Improving Medication Adherence in Bipolar Disorder - A Systematic Review and Meta-Analysis of 30 Years of Intervention Trials

Journal of Affective Disorders 194 (2016) 202–221
Contents lists available at ScienceDirect
Journal of Affective Disorders

journal homepage: www.elsevier.com/locate/jad
Review article
Improving medication adherence in bipolar disorder: A systematic

review and meta-analysis of 30 years of intervention trials
Lindsay MacDonald a, Sarah Chapman a, Michel Syrett b, Richard Bowskill c, Rob Horne a,n
a
Centre for Behavioural Medicine, UCL School of Pharmacy, UK
b
The Roffey Park Institute & Lancaster University (Spectrum Centre for Mental Health Research), UK
c
Brighton and Sussex Medical School, UK & Sussex Partnership NHS Foundation Trust, UK
art ic l e i nf o a b s t r a c t
Article history: Background: Medication non-adherence in bipolar disorder is a significant problem resulting in in-
Received 25 February 2015 creased morbidity, hospitalisation and suicide. Interventions to enhance adherence exist but it is not
Received in revised form clear how effective they are, or what works and why.
27 November 2015
Methods: We systematically searched bibliographic databases for RCTs of interventions to support ad-
Accepted 4 January 2016
herence to medication in bipolar disorder. Study selection and data extraction was performed by two
Available online 20 January 2016
investigators. Data was extracted on intervention design and delivery, study characteristics, adherence
Keywords: outcomes and study quality. The meta-analysis used pooled odds ratios for adherence using random
Bipolar disorder effects models.
Adherence
Results: Searches identified 795 studies, of which 24 met the inclusion criteria, 18 provided sufficient
Interventions
data for meta-analysis. The pooled OR was 2.27 (95% CI 1.45–3.56) equivalent to a two-fold increase in
Meta-analysis
the odds of adherence in the intervention group relative to control. Smaller effects were seen where the
control group consisted of an active comparison and with increasing intervention length. The effects
were robust across other factors of intervention and study design and delivery.
Limitations: Many studies did not report sufficient information to classify intervention design and de-
livery or judge quality and the interventions were highly variable. Therefore, the scope of moderation
analysis was limited.
Conclusions: Even brief interventions can improve medication adherence. Limitations in intervention
and study design and reporting prevented assessment of which elements of adherence support are most
effective. Applying published guidance and quality criteria for designing and reporting adherence in-
terventions is a priority to inform the implementation of cost-effective adherence support.
& 2016 Elsevier B.V. All rights reserved.
Contents
1. Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
2. Objectives. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
3. Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
3.1. Eligibility criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
3.2. Identification of studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
3.3. Selection of trials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
3.4. Data extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
3.5. Data analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
4.1. Intervention content and tailoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
4.2. Delivery channel (provider, duration, frequency of sessions) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
4.3. Control group. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
n
Correspondence to: Centre for Behavioural Medicine, UCL School of Pharmacy,
BMA/Tavistock House, Tavistock Sq., London WC1H 9JP, UK.
E-mail address: r.horne@ucl.ac.uk (R. Horne).
http://dx.doi.org/10.1016/j.jad.2016.01.002
0165-0327/& 2016 Elsevier B.V. All rights reserved.
Downloaded for Anonymous User (n/a) at Rutgers University - NERL from ClinicalKey.com by Elsevier on April 02, 2018.
For personal use only. No other uses without permission. Copyright ©2018. Elsevier Inc. All rights reserved.
L. MacDonald et al. / Journal of Affective Disorders 194 (2016) 202–221 203
4.4. Study and sample characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214

4.5. Adherence assessment and primary outcome . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
4.6. Quality indicators (retention, fidelity, intervention description, study design – risk of bias) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
4.7. Meta-analysis results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
4.7.1. Moderation analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
4.7.2. Sensitivity analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
4.7.3. Publication bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
5. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
5.1. Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
Funding sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
Conflicts of interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
Appendix A. Supplementary material. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
1. Background et al., 2011) and whether it is educational or includes beliefs and

cognitions (Horne et al., 2005; Sajatovic et al., 2004; Gaudiano
Bipolar disorder is among the leading causes of global disability et al., 2008; Leclerc et al., 2013; Desplenter et al., 2006; Fernandez
(Mathers et al., 2008), estimated to affect three million people in et al., 2006). Targeting and delivery variables which we explore
Europe (Wittchen et al., 2011). As a serious, long-term condition, include involving family members (Gaudiano et al., 2008) and
treatment guidelines recommend medication to improve func- patients' illness stage (Lolich et al., 2012; Berk et al., 2010; Rouget
tioning and reduce relapse risk (NICE, 2014). Not unlike other and Aubry, 2007). In addition to content and delivery, process
long-term conditions, non-adherence in bipolar disorder is a variables are key to assessing how interventions were im-
concern, with an estimated 41% of individuals not taking long- plemented such as fidelity (Craig et al., 2008) and content of
term prophylactic medication as prescribed (Lingam and Scott, standard care (De Bruin et al., 2009; Gaudiano et al., 2008).
2002). Discontinuation of medication can have immediate con- We use published guidelines on reporting behavioural inter-
sequences in terms of relapse (Cavanagh et al., 2004) and non- ventions to systematically describe content and delivery (Davidson
adherence is associated with substantial costs, including hospita- et al., 2003; Hoffmann et al., 2014) as well as guidance from the
lisation, suicide and loss of productivity (Hong et al., 2011; Scott Consolidated Standards of Reporting Trials (CONSORT) (Schulz
and Pope, 2002; Baldessarini et al., 2006). Given these high costs, et al., 2010; Boutron et al., 2008) to identify areas where inter-
it is imperative that effective adherence interventions are devel- vention trials are poorly reported. We identify details about the
oped and utilised in clinical care. interventions themselves, specifically; what was delivered, how,
Programmes to improve outcomes in bipolar disorder have and by whom. This review also provides an assessment of meth-
been developed, many explicitly incorporating medication ad- odological quality and reporting of both the studies and the in-
herence (Colom and Vieta, 2006; Basco and Rush, 2005), however terventions (Higgins et al., 2011; Hoffmann et al., 2014) which has
systematic reviews of intervention trials to date have limitations not been included in previous reviews (Crowe et al., 2012; Depp
which reduce their use in identifying effective techniques to im- et al., 2008; Gaudiano et al., 2008; Sajatovic et al., 2004). This will
prove medication adherence. Some focussed only on psychoedu- go some way to identifying, not just which programmes are ef-
cational or psychosocial interventions (Batista et al., 2011; Berk fective, but what characteristics would be of particular interest for
et al., 2010; Crowe et al., 2012; Depp et al., 2008; Gaudiano et al., development in the ongoing endeavour to enhance medication
2008), while others did not examine adherence outcomes (Lolich adherence and improve outcomes for individuals with bipolar
et al., 2012), provide only a narrative review (Busby and Sajatovic, disorder, and limitations of the existing literature.
2010; Colom and Lam, 2005; Miklowitz, 2006), or were not spe-
cific to bipolar disorder (Desplenter et al., 2006; Fernandez et al.,
2006). Together these provide a limited picture of the adherence 2. Objectives
intervention literature in bipolar disorder, as they do not system-
atically quantify the magnitude of intervention effects across stu- To review randomised controlled trials of interventions in-
dies, or test what components of study and intervention design corporating medication adherence for people with a diagnosis of
may be most effective. The effects of inadequately powered studies bipolar disorder. To describe intervention design and delivery,
may also be over-estimated in narrative reviews. Meta-analysis study characteristics, and to evaluate quality of reporting and trial
provides a quantification of effect and we can test whether par- design. To quantitatively synthesise evidence for the effectiveness
ticular components may significantly contribute to the effect. This of interventions relative to control conditions.
review therefore provides a systematic assessment and meta-
analysis which is crucial in order to recommend what clinically
effective, cost and time efficient interventions should be taken 3. Methods
forward and used in clinical practice. The intervention and study
components tested reflect current thinking on promising areas for We conducted a comprehensive search of published literature
improving adherence. Certain intervention features may be im- until the end of October 2012 through abstract databases, clinical
portant, for example; tailoring (Horne et al., 2005; Berk et al., trial registers, hand-searching of citations in previous reviews and
2010; Desplenter et al., 2006; Crowe et al., 2012), content and included studies and by contacting researchers directly. In order to
delivery (Abc Project Team, 2013) the adherence focus, programme ensure the review was current, the original search criteria were re-
length (Gaudiano et al., 2008; Rouget and Aubry, 2007; Batista run in August 2014. We followed the PRISMA statement for
204 L. MacDonald et al. / Journal of Affective Disorders 194 (2016) 202–221
reporting systematic reviews and meta-analyses (Moher et al., psychosis, schizoaffective disorders or a diagnosed substance
2009) (Appendix S1 PRISMA checklist). misuse problem, or who are undergoing compulsory treatment.
3.1. Eligibility criteria 3.2. Identification of studies
Type of studies: Randomised controlled trials of interventions We searched the following databases; CINAHL, EMBASE, Psy-
incorporating medication adherence either directly or indirectly chInfo, PubMed, Sociological abstracts and Cochrane trials, no date
and comparing the intervention/s to an active or passive control or language restrictions were imposed (Appendix S2 for example
(including trials where patients were described as being randomly search strategy). We designed and tested search strategies with
allocated to condition). the assistance of a research librarian.
Participants: Adults over 18 years of age with a diagnosis of Search terms to identify studies were; Adheren*, complian*,
bipolar disorder (all types), or with general psychiatric populations medicat*, medicine, drug, clinical trial, random*, control*. For the
when subgroup analysis was provided for bipolar disorder. Cochrane trials database the following terms were included;
Types of outcome measures: Medication adherence as either a pharmacotherapy or regimen* or educat*. We combined this with
primary outcome, secondary outcome or mediator, measured by search terms; Bipolar disorder (MESH), Manic depressi*). We
subjective or objective methods. subsequently searched the trials registers Clinicaltrials.gov and
Excluded: Studies focussing exclusively on individuals with HTA for trials including 'bipolar disorder’. Finally, we reviewed the
Fig. 1. PRISMA flow chart of selection of studies for the systematic review and meta-analysis.
Table 1a
Summary of included studies – Adherence specific interventions.
Study Country Intervention Control description Intervention Participants Follow-up after Adherence a Adherence Adherence results
retention last interven- primary measurement
tion contact outcome?
Cochran (1984). US Modified CBT delivered in- TAU (affective disorders 86% participants 38 pts (IG ¼ 20, CG¼ 6 months Yes Author designed self Self-report & Informant
dividually by Psychologists clinic, inc. brief medica- completed full 18) with current li- report, informant re- report: no significant
weekly for 6 weeks. tion visits weekly or 6 weeks. thium prescription. port, and Physician differences between IG
bimonthly) Mean age 32, 61% report scales. and CG at post, 3 months
female, 21% married. or 6 months.
Serum lithium levels. Physician report: Sig-

nificant difference be-
tween IG and CG at post-
intervention and
6 months (n/s at
3 months).
L. MacDonald et al. / Journal of Affective Disorders 194 (2016) 202–221

Compliance index Serum levels: n/s at post-
intervention and
3 months. Levels mostly
unavailable at 6 months.
Compliance index: Sig-
nificant difference be-
tween IG and CG at post-
intervention and
6 months (n/s at
3 months). IG less likely
to have major com-
pliance problems.
Dogan and Sa- Turkey Education sessions delivered TAU (no details specified) Not specified 32 pts (IG ¼ 14, 2 months Yes Serum lithium levels Significant difference be-
banciogullari individually by Psychiatric CG¼ 12) long-term tween IG and CG in
(2003). nurses for 2 sessions followed lithium users. Mean baseline to 3 month
by one group session. age 38, 35% female, change in proportion of
73% married. pts low to normal serum
levels. Higher proportion
in IG moving to ‘normal’
range levels than in CG.
Elixhauser et al. US Use of an electronic ad- TAU (psychiatric out- Not specified 93 pts (IG ¼ 42, 3 months Yes Self-report (Morisky) Self report scale mean &
(1990). herence monitoring device patient clinic, usual mail CG¼ 51) with cur- Serum lithium % reporting no missed
and feedback for individuals refill of lithium rent lithium pre- levels. doses – n/s effect of in-
on adherence. Provision of prescriptions) scription. Mean age Prescription refills tervention (monitoring
adherence education for non- þ 49, gender break- (% obtained). or feedback).
adherent individuals by Feedback on com- down not reported. Medication taking Serum lithium – n/s ef-
phone or mail. pliance (from lithium patterns-Daily and fect of intervention
levels) and suggestions period pill counts (% (monitoring or feedback).
for improving com- prescribed doses) Prescription refills (%) –
pliance at visit 2. significant effect of
monitoring with higher
& refill in intervention
group. n/s effect of
feedback.
Harvey and Peet UK Educational video on lithium TAU (not specified) 97% attended IG 60 pts (IG ¼ 30, 5 months Yes Red blood cell lithium Lithium RBC, serum le-
(1991). with illustrated transcript þ sessions CG¼ 30) in remis- levels during fixed- vels and lithium ratio: No
205
viewed by participants Educational sion attending dose regimen. significant difference
citations of previous reviews and studies included in this review
greater reduction in mis-
(n¼ 62) 83.87 (28.66), CG
6 months I (n ¼51) 90.20

sed doses in the IG com-
(22.40), C (n¼ 55) 77.27

significant difference in
(n¼ 41) 95.73 (11.04), C

between change scores
for additional pertinent articles.
Lithium missed days:
(n¼ 39) 81.08 (30.85)

change scores with a
(n¼ 61) 81.15 (32.49)
(35.12) 12 months I
SRTAB 3 months IG
Adherence results
3.3. Selection of trials

for IG and CG.
pared to CG
We downloaded the results of the search strategies into a re-
p ¼ .41.
ferencing management software package before removing dupli-
cates (Endnote x3) and cross-checking the additional searches
with the downloaded citations.
viours SRTAB-self-re-
warning with spouse
doses/ days in each
Two investigators (LM & SC) carried out title and abstract re-
Estimate of missed
days, 0%, 25%, 50%,

port over past 30
Adherence beha-
checked without
view on the 671 studies retrieved and excluded studies if they did
6 week period,
measurement
not meet the inclusion criteria, inter-rater agreement was high

Adherence
75% 100%.
(kappa ¼ 0.99). Four potentially relevant papers, identified by SC
or other.
were excluded after discussion (Valenstein et al., 2011; Colom

et al., 2009; Frank et al., 2008; Simon et al., 2006). We retrieved
papers for the remaining 116 studies and reviewed the full text
Adherence a
articles (Fig. 1: PRISMA flow diagram) (Moher et al., 2009).

outcome?
primary
Twenty-three studies met the inclusion criteria (Tables 1a and b).

