate of adherence IG: magnitude of change was not greater in
and 30 days later 76.5%, CG: 70.4% (p = .06) IG. Average BDI score for IG from 21.2
[95% CI = 0.945–12.966] to 14.5 (SNR) and CG: from 17.8 to
13.2 (SNR)
Hommel et al. IG: 4 weekly educational Pill count from patient’s No differences between IG and Pediatric Ulcerative Colitis Activity Index Case series with pretest/posttest out-
[29] sessions through prescription bottle CG for either drugs. Adherence and Partial Harvey-Bradshaw Index NR comes, Level IV
Skype. CG: baseline carried out telephone for mesalamine: 62% at for both measures
measurements at baseline and baseline to 91% after IG
post-treatment (p = .29). Adherence for 6-MP/
azathioprine: 61% at baseline to
53% after IG (p = 65)
Labovitz et al. IG: daily monitoring by AI Pill count, drug plasma Mean adherence through visual Activated partial thromboplastin time IG: RCT, level II
[11] platform. CG: no daily concentration, and confirmation IG: 90.5 (SNR). 41.7 and CG: 48.4 (SNR). Prothrombin
monitoring visual confirmation Mean cumulative adherence time IG: 35.1 and CG: 32.9 (SNR), INR
through app based on pill count IG: 97.2% IG: 3.4, and CG: 3.1 (SNR)
and CG: 90.6% (SNR). % pa-
tients with plasma samples
above minimum required thera-
peutic range IG: 100% and CG:
50% (SNR)
Stoner et al. IG: received SASED, re- MEMS, which records the Mean adherence at study mid- Craving intensity decreased from baseline RCT, Level II
[45] minders via SMS text time and date when- point IG: 83% and CG: 77% M = 3.8 through days 1–28 M = 2.0 (p
messages and a hyper- ever the lid is opened (p = .35). IG (M = 19 days [95% < .001) and days 29–56 M = 1.6 (p <
link to access adherence. CI = 0.0–44.0]) sustained ad- .001), drinks per drinking day decreased
CG: received SASED equate adherence significantly from baseline alcohol use M = 10.0
prompts/assessments, longer than CG (M = 3 days through days 1–28 M = 4.9 (p < .001)
but not adherence [95% CI = 0.0–8.1]) at mid- and days 29–56 M = 4.1(p < .001)
reminders/assessment study (p = .04), but not at study
end (p = .50)
BDI Beck Depression Inventory; CI confidence interval; INR international normalized ratio; MEMS Medication Event Monitoring System; 6-MP 6-mercaptopurine; NS not significant; NR not reported; RCT randomized controlled trial; SASED smartphone alcohol and side effect diary;
SYSTEMATIC REVIEWS

page 9 of 17
SNOT-20 SinoNasal Outcome Test-20; SNR significance not reported; WeChat/AI platform names of the app.
Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019
SYSTEMATIC REVIEWS

Measures of adherence the following results highlight those studies that re-
In each study, adherence was measured subjectively, ported changes from baseline for adherence and/or
objectively, or both, as illustrated in Tables 1, 2 and clinical outcome(s)/PROM(s) along with the statis-
3, respectively. The value of statistical significance is tical significance of changes.

Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019

reported in these tables only when significant. The Seventeen studies reported adherence measure(s)
values for nonstatistical significant results were rep- in both control and intervention groups along with
resented as NS (nonsignificant). Wherever possible, statistical significance of changes [28–30,32,34,36–
data presented in Tables 1, 2, and 3 are as detailed 47]. Twelve of 17 studies also reported measures of
as the data reported in the respective publications. both adherence and at least one clinical outcome/
Difference or lack of consistency in reporting in the PROM(s) in both control and intervention groups
tables presented in this review is due to differences along with the statistical significance of any changes
among publications, rather than omissions by the [28,30,32,34,36,38–40,42,43,45,47]. The remaining
authors of this systematic review. five studies reported statistical changes of adherence
Eleven studies measured adherence using sub- measure(s) only [29,37,41,44,46]. Three of the five
jective measures only (Table 1) through self-report studies did not report on a clinical outcome/PROM
measures and questionnaire [28,30,31,33,35,36,39– [29,41,46] whereas the other two studies did report
42,46]. Ten studies used a single subjective measure on the change in clinical outcome(s)/PROM(s) but
[28,30,31,33,35,36,39–41,46]. Only one study the change was not significant [37,44].
used two subjective measures, which involved Twelve of 17 studies demonstrated some im-
using the 4-item Morisky Medication Adherence provement [28,30,32,34,36,38,40–45]. Eight of
Scale (MMAS-4) questionnaire and asking partici- these showed significant improvement in all meas-
pants to self-report their number of missed doses ures of adherence [28,30,34,36,40–43] and four
[42]. Among the 10 studies, 4 studies involved self- other studies showed significant improvement in
reporting adherence through the app [28,31,33,35], at least one of the reported measure of adherence
1 study required participants to self-assess their own [32,38,44,45]. It was interesting to note that while
adherence at a point in time [40], 1 study required eight studies showed significant improvement in all
that participants self-report their adherence over adherence measures, seven of these studies demon-
7 days [30] and 4 studies used questionnaires to strated significant improvement in all adherence
measure adherence [36,39,41,46]. measures using subjective methods and only one
Five studies used objective measurements (Table study was able to show significant improvements
3) of adherence only [11,29,34,37,45]. Three of these in all adherence measures using subjective and
five studies used dose or pill counts [29,34,37]. One objective methods. Among the eight studies that
study reported several objective measures, namely showed significant improvement in all measures
a count of remaining doses available, a measure of of adherence, two studies by Guo et al. [36] and
plasma concentration, and an observation of the in- Patel et al. [43] demonstrated significant improve-
jection of the dose [11]. Only one study monitored ments in all measurements of adherence and clinical
adherence through a Medication Event Monitoring outcome/PROM outcomes. Guo et al. [36] dem-
System (MEMS) [45]. onstrated a significant improvement in subjective
The remaining five studies used a combination of adherence, using the 3-item adherence estimator
both subjective and objective measures (Table 2) of and a significant improvement in quality of life.
adherence [32,38,43,44,47]. In these studies, the sub- Patel et al. [43] demonstrated significant improve-
jective measure was self-reported either through an ment in adherence measured subjectively with the
app [47] or by questionnaire, namely the Medication MMAS-4 questionnaire and objectively using pro-
Adherence Rating Scale [38,44], the Morisky self- portion of days covered . Systolic blood pressure im-
reported medication scale [43], and the Simplified proved from baseline while using the app [43].
Medication Adherence Questionnaire [32]. For the
objective measure of medication adherence, three Risk of bias
studies used pharmacy refill rates or dispensing The risk of bias of the included studies was evalu-
records [43,44,47]. Two studies used pill counts ated systematically using the RoB 2 and ROBINS-I
as an objective measure of medication adherence tool and is reported in Tables A2 and A3, respect-
[32,38]. Specifically, the study conducted by Perera ively. Of the 21 included studies, 16 studies were
et al. [44] also assessed adherence to antiretroviral classified as some concerns or moderate risk of bias
therapy by measuring HIV viral load. [11,28,30–32,34,36–39,41,43–47]and 5 were classi-
fied as high or serious risk of bias [29,33,35,40,42].
The impact of using an app to promote adherence and im- There were several confounding factors and issues
prove health outcomes with study design and implementation which may
As per the inclusion and exclusion criteria, no study have contributed to risk of bias. Arguably the most
was excluded based on its rigor or quality. However, serious risk of bias identified was the way in which
to make accurate and comprehensive conclusions medication adherence was measured. For example,

page 10 of 17TBM
 SYSTEMATIC REVIEWS

self-reporting of adherence within an app introduces Another way bias could have been introduced
potential for bias because the measure of adher- was during the recruitment process. Studies that re-
ence to the medicine depends on the participant’s cruited participants from clinical settings may be ex-
adherence to the app, itself. Overestimation in self- posed to less risk of bias than those that recruited

Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019

reporting is common, for example, when comparing participants through electronic means. For example,
self-reports with MEMS data, an average of 30% sur- Goldstein et al. [35] reported recruiting participants
plus has been reported [49]. The risk of bias for one through electronic advertising, which may have led
RCT was classified as serious because of issues with to recruitment of participants that are more likely to
random allocation. Pre-intervention MMAS score be engaged with technology.
appeared to be significantly lower for intervention All articles provided descriptions of sources of
compared with comparator group [42]. Only two funding and role of potential funding. The study
studies reported on participants’ adherence to the conducted by Ernst et al. [33] was funded by the
features of the app [35,41], suggesting the potential app company itself and another study by Labovitz
for bias among studies that did not report on adher- et al. [11] had authors that were employers of the
ence to using the app itself. app company, introducing potential bias. Twelve
This systematic review also identified that there studies were identified that had received funding
are several issues to consider when designing the [28,30,31,35–39,41–43,45] but this funding was
comparator arm in a study aiming to measure the apparently received by a company or organization
impact of an app on medication adherence. For not related to the app under investigation. Seven
example, the comparator arm in some studies in- studies reported receiving no external funding
cluded usual treatment without the use of an app, [29,32,34,40,44,46,47].
whereas in others it included the use of an app with
minimal features [45]. The use of an app with min- DISCUSSION
imal features may be an appropriate and compar- To the best of our knowledge, this is the first review
able intervention for the comparator arm, in that it to systematically evaluate the evidence regarding
does not contain the features of the intervention app the efficacy of apps specifically in improving medi-
designed to support adherence. However, it may cation adherence. This review identified 21 publi-
introduce bias in favor of the comparator group. As cations that assessed the use of an app to improve
suggested by Goldstein et al. [35], giving the com- medication adherence. Seventeen studies provided
parator group an app with minimal features or even the results of statistical tests comparing the effect of
without features may improve adherence because it the app on medication adherence with a comparator
encourages self-monitoring. group. Twelve of these 17 studies demonstrated stat-
However, if the comparator group receives usual istically significant improvements in at least one
care only, then it is challenging to design a study measure of adherence [28,30,32,34,36,38,40–45],
whereby participants are blind to which group they of which 8 studies showed significant improvements
are allocated. This is due to the nature of the interven- in all measures of adherence [28,30,34,36,40–43]
tion whereby participants in the intervention group and 4 studies showed significant improvement in
are provided with an app or even a device containing at least one of the reported measures of adherence
an app, whereas participants in the control group do [32,38,44,45]. Nine of the 12 studies also measured
not receive this app or device [45]. The absence of the both adherence and clinical outcome(s)/PROM(s)
app or device may be noticeable to the control group, [28,30,32,34,36,40,42,43,45]. They reported a
thereby introducing bias, in that participants in both statistically significant improvement in at least one
groups may become aware of their allocation. medication adherence measure and one clinical out-
In general, the recruited sample size of the in- come/PROM. Two of these 12 studies demonstrated
cluded studies was small, ranging from 13 to 209 par- statistically significant improvements in all measures
ticipants, indicating that the results of these studies of adherence and clinical outcome(s)/PROM(s)
were potentially non-generalizable and should be [36,43]. This systematic review identified several
interpreted with caution. Another potential source challenges in synthesizing evidence about the im-
of bias was the time course of medicine therapy. It pact of apps on adherence and clinical outcomes/
is likely that the length of time a person has had a PROMs. These limiting factors may be summar-
disease and the variation in severity at different ized as the use of subjective measures for adherence
time points may influence a person’s adherence to without triangulation with objective measures and
a medicine(s) for that condition. There is evidence the limited use of high-quality study designs. Most
to suggest that interventions designed to promote studies were of small size and incorporated a range
disease management and adherence may have most of confounding factors leading to a moderate-to-
salience to patients who are new to therapy [45]. high risk of bias. Therefore, the findings of this sys-
Nonetheless, transparency regarding the recruit- tematic review suggest that using apps may enhance
ment process allows for consideration of these issues patient’s medication adherence, however, the level
when interpreting the findings. of evidence is not strong.

TBM page 11 of 17
SYSTEMATIC REVIEWS

The high level of variation in the efficacy of identi- [49]. A recent review highlighted that only 25% of
fied apps could have resulted from the heterogeneity studies evaluating text messaging interventions or
apps’ content and features. The wide variation in app mobile phone applications to enhance medication
