Journal of Clinical Epidemiology 67 (2014) 1083e1092

The 2
2 cluster randomized controlled factorial trial design

is mainly used for efficiency and to explore intervention interactions:
a systematic review
Noreen D. Mdegea,*, Sally Brabyna, Catherine Hewittb, Rachel Richardsona,
David J. Torgersonb
a
Department of Health Sciences, University of York, Heslington, York YO10 5DD, UK
b
York Trials Unit, University of York, Heslington, York YO10 5DD, UK
Accepted 2 June 2014; Published online 22 July 2014

Abstract
Objectives: To describe the health care settings, purposes, and study reporting quality of the 2
2 cluster randomized controlled facto-
rial trial design.
Study Design and Setting: This study is a systematic review. We searched Medline, Embase, Cochrane Library, and Web of Knowl-
edge for articles published up to May 2012. Cluster randomized controlled 2
2 factorial trials in health, evaluating at least one complex
intervention, were included. Two authors independently reviewed and extracted data from the studies.
Results: Twenty-nine studies covering a wide range of clinical areas and health care settings were included. The cluster design was
mostly used to minimize contamination. The factorial design was mostly used to assess the effects of two interventions in the same study
and to explore interactions between interventions. However, although most studies explored the presence or absence of intervention inter-
actions, they were often either not sufficiently powered to detect any interactions or did not provide information on whether the study was
sufficiently powered to detect any interactions. There was a considerable variability in the reporting of a number of study characteristics and
methodological aspects. Study quality was also variable within and across studies.
Conclusion: The design has been used in a wide range of health care settings and clinical areas to minimize contamination, assess the
effects of two interventions in the same study, and explore intervention interactions. There is need for improvement on and guidelines for
the reporting of factorial trials. 2014 Elsevier Inc. All rights reserved.
Keywords: Systematic review; 2
2; Factorial; Cluster randomized; Randomized controlled trial; Complex interventions

1. Introduction penalty is a factorial design. Factorial RCTs allow the eval-

uation of more than one intervention in a single study [2].
Randomized controlled trials (RCTs) are the most robust
The simplest form of a factorial design is a 2
2 factorial
method of assessing the effectiveness of an intervention.
in which two interventions are evaluated. Let us suppose we
Most RCTs are simple two-armed designs with a single
have two interventions to compare against a control group.
intervention compared with a single control group [1].
In a factorial design, we would randomly allocate partici-
However, there are many clinical areas where there are pants to one of four groups: receiving both interventions,
more options that need evaluation. We could design multi-
intervention A, intervention B, or no intervention. This
armed studies with three or more interventions under eval-
way we can test whether intervention A is better than no
uation. However, these require increases in total sample
intervention A or whether intervention B is better than no
size, which would usually increase the cost and duration
intervention B. This gives us the same answer as to whether
of a trial. An alternative where there is no sample size
intervention A is better than no intervention A and inter-
vention B is better than no intervention B as does a
three-armed trial, but with a sample size that is at least a
Conflict of interest: The authors declare that they have no conflicts of third lower. However, the lower sample size requirement
interest and financial interests.
* Corresponding author. Tel.: 44-(0)1904321836; fax: 44-(0)
is only true when there is no substantive interaction be-
1904321383. tween the two interventions [1,3], which usually cannot
E-mail address: noreen.mdege@york.ac.uk (N.D. Mdege). be known with certainty when planning the study.
0895-4356/$ - see front matter 2014 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.jclinepi.2014.06.004
1084 N.D. Mdege et al. / Journal of Clinical Epidemiology 67 (2014) 1083e1092
What is new?
 The 2
2 cluster randomized factorial design has
been used in a wide range of health care settings
and clinical areas to minimize contamination, to
assess the effects of two interventions in the same
study, and to explore interactions between
interventions.
 Most studies explored the presence or absence of
intervention interactions but they were often either
not sufficiently powered to detect any interactions
or did not provide information on whether the
study was sufficiently powered to detect any
interactions.
 There is a considerable variability in the reporting
of a number of study characteristics and methodo-
logical aspects.
 There is need for improvement on and guidelines
for the reporting of factorial trials.
With a factorial, because we also have a combination of
interventions A and B, it is also possible to test whether in-
terventions A and B work together in synergy or are antag-
onistic, that is, whether the combined effects are greater
than or less than the additive effects one would observe if
the interventions acted independently of one another
[1,3]. However, powering a trial to detect an interaction
would usually mean that it would actually need to be larger
than a three-armed trial, thus removing one of the key ad-
vantages of a factorial design, which is to evaluate two in-
terventions for the price and the same sample size as
evaluating a single intervention. Although interactions are
often explored in factorial trials, very few of these trials
have sufficient power to demonstrate that an interaction ex-
ists, unless it is very large.
A review by McAlister et al. on the analysis and reporting
of factorial cardiac trials found that most of the trials (36 of
44) used the factorial design for reasons of efficiency
(testing two interventions in the same patient population)
and only 18% were done to assess the incremental benefits
of combining the two interventions [2]. However, in another
review focusing on factorial trials of complex interventions
in community settings, only 21% of the 76 included trials
specified the rationale for using the design [3].
In a factorial trial, randomization can be at the individ-
ual level, cluster level, or a mixture of individual and clus-
ter level. For cluster randomization, clusters, for example,
geographical areas, general practices surgeries, or hospitals,
are randomly allocated to the different groups. Although
other reviews have explored the analysis and reporting of
factorial trials in general or specific settings [2,3], this re-
view aims to describe the application of the 2
2 cluster
randomized controlled factorial design. Cluster designs
can be difficult to undertake, especially of complex inter-
ventions. Therefore, it may make sense to use factorial de-
signs more widely when undertaking cluster trials for the
sake of efficiency. In this review, we have systematically
identified a sample of 2
2 cluster factorial designs that
aim to test at least one complex intervention. We focused
on the 2
2 factorial designs to allow for a manageable
number of articles. We also assumed that because the 2