Data was extracted from 26 papers as four studies reported
Yes
methods and results across two papers each (accounts for where
Follow-up after
the number of citations is higher than the number of studies

last interven-
10 ½ months
tion contact
described).
The search update in August 2014 retrieved an additional 125
studies which were reviewed by the same investigators and five
potentially relevant articles were retrieved. One study was in-
cluded in the review (Javadpour et al., 2013), the remaining four
female, 13% married.
lithium clinic. Mean
were excluded as they either comprised conference abstracts

CG¼ 80) in remis-
Mean age 40, 68%

sion or episodic.
164 pts (IG ¼ 84,
where the data was not available from authors or studies with
age 55 m 67%
Participants
mixed psychiatric samples for which sufficient detail for inclusion

was not available.
female.
3.4. Data extraction

least 4 sessions. 37%
sions, less than 10%

participated in op-
completed no ses-
We designed and piloted a data extraction template before

49% completed at
tional sessions.
making refinements. We extracted data on intervention design

Intervention
and delivery, study characteristics, quality of reporting and trial

retention
design and adherence outcomes.

Intervention design and delivery: The term ‘Psychoeducation’
was applied to heterogeneous interventions. Colom provides a
broad definition; including empowering patients and giving them
TAU (community mental
therapy, counselling, ac-

health centre care, typi-
cally medication man-
agement, psychosocial
the tools to manage, cope and live with bipolar disorder, involving
Control description
intervention at 12
adherence enhancement, early warning sign identification, lifestyle,

cess to intensive
crisis management and communication. It involves changing be-

haviour and attitudes in addition to simply providing information
assistance.
(Colom, 2011). We coded interventions in terms of whether they

weeks.
used the term psychoeducation and whether the description of the

programme included all of these elements of psychoeducation.
Life Goals Programme (LGP) –
health nurse and ‘psychiatric
Other intervention description characteristics were; whether ad-

followed by individual home
monthly unstructured group

Group psychoeducation ses-
counsellor’ in 6 weekly ses-

sions delivered by a mental
sions. Followed by optional

visit after 2 weeks for dis-
cussion with Psychiatrists.
herence was the primary focus of the intervention, provider, format

(i.e. group, individual, family), delivery setting (e.g. specialist af-
fective disorder/ bipolar clinic or unit, university), recipient, number
of intervention contacts, total intervention contact time, duration of
Country Intervention
intervention delivery, intensity of intervention contacts. Guidelines

sessions.
on reporting behavioural intervention trials were used to sum-

marise the descriptions of interventions (Davidson et al., 2003;
Hoffmann et al., 2014). Where the number of intervention contacts,
time or duration varied (as specified by protocol), we used the
average. In coding intervention duration in months, one-contact
US
interventions were classified as 0.03 months, in order that they had

a quantifiable duration for computation of medians.
Table 1a (continued )
Intervention tailoring: categorised (‘yes’, ‘no’ or ‘unclear’) ac-

Peet and Harvey
Sajatovic et al.
cording to inclusion of different forms of tailoring based on

guidelines described by de Bruin and colleagues in the ‘Coding
(2009).
(1991)
manual for Behaviour change techniques’: Individualization (con-

Study
tent individualised to reflect participants' questions or individual

progress), Macro-tailoring (intervention depends on pre-tested
Table 1b
Summary of included studies – multi-focus interventions.
Study Country Intervention Control description Intervention Participants Follow-up Adherence a Adherence measurement Adherence Results
retention after last in- primary
tervention outcome?
contact
Multi-focus interventions (Individual/ couples/ family)

Ball et al. Australia Modular cognitive ther- TAU (sessions as required Not specified 52 pts (IG ¼ 25, 12 months No Serum concentration. Self-report; No sig-
(2006). apy programme in- with GP/ psychiatrist. CG ¼27) with at Self-report (reporting occa- nificant difference be-
corporating Psychoedu- þ least 1 bipolar epi- sions of missing tween proportion with
cation and CBT techni- Clinicians provided with sode in last 18 medication) adequate compliance
ques. Delivered by Psy- educational package on BD months. Mean age in IG and CG, either
chologists over 20 & mood monitoring. 42, 58% female. post-treatment or 12
weekly, individual months.
sessions. Serum levels not re-
ported due to low at-

tendance for
measurement.
Clarkin US Structured Psychoedu- TAU (medication management Not specified 42 pts (IG ¼ 19, 0 (post- Not specified Author developed adherence Significant difference
et al. cation intervention for as part of usual inpatient and CG ¼ 23) married or intervention) scale between IG and CG in
(1998). patients and their spou- outpatient care) living with sig adherence scores at
ses delivered by Social other 4 6 months. post-intervention fol-
workers in 25 sessions Mean age 47, 46% low-up.
over 11 months. female. (nb missing data for 1/
3 CG)
Frank et al US Interpersonal and social Intensive clinical management Not reported for each 82 pts (IG ¼ 60, 0 (post- No Adherence – ratio of blood le- No significant differ-
(1999). rhythm therapy (IPSRT) (sessions focus on education, treatment. 65% overall CG ¼22) in at least intervention) vel to dose prescribed. ence between four dif-
– Psychotherapy and adherence & side-effects), de- entered preventative 3rd discrete affec- ferent treatment stra-
psychoeducation with a livered weekly for min 12 phase. tive episode. Mean tegies in blood levels
focus on life events and weeks, then monthly for age 36, 66% female, and no effect of chan-
social rhythms. In- 2 years. 37% married. ging treatment
dividually delivered strategy.
with ancillary family
education sessions.
IPSRT delivered by
'therapist' and physician
weekly for min 12
weeks, then monthly for
2 years.
Frank et al. US IPSRT delivered by non- Intensive clinical management Not reported for each 175 pts (IG ¼ 132, 0 (post- No Serum levels - coefficient of No significant differ-
(2005). physician clinician (so- (sessions focus on education, treatment. 70% reten- CG ¼43) in at least intervention) variation of mood stabilizer ence between four dif-
cial worker, nurse, or adherence & side-effects), de- tion rate at end of 3rd discrete affec- medication. ferent treatment stra-
psychologist) and a psy- livered weekly for min 12 stabilisation. 53% of tive episode. Mean tegies in coefficient of
chiatrist. Sessions deliv- weeks, then monthly for originally randomised age 35, 57% female, variation of mood sta-
ered weekly until stabi- 2 years. completed 2 year 35% married. bilizer serum levels.
lisation, every other maintenance.
week for 12 weeks
(preventative phase)
and monthly until the
end of the 2-year
(maintenance phase).
207
208
Table 1b (continued )
tervention outcome?
contact
Gilbert US Family Focussed Ther- Individual patient manage- Not specified 53 pts (IG ¼ 19, 12 months No Adherence scale (7 point), No significant effect of
(2000). apy (FFT) involving psy- ment (30 min sessions inc CG ¼18) with manic psychiatrist rated for first 12 group for any time
choeducation, commu- support, problem solving and episodes with con- months, then self report at 24 point.
nication enhancement education, 1–2 family educa- senting family months. Study reported that
training and problem- tion sessions & crisis interven- member. Mean age there was lower ad-
solving skills training. tion as needed. 25, 57% female, 14% herence at 2 years
Sessions delivered to fa- married. than at 1 year follow-

milies by therapists in up.
22 sessions delivered
over 12 months Weekly
for first 3 months, bi-

weekly for 3 months,
monthly for 3 months.
Lam et al. UK Cognitive Therapy (CT) TAU (routine outpatient and Flexible number of 25 pts (IG ¼ 13, 6 months No Medication Compliance Ques- Significant difference
(2000). including Psychoeduca- appropriate MDT input judged sessions - average CG ¼12) experien- tionnaire (MCQ) – self report between mean scores
tion and CBT delivered by clinical team.) sessions 16.3(3.2). 92% cing at least 2 epi- over previous month. Never over 12 months with
by psychologists to in- completed at least sodes in previous misses, missed once or twice, better adherence in IG.
dividuals over 12–20 4 sessions. 2 yrs, or 3 episodes missed 3–7 times, missed
sessions over 6 months. in 5 yrs, but not more than 7, stopped
currently episodic. altogether.
Mean age 39, 52%
female, 32% mar-
ried/cohabiting.
Lam et al., UK Cognitive Therapy (CT) TAU (psychopharmacology & Flexible number of 103 pts (IG ¼51, 2 yrs No Medication Compliance Ques- Self report (MCQ):
2003, including Psychoeduca- regular psychiatric outpatient sessions. 84% retention CG ¼52) experien- tionnaire (MCQ) Medication significant difference
2005 tion and CBT delivered appointments). rate to 6 sessions. cing at least 2 epi- serum levels at 6 months between
to individuals by Psy- Average sessions com- sodes in previous IG and CG with higher
chologists over 12–18 pleted 13.9 (5.5). 2 yrs, or 3 episodes proportion of IG cate-
sessions for the first in 5 yrs, but not gorised as having
6 months and 2 booster currently episodic. ‘good compliance’.
sessions in the second Mean age 44, 56% Self report (MCQ
6 months. female. mean): significant dif-
ference between IG
and CG at 24 months
and 30 months with IG
group reporting better
adherence (n/s differ-
ence at 18 month fol-
low-up).
Serum levels: n/s dif-
ference between pro-
portion classified as
adequate in IG and CG
at 6 months (p ¼ .06).
Javadpour Iran Psychoeducation invol- TAU (standard pharmacother- 89% participated in at 108 pts (IG ¼54, 0 months (post Yes Medication Adherence Rating Significant difference
et al. ving 8 individually de- apy with own psychiatrist) least 4 sessions. Aver- CG ¼54) currently final telephone Scale (MARS) (Thompson between IG and CG
(2013). livered by psychiatrists age 7.3 sessions and euthymic. Mean age follow-up) 2000) 10 items at 6, 12 and 18 over follow-up period
in weekly sessions and 15.3 telephone not specified, 41% months (p ¼0.008) with higher
follow-up phone calls sessions female, 20% adherence in the IG.
for 18 months. married.
Miklowitz US Family Focussed Ther- Crisis management (2 1 hr 10% of IG withdrew 101 pts (IG ¼ 31, 15 months No Patient reported com- Compliance score – n/s
et al. apy (FFT) involving psy- home-based family education before 6 months. CG ¼70) with re- pliance – Checked against group, time or interac-
(2000, choeducation, commu- sessions, crisis intervention as cent bipolar epi- physician and family report. tion effects at 12
2003) nication enhancement needed (9 months), relapse sode (within Serum levels. months.
training and problem- prevention and resolution of 3 months). Mean Compliance (composite)
solving skills training. family conflicts, telephone age 36, 63% female, 3 point rating from other
Sessions delivered at support/ monitoring contact 60% married/ reports; fully nonadherent,
home to families by once per month. cohabiting. partially nonadherent, fully
therapists in 21 sessions adherent.
delivered over 9 months

12 weekly, then bi-
weekly, then 3 monthly.
Rea et al. US Family Focussed Ther- TAU (Weekly medication 79% of IG completed 53 pts (IG ¼ 28, 15 months No Medication compliance - Psy- Compliance score - n/s