features and content and the fact that the U.S. Food adherence used more than one validated method

Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019

and Drug Administration does not currently regu- to overcome the inaccuracy of current individual
late the quality of most medical apps highlights two approaches [55]. It should also be mentioned that
potential problems with the use of apps to support as evident in this review, using a combination of ad-
adherence [50]. First, the accuracy and quality of in- herence measures often results in conflicting results
formation provided within medical and health apps in adherence, necessitating some interpretation and
cannot easily be ascertained [51] and these factors explanation. It is noteworthy that this review iden-
are likely to affect medication adherence and, more tified that the subjective self-reporting of adherence
importantly, patient health outcomes. Second, apps was the most commonly used method, most likely
often involve communicating patient-specific data due to its convenience. However, when subjects self-
over the internet which raises the issue of patient report their medication adherence behavior they
privacy. Grundy et al. [51] reports that up to 80% of tend to underreport nonadherence due to social
mobile health apps transmit user-related information acquiescence and recall bias [56]. Furthermore, the
to online services and 66% of apps sent unencrypted validity and reliability of the questionnaires used to
identifying information over the Internet. The bene- measure medication adherence varies across settings.
fits of secure communication of information between For example, the MMAS is validated for chronic
health providers and patients cannot be ignored. In disease such as hypertension with 93% sensitivity
the past, SMS messaging using the telephone system and 53% specificity [49]. The Simplified Medication
has been shown to improve medication adherence Adherence Questionnaire is shown to have 72% sen-
[52]. Compared to messaging with SMS, however, sitivity and 91% specificity [57] and may only suitable
the messaging facilities within apps are more flex- for HIV-infected individuals [57] and kidney recipi-
ible and customizable [18] and may eventually re- ents [58]. It is important to explore the psychometric
place SMS messaging completely. Yet, a review from properties of measurement instruments to ensure
2015 reported that SMS reminders were still the they yield valid and reliable results [59].
most commonly used tool to improve adherence to Although using objective measures may be at
chronic disease management with 40.2% of studies face value more definitive than subjective meas-
incorporating SMS reminders as compared to only ures, objective methods also have their drawbacks.
23.4% of studies using an app [53]. Although this re- Objective measures including pharmacy refill rates,
view aimed to include studies over the last decade, pill counts, and MEMS assume that the medication
11 of the 21 studies included in this systematic review is taken by the patient [12]. For example, MEMS
were published within the last 2 years, reflecting the only provides an indication of bottle openings,
increased research interest in apps for medication ad- which does not necessarily mean that the medica-
herence, in recent years. Future research regarding tion was taken by the patient. MEMS, despite being
the safety of medical and health-related apps is re- one of the best measures of medication adherence
quired, as well as, formal guidance and regulations was used in only one included study [45]. Similarly,
for those designing and distributing these apps. with pharmacy refill rates, having the medicine
The fact that only 12 of 17 studies showed stat- dispensed does not ensure that it is actually used.
istically significant improvements in at least one Moreover, depending on the electronic monitoring
measure of adherence aligns with previous research or dispensing systems, refill rates may be incomplete
demonstrating the difficulty in improving adherence if the patient goes to several pharmacies.
regardless of the type of intervention [54]. It also In this systematic review, all included studies had
highlights that improving adherence to all health a relatively short duration of follow-up (less than
behaviors probably requires multiple interventions 12 months). Therefore, there is some degree of un-
targeting a range of factors. The involvement of certainty about the sustainability of improvements
health care providers in improving medication ad- in adherence. This is important because many
herence may also be necessary, as patients favor the chronic conditions require long-term, if not life-long
use of an app to communicate with their health care treatment. Well-designed studies are required to de-
provider through video chat, instant messaging, and termine whether any improvement is sustained or
e-mail functions [53]. whether the effect decays after initial enthusiasm for
A wide variety adherence measures were used the app. Future research in this area is paramount
within the studies and contributed to the challenge to determine whether the short-term impact of an
of interpreting the results and reporting on con- app translates into longer-term effects on adherence
clusive, generalizable findings. It is acknowledged and/or clinical outcomes/PROMs.
that there is no gold standard to measure adher- A wide range of apps are available to help im-
ence [12]. The WHO categorizes adherence meas- prove medication adherence, for various conditions,
ures broadly into subjective and objective measures including but not limited to cardiovascular disease