2 factorial is the simplest form of a factorial design, studies
using this design would be more common and are more
likely to be of better methodological, implementational,
and reporting quality than studies using more complex
factorial designs. Therefore, many conclusions and recom-
mendations from the review on how to improve the design,
conduct, and reporting of studies using this design are very
much likely to apply also to the other more complex forms
of the factorial design. The focus on complex interventions
is because of their wide use and evaluation in health ser-
vices, public health practice, and other areas relevant to
health [4] and because rigorously evaluating them is often
expensive both in terms of time and financial investments,
which makes the factorial design very attractive [3].
We have explored (1) the areas of health research in
which the design has been used, (2) general characteristics
of the design, (3) the motivation for using the design, and
(4) whether there is an interaction assessed and reported.
We also appraised the studies on whether sequence genera-
tion was adequate, participants were identified before clus-
ter randomization, recruitment bias was likely by assessing
evidence of cluster imbalances at baseline, sample size
calculation including whether the study was powered to
detect an intervention interaction, and whether the data
analysis took account of the cluster design.
2. Methods
The systematic review followed the principles recom-
mended by the Cochrane Handbook for Systematic Re-
views of Interventions [5] and Centre for Reviews and
Dissemination guidance for undertaking systematic reviews
[6]. The reporting procedures followed the Preferred Re-
porting Items for Systematic Reviews and Meta-Analysis
guidance [7].
2.1. Data sources and search methods
The following sources were searched for studies pub-
lished up to May 2012: Medline, Medline InProcess and
other nonindexed citations (1946 to May 15, 2012), Embase
(1980 to May 15, 2012), Cochrane Library (2012 issue 5),
and Web of Knowledge (1900 to May 15, 2012). The search
strategy comprised two sets combined with the AND oper-
ator: one set to identify RCTs and the other to identify
factorial designs [3]. The following phrases were used for
 N.D. Mdege et al. / Journal of Clinical Epidemiology 67 (2014) 1083e1092 1085

the factorial design: factorial, 2
2 and 2 by 2 [2,3]. The
full search strategy for Medline is shown in Appendix, and
this was adapted accordingly for the other databases. We
also hand searched the reference lists of previous reviews
of factorial design studies [2,3]. We limited the search to
English language. However no publication date limits or
geographic limits were applied.
Reviews and Dissemination [6] to take account of cluster
randomization. Our quality assessment criteria therefore
comprised the following items: accounting for the cluster
design in sample size calculation and in the analysis, ade-
quacy of sequence generation, recruitment of patients
before cluster allocation to study groups, evidence of clus-
ter imbalances, participant attrition, and cluster attrition.

2.2. Eligibility criteria

3. Results
Cluster randomized controlled factorial trials were
eligible. We classified a study as cluster randomized if clus- 3.1. Literature search and study selection
ter randomization was used for at least one of the interven-
tions under investigation. We focused on trials evaluating Fig. 1 shows the flow of articles through the review pro-
two interventions, of which one of the interventions had cess. Of a total of 3,829 records identified, 903 were dupli-
to be a complex intervention and not more than one of cates, and 2,742 were irrelevant, resulting in 184 full-text
the interventions was a drug. The UK Medical Research articles that were assessed for eligibility. One hundred
Council guidance on the definition of complex interven- fifty-five articles were ineligible, of which 41 were explan-
tions was used [8]. We also focused on pragmatic trials, atory trials, 32 not 2
2 factorial design, 47 not cluster
rather than trials of an explanatory or methodological na- design, 12 not randomized, 22 had no complex interven-
ture. We included studies of any population group. tion, and one study was not reported in English.
We excluded studies whose aim was to increase ques- Twenty-nine studies met the inclusion criteria.
tionnaire response and those in which the two interventions
under evaluation were both drugs as these are relatively 3.2. Characteristics of included trials and study design
straightforward designs and we were interested in cluster features
designs of complex interventions.
Twenty-two studies used cluster randomization for both
interventions (cluster-cluster design) [9e30], whereas
2.3. Study selection and data extraction
seven employed a split plot design in which cluster
One reviewer conducted the searches. The title and ab- randomization was used for one intervention and individual
stract of each record obtained from the searches were inde- participant randomization for the other [31e37] (Table 1).
pendently screened by two reviewers. Those records judged In terms of the study setting, eleven were conducted in
as potentially eligible were obtained as full articles which the United Kingdom [10,13e16,19,21,25,26,28,37], seven
were then independently assessed by two reviewers accord-
ing to the prespecified criteria and classified as include,
exclude, or insufficient information to make a decision.
Differences in opinion were resolved through consensus
or otherwise by referral to a third author.
For each eligible study, two reviewers independently ex-
tracted data from the published source. The extracted infor-
mation included study identifiers (lead author and
publication year), nature of intervention, setting and coun-
try, cluster size, number of clusters, motivation for using
the design, whether participants were identified before clus-
ter randomization, whether sample size calculation and
analysis undertaken accounted for the cluster design,
whether an interaction was suspected, reported as being
tested, whether one was present, and whether the study re-
ported being powered to detect an intervention interaction.
Differences in data extraction were resolved through dis-
cussion or through referral to a third reviewer.

