MEDICINE

Review Article

Cluster-Randomized Studies
Part 25 of a Series on Evaluating Scientific Publications

Eva Lorenz, Sascha Köpke, Holger Pfaff, Maria Blettner

C
luster-randomized trials (CRT) are often carried
Summary out to evaluate the kind of complex interventions
that are increasingly being adopted in health ser-
Background: Cluster-randomized trials (CRT) are needed to compare interventions
vices research, for example (1). Complex interventions
that are allocated to entire groups of subjects, rather than to individuals. Publi-
consist of several individual interventions that may inter-
cations about CRT have become steadily more common over the past decade.
act with each other. An example from Germany is a study
Readers of such publications should be able to categorize and interpret the findings
on guideline-based reduction of the use of physical re-
of CRT correctly while considering the methodological requirements applicable to
this type of study. straints (PR) in residential care homes, where information
sheets were distributed, training courses given, and PR
Methods: This review is based on a selection of pertinent literature and on the officers designated in the facilities involved (Box 1).
authors’ expertise. CRT-specific methodological aspects of the planning, The number of publications concerning them-
performance, and interpretation of studies are discussed. selves with CRT has increased continuously over the
past 10 years (the PubMed search term “cluster
Results: Readers of publications on CRT should check whether due consideration
randomised trial” OR “cluster-randomised trial” OR
has been given to correlations within and between the clusters during the planning
“cluster randomized trial” OR “cluster-randomized
of the study. These correlations enable the determination whether persons within a
trial” threw up 54 hits in 2006, 156 in 2011, and 392
cluster resemble each other more closely, or respond more similarly to the study
in 2016), and articles on CRT formed a four times
intervention, than persons drawn from different clusters. It should also be checked
higher proportion of all Medline-indexed studies in
whether the randomization for the study has been carried out with such methods as
2016 than they did in 2006.
stratification and covariate-adjusted randomization. CRT can be analyzed on either
the individual or the cluster level. The rationale for the choice of a cluster-
In a CRT, not individual participants but rather
randomized design should be explained, and intracluster correlation coefficients whole facilities or groups of participants (clusters)
(ICC) should be reported as an aid to the planning of future studies. Particular are allocated to one or more intervention or control
requirements are also described in an extended version of the CONSORT group (2). For instance, in the above-mentioned
guidelines that has been developed specifically for CRT. study on reduction of PR, all study participants under
the care of a given physician received the same
Conclusion: Readers of publications on CRT should be aware of the special require- information and the same ensuing intervention,
ments mentioned above with respect to the design, performance, and analysis of without being influenced by other participants who
this type of study as opposed to individually randomized studies. If no special received the control intervention (3). Formation of
techniques are applied in the design, performance, and analysis of a CRT, or if the clusters was necessary in this study (Box 1) because
assumptions underlying each of these steps have not been properly checked, then the residents of a care home cannot be considered as
the findings of the study may well be misleading. independent of each other.
A study may have to be conducted as a CRT if an
Cite this as:
Lorenz E, Köpke S, Pfaff H, Blettner M: Cluster-randomized studies—part 25 of a intervention is being performed not at individual
series on evaluating scientific publications. Dtsch Arztebl Int 2018; 115: 163–8. level but at the level of whole regions or organi-
DOI: 10.3238/arztebl.2018.0163 zations. Examples of such situations are the restruc-
turing of a hospital, the implementation of
guidelines, or the introduction of a new form of care.
The intervention cannot be withheld from any
individual in the organizational unit for fear of con-
tamination. In the case of the intervention to reduce
physical restraints in residential care homes, individ-
ual randomization is impossible because intervention
addresses the care given in the whole facility.
Institute for Medical Biostatistics, Epidemiology and Informatics, Mainz University Medical Center:
Dr. Lorenz, Prof. Blettner This article describes design considerations,
Department of Teaching and Research in the Care Sector, Institute for Social Medicine and Epidemi-
randomization strategies, statistical methods, and the
ology, University of Lübeck: Prof. Köpke strengths and limitations of CRT. Our aim is to en-
Institute for Medical Sociology, Health Services Research, and Rehabilitation Science, University of able the reader to scrutinize and interpret the results
Cologne: Prof. Pfaff of CRT in critical fashion, taking into account the
Center for Health Services Research Cologne (ZVFK), University of Cologne: Prof. Pfaff methodological requirements.