page 12 of 17TBM
 SYSTEMATIC REVIEWS

and alcohol use disorder. Even though these apps PROMs. Twelve of 17 included studies which re-
aim to improve medication adherence, the level of ported the results of statistical tests demonstrated
engagement with the app should also be measured improvements in adherence. However, the evidence
as a quality indicator, especially when self-reporting available is generally not of high quality and studies

Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019

through the app is used as the adherence measure. varied greatly in terms of their study design, app
A study showed that the effectiveness of non- features, and outcome measures. It is, therefore,
pharmacological interventions depends on the level difficult to definitively draw conclusions regarding
of adherence to the intervention, itself, meaning that the impact of apps on medication adherence and
the effectiveness of an app in improving medication whether improvements in adherence translate into
adherence depends on patients’ engagement with improvements in clinical outcomes/PROMs.
the app [60], highlighting the multifaceted nature
of adherence. It is interesting to note that 14 of 21 Acknowledgement: We would like to thank the Faculty Liaison Librarian,
Edward Luca, at The University of Sydney Medical Sciences Libraries. This
apps explored in this systematic review had a re- research did not receive any specific grant from funding agencies in the
minder feature, which is an approach that was intro- public, commercial, or not-for-profit sectors.
duced long before the creation of apps. It has been
suggested that mobile health apps create a digital Compliance with Ethical Standards
placebo effect whereby the beliefs about technology
Conflict of Interest: There are no conflicts of interest in authorship of this article.
and perceptions of being more connected to health
care providers through the app lead to clinical im-
provements [61]. If the act of using an app improves References
adherence, future studies could explore the effects
of such a placebo app [62].
1. Boutayeb A, Boutayeb S. The burden of non communicable diseases in
developing countries. Int J Equity Health. 2005;4(1):2.
Strengths and limitations 2. Lunenfeld B, Stratton P. The clinical consequences of an ageing world
and preventive strategies. Best Pract Res Clin Obstet Gynaecol.
The inclusion of all studies regardless of quality of 2013;27(5):643–659.
methodological design and quality of apps is both a 3. Armstrong BK, Gillespie JA, Leeder SR, Rubin GL, Russell LM.
strength and limitation. This allowed the authors to Challenges in health and health care for Australia. Med J Aust.
2007;187(9):485–489.
collate the most comprehensive evidence based on 4. Meetoo D. Chronic diseases: The silent global epidemic. Br J Nurs.
the current literature; however, it also posed limita- 2008;17(21):1320–1325.
5. Mondal MN, Shitan M. Factors affecting the HIV/AIDS epidemic: An eco-
tions when attempting to draw definitive conclusions logical analysis of global data. Afr Health Sci. 2013;13(2):301–310.
based on the findings. Furthermore, the heterogen- 6. Sabaté E. Adherence to Long-Term Therapies: Evidence for Action.
Geneva, Switzerland: World Health Organization; 2003.
eity among the studies, app features, adherence, and 7. Haynes R, McDonald HP, Garg A, Montague P. Interventions for helping
clinical outcome(s)/PROM(s) prevented direct com- patients to follow prescriptions for medications. Cochrane Database
parisons through meta-analysis. Syst Rev. 2002;(2):CD000011.
8. Chong WW, Aslani P, Chen TF. Effectiveness of interventions to improve
The exclusion of non-English publications may antidepressant medication adherence: A systematic review. Int J Clin
have contributed to language bias and ultimately Pract. 2011;65(9):954–975.
9. Polonsky WH, Henry RR. Poor medication adherence in type 2 diabetes:
the exclusion of potentially relevant articles pub- Recognizing the scope of the problem and its key contributors. Patient
lished in different languages. Prefer Adherence. 2016;2016(10):1299–1307.
10. Hugtenburg JG, Timmers L, Elders PJ, Vervloet M, van Dijk L. Definitions,
variants, and causes of nonadherence with medication: A challenge for
Implications tailored interventions. Patient Prefer Adherence. 2013;7:675–682.
11. Labovitz DL, Shafner L, Reyes Gil M, Virmani D, Hanina A. Using arti-
This systematic review provides a comprehensive ficial intelligence to reduce the risk of nonadherence in patients on
overview of apps developed to improve medication anticoagulation therapy. Stroke. 2017;48(5):1416–1419.
12. Baysari MT, Westbrook JI. Mobile applications for patient-centered care
adherence and the current evidence regarding their coordination: A review of human factors methods applied to their design,
impact on medication adherence. Although the in- development, and evaluation. Yearb Med Inform. 2015;10(1):47–54.
cluded studies in this systematic review provide 13. Kim SH, Gerver SM, Fidler S, Ward H. Adherence to antiretroviral therapy
in adolescents living with HIV: Systematic review and meta-analysis.
low-quality evidence, as shown in the risk of bias AIDS. 2014;28(13):1945–1956.
analyses, this review adds to the literature, in that 14. Náfrádi L, Galimberti E, Nakamoto K, Schulz PJ. Intentional and unin-
tentional medication non-adherence in hypertension: The role of health
it captures the current availability of apps, impact literacy, empowerment and medication beliefs. J Public Health Res.
of apps and quality of studies in this area. This sys- 2016;5(3):762.
15. Kvarnström K, Airaksinen M, Liira H. Barriers and facilitators to medica-
tematic review highlights avenues for much needed tion adherence: A qualitative study with general practitioners. BMJ Open.
further research in this area, especially as health 2018;8(1):e015332.
16. Jimmy B, Jose J. Patient medication adherence: Measures in daily prac-
technologies rapidly evolve. There is a need for fur- tice. Oman Med J. 2011;26(3):155–159.
ther RCTs with larger sample sizes to allow for a 17. Martin LR, Williams SL, Haskard KB, Dimatteo MR. The challenge of pa-
greater understanding of the effectiveness of apps in tient adherence. Ther Clin Risk Manag. 2005;1(3):189–199.
18. Dayer L, Heldenbrand S, Anderson P, Gubbins PO, Martin BC. Smartphone
improving medication adherence. medication adherence apps: Potential benefits to patients and providers.
J Am Pharm Assoc (2003). 2013;53(2):172–181.
19. Martin W, Sarro F, Jia Y, Zhang Y, Harman M. A survey of app store analysis
CONCLUSION for software engineering. IEEE Trans Softw Eng. 2017;43(9):817–847.
20. Muessig KE, Pike EC, Legrand S, Hightow-Weidman LB. Mobile phone
This systematic review explored the impact of apps applications for the care and prevention of HIV and other sexually trans-
on medication adherence and clinical outcomes/ mitted diseases: A review. J Med Internet Res. 2013;15(1):e1.

TBM page 13 of 17
SYSTEMATIC REVIEWS

21. mHealth App Economics 2017. Germany: Research 2 Guidance; 2017. 42. Mira JJ, Navarro I, Botella F, et al. A Spanish pillbox app for elderly pa-
22. Mendiola MF, Kalnicki M, Lindenauer S. Valuable features in mobile tients taking multiple medications: randomized controlled trial. J Med
health apps for patients and consumers: Content analysis of apps and Internet Res. 2014;16(4):e99.
user ratings. JMIR Mhealth Uhealth. 2015;3(2):e40. 43. Patel S, Jacobus-Kantor L, Marshall L, et al. Mobilizing your medications:
23. Choi A, Lovett AW, Kang J, Lee K, Choi L. Mobile applications to improve An automated medication reminder application for mobile phones and

Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019

medication adherence: Existing apps, quality of life and future directions. hypertension medication adherence in a high-risk urban population. J
Adv Pharmacol Pharmacy. 2015;3(3):64–74. Diabetes Sci Technol. 2013;7(3):630–639.
24. Santo K, Richtering SS, Chalmers J, Thiagalingam A, Chow CK, Redfern J. 44. Perera AI, Thomas MG, Moore JO, Faasse K, Petrie KJ. Effect of a smart-
Mobile phone apps to improve medication adherence: A systematic phone application incorporating personalized health-related imagery
stepwise process to identify high-quality apps. JMIR Mhealth Uhealth. on adherence to antiretroviral therapy: A randomized clinical trial. AIDS
2016;4(4):e132. Patient Care STDS. 2014;28(11):579–586.
25. Kim BY, Sharafoddini A, Tran N, Wen EY, Lee J. Consumer mobile apps 45. Stoner SA, Arenella PB, Hendershot CS. Randomized controlled trial of
for potential drug-drug interaction check: Systematic review and con- a mobile phone intervention for improving adherence to naltrexone for
tent analysis using the Mobile App Rating Scale (MARS). JMIR Mhealth alcohol use disorders. PLoS One. 2015;10(4):e0124613.
Uhealth. 2018;6(3):e74. 46. Walker CM, Elder BL, Hayes KS. The role of a self-directed technology
26. Higgins JPT, Sterne JAC, Savovic J, et al. A revised tool for assessing to improve medication adherence in heart failure patients. J Nurse Pract.
risk of bias in randomized trials. Cochrane Database of Syst Rev. 2014;10(10):856–863.
2016;10(suppl 1):29–31. 47. Leonard S, Anderson LM, Jonassaint J, Jonassaint C, Shah N. Utilizing a
27. Sterne JAC, Hernán MA, Reeves BC, et al. ROBINS-I: A tool for as- novel mobile health “selfie” application to improve compliance to iron
sessing risk of bias in non-randomised studies of interventions. BMJ. chelation in pediatric patients receiving chronic transfusions. J Pediatr
2016;355:i4919. Hematol Oncol. 2017;39(3):223–229.
28. Dietrich JE, Yee DL, Santos XM, et al. Assessment of an electronic inter- 48. NHMRC additional levels of evidence and grades for recommenda-
vention in young women with heavy menstrual bleeding. J Pediatr tions for developers of guidelines. 2009. Available at https://www.
Adolesc Gynecol. 2017;30(2):243–246. mja.com.au/sites/default/files/NHMRC.levels.of.evidence.2008-09.pdf.
29. Hommel KA, Hente E, Herzer M, Ingerski LM, Denson LA. Telehealth be- Accessibility verified May 14, 2019.
havioral treatment for medication nonadherence: A pilot and feasibility 49. Lam WY, Fresco P. Medication adherence measures: An overview.
study. Eur J Gastroenterol Hepatol. 2013;25(4):469–473. Biomed Res Int. 2015;2015:217047.
30. Johnson KB, Patterson BL, Ho YX, et al. The feasibility of text reminders 50. McCarthy M. FDA will not regulate most mobile medical apps. BMJ.
to improve medication adherence in adolescents with asthma. J Am Med 2013;347:f5841.
Inform Assoc. 2016;23(3):449–455. 51. Grundy QH, Wang Z, Bero LA. Challenges in assessing mobile health app
31. Ammenwerth E, Woess S, Baumgartner C, et al. Evaluation of an inte- quality: A systematic review of prevalent and innovative methods. Am J
grated telemonitoring surveillance system in patients with coronary Prev Med. 2016;51(6):1051–1059.
heart disease. Methods Inf Med. 2015;54(5):388–397. 52. Hall AK, Cole-Lewis H, Bernhardt JM. Mobile text messaging for
32. Anglada-Martínez H, Martin-Conde M, Rovira-Illamola M, et al. Feasibility health: A systematic review of reviews. Annu Rev Public Health.
and preliminary outcomes of a web and smartphone-based medica- 2015;36(1):393–415.
tion self-management platform for chronically ill patients. J Med Syst. 53. Hamine S, Gerth-Guyette E, Faulx D, Green BB, Ginsburg AS. Impact
2016;40(4):99. of mHealth chronic disease management on treatment adherence
33. Ernst LL, Harden CL, Pennell PB, et al. Medication adherence in and patient outcomes: A systematic review. J Med Internet Res.
women with epilepsy who are planning pregnancy. Epilepsia. 2015;17(2):e52.
2016;57(12):2039–2044. 54. Costa E, Giardini A, Savin M, et al. Interventional tools to improve
34. Feng S, Liang Z, Zhang R, et al. Effects of mobile phone WeChat services medication adherence: Review of literature. Patient Prefer Adherence.
improve adherence to corticosteroid nasal spray treatment for chronic 2015;9:1303–1314.
rhinosinusitis after functional endoscopic sinus surgery: A 3-month 55. Anglada‐Martinez H, Riu‐Viladoms G, Martin‐Conde M, Rovira‐
follow-up study. Eur Arch Otorhinolaryngol. 2017;274(3):1477–1485. Illamola M, Sotoca‐Momblona J, Codina‐Jane C. Does mHealth increase
35. Goldstein CM, Gathright EC, Dolansky MA, et al. Randomized controlled adherence to medication? Results of a systematic review. Int J Clin Pract.
feasibility trial of two telemedicine medication reminder systems for older 2015;69(1):9–32.
adults with heart failure. J Telemed Telecare. 2014;20(6):293–299. 56. Stirratt MJ, Dunbar-Jacob J, Crane HM, et al. Self-report measures of
36. Guo Y, Chen Y, Lane DA, Liu L, Wang Y, Lip GYH. Mobile health tech- medication adherence behavior: Recommendations on optimal use.
nology for atrial fibrillation management integrating decision sup- Transl Behav Med. 2015;5(4):470–482.
port, education, and patient involvement: mAF App trial. Am J Med. 57. Knobel H, Alonso J, Casado JL, et al.; GEEMA Study Group.
2017;130(12):1388–1396.e1386. Validation of a simplified medication adherence questionnaire in
37. Hammonds T, Rickert K, Goldstein C, et al. Adherence to antidepressant a large cohort of HIV-infected patients: The GEEMA Study. AIDS.
medications: A randomized controlled trial of medication reminding in 2002;16(4):605–613.
college students. J Am Coll Health. 2015;63(3):204–208. 58. Ortega FS, Sánchez JP, Pérez MV, Pereira PP, Muñoz MC, Lorenzo DA.
38. Johnston N, Bodegard J, Jerstrom S, et al. Effects of interactive patient smart- Validation on the simplified medication adherence questionnaire
phone support app on drug adherence and lifestyle changes in myocardial (SMAQ) in renal transplant patients on tacrolimus. Nefrologia.
infarction patients: A randomized study. Am Heart J. 2016;178:85–94. 2011;31(6):690–696.
39. Kim JY, Wineinger NE, Steinhubl SR. The influence of wireless self- 59. Mokkink LB, Terwee CB, Patrick DL, et al. The COSMIN checklist for as-
monitoring program on the relationship between patient activation and sessing the methodological quality of studies on measurement proper-
health behaviors, medication adherence, and blood pressure levels in ties of health status measurement instruments: An international Delphi
hypertensive patients: A substudy of a randomized controlled trial. J Med study. Qual Life Res. 2010;19(4):539–549.
Internet Res. 2016;18(6):e116. 60. Inouye SK, Bogardus ST, Jr, Williams CS, Leo-Summers L, Agostini JV.
40. Kim MY, Lee SY, Jo EJ, et al. Feasibility of a smartphone applica- The role of adherence on the effectiveness of nonpharmacologic inter-
tion based action plan and monitoring in asthma. Asia Pac Allergy. ventions: Evidence from the delirium prevention trial. Arch Intern Med.
2016;6(3):174–180. 2003;163(8):958–964.
41. Mertens A, Brandl C, Miron-Shatz T, et al. A mobile application improves 61. Torous J, Firth J. The digital placebo effect: Mobile mental health meets
therapy-adherence rates in elderly patients undergoing rehabilitation a clinical psychiatry. Lancet Psychiatry. 2016;3(2):100–102.
crossover design study comparing documentation via iPad with paper- 62. Anthes E. Mental health: There’s an app for that. Nature.
based control. Medicine (Baltimore). 2016;95(36):e4446. 2016;532(7597):20–23.