2.4. Study quality assessment

We adapted the study quality criteria for RCTs recom-
mended by the Cochrane Handbook for Systematic Re-
views of Interventions [5] as well as the Centre for Fig. 1. Flow of articles through the systematic review process.
1086 N.D. Mdege et al. / Journal of Clinical Epidemiology 67 (2014) 1083e1092

Table 1. Characteristics of the included studies

Sample size;
Randomization unit cluster size;
Study Country (cluster-cluster/split plot) number of clusters Participants
Amsallem 2007 France Cluster-cluster (departmental 72; !10; 22 Cardiologists
level)
Bahrami 2004 UK Cluster-cluster (dental practices) 3,342; d; 51 16- to 24-year-old dental patients
Blalock 2002 USA Split plot (county level, then 714; d; 12 Adult women aged 40e56 years
individual level)
Brock 2008 Canada Cluster-cluster (family physician 1,029; d; 150 Adult men aged O18 years with
level) erectile dysfunction.
Cals 2009 The Netherlands Cluster-cluster (general practices 431; d; 20 Adult GP patients with lower
[GP]) respiratory tract infection
Cheater 2006 UK Cluster-cluster (family practices) 1,078; d; 157 GP patients aged O16 years and
urine incontinent
Clarkson 2008 UK Cluster-cluster (dentist 3,682; |25; 149 Children between the ages of 12e
practitioner level) 14 years
Daucourt 2003 France Split plot (hospital level, then 1,412; d; 6 hospitals, Patients requiring thyroid function
ward level) 67 wards tests
Eccles 2001 UK Cluster-cluster (GP) 247; d; 247 General practitioners
Fletcher 2004 UK Cluster-cluster (GP) 43,219; d; 109 Primary care adult patients aged
75 years
Herbert 2004 Canada Cluster-cluster (practice-based 200; d; 28 Physicians in active practice in
small group) groups of which O3 and no
fewer than 50% agreed.
Jafar 2009 Pakistan Cluster-cluster (community) 1,341; d; 12 Adults (40 years; blood
pressure  140/90)
Jiwa 2006 UK Cluster-cluster (GP) 716; d; 44 Adult cases of significant
pathologydcancer; large polyps
(3 cm); moderate-to-severe
diventricular disease, and
inflammatory bowel disease
Jones 2009 and Weymiller USA Split plot (endocrinologist/ 98; d; 21 Adult diabetic patients
2010 individual patient)
Kinney 2003 USA Cluster-cluster (area agencies 2,222; d; 16 Adults
on aging)
Langham 2002 UK Cluster-cluster (GP) 1,261; d; 17 All patients with cardiovascular
disease (CVD)
Larsen 2004 Denmark Cluster-cluster (blocks within a 9,605; d; 4 Adults (66 years)
municipality)
Ling 2009 USA Split plot (primary cancer 599; d; 10 Adults (50e79 years)
physician group practice/
individual patients)
Mannheimer 2006 USA Cluster-cluster (by site or group 928; d; 38 Adults
of sites)
Ravaud 2004 France Cluster-cluster (rheumatologist) 2,957; d; 867 Adults
Richards 2001 UK Cluster-cluster (GP) 6,133; d; 24 Female primary care patients aged
50e64 years and eligible for
breast screening
Riemersma-van der Lek The Netherlands Split plot (group care facilities/ 189; d; 12 Residents of Dutch homes for the
2008 individual patients) elderly. Not all had dementia.
Scholes 2006 USA Split plot (primary care clinics/ 12; d; 23 Different populations for second
individual patients) randomizationdsmaller subset
of population. Primary
intervention looked at women
aged 14e25 years. Subset
population women aged 14e
20 years
Thomas 2006 UK Cluster-cluster (primary care 370; d; 85 GP practices in NHS Grampian
practices)
Twardella and Brenner Germany Cluster-cluster (medical 587; 1e30; 82 Patients in primary care aged
2007 practices) 36-75 years smoking at least 10
a day.
Watson 2002 Scotland Cluster-cluster (community 108 baseline and Simulated visits to pharmacies
pharmacies) 276 postintervention
visit; d; 60
(Continued )
 N.D. Mdege et al. / Journal of Clinical Epidemiology 67 (2014) 1083e1092
Table 1. Continued
Randomization unit
Study Country (cluster-cluster/split plot)
Watson 2008 UK Split plot (GP/individual patients) 200; d; 91
Wesson 2008 Kenya Cluster-cluster (public health
clinics)
Zillich 2008 USA Cluster-cluster (primary care
prescribers)
in the United States [20,23,30,31,33,34,36], three in France
[9,24,32], two in Canada [11,17], two in The Netherlands
[12,35], and one each in Pakistan [18], Denmark [22], Ger-
many [27], and Kenya [29]. The clusters varied widely and
included family and/or general practitioner practices, hospi-
tals, wards within hospitals, communities, and pharmacies.