Deutsches Ärzteblatt International | Dtsch Arztebl Int 2018; 115: 163–8 163
MEDICINE

BOX 1

Cluster-randomized trials (CRT) in residential care homes for the elderly

Despite legal restrictions and evidence of deficiencies in effectiveness and safety, physical restraints (PR) are often employed in care homes. In
the framework of a CRT, a guideline-based complex intervention was carried out to reduce the use of PR, including distribution of information
materials, training courses, and appointment of PR officers in each facility. The intervention addressed processes in the whole facility. Randomi-
zation at the level of individual residents is therefore not feasible owing to the danger of contamination. The primary endpoint was the prevalence
of PR 6 months after commencement of the intervention, as determined in the course of direct, unannounced visits by blinded external study
assistants.
Sample calculation: For reduction of PR prevalence, an change from 33% to 21% in 6 months was assumed, with power of 90% at a signifi-
cance level of 5% in a two-sided test. On the basis of earlier studies the intracluster correlation coefficient (ICC) was estimated at 0.034, resulting
in a design factor of 5.0 and thus a necessary sample size of 2824 residents in 34 care homes. The mean cluster size was 83 residents. In
anticipation of a drop-out rate of 5% of facilities and 2% of residents (not including deaths and moves away), the final sample was fixed at 3060
residents in 36 care homes.
Avoidance of bias: Prospective publication of the study protocol, stratified concealed block randomization after acquisition of the baseline data
of all participating residents, inclusion of all persons resident in each facility at the time of the survey by waiving individual consent, intensified
(and successful) efforts to avoid loss of clusters, blinded documentation of the primary endpoint by external study assistants including verification
of successful blinding.
Results: All care homes completed the study and all residents were included in the analyses. The prevalence of PR went down from 30.6% to
29.1% in the control group and from 31.5% to 22.6% in the intervention group over the 6-month study period.
Conclusion: The guideline-based complex intervention reduced the use of PR in care homes.

(Modified from Köpke et al. 2012 [18])

Planning Case number planning necessarily includes

Power and case number calculation deciding on the number of clusters and the number of
The formation of clusters decreases the effective individuals per cluster. The latter is often determined
sample size and thus the statistical power of CRT by the basic investigation unit and may vary widely
compared with individually randomized trials, be- (e.g., number of residents in a care home for the
cause persons within an organizational unit resemble elderly). In addition to the usual assumptions in case
each other more strongly than persons from different number calculation, an assumption of the anticipated
organizational units. Similarity within and between ICC is needed. The required sample size increases
clusters is quantified with the aid of the intracluster with increasing ICC.
correlation coefficient (see equation in Box 2). For the One must also take into consideration that the
purposes of calculation the clusters are assumed to be power can no longer be increased very much when
the same size; extension to take account of unequal the number of individuals per cluster exceeds 1/ICC
cluster sizes is possible (4). (5). In other words, it is not helpful to include a lot of
The resulting necessity to enlarge the sample size is “large” clusters in one’s study. In this case a random
known as the design effect (DE) (see equation in sample can be selected within a cluster, meaning that
Box 2) (5). Thus, a DE of 1.2 means that the cluster fewer individuals are included than are available in
size, assuming equally large clusters, has to be that cluster. With an ICC of 0.05, a total of around 20
increased by 20% compared with individual ran- (1/0.05) persons per cluster suffices to achieve the de-
domization. If there is no correlation within the sired power. There are often no valid data on the ICC,
cluster (i.e., intracluster correlation coefficient [ICC] in which case a literature-based figure or a realistic
= 0), then DE = 1 and the sample size of the CRT value based on earlier studies should be used. Experi-
corresponds to that of a study with individual ran- ence shows that an ICC of around 0.05 can be
domization. Conversely, if all elements in a cluster assumed for studies of primary care (6); in commu-
react to the intervention in the same way (i.e., nity-randomized studies the ICC is usually lower
ICC = 1), then the number of clusters needed is the (0.01 or often even 0.001 [2]). The resulting DE is
same as the number of individuals in an individually nevertheless large in large clusters.
randomized study (Table). Furthermore, the necessary number of cases
● If the DE is ignored when planning a CRT, type 2 depends on the size of the clusters: 100 clusters each
error increases (Box 3). containing 10 probands lead to greater statistical
● Ignoring the DE and the resulting decrease in power than 10 clusters of 100 probands each. Recruit-
variance in the course of analysis at cluster level ment of additional clusters yields greater effective
leads to an increase in type 1 error (Box 3). case numbers than recruitment of more individuals in