page 14 of 17TBM
TBM
APPENDIX

Table A1. General characteristics of included studies

Reference year, country Mobile application Medical condition Study design and study duration Number of participants (n =)a

Ammenwerth et al. 2015, MyCor Coronary heart disease Evaluation study. 4.5 months Recruited (25)
Austria[31]
Anglada-Martinez et al. 2016, Medplan Hypertension, dyslipidemia, heat Single arm prospective pre–post intervention study. 6 months Completed (42)
Spain[32] failure, HIV infection
Dietrich et al. 2017, America[28] iPeriod HMB, BD Prospective cohort study. 3 months IG (23) and CG (22)
Ernst et al. 2016, America[33] WEPOD app Epilepsy Four-center prospective observational study. 12 months if pregnancy not Analyzed (66)
achieved
Feng et al. 2016, China[34] WeChat Chronic rhinosinusitis Two-arm randomized, follow-up investigation. 90 days IG (16) and CG (13)
Goldstein et al. 2014, America[35] iRx Reminder LLC Systolic and diastolic HF Randomized controlled feasibility trial. 28 days IG (26) and CG (29)
Guo et al. 2017, China[36] mAF App AF Prospective cluster randomized design pilot study. 3 months IG (71) and CG (96)
Hammonds et al. 2015, NR Depression, anxiety, bipolar disorder, Open-label randomized parallel-group clinical trial. 1 month IG: (30) and CG: (27)
America[37] and some conditions NR
Hommel et al. 2013, America[29] Skype IBD Single-arm pilot and feasibility and clinical trial (pre–post intervention). Received treatment (9)
5 months
Johnson et al. 2015, America[30] MMH Asthma Block randomized controlled study. 3 weeks IG (46) and CG (43)
Johnston et al. 2016, Sweden[38] NR MI Multicenter, randomized study. 6 months IG (85) and CG (77)
Kim et al. 2016, America[39] HealthyCircle Hypertension Prospective, randomized controlled, Two-group, pre–post intervention. IG (52) and CG (43)
6 months
Kim et al. 2016, Korea[40] SnuCare Asthma Randomized study. 8 weeks IG (22) and CG (22)
Labovitz et al. 2017, America[11] AI platform Ischemic stroke Randomized parallel-group, controlled single-site study. 12 weeks IG (15) and CG (12)
Leonard et al. 2017, America[47] ITP mobile application β-thalassemia and SCD Pilot study with a pre–post design for comparison. 6 months Completed (10)
Mertens et al. 2016, Germany[41] Medication Plan Coronary heart disease Observational study with cross over design. 28 days Enrolled (24).
Mira et al. 2014, Spain[42] ALICE NR Single-blind RCT with two groups (control and experimental) and pre and IG (51) and CG (48)
post assessments. 3 months
Patel et al. 2013, America[43] Pill Phone application Hypertension Pilot and feasibility trial (pre–post, sequential design). 10 months Completed (46)
Perera et al. 2014, Auckland[44] NR HIV RCT. 3 months IG (16) and CG (11)
Stoner et al. 2015, America[45] NR Alcohol use disorder RCT. 8 weeks IG (17) and CG (20)
Walker et al. 2014, America[46] MedActionPlan HF Pre/post intervention exploratory design. 3 months IG (28) and CG (33)
AF atrial fibrillation; BD bleeding disorder; CG comparator group; HF heart failure; HIV human immunodeficiency virus; HMB heavy menstrual bleeding; IBD inflammatory bowel disease; IG intervention group; MI myocardial infarction; MMH: MyMediHealth (app); NR not reported; RCT randomized controlled
trial;; SCD sickle cell disease.
a
Where reported.
SYSTEMATIC REVIEWS