(2003). apy (FFT) involving psy- management sessions with full 9 months of CG ¼25) recently chiatrist rating using standar- group, time or interac-
choeducation, commu- psychiatrist for 1 year, then treatment. hospitalised for dised form 7 point scale 1(full tion effects at 12
nication enhancement every two weeks then mania episode. compliance 7(discontinued months. n/s difference
training and problem- monthly. Mean age 26, 57% against medical advice). Com- between groups in
solving skills training. þ female, 15% posed of pt report, psychiatrist mean compliance dur-
Sessions delivered to fa- Individual medication man- married. observation, medication blood ing post-treatment fol-
milies by therapists in agement sessions with psy- levels. low-up or any time
21 sessions delivered chiatrist for 1 year points.
over 9 months 12 þ
weekly, then bi-weekly, Individually focussed pa-
then 3 monthly. tient treatment. 30 min
sessions x 21 over
9 months.
Zaretsky Canada Psychoeducation (7 ses- TAU (standard outpatient 72% of IG completed at 79 pts (IG ¼ 40, 7 ½ months No Number of missed doses per At follow-up there was
et al. sions) followed by CBT care usual pharmacother- least 6 sessions of CG ¼39) currently month (pt interview) high adherence in both
(2008). (13 sessions) delivered apy, and naturalistic mon- psychoeducation and euthymic. Mean age the IG and CG, but no
individually by psychia- itoring, no additional CBT) 9 sessions of CBT. 41, gender break- difference between
trists over a period of 20 þ down not specified. groups.
weeks. 7 PE sessions.
Multi-focus interventions (Group interventions)
tervention outcome?
contact
Bordbar Iran One session of group TAU (routine psycho-education 100% of families as- 60 pts (IG30, 12 months No Months using medication, Significant difference in
et al. psychoeducation for fa- & pharmacotherapy for pts signed to IG com- CG ¼30) in acute questionnaire by blinded time using medication
(2009). mily members prior to from own psychiatrist. pleted the manic episode, o5 home visit team. at 6, 9 and 12 months
patient discharge. intervention. yrs onset. Mean age with IG continuing for
30, 25% female, 42% longer.
married
Castle et al. Australia Psychoeducation and TAU (Own GP/ psychiatrist 90% retention. 20 pts (IG ¼ 10, 0 (post- No MARS No significant differ-
(2007). CBT-type group therapy care) & weekly phone calls CG ¼10) without intervention) ence between pre- and
delivered weekly over controlling for extra contact current severe post intervention
12 weeks by Research time. symptoms. Mean change scores.
Assistants. Incorporated age 44, 82% female,
personalised Collabora- 82% married/ with
tive Therapy Journal partner.
workbook, homework
209
exercises and phone
210
Table 1b (continued )
tervention outcome?
contact
calls.
Colom et al. Spain Group psychoeducation TAU (4 weekly psychiatrist 73% adhered to psy- 120 pts (IG ¼ 60, 5 yrs No Composite measure Plasma lithium con-
(2003, (as part of the Barcelona appointments with mood choeducation (did not CG ¼60) in remis- centrations: Sig-
2009). Bipolar Disorders Pro- support but not formal miss more than sion. Mean age 34, nificant difference be-
gramme) involving edu- psychotherapy or psychoe- 5 sessions). 63% female. tween IG and CG at
cation, exercises and ducation, standard 2 yrs.
discussion. Delivered by psychopharmacology. Combined measure:
psychologists in 21 þ Significant difference
weekly sessions. Weekly group meetings between IG and CG at

without instruction (to 5 yrs, fewer IG classi-
control for supportive ef- fied as poorly
fects of group meetings) adherent.
Eker and Turkey Group psychoeducation TAU (outpatient mood dis- 4 pts did not attend 71 pts (IG¼ 36, 0 (post- Yes Composite measure Combined measure:
Harkin delivered weekly over orders clinic) regularly (retention CG ¼35) in remis- intervention) Medication Adherence Rat- significant difference in
(2012). 6 weeks by mental þ detail not provided) sion. Mean age 36, ing Scale (MARS). proportion of pts clas-
health nurses. 10–15 min medication 54% female, 49% McEvoy treatment observa- sified as adherent from
training from doctor. married. tion form. baseline to 6 weeks,
with greater proportion
of IG pts classified as
adherent at 6 weeks.
Lenz Austria Cognitive Psychoeduca- TAU (not specified) Not specified 100 pts (IG¼ 52, 3 months Not specified Medication Compliance Ques- Significant decrease
(2010). tive therapy (CPT) – þ CG ¼48) with mini- tionnaire (self-report) after 12 months for
psychoeducation and Self-help book, 3 group mum 2 episodes in control group in % re-
CBT-type techniques sessions & booster sessions last 3 years or porting good com-
delivered in 14 weekly 6 and 9 months. 3 episodes in last pliance (no change in
group sessions. Ad- 5 years, but not intervention group.
ditionally 8 hours of currently episodic.
group sessions with sig- Mean age 40, 59%
nificant others and female.
booster sessions 6 and
9 months after baseline.
D'souza Australia Systematic Illness Man- TAU (community based case Not specified 58 pts (IG ¼ 27, 11 months No Medication Adherence Scale Significant difference in
et al. agement Skills En- management; 45 min weekly CG ¼31) recently in (ARS) (pill count and need for mean adherence scores
(2010). hancement Programme- review with clinician & remission. Mean repeat prescription) with better adherence
Bipolar Disorder (SIM- monthly medical review). age 40, 52% female. in IG at follow-up.
SEP-BD). Group psy-
choeducation sessions
with companion-patient
dyads delivered weekly
over 12 sessions.
Reinares Spain Psychoeducation group TAU (standard psychiatric care; 95% attended at least 113 pts (IG ¼57, 12 months No Medication compliance - pt n/s within-group com-
et al. sessions delivered to outpatient follow-up, pharma- 8 out of 12 sessions. CG ¼56) euthymic report, caregiver report, plas- parisons between the
(2008). caregivers only (not pa- cotherapy, advice to contact for at least ma concentrations (described baseline and final
tients) including clinician as needed, no 3 months. Mean age in Colom 2000) assessment.
characteristics e.g. motivational stage or adherence level), Atten-
herent before and after

CG in terms of propor-
ence between IG and
tion of patients cate-

tion/Meso-tailoring (amount of intervention contacts depends on
No significant differ-
gorised as non-ad-
participant need), Micro-tailoring (intervention components are
tailored to the participant e.g. specific action plans generated) and
Participation (the participant is actively involved by providing
the study.
input or making decisions) (De Bruin et al., 2009).
Control group: Due to differences in the intensity of standard
care, treatment as usual (TAU) was coded as either 1. General care-
outpatient psychiatry or GP appointments/medication manage-
compliance ¼difference be-
Serum lithium levels. Non
ment (no specific education or psychological support mentioned)

tween tests of more than
Note: TAU ¼Treatment as usual, CBT ¼Cognitive Behaviour Therapy, IG¼ Intervention group, CG¼ Control group, RBC ¼ Red blood count, SRTAB¼ Self-reported treatment adherence behaviours.
or 2. Intensive support- structured specialist support. Additional
components were also coded as being included or not; additional
GP training, additional education/psychoeducation, additional
sessions (not educational but attention matched) and receiving
0.3 mmol/l
feedback on adherence.
Study and sample characteristics: date of publication, number
of participants randomised (intervention and control groups),
length of follow-up assessment (both from baseline and last in-
tervention contact point), primary outcome (whether adherence
was specified as a primary outcome or not), country of study, di-
Yes
agnostic assessment of bipolar (which assessment measure or

criteria was used to determine the diagnosis), participants’ illness
state at the point of recruitment (depression, mania, euthymia or
mixed), gender, ethnicity.
6 months
Adherence outcome: Mode of measurement – 1. subjective

(self-report, physician report, informant report), 2. objective (e.g.
electronic monitoring, pill counts), 3. biological (e.g. lithium serum
34, 54% female, 40%
der breakdown not
levels) and 4. composite. Target medication – 1. Lithium only, 2.

Mean age 50, gen-
criteria specified).
partner (no other
CG ¼20) with a
Multiple medications reported/analysed together, 3. Multiple

39 pts (IG ¼ 19,
medications – reported/analysed separately. Data was extracted on

specified.
married.
every reported adherence assessment at each time point. Eighteen

studies provided sufficient data in the published report in order to
compute standardised effect sizes for the meta-analysis. For six
studies, authors were contacted for additional data and this was
provided for one study (Eker and Harkin, 2012), data was un-
available for four (Frank et al., 1999, 2005; Peet and Harvey, 1991;
TAU (not specified, but referred Not specified
Harvey and Peet, 1991; Gilbert, 2000; Javadpour et al., 2013) and in
one study the data were medication continuation months (Bord-
bar et al., 2009) which was not comparable with the other studies
and was not included in the meta-analysis.
Quality of intervention reporting: coded using the Template for
Intervention Description and Replication (TIDieR) checklist (Hoff-
from university outpatient
systematic psychotherapy.
mann et al., 2014). This comprises 12 intervention items; Brief

name; Why (rationale, theory or goal); What (materials used);
What (procedures used); Who provided; How (mode of delivery);
Where (location); When and how much; Tailoring; Modifications;
How well (planned); and How well (actual). For each intervention,
each item was coded as being either adequately reported, in-
clinic.)
adequately reported, not reported or not applicable to the

intervention.
gists in weekly sessions
Netherlands Group psychoeducation
Fidelity: coded as formal assessment using coding systems by

sessions for partners of
Worker over 5 sessions

Delivered by psycholo-
patients delivered by a
Psychiatrist and Social
communication skills.
independent raters, informal assessment (where fidelity was re-

over an unspecified
ported to have been conducted, but formal scoring or rating was

over 12 weeks.
not used), or no evidence of fidelity assessment.

Intervention retention: the proportion of participants who took
duration.
part in intervention sessions, completed the programme, or were

classed as adhering to the protocol according to cut-offs specified
in the study.
Quality of trial design: assessed using the Cochrane collabora-
tion tool for assessing risk of bias in; selection, performance, de-
tection, attrition and reporting (Higgins et al., 2011). Each study
was classified into ‘low risk of bias’, ‘high risk of bias’ or ‘unclear’
for; Random sequence generation, Allocation concealment, Blind-
Van Gent
(1991).
Zwart
ing of participants, Blinding of personnel, Blinding of adherence

and
outcome assessments, Incomplete adherence outcome data, and

Selective adherence outcome reporting. As studies frequently used
more than one outcome measure, for the purposes of the narrative models were used in this analysis as we have no reason to assume
summary, the risk of bias in; blinding of outcome assessment, systematically different variance between groups of studies by
incomplete outcome data and selective reporting categorised moderator variable and there are a small number of studies within
studies according to the lowest risk of bias. For the meta-analysis, each category (Borenstein et al., 2011).
the risk of bias was assessed for the outcome measure and time- Sensitivity analysis was performed in order to detect whether
point specified in each sub-analysis. the overall effect was robust to the exclusion of small studies,
Data extraction for all studies was conducted by LM, with a outliers, studies with low retention and high risk of bias, and
second data extraction by two investigators (SC & another in- across the type of adherence measure and timepoint of the follow-
dependent researcher). Inter-rater reliability was assessed for 15 up. We excluded studies with small sample sizes for the first
items by calculating Cohens Kappa values. Kappa values were; sensitivity analysis (using a recommended cut-off where the
target medications (k ¼0.91), intervention focus (k ¼0.66), setting sample size of either the intervention or control group was o35
(k ¼0.49), intervention tailoring (5 items, k¼ 0.11 to 0.78), control (Coyne et al., 2010)), resulting in the inclusion of only 5 studies
group (1 item, k ¼0.66) risk of bias (6 items, k ¼0.13–0.83). There (Sajatovic et al., 2009; Eker and Harkin, 2012; Lenz, 2010; Colom
was good agreement except for the TIDieR tailoring item (54%), for et al., 2003; Reinares et al., 2008). We excluded one study from the
two items on intervention tailoring (Individualisation 38%, Micro- primary analysis due to its extremely large effect size and there-
tailoring 50%) and for two risk of bias items (Personnel blinding fore being classed as an outlier (Dogan and Sabanciogullari, 2003).
42%, Selective reporting 54%). Disagreements in coding were re- We selected the longest follow up point outcome for serum levels,
solved through discussions between the three investigators. composite measures and self-report and computed effect sizes for
each outcome category. We computed effect sizes for study out-
3.5. Data analysis comes grouped into 1–3 months post-intervention, 6 months
and 12 months. We included only those studies with satisfactory
We organised and processed extracted data using IBM SPSS retention of participants at follow-up assessment using a slightly
statistics (v21) and Comprehensive Meta Analysis (v2) software. less stringent cut-off than that used in a previous review of ad-
We included studies in the meta-analysis where it was possible to herence interventions (80%) (Haynes et al., 2008) of greater than
pool the adherence data and compute standardised effect sizes. 75% in both intervention and control groups. Finally we assessed
Studies reported different statistics including t-tests, chi squares, whether the outcome was affected by the quality of the evidence
ANOVAs, means, sd and p values. In order to compare data, stan- (including studies judged to have low risk of bias).
dardised odds ratios with 95% confidence intervals were computed We conducted analysis to assess the risk of publication bias as
from the proportions of participants classified as adherent in the it is possible that studies reporting a statistically significant result
control and interventions groups at follow-up. A random effects may be more likely to be published and therefore a meta-analysis
model was selected as it is more conservative than a fixed effects may over-estimate the real effect size of all studies conducted
model and should be used when analysing real world data (Hunter (published or unpublished) (Rothstein et al., 2006). We examined
and Schmidt, 2000; Field, 2003). a funnel plot for symmetry which displays the effect size of the
Where data was reported on multiple outcome measures and published studies against the variance; this would be asymme-
time points, effect sizes were computed for each (the outcome and trical if fewer studies with a large variance (typical of smaller
timepoint selected is specified in each analysis). The order of studies) and large effect size were published, suggesting that small
preference for selecting adherence outcomes was in descending studies are more likely to be published if consistent with the in-
order of priority; composite measure, serum lithium, self-report, tervention being effective. However, as funnel plots can be mis-
informant report. This judgement was based on review papers leading, particularly when the number of studies is small or the
recommending that a combination of measures is the most valid studies are heterogeneous (Walker et al., 2008), we conducted
method for adherence in psychiatric research (Sajatovic et al., additional checks of bias by applying a trim and fill method (Duval
2010). and Tweedie, 2000), running a classic fail safe analysis to compute
The chi-squared statistic (Q) was used to test for the presence the number of studies needed to produce a null result (Rosenthal,
of heterogeneity across studies. The level of heterogeneity is ex- 1979) and Egger's test to measure funnel plot asymmetry (Egger
pressed using the I2 statistic, which describes the percentage of et al., 1997).
variation due to heterogeneity as opposed to by chance. Cochrane
guidance provides cut-offs for interpretation of the magnitude of
heterogeneity, where 30–60% represents potentially moderate 4. Results
heterogeneity and 50–90% represents substantial heterogeneity
(Higgins et al., 2011). Studies are described in Tables 1a and b and intervention
Moderation analysis was conducted to assess if adherence characteristics across all studies are summarised in Tables 2a
outcome varied according to study characteristics (where possible and b.
i.e. where sub-categories contained two or more studies). This was
conducted for the following categorical potential moderators; in- 4.1. Intervention content and tailoring
tervention focus (adherence primary vs Multi-focus interventions),
content (all elements of psychoeducation vs other interventions), Five interventions had adherence as the primary focus of the
delivery setting (specialist clinic/department vs other settings), intervention content (Cochran, 1984; Elixhauser et al., 1990; Peet
delivery format (individual vs group), adherence as a primary and Harvey, 1991; Harvey and Peet, 1991; Sajatovic et al., 2009;
outcome, adherence – what was measured (mixed medications Dogan and Sabanciogullari, 2003), 18 were multi-focus pro-
assessed globally vs Lithium only), intervention tailoring, use of grammes, (Colom et al., 2003; Clarkin et al., 1998; D'souza et al.,
manual, assessment of fidelity, baseline illness state and study 2010; Eker and Harkin, 2012; Lam et al., 2000, 2003, 2005; Mik-
comparison group. Meta-regression analysis was conducted for lowitz et al., 2000, 2003; Rea et al., 2003; Gilbert, 2000; Javadpour
the following continuous moderator variables: date of publication, et al., 2013; Ball et al., 2006; Castle et al., 2007; Frank et al., 1999,
intervention contacts (as per protocol) intervention duration 2005; Zaretsky et al., 2008; Lenz, 2010; Van Gent and Zwart, 1991;
(months), total intervention contact time, and follow-up duration, Reinares et al., 2008). We were unable to classify one study due to
both from baseline and last intervention contact. Random effects the limited information provided in the paper (Bordbar et al.,
Table 2a Table 2b
Intervention components/delivery. Study design.
Intervention component Number of studies (k) Study Component Number of studies (k)
Content Eligibility assessment