164 Deutsches Ärzteblatt International | Dtsch Arztebl Int 2018; 115: 163–8
 MEDICINE

BOX 2 TABLE

Effective sample size and power at constant total sample size with different
Equations numbers of clusters, numbers of patients per cluster, intracluster correlation
coefficients, and design effects
● Intracluster correlation coefficient (ICC)
The ICC is defined as Clusters k Patients m Total (mk) ICC DE ESS Power
t-test*
144 1 144 0.00 1.00 144 85
where is the variability among clusters and is 72 2 144 0.01 1.01 142 84
the variability within clusters. 72 2 144 0.05 1.05 137 82
36 4 144 0.01 1.03 140 82
● Design effect (DE)
The DE is defined as 36 4 144 0.05 1.15 125 78
18 8 144 0.01 1.07 135 78
where is the average cluster size. 18 8 144 0.05 1.35 107 68
8 18 144 0.01 1.17 123 67
● Randomization scheme 8 18 144 0.05 1.85 78 48
The number of possible randomization schemes is given
as , where n is the the total number of clusters and k * Power in % for demonstration of an effect size of 0.5 between clusters assigned equally to intervention and
the number of clusters per study arm. control ( two-sided t-test in studies with two arms of the same size: α = 0.05; standard deviation SD = 1).
DE, Design effect; ESS, effective sample size; ICC, intracluster correlation coefficient; k, number of clusters;
m, number of patients per cluster; mk, total number of patients included

clusters (7). As an ad-hoc approach, the number of pend on the intervention arm. Equally, the patients’
cases for individual randomized trials can be motivation to take part may be affected by previous
calculated and multiplied by the DE. The formula has knowledge of the various interventions that are to be
to be expanded in the case of extreme variation in compared. Brierley et al. published a review of sus-
cluster size (8). ceptibility to recruitment bias (16). To avoid this
A large number of strategies for case number plan- source of distortion, the recruitment of study partici-
ning in CRT have been published, and implemented pants should be completed before randomization. Be-
with the aid of various statistical analysis softwares cause the study staff and patients often cannot be
such as R and Stata (9–13). blinded, at least the documentation of the primary
outcome parameter should be accomplished by
Potential pitfalls in planning and sources of bias in others.
cluster-randomized trials
The Cochrane Handbook (Higgins & Green 2011, [14]) Trial conduct
lists four specific potential sources of distortion in the Randomization
context of CRT: Equal distribution of potential influencing factors and
● Recruitment bias sources of disturbance is a precondition for being able
● Baseline imbalance among groups to attribute observed effects to an intervention. The
● Loss of clusters units of randomization in a CRT can be care homes (see
● Incorrect analysis example), hospital groups, hospitals, hospital wards,
Distortion can arise as early as the recruitment doctors’ offices, schools, or whole local communities.
stage, if participants cannot be followed-up for the These groups do not arise by chance but as the result of
whole duration of the study or the intention-to-treat social, geographic, or other interacting factors. Various
(ITT) analysis is not carried out. To avoid this source randomizing strategies exist to nevertheless ensure
of bias, it should be ensured that data can be acquired even distribution.
from all members of the randomized clusters (or of
the random sample). In the event of incomplete Simple randomization
follow-up, techniques for dealing with missing data In simple (unrestricted) randomization, the clusters are
should be employed (15). assigned randomly to the treatment and control arms. In
Allocation concealment (blinding; see Box 3) is the case of a small number of clusters of varying size, this
often not feasible in a CRT, where bias can arise may result in wide discrepancies in sample size. Consider-
through intervention assignment. For instance, the able imbalance of study participants’ characteristics can
motivation of study staff to recruit patients can de- arise both at cluster level and at the individual level.

Deutsches Ärzteblatt International | Dtsch Arztebl Int 2018; 115: 163–8 165
MEDICINE