Most studies recruited adult patients.
Of the 29 included studies, 23 described the studies as
randomized in both the title and the abstract, whereas this
was stated only in the abstract for three studies, only in
the title for two studies, and neither the title nor the abstract
for one. The cluster design was specified in both the title
and the abstract by 13 studies, only in the title by one, only
in the abstract by seven, and neither in the title nor in the
abstract in the remaining eight. Only five studies were
described as a factorial design in both the title and the ab-
stract, whereas this was stated only in the abstract for 22
studies, only in the title for one, and neither in the title
nor in the abstract for one.
3.3. Areas of health research in which the design has
been used
The included studies covered a wide range of clinical
conditions, areas of health research, and interventions.
These are shown in Table 2. The interventions generally
aimed to improve at least one of the following: knowledge
transfer, guideline implementation, disease prevention,
health care professional practice and quality of care, refer-
rals, treatment decisions, health care utilization, and the
management of specific diseases.
3.4. Motivation for using the design
Twenty-three studies gave at least one reason for using the
factorial cluster randomized controlled design. Ten studies
gave a reason for cluster randomization, and the most cited
was to minimize contamination [9,12,13,16,18,23,33]. The
other reasons included to minimize the Hawthorne effect
[22], to optimize the pragmatic nature of the study [12], to
obtain reliable and unbiased results [15], to evaluate health it was 28% [27]; for another it was very high (76.9%) as out
1087
Sample size;
cluster size;
number of clusters Participants
Patients over 16 years consulting
with pain in one or both
shoulders for 12 months
611; d; 45 Family planning (FP) providers and
community-based distribution
agents at rural FP clinics
284; 5.9 (standard Continuously enrolled Medicaid
deviation [SD], 8.4, members at high risk of CVD, not
range, 1e51) per receiving statin medication in
practice and 2.2 (SD, the 18 months before the
2.0, range, 1e13) intervention
per provider; 48
system interventions at a population level [18], and the nature
of the intervention which meant randomization had to be at
cluster level [27]. Twenty studies gave at least one reason for
using a factorial design and the reasons were to assess the
effects of two interventions independently in the same
study [10,12e16,19e24,26,28,29,31e35] and to explore
interactions between the interventions [10,12e15,20,21,24,
29,31e33,35].
3.5. Assessment of intervention interaction effects
Results for the assessment of interaction effects are summa-
rized in Table 3. Eight studies specified an a priori interaction
hypothesis [11,12,14,17,18,23,25,28]; however, none of these
studies clearly specified the rationale for that hypothesis. Thir-
teen studies stated exploring interactions as the rationale for
using a factorial design [10,12e15,20,21,24,29,31e33,35],
of these, five studies assumed no interaction in the sample size
calculation [12,13,15,24,29], one study included interaction
effects in the sample size calculation [35], one study reported
not being statistically powered to detect any interactions [14],
and it was unclear in six studies whether the study had been
powered to explore interactions [10,20,21,31e33]. Of the 21
studies that assessed the presence or absence of interactions
[9,12e16,18,20,21,23,25,26,28e30,32e37], only three
studies were reported to be statistically powered to detect an
interaction [9,18,35] and five assumed no interaction in
the sample size calculation [12,13,15,26,29]. Statistically sig-
nificant interaction effects were observed in 4 of the 21 studies
assessing the presence or absence of an interaction [13,
18,20,35].
3.6. Study quality
The results from the study quality assessment are sum-
marized in Table 3. In addition, one study that showed ev-
idence of cluster imbalances adjusted for them in all
subsequent analysis [25]. Cluster attrition occurred in eight
studies. For five of these eight studies, cluster attrition rates
ranged from about 1% to 11% [10,11,14e16]; for one study
1088 N.D. Mdege et al. / Journal of Clinical Epidemiology 67 (2014) 1083e1092