All elements of Psychoeducation 13 DSMa 17
CBT/ CBT-type techniques 9 Research Diagnostic criteria 2
Social/ family therapy 6 MINIb 2
Communication skills training 4 SADSc 3
Education 3 Not specified 1
Psychotherapy 2
Medication packaging technology 1
Control/ Comparison group
TAUd only (general care) 6
Tailoring TAU only (intensive support) 3
Micro-tailoring 16 TAUþ additional content (sessions or materials) 3
Participation 16 TAUþ attention matched sessions 3
Individualisation 6 TAUþ clinician training 1
Attention tailoring 5 TAUþ brief medication training 1
Macro-tailoring 1 TAUþ compliance feedback 1
Alternative intervention 4
Unclear 2
Intervention focus
Multi-focus 18
Adherence as primary focus 5 Adherence assessment mode
Unclear 1 Self-report 11
Caregiver report 3
Physician report 4
Provider Objective (pill counts/ MPR) 2
Psychologists 5 Biological measures 9
Psychiatrist 6 Composite measure 7
Mental health nurses 3
Social workers 3
'Therapists’ 3 Adherence – medications analysed
Other (research assistants/ trainees/ psychiatric 4 Mixed medications – reported globally 17
counsellor) Lithium only 6
Mixed medications – reported separately 1
Format
Groups 8 Primary outcome
Individual 9 Relapse/ recurrence/ Symptoms 14
Family/ couples-based 4 Adherence 8
Group and individual 1 Wellbeing/ functioning 7
Electronic device, telephone & mail 1 Coping with prodromes 1
Group & telephone 1 Attitudes towards treatment 1
Not specified 2
n. participants, median (IQR) 66 (45–102)
Delivery setting n. participants IG, median (IQR) 31 (19–54)
Hospital outpatient psychology/ psychiatry 5 Follow-up from last intervention contact, median 6 (0.5–12)
department (IQR) mths
Specialist affective disorder/ bipolar clinic/ unit 5
a
University psychology/ psychiatry department (not 5 Diagnostic and Statistical Manual of Mental Disorders.
b
specialist BD) MINI International Neuropsychiatric Inventory.
c
Other 3 Schedule for Affective Disorders and Schizophrenia.
d
Actual delivery location unclear 6 Treatment as usual.
Recipient 2010; Eker and Harkin, 2012; Colom et al., 2003; Sajatovic et al.,
Patient only 11 2009; Lam et al., 2000, 2003, 2005).
Family/significant others & patient 10
As described, four interventions appeared to be information
Family/significant others only 3
provision only without a specific behavioural or psychosocial
element (Bordbar et al., 2009; Dogan and Sabanciogullari, 2003;
Peet and Harvey, 1991; Harvey and Peet, 1991; Javadpour et al.,
2009). 2013). Three had a specific focus on family communication (Mik-
Fifteen interventions were described as being or including lowitz et al., 2000, 2003; Rea et al., 2003; Gilbert, 2000). Elix-
psychoeducation (Zaretsky et al., 2008; Ball et al., 2006; Lenz, hauser et al. (1990) used an electronic adherence monitoring de-
2010; Miklowitz et al., 2000, 2003; Rea et al., 2003; Reinares et al., vice, adherence feedback and education.
2008; Gilbert, 2000; Javadpour et al., 2013; Bordbar et al., 2009; Fifteen interventions included more than one type of tailoring.
Van Gent and Zwart, 1991; Clarkin et al., 1998; D'souza et al., 2010; Sixteen interventions used micro-tailoring of components of the
Eker and Harkin, 2012; Colom et al., 2003; Sajatovic et al., 2009). intervention such as action planning in relation to participants
Five classified themselves as cognitive-behavioural therapy (CBT) lifestyle and goals or information on risk tailored to individual
or cognitive therapy (Zaretsky et al., 2008; Lam et al., 2000, 2003, characteristics or history (Ball et al., 2006; Frank et al., 1999; Gil-
2005; Ball et al., 2006; Cochran, 1984). When coding interventions bert, 2000; Lam et al., 2000, 2003, 2005; Eker and Harkin, 2012;
based on the content described in the paper, 13 clearly included all Rea et al., 2003; Castle et al., 2007; Cochran, 1984; Colom et al.,
elements of psychoeducation (Ball et al., 2006; Castle et al., 2007; 2003; D'souza et al., 2010; Miklowitz et al., 2000, 2003; Reinares
Lenz, 2010; Miklowitz et al., 2000, 2003; Rea et al., 2003; Gilbert, et al., 2008; Sajatovic et al., 2009; Van Gent and Zwart, 1991;
2000; Javadpour et al., 2013; Clarkin et al., 1998; D'souza et al., Zaretsky et al., 2008).
Sixteen studies involved participation, where the participants weeks then monthly telephone follow-up until 18 months. Booster
had active input, for example, identifying their own difficulties sessions after the main intervention were a feature of three studies
with adherence, deciding on strategies and evaluating the out- (Lam et al., 2000, 2003; Lenz, 2010).
comes of strategies used (Ball et al., 2006; Frank et al., 1999; Gil-
bert, 2000; Lam et al., 2000, 2003, 2005; Eker and Harkin, 2012; 4.3. Control group
Rea et al., 2003; Castle et al., 2007; Cochran, 1984; Colom et al.,
2003; D'souza et al., 2010; Miklowitz et al., 2000, 2003; Reinares Thirteen studies compared the intervention to TAU or standard
et al., 2008; Sajatovic et al., 2009; Van Gent and Zwart, 1991; care only, eight compared standard care plus an additional com-
Zaretsky et al., 2008; Javadpour et al., 2013). ponent and four described the comparison group as another form
Six interventions used individualization, where the content of of therapy. TAU was classified into low intensity general care or
sessions was individualised in response to participants needs, their intensive support (structured specialist support). The comparison
questions or queries (Ball et al., 2006; Dogan and Sabanciogullari, group comprised only low intensity general care in six studies
2003; Eker and Harkin, 2012; Rea et al., 2003; Peet and Harvey, (Clarkin et al., 1998; Lam et al., 2000, 2003, 2005; Reinares et al.,
1991; Harvey and Peet, 1991; Gilbert, 2000). 2008; Peet and Harvey, 1991; Harvey and Peet, 1991; Javadpour
Five interventions were tailored by varying the level of inter- et al., 2013) and only high intensity in three studies (Bordbar et al.,
vention contacts depending on participants’ needs or preferences 2009; D'souza et al., 2010; Sajatovic et al., 2009). Two studies did
(Frank et al., 1999, 2005; Gilbert, 2000; Lam et al., 2000, 2003, not provide detail on what TAU constituted (Van Gent and Zwart,
2005) with the decisions either based on participant choice (k ¼1) 1991; Dogan and Sabanciogullari, 2003).
(Gilbert, 2000), HCP/ clinical need (k ¼2) (Lam et al., 2000, 2003) Where the comparison group received additional components,
or both patient and practitioner choice (k ¼ 2) (Frank et al., 1999, these included additional clinician training (Ball et al., 2006), very
2005). brief medication training (Eker and Harkin, 2012) and attention
Only one intervention used tailoring at a macro-level, i.e. de- matched group therapy sessions or phone calls to control for
termining the intervention to be received by pre-tested char- contact time (Castle et al., 2007; Cochran, 1984; Colom et al.,
acteristics, in this case the levels of symptoms, where participants 2003). Additional education or psychoeducation sessions were
must be classed as asymptomatic before proceeding (Frank et al., provided to control groups, for example seven sessions of psy-
1999) and one tailored feedback according to a mid-point assess- choeducation (the IG received the same and the addition of 13
ment of adherence (Elixhauser et al., 1990). sessions of CBT) (Zaretsky et al., 2008), individual patient treat-
ment sessions with a therapist and medication management ses-
4.2. Delivery channel (provider, duration, frequency of sessions) sions with a psychiatrist for a year (Rea et al., 2003), the provision
of a self-help book and additional group sessions (Lenz, 2010).
Psychologists (k¼ 5) and psychiatrists (k ¼6) were the most The electronic monitoring device intervention provided
common intervention providers, but also mental health nurses by Elixhauser et al., 1990 was compared against TAU and lithium
(k ¼3), social workers (k ¼3) and ‘therapists’ (k¼ 3) (Table 2a). levels feedback with suggestions for improving compliance.
Five interventions were delivered through specialist affective, For the four studies where the comparison group received an
mood or bipolar disorder clinics (Cochran, 1984; Colom et al., alternative intervention, this took the form of a programme of
2003; Eker and Harkin, 2012; Harvey and Peet, 1991; Reinares therapy differing in content or focus: Intensive Clinical Manage-
et al., 2008). Other settings included hospital outpatient mental ment (ICM), Interpersonal and Social Rhythm Therapy (IPSRT),
health departments and University psychology or psychiatry de- ‘Crisis management’, Family Focussed Therapy and ‘Individual
partments (not specifying specialism in affective disorders) patient management’ (FFT) (Frank et al., 1999, 2005; Miklowitz
(Table 2a). In six cases, the exact location of delivery of the inter- et al., 2000, 2003; Gilbert, 2000).
vention was not specified (Clarkin et al., 1998; D'souza et al., 2010;
Lenz, 2010; Van Gent and Zwart, 1991; Zaretsky et al., 2008; Castle 4.4. Study and sample characteristics
et al., 2010), however it might be assumed that they were con-
ducted at the outpatient clinics or University departments where The median number of participants in the studies was n ¼66
authors were based. (IQR ¼45–102) and participants randomised to intervention
Most interventions targeted either the patient alone (k ¼11) or groups was n¼ 31 (IQR ¼19–54). The length of time of follow-up
the patient and their family (k ¼10), three interventions targeted varied across studies, participants were followed up for a median
the family or patients significant others only (Bordbar et al., 2009; of 6 months (IQR ¼0.5–12) after last intervention contact
Van Gent and Zwart, 1991; Reinares et al., 2008) (Table 2a). In- (Table 2b).
terventions were delivered to only groups (k ¼8), solely to in- Over one-third of studies (k ¼9) were based in the United
dividuals (k¼ 9), or delivered to the patient and family members or States, three in the UK, three in Australia, two in Spain, two in
as a couple (k ¼4). Turkey and Iran and one each in Canada, Austria and the Nether-
The median number of intervention contacts as per protocol lands (Table 1a and b).
was 17 (range 1–56), with a mean estimated contact time of 19.7 h To determine eligibility different diagnostic criteria were used;
(sd 11.67). The median duration of interventions was 6 months, DSM (k ¼17); Research Diagnostic criteria (k ¼2); MINI (k¼ 2);
ranging from a single intervention session to 2 ½ years. The spa- Schedule for Affective Disorders and Schizophrenia (SADS (k ¼2).
cing of intervention contacts was highly variable including, single- In 11 studies researchers recruited patients who were euthymic,
sessions (Bordbar et al., 2009) or two contacts (Elixhauser et al., five during episodes, three during manic episodes only, one re-
1990; Peet and Harvey, 1991). Most were delivered weekly for at cruited participants in all states, four did not report this (Table 2b).
least 6 sessions (k ¼17). Seven interventions started with weekly Data was available in 21 studies for the proportion of males and
sessions before reducing the frequency. The family-focussed females recruited to the study. Overall there was a mean of 55.72%
therapy programmes were conducted for 12 weekly sessions, re- (sd ¼12.21) females.
ducing to bi-weekly, then 3 monthly (Rea et al., 2003; Miklowitz
et al., 2000). Clarkin et al. (1998) marital therapy ran weekly for 10 4.5. Adherence assessment and primary outcome
sessions, then bi-monthly for remaining 15 sessions. Javadpour
et al. (2013) conducted weekly face to face sessions for eight Thirteen studies reported the use of more than one assessment
method, of which seven combined these into one adherence score replication (TIDieR items 1–9) (Colom et al., 2003; Miklowitz et al.,
or categorisation. Seven studies used a composite measure of ad- 2000). Adequate reporting was most common for: Brief name
herence, combining a number of assessment methods which were (100%), Why (rationale, theory or goal) (92%) and What (proce-
then either rated for the level of adherence, or defined by pre-set dures used) (88%). Less well described items were Where (location)
criteria (Table 1 a and b). Subjective techniques included self-report (54%), When and how much (58%) and What (materials used) (58%)
(k¼11), informant report (k¼ 3) and physician report (k¼4). Ob- (Appendix S3).
jective techniques (other than biological measures) were used in The method of randomisation was judged to have low risk of
two studies (D'souza et al., 2010; Elixhauser et al., 1990), nine stu- bias in nine studies, high risk in one (Dogan and Sabanciogullari,
dies used biological measures for example serum medication levels. 2003) but was inadequately described to make an assessment in
Seventeen studies either assessed and reported adherence 14 studies. Three demonstrated low risk of bias in allocation
overall for all medications participants were prescribed (Ball et al., concealment, (Miklowitz et al., 2000; Frank et al., 2005; Lam et al.,
2006; Bordbar et al., 2009; Castle et al., 2007; Clarkin et al., 1998; 2003), one demonstrated high risk of bias (Dogan and Sabancio-
D'souza et al., 2010; Eker and Harkin, 2012; Lam et al., 2000, 2003; gullari, 2003) and in 20 studies was judged to be unclear. De-
Sajatovic et al., 2009; Lenz, 2010; Gilbert, 2000; Javadpour et al., scriptions of blinding of study personal was inadequately de-
2013), or assessed them separately and then reported a overall scribed to judge risk of bias in nine studies and in 14 studies there
adherence result (Frank et al., 2005; Miklowitz et al., 2000; Rea was high risk of bias. Studies were judged to have a high risk of
et al., 2003; Reinares et al., 2008; Zaretsky et al., 2008). Within bias in participant blinding in 10 cases. Only one study each were
these 17 studies, data was collected and reported for mood sta- judged to have low risk of bias for personnel blinding (Castle et al.,
bilisers only in four (Ball et al., 2006; Frank et al., 2005; Lenz, 2010; 2007) and participant blinding (Reinares et al., 2008) (Appendix
Gilbert, 2000), for mixed medications such as mood stabilisers, S4). In terms of adherence outcome assessment, 10 studies in-
anti-depressants and benzodiazepines in nine studies (Bordbar cluded an outcome measure with low risk of bias, however, 12
et al., 2009; Clarkin et al., 1998; D'souza et al., 2010; Lam et al., studies only included measures where there was a high risk of bias
2000, 2003; Miklowitz et al., 2000; Rea et al., 2003; Reinares et al., (for example self-report).
2008; Zaretsky et al., 2008). Four studies did not report what
medications were assessed (Eker and Harkin, 2012; Sajatovic et al., 4.7. Meta-analysis results
2009; Javadpour et al., 2013; Castle et al., 2007). Only one study
measured and reported results individually for different mood Across 18 studies, there was a significant effect of interventions
stabilisers, and this was only at the two year follow-up paper versus control on adherence, OR 2.27, 95% CI [1.45, 3.56],
(Colom et al., 2005). Six studies only reported on lithium pre- po 0.001). Adherence outcomes were better in the intervention
scriptions (Cochran, 1984; Dogan and Sabanciogullari, 2003; Elix- group than the control group in 15 studies, as indicated by positive
hauser et al., 1990; Frank et al., 1999; Harvey and Peet, 1991; Van effect sizes, while in 3 studies adherence outcomes were not im-
Gent and Zwart, 1991). proved relative to control (Fig. 2). There was substantial hetero-
The primary outcome included adherence (k ¼8) (three studies geneity between the studies Q(17) ¼36.96 (p ¼0.003) I2 ¼54.00.
had adherence as the only primary outcome and five had multiple
primary outcomes including adherence), relapse/ recurrence, 4.7.1. Moderation analysis
symptoms (k ¼14), wellbeing/ functioning (k ¼7), coping with Moderation analysis was conducted to assess whether inter-
prodromes (k ¼1) and attitudes towards treatment (k¼1), and was vention and study characteristics (intervention focus, content,
not specified for two studies (Clarkin et al., 1998; Lenz, 2010). delivery, setting, tailoring, fidelity, primary outcome and target
medication, baseline illness state, control group, publication year
4.6. Quality indicators (retention, fidelity, intervention description, and follow-up time) contributed to the variability in effect sizes
study design – risk of bias) (Tables 3 and 4). Intervention contact time (hours) had a sig-
nificant impact with increasing contact time being associated with
Many studies reported a good participant retention rate, for a reduction in effect B ¼ 0.08 ( 0.14, 0.02); Q¼ 6.12 p ¼.013
example, 76% completing the full programme of nine months of (Tables 4 and 5). If TAU was supplemented with an additional
treatment (Rea et al., 2003), 73% of the intervention group ad- component for the control group, the intervention effect was
hering to the programme (missed no more than five out of 21 smaller (p ¼0.034) (Table 4). None of the other variables sig-
sessions) (Colom et al., 2003). Sajatovic and colleagues (Sajatovic nificantly moderated the intervention effect.
et al., 2009) found that 37% never participated in the sessions, 49%
did complete at least four of the six sessions. This data was not 4.7.2. Sensitivity analysis
reported for nine studies (Table 1 a and b). Where the programme Sensitivity analyses showed that the effect of interventions on
was delivered to caregivers, a good retention rate was reported, for adherence remained similar when small studies, studies with low
example 95% of caregivers attending at least eight out of 12 ses- retention and outliers were excluded, and across all follow-up
sions (Reinares et al., 2008). time points (Table 5). When only studies using serum levels were
Eighteen studies specified that the intervention was manua- included there was no significant effect of intervention however
lised. Only nine mentioned a fidelity assessment. For five this was the effect size for interventions was similar to that seen in the
formal assessment by independent raters using published or overall analysis.
adapted coding systems (Frank et al., 2005; Miklowitz et al., 2000, Highest OR were found by including those reporting outcomes
2003; Rea et al., 2003; Ball et al., 2006). The outcome of the fidelity from composite adherence measures (2.50) followed by serum
assessments was that there was good adherence to the manuals, levels (1.91) and self-report measures (1.71). Significant hetero-
compliance with the principles of the therapy and competent geneity was present when ORs from studies using composite
delivery. Fidelity assessments not formally rated were used in four measures were included (Q(7)¼ 27.35, p o.001). Study ORs were
studies (Clarkin et al., 1998; Sajatovic et al., 2009; Lam et al., 2000, not significantly heterogeneous within the serum level or self-
2003) for example, Sajatovic and colleagues (Sajatovic et al., 2009) report outcome measures.
reported that fidelity was ensured by the co-principal investigator When selectively analysing only higher quality studies, inter-
delivering feedback at the end of each session. ventions with low risk of bias in blinding of outcome measure
Only two studies provided all detail required to allow were effective (OR ¼1.49, CI ¼0.60, 3.70) as were those with low
Fig. 2. Forest plot of odds ratios and 95% confidence intervals for interventions versus control.
risk in selective reporting (OR ¼1.66, CI ¼0.56, 4.91) and in- moderation and sensitivity analysis which could be conducted.
complete outcome data (OR ¼ 2.99, CI ¼1.44, 6.21). Most interventions involved psychoeducational techniques
which appear to be effective; these take into account an in-
4.7.3. Publication bias dividuals’ knowledge, beliefs and attitudes. These are promising
The funnel plot was symmetrical suggesting the absence of areas of focus for interventions (Sajatovic et al., 2004; Gaudiano
publication bias (Appendix S5). Duval and Tweedie's ‘trim and fill’ et al., 2008; Leclerc et al., 2013) and are included in re-
method suggested that one study would need to be added or recommendations for improving adherence in the UK National
moved for the funnel plot to be symmetrical. The imputed OR after Health Service (Horne et al., 2005). Taking into account partici-
this procedure was 2.18 (1.39, 3.42) i.e. very similar to our original pants’ own beliefs and opinions about bipolar and its treatment
estimate of the effect of interventions on adherence (OR 2.27) (see may foster better engagement (Berk et al., 2010; Clatworthy et al.,
Appendix S5 for a funnel plot with the imputed study). Using 2009). However, we would need additional evidence to tease out
Egger's test for asymmetry, this was not significant indicating no the mechanisms behind how these interventions might have an
evidence for publication bias, the B0(bias) ¼0.86, 95% CI ( 1.74, effect because description of intervention content was often poor.
3.47), with t¼0.70, df ¼ 16, p ¼0.246. The fail-safe N was calculated The way psychoeducation or CBT were operationalised varied
and 117 studies would need to be included, i.e. 6.5 missing studies across studies, therefore accurate moderation analysis comparing
for every observed study, for the intervention effect to be nullified. intervention types is challenging. We were careful, to scrutinise
The results of these tests indicate that publication bias does not the content provided, rather than rely on the broad title or name
present a threat to the results of this meta-analysis. given to each intervention.
Many of the interventions are complex and resource intensive
and identification of the active components, mechanisms of action
5. Discussion and effective dose of intervention will be needed to allow for
better integration into clinical care (Rouget and Aubry, 2007; Ba-
This review brings together the published literature on ran- tista et al., 2011). We recommend that future trials conducted use
domised trials of interventions to enhance medication adherence published taxonomies of evidence-based behaviour change tech-
in bipolar disorder. There is strong evidence that interventions can niques in planning, delivery and reporting (Michie et al., 2013,
improve medication adherence. The pooled OR was 2.27 (95% CI 2011). Many studies used multiple methods to measure adherence
1.45, 3.56) equivalent to a two-fold increase in the odds of ad- conforming to the recommendation that studies use more than
herence in the intervention group relative to control. The effects one type of measure as has been recommended (Horne et al.,
appear to be durable, as there was no significant change in effect 2005; Sajatovic et al., 2010). Sensitivity analysis demonstrated that
of the interventions when patients were followed up at longer a positive effect of the interventions was present regardless of type
post-intervention intervals. Studies with a two-year follow-up still of measure used.
reported positive effects on adherence. Smaller effects on adherence were found when the comparison
Brief interventions tending to specifically focus on adherence group received additional therapy sessions than when this group
were more effective in improving adherence than longer inter- received only TAU. This confirms empirically the conclusion by
ventions where medication adherence was combined with other Gaudiano and colleagues (2008) that interventions compared to
aspects of self-management such as mood management and life- TAU, predominantly CBT and psychoeducation, had greater evi-
style changes. Smaller effects on adherence were found when the dence for effectiveness than for those using an active control,
comparison group received additional therapy sessions compared which tended to be Family-focussed Therapy and Interpersonal
to those receiving just TAU. When we analysed only studies with and Social Rhythm Therapy interventions. Comparing groups re-
low risk of bias, i.e. higher quality studies, interventions remained ceiving different ‘active’ interventions e.g. in factorial designs, is
effective, indicating that effectiveness is not due to poor quality essential to identify the effective components of the intervention
studies. No other significant moderator variables were associated (Collins et al., 2007) and if the intervention is effective beyond
with intervention effectiveness. However, both intervention and attentional ‘Hawthorne’ effects. However, TAU comparison groups
study design were often poorly reported, limiting the extent of are also extremely useful as, in practice, the content and attention
Table 3
Moderation analysis-intervention characteristics.
Intervention characteristics k OR (95%CI) Heterogeneity (within group) Heterogeneity (between group)
Q(df), p, I2 Q(df), p
Intervention focus 0.50(1), 0.479