BOX 3
participants’ relevant characteristics. The aim is to
make the arms of the trial as homogeneous as possible.
Definitions of central terms In the case of a small number of clusters this balance
runs against the principle of randomness and may lead
● Type 1 error: A type 1 error is present when the null hypothesis is wrongly to an increased risk of selection bias. In minimization
rejected. the number of covariables for stratification is limited,
● Type 2 error: A type 2 error is present when the null hypothesis is wrongly so that the variables that are considered can also be
retained. modeled when it comes to analysis. Clusters are deter-
ministically assigned to an intervention or the control
● Blinding: This means that study participants, study physicians, other study group according to relevant variables. In this way ob-
staff, or assessors are not aware of the assigned intervention and are servable confounders can be balanced between the
therefore not influenced by that knowledge. study arms.
● Allocation concealment: With concealed allocation, the persons who carry out
randomization do not know to which study group the next participant will be Covariable randomization
allotted. Another approach is covariable-restricted random-
ization, in which clusters are allotted to the study arms
in equal numbers according to the distribution of rel-
evant basic variables (19–21). For constant variables
one takes account of aggregated data such as mean
Matching values within clusters or strata. Data from the basic
To prevent blatant mismatching of the clusters in the data acquisition stage must already be available at the
intervention and control groups from the outset, the time of randomization. A randomization scheme is
participating clusters are paired with regard to factors selected randomly from among those that result in bal-
such as age, sex, cultural background, socioeconomic anced study arms with regard to predefined relevant
status, and occupation. In our example, randomization properties and exposures. Because the final ran-
would be preceded by formation of “cluster pairs,” domization scheme is selected from the group of all
each comprising two care homes with similar age theoretically possible schemes (see the equation for the
structure and sex distribution (Figure). In each cluster number of possible randomization schemes in Box 2),
pair, one cluster is randomly selected for the interven- randomness of assignment to intervention or control is
tion, thus guaranteeing that the two arms of the trial largely preserved.
are balanced. However, this means that whenever a The evaluation of CRT takes place on at least two
cluster leaves the study (loss to follow-up), the paired levels, namely the cluster level and the individual (pa-
cluster has to be excluded. To alleviate this problem, tient) level. In the multi-level models the statistical
matching can be set aside at the data analysis stage (17). model is expanded by adding a random component
for the variation of the clusters (21, 22). This takes
Stratification account of the ICC resulting from the design. A lucid
In the case of stratification the study population is account of how to carry out a multi-level analysis is
divided into disjoint groups (“strata”). In the study by provided by Ansmann et al. (23).
Köpke et al., study regions 1 and 2 were stratified for
randomization (Figure) (18). Each stratum is homo- Presentation
geneous with regard to relevant characteristics, but the The use of CRT has increased steeply in the past 15
strata may differ very widely from one another. years. This has led to expansion of the CONSORT
Clusters for the intervention and control arms are guidelines (www.consort-statement.org/) on the
chosen randomly to form equally sized blocks in each publication of trials of this type, because CRT design
stratum. The number of strata should be kept low so presents specific methodological challenges (24, 25).
that balanced blocks result. This requirement often The principal extensions of the CONSORT guidelines
stands in opposition to the frequent need for ran- with regard to CRT are as follows:
domization to take account of a large number of ● The reasons for deciding to perform a CRT should
variables by differentiation of cluster and individual be explicitly laid out.
level. For example, stratification in a geographic re- ● The provision for the influence of clustering in the
gion with four values and two funding bodies would individual phases of the study from case number
involve division of the clusters into eight strata. Such a calculation through randomization to analysis
strategy can lead to underoccupation of individual should be described.
cells. ● The ICC should be presented as a basis for case
number planning in future studies.
Minimization The reporting of CRT in medical research currently
The minimization method represents a compromise be- displays major deficits. Therefore, it is very important
tween balance and (true) randomization. Individual that authors plan their studies in accordance with the
clusters are allocated sequentially to the intervention expanded CONSORT guidelines or, for example, the
arm and the control arm while taking account of the stepped wedge design (26).