Table 2. Interventions and areas of health research

Study Intervention description Area of health research
Amsallem 2007 Two modes of evidence-based knowledge transfer (active and Knowledge transfer among physicians
passive modes)
Bahrami 2004 Two different guideline implementation strategies: audit and Implementation of clinical guidelines on
feedback (A&F), and computer-aided learning (CAL) package unerupted third molars in primary dental care
Blalock 2002 A tailored educational intervention and a community-based Osteoporosis prevention
intervention on calcium intake and exercise activity
Brock 2008 Education of the primary care physician and of the patient in the Practitioner and patient education on erectile
treatment of erectile dysfunction with vardenafil function
Cals 2009 Point of care testing for C reactive protein (disease approach) and Antibiotic use in lower respiratory tract infection
training in enhanced communication skills
Cheater 2006 A&F and educational outreach in improving nurses practice and Nursing practice on urinary incontinence
patient outcomes
Clarkson 2008 A fee for applying sealant and a 1 day educational workshop on Improving fissure-sealing rates
evidence-based practice
Daucourt 2003 Two guideline implementation interventions: a memorandum pocket Implementation of thyroid function testing
card and a test request form guidelines
Eccles 2001 A&F and educational reminder messages on general practitioner Primary care radiology referrals
requests for radiological tests
Fletcher 2004 Multidimensional assessment and management strategies for older Assessment and management of older people
people: of universal vs. targeted assessment and management by
hospital outpatient vs. primary care teams
Herbert 2004 Case-based educational modules and personal prescribing feedback General practitioner prescribing for hypertension
in primary care
Jafar 2009 Family-based home health education (HHE) delivered by trained Hypertension management and control
community health workers and annual training of general
practitioners in hypertension management
Jiwa 2006 Electronic interactive referral proforma and educational outreach Patient referral by general practitioners to
visits specialist for lower bowel symptoms.
Jones 2009 A decision aid (statin choice or pamphlet about cholesterol) and two Treatment decisions by patients with
Weymiller 2010 modes of delivery of decision aid (delivered either during the diabetes mellitus
office visit [by the clinician] or before the visit [by a researcher])
Kinney 2003 Normative Treatment Planning and a Client Feedback System Quality improvement in community-based
long-term care
Langham 2002 Training in information management and evidence-based medicine Recording, prescribing, and control of
cardiovascular disease (CVD) risk factors
in primary care
Larsen 2004 Environment and Health Program and calcium and vitamin D Osteoporotic fracture prevention in
supplementation community-dwelling residents
Ling 2009 Tailored vs. nontailored physician recommendation letter and Colorectal screening in primary care practice
enhanced vs. nonenhanced physician office and patient
management intervention
Mannheimer 2006 Medication management and medication alarms Antiretroviral medication adherence
Ravaud 2004 Home-based exercise program and standardized evaluation tool Osteoathritis
Richards 2001 General practitioner letter and prompt in patient notes for increasing Breast screening
breast screening uptake
Riemersma-van der Lek Dim or bright light with and without melatonin for cognitive and Gerontology/neurology/neuropsychology
2008 noncognitive function in elderly care home residents
Scholes 2006 Clinic- and patient-level interventions to improve chlamydia Public health
screeningdenhanced guideline implementation and chart prompt
Thomas 2006 Feedback on practice test request rates and brief educational Primary care/hematology
reminders to decrease laboratory test requests from general practices
Twardella and Brenner Training for general practitioner plus general practitioner payment vs. Public health
2007 training for general practitioner plus reimbursement of pharmacy
costs for participants to improve smoking cessation rates
Watson 2002 Guideline dissemination strategies: educational outreach and/or Pharmacy
attendance at a continuing professional education session for
appropriate identification and treatment of candidiasis by
pharmacists and pharmacy staff
Watson 2008 Lignocaine and/or corticosteroid and general practitioner training for Primary care
acute shoulder pain in primary care
Wesson 2008 Two different types of outreach interventions and detailing design Family planning
to increase intrauterine device use
Zillich 2008 Academic detailing by specially trained nurses and telephonic care CVD
management for patients to improve statin prescribing
 N.D. Mdege et al. / Journal of Clinical Epidemiology 67 (2014) 1083e1092 1089

Table 3. Assessment of intervention interaction effects assessment

Number judged Number unclear (ie, not
Number judged as satisfying as not satisfying enough information for
Criterion the quality criterion the quality criterion a judgment)
Assessment of intervention interaction effects
A priori interaction hypothesis 8 (7 No interaction and 1 synergy) 21 0
Interaction stated as rationale for using 13 7 9
factorial design
Assessed the presence or absence of 21 8 0
interactions
Interaction included in sample size calculation
Those stating interaction as rationale for 6 (5 No interaction and 1 specified magnitude) 1 6
using factorial design (n 5 13)
Those assessing presence or absence of 8 (5 No interaction, 2 specified magnitude, and 2 11
interactions (n 5 21) 1 magnitude not specified)
Study quality
Accounting for cluster design in sample size 19 3 7
calculation
Adequate sequence generation 17 0 12
Recruiting participants before allocation of 10 (2 Recruiting participants before allocation of 18 1
clusters/recruiting all participants in clusters, 7 recruiting all participants in
cluster/not relevant cluster, and 1 not relevant)
No evidence of cluster imbalances 6 13 10
Cluster design accounted for in the analysis 24 3 2