Multi-focus 14 2.09 (1.23,3.54) 30.43(13), 0.004, 57.27%
Adherence as primary focus 4 3.12 (1.18,8.25) 6.02(3), 0.111, 50.17%
Content 0.25(1), 0.617

Psychoeducation – all elements 11 2.14 (1.18,3.89) 27.82(10), 0.002, 64.06%
Other 4 3.03 (0.89,10.23) 5.59(3), 0.133, 46.35%
Delivery setting 1.12(1), 0.289

Specialist 4 3.17 (1.09, 9.21) 9.28(3), 0.026, 67.66
General 11 1.65 (0.94, 2.92) 18.80(10), 0.043, 46.79
Delivered to partner/family only 2 1.31 (0.20,8.55) 0.73(1), 0.400, 0.00% 0.13(1), 0.719
Delivered to partner/family & pt 4 1.98 (0.57,6.93) 16.60(3), 0.001, 81.93%
Individual and/or group only 12 2.66 (1.45,4.85) 17.89(11), 0.084, 38.49% 0.29(1) 0.588
Family/partner & pt involved 4 1.94(0.75,5.06) 16.60(3), 0.001, 81.93%
Group & individual 2 1.75 (0.29,10.41) 3.99(1), 0.046, 74.95% 0.22(2) 0.896
Group only 4 2.77 (1.23,6.23) 8.13(3), 0.043, 63.12%
Individual 6 2.69 (1.23,5.90) 4.98(5), 0.418, 0.00%
Tailoring – Individualisation 1.06(1) 0.303

Yes 4 3.56 (1.84,6.89) 7.44(3), 0.059, 59.68%
No/Unclear 14 1.95 (1.41,2.69) 26.94(13), 0.013, 51.75%
Tailoring – Attention (individual level) 0.89(1) 0.346

Yes 2 4.58 (1.43,14.66) 0.09(1), 0.763, 0.00%
No/Unclear 16 2.08 (1.54,2.81) 35.22(15), 0.002, 57.41%
Tailoring – Micro (individual level) 0.14(1) 0.714

Yes 14 2.18 (1.57,3.01) 30.78(13), 0.004, 57.77%
No/Unclear 4 2.24 (1.17,4.30) 6.17(3), 0.104, 51.37%
Tailoring – Participation 0.82(1) 0.365

Yes 13 1.97 (1.64,2.93) 27.44(12), 0.007, 56.27%
No/Unclear 5 2.89 (1.67, 5.00) 8.17(4), 0.086, 51.05%
Manualised 0.76(1), 0.384

Yes 13 2.03 (1.21,3.42) 23.86(12), 0.021, 49.70%
No 5 3.26 (1.29,8.27) 11.58(4), 0.021, 65.45%
Intervention fidelity assessed 5.42(2), 0.066

Formal 3 0.98 (0.397,2.425) 4.80(2), 0.091, 58.32%
Informal 4 4.25 (1.825,9.903) 0.82(3), 0.844, 0.00%
No fidelity assessment reported 11 2.34 (1.372,3.976) 20.17(10), 0.028, 50.42%
k B, (95% CI) Q; p
Intervention contacts (per protocol) Range 2–27 18 0.04 ( 0.10,0.01) 2.21; 0.137
Intervention duration (months) 17 0.11 ( 0.24,0.02) 2.86; 0.091
Range 1.4–12
Intervention contact time (hours) 12 0.08 ( 0.14,0.02) 6.12; 0.013
Range 6–31.5
participants' receive are components of the intervention and using moved on to other aspects of self-management, whereas the three
TAU can establish whether the intervention can improve outcomes shortest interventions all had adherence as a primary outcome
over current standard care (Freedland et al., 2011). and content focused firmly on adherence (Cochran, 1984; Sajatovic
Smaller effects on adherence were found in longer interven- et al., 2009; Eker and Harkin, 2012). These findings are consistent
tions. But these were typically multi-focused interventions that with Gaudiano and colleagues (2008) suggestion that making an
did not have adherence as the primary outcome and aimed to intervention's primary focus adherence tends to increase effects
improve clinical outcomes through a range of determinants. Ses- on adherence, and suggest that resource intensive interventions
sions frequently included medication components, but then may be unnecessary for adherence improvement. In clinical care,
Table 4
Moderation analysis – Study characteristics.
Study characteristics k OR (95%CI) Heterogeneity (within group) Q Heterogeneity (between group) Q

(df), p, I2 (df), p
Adherence a primary outcome 1.29(1), 0.256

No (including ‘missing’) 12 1.91 (1.12,3.27) 21.79(11), 0.026, 49.53%
Yes 6 3.34 (1.40,7.45) 12.07(5), 0.034, 58.58%
Target medication 0.002(1), 0.968

Mixed medication – reported/ assessed together 13 2.43 (1.45,4.01) 28.13(12), 0.005, 57.34%
Lithium only 4 2.51 (0.65,9.63) 6.54(3), 0.008, 54.14%
Baseline state 0.21(1), 0.649

Episodic included 5 2.05 (0.88,4.75) 15.95(4), 0.003, 74.93%
Euthymia only 10 2.62 (1.38,4.96) 15.21(9), 0.085, 40.81%
TAU type 0.14(1) 0.710

General care 11 2.46 (1.66,3.65) 17.82(10), 0.058, 43.87%
Intensive care 5 2.02 (1.25,3.26) 17.42(4), 0.002, 77.03%
TAUþ education/ psychoeducation 4.47(1) 0.034

Yes 4 0.92 (0.50,1.70) 4.42(3), 0.220, 32.11%
No/ Unclear 14 2.81 (2.02,3.90) 22.75(13), 0.045, 42.86%
K B, (95% CI); Heterogeneity Q; p

Publication year 18 39.88 ( 160.74,80.98) 0.44, 0.509
Follow-up duration (mths since last intervention contact) 18 0.02 ( 0.05,0.01) 1.66; 0.197
Range 0–60
Follow-up duration(mths from baseline (excluding van 17 0.03 ( 0.05,0.00) 3.06; 0.080
Gent) Range 1.4–65
Table 5
Sensitivity analysis.
Heterogeneity
Comparison k OR 95% CI p I2 Q (df) p
Small studies excluded 5 2.57 1.31,5.06 0.006 53.72 8.64(4) 0.071