166 Deutsches Ärzteblatt International | Dtsch Arztebl Int 2018; 115: 163–8
 MEDICINE

Discussion FIGURE
The first step in planning a study is to decide whether it
can be performed with individual randomization or
whether a CRT is necessary. An acceptable reason for
carrying out a CRT is that the intervention is being per-
formed in clusters and there would be a risk of contam-
ination in an individually randomized trial. Alternative
study types for organizational interventions are the
stepped wedge and crossover designs, in which the
clusters are included in analysis both as intervention Intervention Control
and as control entities.
The planning and conduct of CRT presents special
challenges differing from the requirements for indi-
vidually randomized trials. The clustering must be re-
tained at all stages, from case number planning
through analysis techniques to reporting. Study con-
duct also involves specific challenges, e.g., regarding
selection bias and information bias. Matching and stratification: The clusters shaded dark green (region 1) and light green
Whether conclusions should be drawn at the indi- (region 2) represent the two study regions in our example of a study of care homes. The
vidual patient level or at cluster level is determined by different-colored contours (red, orange, yellow) of the clusters show homogeneous cluster
the choice of design and analysis technique (5). To pairs. The various colors of the individuals in each cluster (light red, medium red, dark red)
identify the various age groups.
increase analytical precision, strict inclusion and ex-
Matching in cluster-randomized studies: In the matching process, the maximally homogeneous
clusion criteria have to be defined. This can be cluster pairs are divided randomly between the intervention and control arms of the study with
achieved, for example, by recruiting physicians’ of- regard to the predefined participant characteristics (in this example: age).
fices of similar size or physicians with similar profes- Stratification in cluster-randomized studies: In stratified randomization, care homes from a
sional experience. It is always important to consider region are randomly selected from the whole set of care homes in that region and divided
the benefit of an intervention not only at cluster level equally between the two study arms.
but also at the level of individual patients, e.g.,
improvement in quality of life by reduction of PR or
improvement in the reputation of the care home with Acknowledgments
We are grateful to Prof. Lena Ansmann and Michael Swora for their
fewer complaints from relatives. valuable comments which helped to improve the quality of this article.
When planning a study, the study-specific ICC can
be estimated on the basis of a baseline survey. More-
over, stratification variables that are already relevant Conflict of interest statement
The authors declare that no conflict of interest exists.
can be identified. Because CRT are frequently used to
evaluate “complex” interventions (27), one should
adhere to the corresponding guidelines, such as those Manuscript submitted on 12 June 2017, revised version accepted on
26 October 2017
of the UK Medical Research Council (MRC) (28).
It is often argued that the conduct of CRT is associ-
Translated from the original German by David Roseveare
ated with less administrative effort, e.g., in connec-
tion with the acquisition of aggregated data. On the
other hand, the consent of the study participants must References
be obtained at two levels, because although the inter- 1. Hayes RJ, Moulton LH: Cluster randomized trials. 1 ed: Chapman and
Hall/CRC Press, Boca Raton, FL; 2009.
vention is carried out at cluster level, when it comes 2. Donner A, Klar N: Design and analysis of cluster randomization trials
to analysis there are frequently interesting parameters in health research. London: Arnold Publishers Limited 2000.
at the individual level. For large communities, it may 3. Bland JM, Kerry SM: Trials randomised in clusters. BMJ 1997; 315:
600.
be logistically challenging or even impossible to ob-
4. Eldridge SM, Ashby D, Kerry S: Sample size for cluster randomized
tain informed consent for all individual study partici- trials: effect of coefficient of variation of cluster size and analysis
pants (5, 29). However, this should not necessarily be method. Int J Epidemiol 2006; 35: 1292–300.
viewed as a limitation of ethical requirements, pro- 5. Donner A, Klar N: Pitfalls of and controversies in cluster randomization
trials. Am J Public Health 2004; 94: 416–22.
vided there is sufficient justification (30). The level at
6. Campbell MJ: Cluster randomized trials in general (family) practice
which consent is necessary depends on the interven- research. Stat Methods Med Res 2000; 9: 81–94.
tion, on the study-specific data protection regulations, 7. Flynn TN, Whitley E, Peters TJ: Recruitment strategies in a cluster
and on the specific requirements of the ethics com- randomized trial—cost implications. Stat Med 2002; 21: 397–405.
8. Kerry SM, Bland JM: Unequal cluster sizes for trials in English and
mittee involved. In some situations it may be justified Welsh general practice: implications for sample size calculations. Stat
to go ahead in the absence of informed consent, for Med 2001; 20: 377–90.
instance if the intervention only tangentially affects 9. Campbell MK, Thomson S, Ramsay CR, MacLennan GS, Grimshaw
individual persons. An example is the introduction of JM: Sample size calculator for cluster randomized trials. Comput Biol
Med 2004; 34: 113–25.
new rules regarding hygiene, which does not require 10. Donner A, Birkett N, Buck C: Randomization by cluster. Sample size
the agreement of all patients. requirements and analysis. Am J Epidemiol 1981; 114: 906–14.