of the 91 recruited clusters, 51 did not recruit any partici-
pants and 19 withdrew from the study [37]; and for one
the attrition rate was not clearly specified [35]. Participant
follow-up rates, or the proportion of participants for which
outcome data were available at the end of the follow-up
period, were 100% for seven studies [12,17,19,26,28,
30,36], 75e99% for 16 studies [9,11,14,15,18,21,23e25,27,
28,31e34,37], 50e74% for two studies [10,13], !50% for
three studies [16,20,35], and data on follow-up rates were
not reported for the remaining two.
4. Discussion
The 2
2 cluster randomized factorial design has been
used in a wide range of health care settings and clinical
areas: mostly for educational interventions and patient
assessment and disease and/or condition management stra-
tegies. The design has mostly been used in developed coun-
tries and with cluster randomization for both interventions
rather than the split plot design. The most commonly cited
reason for the factorial design was efficiency (ie, to assess
the effects of two interventions independently in the same
study), followed by exploring interactions between the in-
terventions, as has been observed in other reviews [2,3].
The cluster design is chosen mostly as a means of mini-
mizing contamination.
Some researchers have recommended reporting of esti-
mates of interaction between interventions for factorial
design trials as desirable as unrecognized interactions could
distort results and their interpretation [38,39]. From our re-
view, however, most studies either reported not being suffi-
ciently powered to detect any interaction or did not provide
information on the matter. Only 4 of the 21 studies (19%)
that assessed the presence or absence of interactions actu-
ally found a statistically significant interaction. In their re-
view, McAlister et al. reported that only 2 of 29 trials
actually found a statistically significant interaction and
seven had clinically significant interaction ratios (1.25
or 0.8) [2]. They concluded that investigators conducting
factorial design trials are appropriately choosing to test on-
ly those interventions that do not have potential for substan-
tive interactions. For our present review, however, it is
difficult to conclude whether this is the case or whether
the lack of interactions was due to the majority of studies
being underpowered to detect any interaction. In relation
to this, we recommend that investigators specify their a pri-
ori hypothesis of whether an interaction exists or does not
exist between interventions, as well as a rationale for such
a hypothesis. A factorial design should not be used if a sub-
stantive intervention interaction is suspected (or sought) un-
less the trial is adequately powered. It is a good design if
one is confident there is no interaction or alternatively if
the study specifically wants to estimate the interaction
and powers the study accordingly.
Torgerson suggests that when considering cluster
randomization over individual randomization as a way to
minimize contamination, researchers have to be certain that
for their trial, contamination is real rather than a theoretical
possibility [40]. However, many trial reports do not give de-
tails on why contamination is a real possibility. For
example, of the seven studies that cited the possibility of
contamination, only one gave details: once general practi-
tioners within a practice had been trained in new communi-
cation skills they could not switch at random between using
these skills and usual consulting practice [12].
One of the problems associated with cluster randomiza-
tion is the possibility of recruitment bias where the person
1090 N.D. Mdege et al. / Journal of Clinical Epidemiology 67 (2014) 1083e1092
recruiting participants for the study can use their foreknowl-
edge of treatment allocation to select patients to enroll into
the trial [41]. Moreover, in cluster randomized trials in
which participants are able to opt out of or avoid the trial
intervention, participants can differentially refuse consent
to participate in the trial. This could result in differences
in participant characteristics between the various arms of
the trial, thereby defeating the objective of randomization.
Methods of minimizing the possibility of selection bias in
cluster randomized trials have been proposed and these
include identifying participants before randomization or uti-
lizing an independent person, preferably blind to allocation,
to recruit participants [41]. A review of 113 cluster random-
ized trials published in MEDLINE in 2008 found that about
40% of those trials had recruited participants before cluster
randomization [42] compared with only approximately 7%
(2 of 29) for our present review. Moreover, for the studies
included in our review, in which cluster imbalances were
evident (13 studies), they were rarely adjusted for in the
analysis. There is therefore a need to improve on incorpo-
rating methods for minimizing cluster imbalances and their
impact when designing and analyzing 2
2 cluster random-
ized factorial studies. Our review also demonstrates the need
to improve on the reporting of any cluster imbalances and
methods used for sequence generation.
Although almost all studies specified they were random-
ized and factorial designs in the title and/or abstract, approx-
imately a third (8 of 29) did not specify that they were a
cluster design. Six of the eight studies that did not specify
the cluster design were published between 2006 and 2009,
which is after the first CONSORT statement extension to
cluster randomized controlled trials (CRCTs) published in
2004, which recommended the identification of CRCTs as
cluster randomized in the title or abstract of reports [43].
Other reviews have also found that the majority of factorial
design studies identify themselves as such in the title or ab-
stract [2,3], whereas the percentage of CRCTs that clearly
identified themselves as cluster randomized in the title or ab-
stract is low [44]. Using currently existing search strategy for
identifying RCTs in electronic databases such as MEDLINE