Outlier excluded 17 2.18 1.39,3.41 0.001 53.98 34.77(16) 0.004
Follow-up assessment retention o 75% excluded 9 3.16 1.67,5.96 o.001 50.78 16.25(8), 0.039
Adherence measure (final timepoint assessed is used)
Serum levels 6 1.91 1.10,3.32 0.021 1.78 5.09(5) 0.405
Serum levels a 5 1.77 1.03,3.05 0.040 0.00 2.58(4) 0.631
Composite measures a 8 2.50 1.11,5.66 0.028 74.41 27.35(7) o 0.001
Self report a 9 1.71 1.05,2.14 0.033 40.43 13.43(8) 0.098
Timepoint assessed
1–3 months 7 1.87 0.86,4.08 0.117 69.80 19.87(6) 0.003
1–3 months a 6 1.63 0.76,3.51 0.213 70.39 16.89(5) 0.005
6 months a 7 1.91 1.15,3.19 0.013 24.59 7.96(6) 0.241
12 months a 11 1.97 1.23,3.16 0.005 52.16 20.90(10) 0.022
Low risk of biasb
Blinding of outcome measure 7 1.49 0.60,3.70 0.393 57.39 14.08(6) 0.029
Incomplete outcome data 9 2.99 1.44,6.21 0.003 54.82 17.71(8) 0.024
Selective reporting 2 1.66 0.60,4.91 0.362 25.60 1.34(1) 0.246
Note: Overall pooled OR across all studies ¼2.27 (95% CI 1.45, 3.56).
a
Dogan and Sabanciogullari, 2003 excluded or not present in analysis
b
Risk of bias analysis for risk measures where at least 2 studies were in low risk category (where risk was unclear, studies were excluded).
the briefest interventions may be more easily adopted. However, moderator variables except those mentioned above, significantly
we recognise that many of the longer interventions also had im- distinguished between the effectiveness of studies. Using the TI-
pacts on adherence, and that these more complex interventions DieR checklist (Hoffmann et al., 2014), many items were not ade-
may address other determinants clinical outcomes in bipolar dis- quately reported and, as stated moderation analysis was therefore
order not explored in this review. limited. For example, the setting where interventions were deliv-
There was a great deal of heterogeneity between the studies, in ered was often not stated, and so we could not fully test the impact
terms of setting, content, adherence outcome, follow-up and of intervention setting.
quality concurring with previous narrative reviews (Desplenter We found only two studies reported sufficient detail to allow
et al., 2006; Crowe et al., 2012; Reinares et al., 2014; Leclerc et al., for replication using the TIDieR checklist (Hoffmann et al., 2014).
2013). We found no intervention, study design or delivery This is consistent with previous findings, Hoffman et al. judged
that only 39% of non-pharmacological interventions published in However, it is striking that many of the studies in this review were
2009 were adequately reported (Hoffmann et al., 2013). We re- not sufficiently powered to detect moderate to large effects.
commend that TIDieR is used by researchers to ensure consistency The measurement of treatment adherence in psychiatry is a
of reporting. complex issue (Sajatovic et al., 2010). This review is limited by the
We systematically assessed study quality and its impact on our heterogeneity of adherence outcomes. To compute an overall ef-
conclusions. This was limited by inadequate descriptions of risk of fect, odds ratios were used as most studies reported the propor-
bias indicators for many studies. For example, insufficient in- tion of participants who were classified as adherent/ non-ad-
formation was provided to judge risk of bias for over half of the herent, even though the cut-offs for adherence differed. There is
studies in terms of method of randomisation, allocation conceal- no agreed clinically optimal adherence level for bipolar disorder
ment, participant blinding and selective reporting. When exclud- and further research is needed to determine whether a clinically
ing studies where follow-up retention was less than 75% overall appropriate standard can be identified (Velligan et al., 2010).
effects of the intervention were still significant. When we included Clinical outcomes are the critical end-point, and interventions
only studies where the outcome assessments were blinded and should ultimately be judged on these. We know adherence is one
where studies did not selectively report outcomes, the effective- important contributor to clinical outcomes. Multi-component in-
ness of these interventions was lower than when all studies were terventions target adherence and other determinants of clinical
included. This fits with the hypothesis that poor quality may in- outcomes. Our review shows that brief, focussed interventions
flate effects. Despite poor reporting, we found no evidence of would be the best at targeting adherence specifically. Further in-
publication bias, and as adherence interventions are resource in- vestigation is needed to establish how multi-component inter-
tensive, it would be unlikely that there would be the number of ventions, although not as effective (but still having a positive ef-
unpublished studies (k ¼117) needed to nullify the overall effect of fect) on adherence, impact on clinical outcomes.
interventions found in this review.
Lack of evidence on interventions for specific clinical groups
limits our ability to establish what works for from whom. The
6. Conclusions
majority of the interventions in this and previous reviews have
focussed on euthymic patients and are therefore not generalizable
This review synthesized evidence for adherence interventions
to those in acute phases (Colom and Lam, 2005; Crowe et al., 2010).
in bipolar disorder, including psycho-educative approaches and
Patients who have experienced greater number of bipolar episodes
other techniques and therapies. We found strong evidence that
may be less likely to benefit from long duration CBT (Scott et al.,
adherence to medication for bipolar disorder can be improved and
2006). In this review, data were not available on number of previous
that brief interventions focusing specifically on adherence were
episodes for us to test this as a hypothesis for adherence inter-
more effective in achieving this than longer interventions.
ventions. In addition, it is important to further investigate inter-
These types of interventions should be incorporated in routine
vention effects on individuals who are inpatients, or who have co-
clinical practice as it would be likely to lead to improved and
morbidities (Desplenter et al., 2006; Lolich et al., 2012; Berk et al.,
sustained adherence for patients. Adherence interventions could
2010; Rouget and Aubry, 2007). Treatment guidelines recommend
positively impact on outcomes for people with a diagnosis of bi-
medication to treat episodes, reduce relapse risk and prevent es-
polar disorder through reducing relapse and risk of hospitalisation.
calation of episodes (NICE, 2014), and so it might be expected that
Our capacity to draw conclusions about exactly what works and
improvements in adherence may lead to better clinical outcomes,
for whom is limited by the strength of reporting, specifically the
such as improved quality of life and lower risk of relapse. However,
descriptions of the intervention content and delivery. However our
simple, direct relationships between levels of adherence and clinical
findings suggest that interventions should include psychoeduca-
outcomes cannot be established due to complex treatments, com-
tional techniques that take account of individuals' knowledge and
bined medication regimens, and the potential impact of clinical
factors such as history of episodes. We therefore do not know beliefs. The growing body of evidence and improved trial and in-
whether the improvements on adherence seen in the intervention tervention reporting will allow us to draw more conclusive re-
groups would equate to clinical improvements. In future, adherence commendations as to whether these results can be generalised to
intervention studies should explore the role of improvements in more naturalistic clinical settings.
adherence in clinical outcomes to build a more comprehensive
picture of effective interventions for clinicians.
Contributors
5.1. Limitations
L. MacDonald designed and conducted the database searches,
Limitations of the published studies are reflected in this review, study selection and data extraction. She conducted data analysis,
i.e. poor reporting and quality, limiting conclusions which can be produced the first draft and subsequent revisions of the
drawn. Some of these limitations may be impossible to overcome manuscript.
in the context of psychosocial interventions, for example blinding S. Chapman designed the search strategy, conducted study se-
of personnel and participants in intervention delivery may be lection and data extraction as well as data analysis. She con-
difficult or impossible (Davidson et al., 2003). Other study lim- tributed towards revisions of the manuscript
itations could be overcome by the use of existing guidelines such M. Syrett advised on the review design, contributed to the in-
as CONSORT statements in order that risk of bias can be assessed terpretation of the results and revisions of the manuscript.
reliably (Boutron et al., 2008). R. Bowskill advised on the review design, study selection cri-
For the effect sizes selected for comparison, half (9/18) were teria and data extraction. He contributed towards interpretation
based on results for which over 25% of follow-up data was not and implications of the results and revisions of the manuscript.
available for one or both of the treatment groups. Exclusion of R. Horne originally conceived the review, advised on the review
these studies resulted in an increase in overall OR, indicating that design, study selection criteria and data extraction as well as on
studies with poorer follow-up rates were not biased to report data analysis. He contributed towards revisions of the manuscript.
better results. Sensitivity analysis also demonstrated that the ef- All authors were involved with revising the manuscript and all
fects did not arise purely from studies with small samples. have approved the final article.
Funding sources Castle, D., Berk, M., Berk, L., Lauder, S., Chamberlain, J., Gilbert, M., 2007. Pilot of
group intervention for bipolar disorder. Int. J. Psychiatry Clin. Pract. 11,
279–284.
This review was conducted using funds from a National In- Castle, D., White, C., Chamberlain, J., Berk, M., Berk, L., Lauder, S., Murray, G.,
stitute for Health Research; Research for Patient Benefit pro- Schweitzer, I., Piterman, L., Gilbert, M., 2010. Group-based psychosocial inter-
gramme grant (Reference number PB-PG-1207-15248). R. Horne vention for bipolar disorder: randomised controlled trial. Br. J. Psychiatry 196,
383–388.
was supported by an NIHR Senior Investigator award (NF-SI-0510- Cavanagh, J., Smyth, R., Goodwin, G.M., 2004. Relapse into mania or depression
10197). S. Chapman is funded by a UCL Excellence Fellowship. following lithium discontinuation: a 7-year follow-up. Acta Psychiatr. Scand.
The funders had no involvement with the study design, data 109, 91–95.
Clarkin, J.F., Carpenter, D., Hull, J., Wilner, P., Glick, I., 1998. Effects of psychoedu-
collection, analysis and interpretation, report writing or the deci- cational intervention for married patients with bipolar disorder and their
sion to submit the article for publication. spouses. Psychiatr. Serv. 49, 531–533.
Clatworthy, J., Bowskill, R., Parham, R., Rank, T., Scott, J., Horne, R., 2009. Under-
standing medication non-adherence in bipolar disorders using a Necessity-
Concerns Framework. J. Affect. Disord. 116, 51–55.
Conflicts of interest Cochran, S.D., 1984. Preventing medical noncompliance in the outpatient treatment
of bipolar affective disorders. J. Consult. Clin. Psychol. 52, 873–878.
Collins, L.M., Murphy, S.A., Strecher, V., 2007. The multiphase optimization strategy
L. MacDonald and S Chapman have undertaken consultancy for (MOST) and the sequential multiple assignment randomized trial (SMART):
a UCL-business spin-out company (SOS). new methods for more potent eHealth interventions. Am. J. Prev. Med. 32,
R. Bowskill reports personal fees from Medical Director Priory S112–S118.
Colom, F. 2011. Keeping Therapies Simple: Psychoeducation in the Prevention Of
Hospital Brighton and Hove outside the submitted work.
Relapse in Affective Disorders.
M. Syrett has no potential conflicts of interest to disclose. Colom, F., Lam, D., 2005. Psychoeducation: improving outcomes in bipolar disorder.
R. Horne has undertaken speaker engagements with honoraria Eur. Psychiatry 20, 359–364.
Colom, F., Vieta, E., 2006. Psychoeducation Manual For Bipolar Disorder. Cambridge
with the following companies; Abbvie, Amgen, Biogen Idec, Gilead
University Press, Cambridge.
Sciences, GlasxoSmithKline, Janssen, Pfizer, Roche and Shire Colom, F., Vieta, E., Martinez-Aran, A., Reinares, M., Goikolea, J.M., Benabarre, A.,
Pharmaceuticals. He is a founder and shareholder of a UCL-busi- Torrent, C., Comes, M., Corbella, B., Parramon, G., Corominas, J., 2003. A ran-
ness spin out company providing consultancy on medication-re- domized trial on the efficacy of group psychoeducation in the prophylaxis of
recurrences in bipolar patients whose disease is in remission. Arch. General
lated behaviours to healthcare policy makers, provider and Psychiatry 60, 402–407.
industry. Colom, F., Vieta, E., Sanchez-Moreno, J., Martinez-Aran, A., Reinares, M., Goikolea, J.
M., Scott, J., 2005. Stabilizing the stabilizer: group psychoeducation enhances
the stability of serum lithium levels. Bipolar Disord. 7, 32–36.
Colom, F., Vieta, E., Sanchez-Moreno, J., Palomino-Otiniano, R., Reinares, M., Goi-
Acknowledgements kolea, J.M., Benabarre, A., Martinez-Aran, A., 2009. Group psychoeducation for
We would like to thank Sanaya Pardiwalla for her assistance in coding papers stabilised bipolar disorders: 5-year outcome of a randomised clinical trial. Br. J.
for the review and Peter Field for his advice on designing search strategies. We are Psychiatry 194, 260–265.
very grateful to the authors who responded to our enquiries for additional in- Coyne, J.C., Thombs, B.D., Hagedoorn, M., 2010. Ain't necessarily so: review and
formation on published papers included in the review and for providing additional critique of recent meta-analyses of behavioral medicine interventions in health
data. Finally, we would like to thank Dr Angel Chater and Professor John MacDonald psychology. Health Psychol. 29, 107.
for their time in reviewing the manuscript and providing valuable feedback. Craig, P., Dieppe, P., Macintyre, S., Michie, S., Nazareth, I., Petticrew, M., 2008. De-
veloping and evaluating complex interventions: the new Medical Research
Council guidance. BMJ 337, a1655.
Crowe, M., Porter, R., Inder, M., Lacey, C., Carlyle, D., Wilson, L., 2012. Effectiveness