page 15 of 17
Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019
SYSTEMATIC REVIEWS

Table A2. Risk of bias of randomized controlled trials using RoB 2 tool

Study Overall risk of bias and direction Reason for assessment of risk other than low

Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019

Feng et al. [34] Some concerns +/− Some participants had missing data due to noncompliance
to treatment
Goldstein et al. [35] High risk +/− Adherence measured for the comparator arm was very dif-
ferent to intervention (pillbox bin openings)
No baseline adherence measure in IG and CG
Hammonds et al. [37] Some concerns +/− Method of randomization was not mentioned in the study
Johnston et al. [38] Some concerns +/− Pill count results (a prespecified analysis plan) were not re-
ported due to lack of participants
Kim et al. [40] High risk + Significant baseline differences between CG and IG that
could affect adherence results
The measurement of adherence by self-assessment is not a
validated tool
Kim et al. [39] Some concerns +/− This study used a subjective adherence measure as a sole
method of measuring adherence. Although participants
were blind to their assigned group before enrollment,
participants were aware of the intervention received.
Adherence results may be influenced by the knowledge
of intervention
Labovitz et al. [11] Some concerns +/− Statistical significance (p value) was not reported for any
adherence and/or clinical outcome
Mertens et al. [41] Some concerns +/− This study employs a crossover design where participants
experienced the interventional and comparative phases
alternatively. It is unclear whether participants had a suf-
ficient wash out period between two phases to prevent
carryover effects
Mira et al. [42] High risk + Random allocation. Pre-intervention MMAS score appeared
to be significantly lower for intervention compared with
comparator group (statistical significance of difference
not reported)
Perera et al. [44] Some concerns +/− No blinding of participants where participants’ assess-
ment could be influenced by their knowledge of the
intervention
Stoner et al. [45] Some concerns +/− High number of participants were lost to follow-up/
discontinued
+ indicates bias in the direction of intervention, − in the direction of comparator. CG comparator group; IG intervention group; MMAS Morisky Medication Adherence Scale.

page 16 of 17TBM
 SYSTEMATIC REVIEWS

Table A3. Risk of bias of pseudorandomized controlled trials and nonrandomized controlled trials using ROBINS-I tool

Reason for assessment of risk other than low (Pre-I:

Study Overall risk of bias and direction pre-intervention; I: intervention; Post-I: post-intervention)

Downloaded from https://academic.oup.com/tbm/advance-article-abstract/doi/10.1093/tbm/ibz125/5544100 by Nottingham Trent University user on 08 August 2019

Ammenwerth et al. [31] Moderate +/− I: No blinding of MyCor clinicians, documented data could
be accessed for discussion during follow-up
Anglada-Martinez et al. Moderate + Pre-I: Population limited to owning a smartphone
[32] I: No blinding of health providers
Ernst et al. [33] Serious + I: Participants were given the option to use paper diaries
(control), and one participant chose this option
Guo et al. [36] Moderate + I: No blinding of health providers, leading to a preference
of using non-vitamin K anticoagulants in the interven-
tion group
I: Confounding due to the educational program as a
co-intervention in the intervention group
Hommel et al. [29] Serious +/− Post-I: The measurements for disease severity that were
set as outcome measures were not reported at the end
of the study
Dietrich et al. [28] Moderate + I: IG participants were allowed to explore BD websites,
whereas CG participants did not have access
Johnson et al. [30] Moderate − I: CG received additional online education other than usual
care
I: No blinding of health provider
Leonard et al. [47] Moderate + I: No blinding of health providers
Patel et al. [43] Moderate + I: Effects of patient education from the nurses in the inter-
vention group
Walker et al. [46] Moderate + Pre-I: Participants were purposively selected from a popu-
lation (Heart Failure Disease Management Program)
I: No blinding of health providers
+ indicates bias in the direction of intervention, − in the direction of comparator. BD bleeding disorder; CG comparator group; IG intervention group.

TBM page 17 of 17