yields high sensitivity for identifying CRCTs [44]. However,
specific reference to cluster randomized trial in the title or
abstract of trial reports it also important as it facilitates sim-
ple and efficient identification of that subset of trials when
required [44], such as for our study in which we were inter-
ested in the cluster randomized controlled factorial design
and not just randomized controlled factorial design.
Factorial design studies are difficult to identify through
electronic searches because of the absence of MeSH terms
or consistent text words [2,3]. Our search might therefore
have missed some studies, particularly those that did not
use any design-specific words or phrases in their title or ab-
stract identifying them as 2
2 factorial cluster random-
ized controlled designs. We however supplemented the
electronic search with hand searching of previous reviews
of factorial design studies.
5. Conclusion
In conclusion, the 2
2 cluster randomized factorial
design has been used for evaluating complex interventions
in a wide range of health care settings and clinical areas to
minimize contamination, assess the effects of two interven-
tions in the same study, and explore intervention interac-
tions. However, although most studies explored the
presence or absence of intervention interactions, they were
often either not sufficiently powered to detect any interac-
tions or did not provide information on whether the study
was sufficiently powered to detect an interaction. There is
need for improvement on the following: (1) clear provision
of information on whether the study was powered enough
to detect an interaction if one is suspected or if the study
intends to estimate the interaction, (2) the inclusion of data
on whether any cluster imbalances were present at baseline
in the study reports, and (3) incorporating methods for
minimizing cluster imbalances and their impact during
study design and analysis. The variability in the reporting
of some methodological aspects such as motivation for
using the design, a priori hypothesis for intervention
interactions including a rationale, and other aspects such
as sequence generation supports calls from other
researchers for the need for guidelines for reporting of
factorial trials [2,3].
Appendix
Search strategy for MEDLINE (OVID SP)
1946ePresent
1. factorial.mp.
2. 2 by 2.mp.
3. 2 x 2.mp.
4. or/1-3
5. randomized controlled trial.pt.
6. controlled clinical trial.pt.
7. (randomized or randomised or rct$).ti,ab.
8. placebo$.ti,ab.
9. (crossover$ or cross over$ or cross-over$ or (doubl$
adj blind$) or (singl$ adj blind$)).ti,ab.
10. randomly.ti,ab.
11. trial$.ti,ab.
12. clinical trials as topic.sh.
13. or/32-39
14. 4 and 13
References
[1] Torgerson DJ, Torgerson CJ. Designing randomised trials in health,
education and the social sciences: an introduction. New York, NY:
Palgrave Macmillan; 2008.
[2] McAlister FA, Straus SE, Sackett DL, Altman DG. Analysis and report-
ing of factorial trialsda systematic review. JAMA 2003;289:2545e53.
[3] Montgomery AA, Astin MP, Peters TJ. Reporting of factorial trials of
complex interventions in community settings: a systematic review.
Trials 2011;12:179.
 N.D. Mdege et al. / Journal of Clinical Epidemiology 67 (2014) 1083e1092
[4] Victora CG, Habicht J-P, Bryce J. Evidence-based public health:
moving beyond randomized trials. AJPH 2004;94:400e5.
[5] Higgins JPT, Green S, editors. Cochrane handbook for systematic re-
views of interventions 5.1.0 [updated March 2011]. The Cochrane
Collaboration; 2008. Available at. www.cochrane-handbook.org. Ac-
cessed December 3, 2011.
[6] Centre for Reviews and Dissemination. Systematic reviews: CRDs
guidance for undertaking reviews in health care. York, UK: Univer-
sity of York; 2009.
[7] Moher D, Liberati A, Tetzlaff J, Altman DG, The PRISMA group.
Preferred Reporting Items for Systematic Reviews and Meta-
Analyses: the PRISMA Statement. BMJ 2009;339:b2700.
[8] Craig P, Dieppe P, Macintyre S, Michie S, Nazareth I, Petticrew M,
et al. Developing and evaluating complex interventions: the new
Medical Research Council guidance. BMJ 2008;337:a1655.
[9] Amsallem E, Kasparian C, Cucherat M, Chabaud S, Haugh M,
Boissel JP, et al. Evaluation of two evidence-based knowledge trans-
fer interventions for physicians. A cluster randomized controlled
factorial design trial: the CardioDAS Study. Fundam Clin Pharmacol
2007;21:631e41.
[10] Bahrami M, Deery C, Clarkson JE, Pitts NB, Johnston M, Ricketts I,
et al. Effectiveness of strategies to disseminate and implement clin-
ical guidelines for the management of impacted and unerupted third
molars in primary dental care, a cluster randomised controlled trial.
Br Dent J 2004;197:691e6.
[11] Brock G, Carrier S, Alarie P, Pommerville P, Casey R, Harris S, et al.
The effect of physician and patient education when combined with
vardenafil treatment in Canadian males with erectile dysfunction:
an open-label, factorial-designed, cluster randomized clinical trial.
J Sex Med 2008;5:705e15.
[12] Cals JWL, Butler CC, Hopstaken RM, Hood K, Dinant G-J. Effect of
point of care testing for C reactive protein and training in communi-
cation skills on antibiotic use in lower respiratory tract infections:
cluster randomised trial. BMJ 2009;338:b1374.
[13] Cheater FM, Baker R, Reddish S, Spiers N, Wailoo A, Gillies C, et al.
Cluster randomized controlled trial of the effectiveness of audit and
feedback and educational outreach on improving nursing practice
and patient outcomes. Med Care 2006;44:542e51.
[14] Clarkson JE, Turner S, Grimshaw JM, Ramsay CR, Johnston M,
Scott A, et al. Changing clinicians behavior: a randomized
controlled trial of fees and education. J Dent Res 2008;87:640e4.
[15] Eccles M, Steen N, Grimshaw J, Thomas L, McNamee P, Soutter J, et al.
Effect of audit and feedback, and reminder messages on primary-care
radiology referrals: a randomised trial. Lancet 2001;357:1406e9.
[16] Fletcher AE, Price GM, Ng ES, Stirling SL, Bulpitt CJ, Breeze E,
et al. Population-based multidimensional assessment of older people
in UK general practice: a cluster randomised factorial trial. Lancet