Appendix A. Supplementary material of interventions to improve medication adherence in bipolar disorder. Aust. N.
Z. J. Psychiatry 46, 317–326.
Crowe, M., Whitehead, L., Wilson, L., Carlyle, D., O’brien, A., Inder, M., Joyce, P., 2010.
Supplementary data associated with this article can be found in Disorder-specific psychosocial interventions for bipolar disorder–a systematic
the online version at http://dx.doi.org/10.1016/j.jad.2016.01.002. review of the evidence for mental health nursing practice. Int. J. Nurs. Stud. 47,
896–908.
D’souza, R., Piskulic, D., Sundram, S., 2010. A brief dyadic group based psychoe-
ducation program improves relapse rates in recently remitted bipolar disorder:
References a pilot randomised controlled trial. J. Affect. Disord. 120, 272–276.
Davidson, K.W., Michael Goldstein, M., Kaplan, R.M., Kaufmann, P.G., Knatterud, G.
L., Spring, B., Trudeau, K.J., Whitlock, E.P., 2003. Evidence-based behavioral
Abc Project Team. 2013. Ascertaining barriers for compliance: policies for safe, ef- medicine: what is it and how do we achieve it? Ann. Behav. Med. 26, 161–171.
fective and cost-effective use of medicines in Europe. Final report of the ABC De Bruin, M., Viechtbauer, W., Hospers, H.J., Schaalma, H.P., Kok, G., 2009. Standard
project. [Online]. Medical University of Lodz. Available: 〈http://abcproject.eu/ care quality determines treatment outcomes in control groups of HAART-ad-
img/ABC%20Final.pd〉 (Accessed 19.03.14). herence intervention studies: implications for the interpretation and compar-
Baldessarini, R.J., Tondo, L., Davis, P., Pompili, M., Goodwin, F.K., Hennen, J., 2006. ison of intervention effects. Health Psychol. 28, 668–674.
Decreased risk of suicides and attempts during long-term lithium treatment: a Depp, C.A., Moore, D.J., Patterson, T.L., Lebowitz, B.D., Jeste, D.V., 2008. Psychosocial
meta-analytic review. Bipolar Disord. 8, 625–639. interventions and medication adherence in bipolar disorder. Dialogues Clin.
Ball, J., Mitchell, P., Corry, J., Skillecorn, A., Smith, M., Malhi, G., 2006. A randomized Neurosci. 10, 239–250.
controlled trial of cognitive therapy for bipolar disorder: Focus on long-term Desplenter, F.A., Simoens, S., Laekeman, G., 2006. The impact of informing psy-
change. J. Clin. Psychiatry 67, 277–286. chiatric patients about their medication: a systematic review. Pharm. World Sci.
Basco, M.R., Rush, A.J., 2005. Cognitive-Behavioral Therapy for Bipolar Disorder. 28, 329–341.
Guilford Press, New York. Dogan, S., Sabanciogullari, S., 2003. The effects of patient education in lithium
Batista, T.A., Baes, C.V.W., Juruena, M.F., 2011. Efficacy of psychoeducation in bipolar therapy on quality of life and compliance. Arch. Psychiatr. Nurs. 17, 270–275.
patients: systematic review of randomized trials. Psychol. Neurosci. 4, 409–416. Duval, S., Tweedie, R., 2000. Trim and fill: a simple funnel plot-based method of
Berk, L., Hallam, K.T., Colom, F., Vieta, E., Hasty, M., Macneil, C., Berk, M., 2010. testing and adjusting for publication bias in meta analysis. Biometrics 56,
Enhancing medication adherence in patients with bipolar disorder. Hum. Psy- 455–463.
chopharmacol.: Clin. Exp. 25, 1–16. Egger, M., Smith, G.D., Schneider, M., Minder, C., 1997. Bias in meta-analysis de-
Bordbar, M., Soltanifar, A., Talaei, A., 2009. Short-term family-focused psycho- tected by a simple, graphical test. BMJ 315, 629–634.
educational program for bipolar mood disorder in Mashhad. Iran. J. Med. Sci. Eker, F., Harkin, S., 2012. Effectiveness of six-week psychoeducation program on
34, 104–109. adherence of patients with bipolar affective disorder. J. Affect. Disord. 138,
Borenstein, M., Hedges, L.V., Higgins, J.P., Rothstein, H.R., 2011. Introduction to 409–416.
Meta-Analysis. John Wiley & Sons, New York. Elixhauser, A., Eisen, S., Romeis, J., Homan, S., 1990. The effects of monitoring and
Boutron, I., Moher, D., Altman, D.G., Schulz, K.F., Ravaud, P., 2008. Extending the feedback on compliance. Medical Care 28, 882–893.
CONSORT statement to randomized trials of nonpharmacologic treatment: Fernandez, R.S., Evans, V., Griffiths, R.D., Mostacchi, M.S., 2006. Educational inter-
explanation and elaboration. Ann. Intern. Med. 148, 295–309. ventions for mental health consumers receiving psychotropic medication: a
Busby, K.K., Sajatovic, M., 2010. REVIEW: Patient, treatment, and systems-level review of the evidence. Int. J. Ment. Health Nurs. 15, 70–80.
factors in bipolar disorder nonadherence: a summary of the literature. CNS Field, A.P., 2003. The problems in using fixed-effects models of meta-analysis on
Neurosci. Ther. 16, 308–315. real-world data. Underst. Stat.: Stat. Issues Psychol. Edu. Soc. Sc. 2, 105–124.
Frank, E., Kupfer, D.J., Thase, M.E., Mallinger, A.G., Swartz, H.A., Fagiolini, A.M., Michie, S., Van Stralen, M.M., West, R., 2011. The behaviour change wheel: a new
Grochocinski, V., Houck, P., Scott, J., Thompson, W., Monk, T., 2005. Two-year method for characterising and designing behaviour change interventions. Im-
outcomes for interpersonal and social rhythm therapy in individuals with bi- plement. Sci. 6, 42.
polar I disorder. Arch. Gen. Psychiatry 62, 996–1004. Miklowitz, D.J., 2006. A review of evidence-based psychosocial interventions for
Frank, E., Soreca, I., Swartz, H.A., Fagiolini, A.M., Mallinger, A.G., Thase, M.E., Gro- bipolar disorder. J. Clin. Psychiatry 67, 28–33.
chocinski, V.J., Houck, P.R., Kupfer, D.J., 2008. The role of interpersonal and Miklowitz, D.J., George, E.L., Richards, J.A., Simoneau, T.L., Suddath, R.L., 2003. A
social rhythm therapy in improving occupational functioning in patients with randomized study of family-focused psychoeducation and pharmacotherapy in
bipolar I disorder. Am. J. Psychiatry 165, 1559–1565. the outpatient management of bipolar disorder. Arch. Gen. Psychiatry 60,
Frank, E., Swartz, H.A., Mallinger, A.G., Thase, M.E., Weaver, E.V., Kupfer, D.J., 1999. 904–912.
Adjunctive psychotherapy for bipolar disorder: effects of changing treatment Miklowitz, D.J., Simoneau, T.L., George, E.L., Richards, J.A., Kalbag, A., Sachs-Ericsson,
modality. J. Abnorm. Psychol. 108, 579–587. N., Suddath, R., 2000. Family-focused treatment of bipolar disorder: 1-year
Freedland, K.E., Mohr, D.C., Davidson, K.W., Schwartz, J.E., 2011. Usual and unusual effects of a psychoeducational program in conjunction with pharmacotherapy.
care: existing practice control groups in randomized controlled trials of beha- Biol. Psychiatry 48, 582–592.
vioral interventions. Psychosom. Med. 73, 323–335. Moher, D., Liberati, A., Tetzlaff, J., Altman, D.G., 2009. Preferred reporting items for
Gaudiano, B.A., Weinstock, L.M., Miller, I.W., 2008. Improving treatment adherence systematic reviews and meta-analyses: the PRISMA statement. BMJ 339, b2535.
in bipolar disorder: a review of current psychosocial treatment efficacy and NICE, 2014. Bipolar Disorder: Assessment and Management (CG185). National In-
recommendations for future treatment development. Behav. Modif. 32, stitute for Health and Care Excellence, London.
267–301. Peet, M., Harvey, N.S., 1991. Lithium maintenance: 1. A standard education pro-
Gilbert, E.A., 2000. Examining the Therapeutic Process of Family-focused Treatment gramme for patients. Br. J. Psychiatry 158, 197–200.
for Bipolar Disorder. University of California, Los Angeles. Rea, M.M., Tompson, M.C., Miklowitz, D.J., Goldstein, M.J., Hwang, S., Mintz, J., 2003.
Harvey, N.S., Peet, M., 1991. Lithium maintenance: 2. Effects of personality and Family-focused treatment versus individual treatment for bipolar disorder:
attitude on health information acquisition and compliance. Br. J. Psychiatry 158, results of a randomized clinical trial. J. Consult. Clin. Psychol. 71, 482–492.
200–204. Reinares, M., Colom, F., Sánchez-Moreno, J., Torrent, C., Martínez-Arán, A., Comes,
Haynes, R.B., Ackloo, E., Sahota, N., Mcdonald, H.P., Yao, X., 2008. Interventions for M., Goikolea, J.M., Benabarre, A., Salamero, M., Vieta, E., 2008. Impact of care-
enhancing medication adherence. Cochrane Libr., 2. giver group psychoeducation on the course and outcome of bipolar patients in
Higgins, J.P., Altman, D.G., Gøtzsche, P.C., Jüni, P., Moher, D., Oxman, A.D., Savović, J., remission: a randomized controlled trial. Bipolar Disord. 10, 511–519.
Schulz, K.F., Weeks, L., Sterne, J.A., 2011. The Cochrane Collaboration's tool for Reinares, M., Sánchez-Moreno, J., Fountoulakis, K.N., 2014. Psychosocial interven-
assessing risk of bias in randomised trials. BMJ, 343. tions in bipolar disorder: what, for whom, and when. J. Affect. Disord. 156,
Hoffmann, T.C., Erueti, C., Glasziou, P.P., 2013. Poor description of non-pharmaco- 46–55.
logical interventions: analysis of consecutive sample of randomised trials. BMJ: Rosenthal, R., 1979. The file drawer problem and tolerance for null results. Psychol.
B. Med. J., 347. Bull. 86, 638–641.
Hoffmann, T.C., Glasziou, P.P., Boutron, I., Milne, R., Perera, R., Moher, D., Altman, D. Rothstein, H.R., Sutton, A.J., Borenstein, M., 2006. Publication Bias in Meta-Analysis:
G., Barbour, V., Macdonald, H., Johnston, M., Lamb, S.E., Dixon-Woods, M., Prevention, Assessment and Adjustments Wiley.com.
Mcculloch, P., Wyatt, J.C., Chan, A.W., Michie, S., 2014. Better reporting of in- Rouget, B.W., Aubry, J.-M., 2007. Efficacy of psychoeducational approaches on bi-
terventions: template for intervention description and replication (TIDieR) polar disorders: a review of the literature. J. Affect. Disord. 98, 11–27.
checklist and guide. Bmj-Br. Med. J., 348. Sajatovic, M., Davies, M., Hrouda, D.R., 2004. Enhancement of treatment adherence
Hong, J., Reed, C., Novick, D., Haro, J.M., Aguado, J., 2011. Clinical and economic among patients with bipolar disorder. Psychiatr. Ser. 55, 264–269.
consequences of medication non-adherence in the treatment of patients with a Sajatovic, M., Davies, M.A., Ganocy, S.J., Bauer, M.S., Cassidy, K.A., Hays, R.W., Safavi,
manic/mixed episode of bipolar disorder: Results from the European Mania in R., Blow, F.C., Calabrese, J.R., 2009. A comparison of the life goals program and
Bipolar Longitudinal Evaluation of Medication (EMBLEM) Study. Psychiatry Res. treatment as usual for individuals with bipolar disorder. Psychiatr. Ser. 60,
190, 110–114. 1182–1189.
Horne, R., Weinman, J., Barber, N., Elliott, R., Morgan, M., 2005. Concordance, Ad- Sajatovic, M., Velligan, D.I., Weiden, P.J., Valenstein, M.A., Ogedegbe, G., 2010.
herence and Compliance in Medicine Taking. NCCSDO. Measurement of psychiatric treatment adherence. J. Psychosom. Res. 69,
Hunter, J.E., Schmidt, F.L., 2000. Fixed effects vs. random effects meta-analysis 591–599.
models: implications for cumulative research knowledge. Int. J. Sel. Assess. 8, Schulz, K.F., Altman, D.G., Moher, D., 2010. CONSORT 2010 statement: updated
275–292. guidelines for reporting parallel group randomised trials. BMJ 340, c332.
Javadpour, A., Hedayati, A., Dehbozorgi, G.R., Azizi, A., 2013. The impact of a simple Scott, J., Paykel, E., Morriss, R., Bentall, R., Kinderman, P., Johnson, T., Abbott, R.,
individual psycho-education program on quality of life, rate of relapse and Hayhurst, H., 2006. Cognitive–behavioural therapy for severe and recurrent
medication adherence in bipolar disorder patients. Asian J. Psychiatry 6, bipolar disorders Randomised controlled trial. Br. J. Psychiatry 188, 313–320.
208–213. Scott, J., Pope, M., 2002. Self-reported adherence to treatment with mood stabi-
Lam, D.H., Bright, J., Jones, S., Hayward, P., Schuck, N., Chisholm, D., Sham, P., 2000. lizers, plasma levels, and psychiatric hospitalization. Am. J. Psychiatry 159,
Cognitive therapy for bipolar illness – a pilot study of relapse prevention. 1927–1929.
Cognit. Ther. Res. 24, 503–520. Simon, G., Ludman, E., Bauer, M., Unutzer, J., Operskalski, B., 2006. Long-term ef-
Lam, D.H., Hayward, P., Watkins, E.R., Wright, K., Sham, P., 2005. Relapse prevention fectiveness and cost of a systematic care program for bipolar disorder. Arch.
in patients with bipolar disorder: cognitive therapy outcome after 2 years. Am. Gen. Psychiatry 63, 500–508.
J. Psychiatry 162, 324–329. Valenstein, M., Kavanagh, J., Lee, T., Reilly, P., Dalack, G., Grabowski, J., Smelson, D.,
Lam, D.H., Watkins, E.R., Hayward, P., Bright, J., Wright, K., Kerr, N., Parr-Davis, G., Ronis, D., Ganoczy, D., Woltmann, E., Metreger, T., Wolschon, P., Jensen, A.,
Sham, P., 2003. A randomized controlled study of cognitive therapy for relapse Poddig, B., Blow, F., 2011. Using a pharmacy-based intervention to improve
prevention for bipolar affective disorder: outcome of the first year. Arch. Gen. antipsychotic adherence among patients with serious mental illness. Schizophr.
Psychiatry 60, 145–152. Bull. 37, 727–736.
Leclerc, E., Mansur, R.B., Brietzke, E., 2013. Determinants of adherence to treatment Van Gent, E.M., Zwart, F.M., 1991. Psychoeducation of partners of bipolar-manic
in bipolar disorder: a comprehensive review. J. Affect. Disord. 149, 247–252. patients. J. Affect. Disord. 21, 15–18.
Lenz, G., 2010. S29-02 Cognitive-psychoeducative group-therapy vs TAU with ad- Velligan, D.I., Weiden, P.J., Sajatovic, M., Scott, J., Carpenter, D., Ross, R., Docherty, J.
ditional information group: a randomized controlled study. Eur. Psychiatry 25, P., 2010. Assessment of adherence problems in patients with serious and per-
19. sistent mental illness: recommendations from the expert consensus guidelines.
Lingam, R., Scott, J., 2002. Treatment non-adherence in affective disorders. Acta J. Psychiatr. Pract. 16, 34–45.
Psychiatr. Scand. 105, 164–172. Walker, E., Hernandez, A.V., Kattan, M.W., 2008. Meta-analysis: its strengths and
Lolich, M., Vazquez, G.H., Alvarez, L.M., Tamayo, J.M., 2012. Psychosocial interven- limitations. Clevel. Clin. J. Med. 75, 431–439.
tions in bipolar disorder: a review. Actas Españolas de Psiquiatría 40, 84–92. Wittchen, H.-U., Jacobi, F., Rehm, J., Gustavsson, A., Svensson, M., Jönsson, B., Olesen,
Mathers, C.D., Fat, D.M., Boerma, J., 2008. The Global Burden of Disease: 2004 J., Allgulander, C., Alonso, J., Faravelli, C., 2011. The size and burden of mental
Update. World Health Organization, Geneva. disorders and other disorders of the brain in Europe 2010. Eur. Neuropsycho-
Michie, S., Johnston, M., Abraham, C., Francis, J., Eccles, M.P., 2013. The behavior pharmacol. 21, 655–679.
change technique taxonomy (v1) of 93 hierarchically clustered techniques: Zaretsky, A., Lancee, W., Miller, C., Harris, A., Parikh, S., 2008. Is cognitive-beha-
building an international consensus for the reporting of behavior change in- vioural therapy more effective than psychoeducation in bipolar disorder? Can. J.
terventions. Ann. Behav. Med. 46, 81–95. Psychiatry 53, 441–448.