Deutsches Ärzteblatt International | Dtsch Arztebl Int 2018; 115: 163–8 167
MEDICINE

Key messages
● Cluster-randomized trials (CRT) are frequently used when interventions are to be carried out at the level of whole groups rather
than single individuals.
● As a result of cluster formation, persons within a group often have more characteristics in common than persons in different
groups . The so-called intracluster correlation coefficient should be reported as measure of similarity of individuals between and
within clusters.
● The results of CRT can be distorted by recruitment bias, baseline imbalance, loss of clusters, and incorrect analysis.
● Blinding of the participants and study personnel in a CRT is frequently not feasible. This may result in differing motivation and
thus become a source of recruitment bias. If at all possible, therefore, recruitment of study participants should be complete before
randomization.
● In CRT where the study staff are not blinded, the outcome parameters should be documented by an external person.

11. Feng Z, Diehr P, Peterson A, McLerran D: Selected statistical issues in 23. Ansmann L, Kuhr K, Kowalski C, für die Arbeitsgruppe Organisations-
group randomized trials. Annu Rev Public Health 2001; 22: 167–87. bezogene Versorgungsforschung des DNVF: Mehrebenenanalysen in
12. Guittet L, Giraudeau B, Ravaud P: A priori postulated and real power der organisationsbezogenen Versorgungsforschung – Nutzen, Voraus-
in cluster randomized trials: mind the gap. BMC Med Res Methodol setzungen und Durchführung. Gesundheitswesen 2017; 79: 203–9.
2005; 5: 25. 24. Campbell MK, Elbourne DR, Altman DG: CONSORT statement:
13. Kerry SM, Bland JM: Sample size in cluster randomisation. BMJ 1998; extension to cluster randomised trials. BMJ 2004; 328: 702–8.
316: 549. 25. Campbell MK, Piaggio G, Elbourne DR, Altman DG: Consort 2010
14. Higgins J, Green S (eds.): Cochrane handbook for systematic reviews statement: extension to cluster randomised trials. BMJ 2012; 345:
of interventions. Version 5.1.0 [updated March 2011]. The Cochrane e5661.
Collaboration 2011. http://handbook.cochrane.org (last accessed on 1 26. Bland JM: Cluster randomised trials in the medical literature: two
December 2017). bibliometric surveys. BMC Med Res Methodol 2004; 4: 21.
15. Giraudeau B, Ravaud P: Preventing bias in cluster randomised trials.
PLoS Medicine 2009; 6: e1000065. 27. Mühlhauser I, Lenz M, Meyer G: Bewertung von komplexen Interven-
tionen: Eine methodische Herausforderung. Dtsch Arztebl 2012; 109:
16. Brierley G, Brabyn S, Torgerson D, Watson J: Bias in recruitment to A 22–3.
cluster randomized trials: a review of recent publications. J Eval Clin
Pract 2012; 18: 878–86. 28. Craig P, Dieppe P, Macintyre S, Michie S, Nazareth I, Petticrew M:
Developing and evaluating complex interventions: the new Medical
17. Diehr P, Martin DC, Koepsell T, Cheadle A: Breaking the matches in a Research Council guidance. BMJ 2008; 337: a1655.
paired t-test for community interventions when the number of pairs is
small. Stat Med 1995; 14: 1491–504. 29. Giraudeau B, Caille A, Le Gouge A, Ravaud P: Participant informed
18. Köpke S, Muhlhauser I, Gerlach A, et al.: Effect of a guideline-based consent in cluster randomized trials: review. PLoS One 2012; 7:
multicomponent intervention on use of physical restraints in nursing e40436.
homes: a randomized controlled trial. JAMA 2012; 307: 2177–84. 30. Sim J, Dawson A: Informed consent and cluster-randomized trials. Am
19. Ivers N, Halperin I, Barnsley J, et al.: Allocation techniques for balance J Public Health 2012; 102: 480–5.
at baseline in cluster randomized trials: a methodological review. Trials
2012; 13: 120.
20. Lorenz E, Gabrysch S: ccrand: Covariate-constrained randomization Corresponding author
routine for achieving baseline balance in cluster-randomized trials. Dr. sc. hum. Eva Lorenz
Stata J 2017; 17: 503–10. Institut für Medizinische Biometrie, Epidemiologie und Informatik (IMBEI)
21. Moulton LH: Covariate-based constrained randomization of group- Universitätsmedizin der Johannes Gutenberg Universität Mainz
randomized trials. Clin Trials (London, England) 2004; 1: 297–305. Obere Zahlbacher Str. 69
22. Diez-Roux AV: Bringing context back into epidemiology: variables and 55131 Mainz, Germany
fallacies in multilevel analysis. Am J Public Health 1998; 88: 216–22. [email protected]

168 Deutsches Ärzteblatt International | Dtsch Arztebl Int 2018; 115: 163–8