2004;364:1667e77.
[17] Herbert CP, Wright JM, Maclure M, Wakefield J, Dormuth C, Brett-
MacLean P, et al. Better Prescribing Project: a randomized controlled
trial of the impact of case-based educational modules and personal
prescribing feedback on prescribing for hypertension in primary care.
Fam Pract 2004;21:575e81.
[18] Jafar TH, Hatcher J, Poulter N, Islam M, Hashmi S, Qadri Z, et al.
Community-based interventions to promote blood pressure control
in a developing country: a cluster randomized trial. Ann Intern
Med 2009;151:593e601.
[19] Jiwa M, Skinner P, Coker AO, Shaw L, Campbell MJ, Thompson J.
Implementing referral guidelines: lessons from a negative outcome
cluster randomised factorial trial in general practice. BMC Fam Pract
2006;7:65.
[20] Kinney ED, Kennedy J, Cook CA, Freedman JA, Lane KA, Hui SL.
A randomized trial of two quality improvement strategies imple-
mented in a statewide public community based, long-term care pro-
gram. Med Care 2003;41:1048e57.
[21] Langham J, Tucker H, Sloan D, Pettifer J, Thom S, Hemingway H.
Secondary prevention of cardiovascular disease: a randomised trial
1091
of training in information management, evidence-based medicine,
both or neither: the PIER trial. Br J Gen Pract 2002;52:818e24.
[22] Larsen ER, Mosekilde L, Foldspang A. Vitamin D and calcium sup-
plementation prevents osteoporotic fractures in elderly community
dwelling residents: a pragmatic population-based 3-year intervention
study. J Bone Miner Res 2004;19:370e8.
[23] Mannheimer SB, Morse E, Matts JP, Andrews L, Child C,
Schmetter B, et al. Sustained benefit from a long-term antiretroviral
adherence intervention. Results of a large randomized clinical trial. J
Acquir Immune Defic Syndr 2006;43(Suppl 1):S41e7.
[24] Ravaud P, Giraudeau B, Logeart I, Larguier JS, Rolland D, Treves R,
et al. Management of osteoarthritis (OA) with an unsupervised home
based exercise programme and/or patient administered assessment
tools. A cluster randomised controlled trial with a 2
2 factorial
design. Ann Rheum Dis 2004;63:703e8.
[25] Richards SH, Bankhead C, Peters TJ, Austoker J, Hobbs FD,
Brown J, et al. Cluster randomised controlled trial comparing the
effectiveness and cost-effectiveness of two primary care interventions
aimed at improving attendance for breast screening. J Med Screen
2001;8:91e8.
[26] Thomas RE, Croal BL, Ramsay C, Eccles M, Grimshaw J. Effect of
enhanced feedback and brief educational reminder messages on lab-
oratory test requesting in primary care: a cluster randomised trial.
Lancet 2006;367:1990e6.
[27] Twardella D, Brenner H. Effects of practitioner education, practi-
tioner payment and reimbursement of patients drug costs on smoking
cessation in primary care: a cluster randomised trial. Tob Control
2007;16:15e21.
[28] Watson MC, Bond CM, Grimshaw JM, Mollison J, Ludbrook A,
Walker AE. Educational strategies to promote evidence-based com-
munity pharmacy practice: a cluster randomized controlled trial
(RCT). Fam Pract 2002;19:529e36.
[29] Wesson J, Olawo A, Bukusi V, Solomon M, Pierre-Louis B,
Stanback J, et al. Reaching providers is not enough to increase
IUD use: a factorial experiment of academic detailing in Kenya. J
Biosoc Sci 2008;40:69e82.
[30] Zillich AJ, Ackermann RT, Stump TE, Ambuehl RJ, Downs SM,
Holmes AM, et al. An evaluation of educational outreach to improve
evidence-based prescribing in Medicaid: a cautionary tale. J Eval
Clin Pract 2008;14:854e60.
[31] Blalock SJ, DeVellis BM, Patterson CC, Campbell MK,
Orenstein DR, Dooley MA. Effects of an osteoporosis prevention
program incorporating tailored educational materials. Am J Health
Promot 2002;16:146e56.
[32] Daucourt V, Saillour-Glenisson F, Michel P, Jutand MA, Abouelfath A.
A multicenter cluster randomized controlled trial of strategies to
improve thyroid function testing. Med Care 2003;41:432e41.
[33] Jones LA, Weymiller AJ, Shah N, Bryant SC, Christianson TJH,
Guyatt GH, et al. Should clinicians deliver decision aids? Further
exploration of the statin choice randomized trial results. Med Decis
Making 2009;29:468e74.
[34] Ling BS, Schoen RE, Trauth JM, Wahed AS, Eury T, Simak DM,
et al. Physicians encouraging colorectal screening: a randomized
controlled trial of enhanced office and patient management on
compliance with colorectal cancer screening. Arch Intern Med
2009;169:47e55.
[35] Riemersma-van dar Lek RF, Swaab DF, Twisk J, Hol EM,
Hoogendijk WJ, Van Someren EJ. Effect of bright light and melatonin
on cognitive and noncognitive function in elderly residents of group care
facilities: a randomized controlled trial. JAMA 2008;299:2642e55.
[36] Scholes D, Grothaus L, McClure J, Reid R, Fishman P, Sisk C, et al.
A randomized trial of strategies to increase chlamydia screening in
young women. Prev Med 2006;43:343e50.
[37] Watson J, Helliwell P, Morton V, Adebajo A, Dickson J, Russell I,
et al. Shoulder acute pain in primary healthcare: is retraining effec-
tive for GP principals? SAPPHIREda randomized controlled trial.
Rheumatology 2008;47:1795e802.
1092 N.D. Mdege et al. / Journal of Clinical Epidemiology 67 (2014) 1083e1092

[38] Lubsen J, Pocock SJ. Factorial trials in cardiology: pros and cons. Eur
Heart J 1994;15:585e8.
[39] Brittian W, Wittes J. Factorial designs in clinical trials: effects of non-
compliance and subadditivity. Stat Med 1989;8:161e71.
[40] Torgerson DJ. Contamination in trials: is cluster randomisation the
answer? BMJ 2001;322:355e7.
[41] Hahn S, Puffer S, Torgerson DJ, Watson J. Methodological bias in
cluster randomised trials. BMC Med Res Methodol 2005;5:10.
[42] Giraudeau B, Caille A, Le Gouge A, Ravaud P. Participant informed
consent in cluster randomized trials: review. PLoS One 2012;7:e40436.
[43] Campbell MK, Elbourne DR, Altman DG, CONSORT group. CONSORT
statement: extension to cluster randomised trials. BMJ 2004;328:702e8.
[44] Taljaard M, McGowan J, Grimshaw JM, Brehaut JC, McRae A,
Eccles MP, et al. Electronic search strategies to identify reports of cluster
randomized trials in MEDLINE: low precision will improve with adher-
ence to reporting standards. BMC Med Res Methodol 2010;10:15.