Improving Medication Adherence in Bipolar Disorder - A Systematic Review and Meta-Analysis of 30 Years of Intervention Trials

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Improving Medication Adherence in Bipolar Disorder - A Systematic Review and Meta-Analysis of 30 Years of Intervention Trials

Transféré par

Droits d'auteur :

Formats disponibles

Journal of Affective Disorders 194 (2016) 202–221

Contents lists available at ScienceDirect

Journal of Affective Disorders

Improving medication adherence in bipolar disorder: A systematic

4.4. Study and sample characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214

1. Background et al., 2011) and whether it is educational or includes beliefs and

3.1. Eligibility criteria 3.2. Identiﬁcation of studies

Serum lithium levels. Physician report: Sig-

L. MacDonald et al. / Journal of Affective Disorders 194 (2016) 202–221

citations of previous reviews and studies included in this review

greater reduction in mis-

(n¼ 62) 83.87 (28.66), CG

6 months I (n ¼51) 90.20

(22.40), C (n¼ 55) 77.27

(n¼ 41) 95.73 (11.04), C

Lithium missed days:

(n¼ 39) 81.08 (30.85)

(n¼ 61) 81.15 (32.49)

3.3. Selection of trials

days, 0%, 25%, 50%,

not meet the inclusion criteria, inter-rater agreement was high

were excluded after discussion (Valenstein et al., 2011; Colom

articles (Fig. 1: PRISMA ﬂow diagram) (Moher et al., 2009).

Twenty-three studies met the inclusion criteria (Tables 1a and b).

the number of citations is higher than the number of studies

were excluded as they either comprised conference abstracts

Mean age 40, 68%

mixed psychiatric samples for which sufﬁcient detail for inclusion

3.4. Data extraction

sions, less than 10%

We designed and piloted a data extraction template before

making reﬁnements. We extracted data on intervention design

and delivery, study characteristics, quality of reporting and trial

design and adherence outcomes.

therapy, counselling, ac-

adherence enhancement, early warning sign identiﬁcation, lifestyle,

crisis management and communication. It involves changing be-

(Colom, 2011). We coded interventions in terms of whether they

used the term psychoeducation and whether the description of the

health nurse and ‘psychiatric

Other intervention description characteristics were; whether ad-

monthly unstructured group

counsellor’ in 6 weekly ses-

sions. Followed by optional

herence was the primary focus of the intervention, provider, format

intervention delivery, intensity of intervention contacts. Guidelines

on reporting behavioural intervention trials were used to sum-

interventions were classiﬁed as 0.03 months, in order that they had

Intervention tailoring: categorised (‘yes’, ‘no’ or ‘unclear’) ac-

cording to inclusion of different forms of tailoring based on

manual for Behaviour change techniques’: Individualization (con-

tent individualised to reﬂect participants' questions or individual

Multi-focus interventions (Individual/ couples/ family)

L. MacDonald et al. / Journal of Affective Disorders 194 (2016) 202–221

Sessions delivered to fa- married. than at 1 year follow-

L. MacDonald et al. / Journal of Affective Disorders 194 (2016) 202–221

delivered over 9 months

L. MacDonald et al. / Journal of Affective Disorders 194 (2016) 202–221

L. MacDonald et al. / Journal of Affective Disorders 194 (2016) 202–221

characteristics e.g. motivational stage or adherence level), Atten-

herent before and after

tion of patients cate-

ment (no speciﬁc education or psychological support mentioned)