Planning a cluster randomized controlled trial Methodological issues

Christie, J., O'Halloran, P., & Stevenson, M. (2009). Planning a cluster randomized controlled trialMethodological issues. Nursing Research, 58, 128-134.

Published in:Nursing Research

Queen's University Belfast - Research Portal:Link to publication record in Queen's University Belfast Research Portal

General rightsCopyright for the publications made accessible via the Queen's University Belfast Research Portal is retained by the author(s) and / or othercopyright owners and it is a condition of accessing these publications that users recognise and abide by the legal requirements associatedwith these rights.

Take down policyThe Research Portal is Queen's institutional repository that provides access to Queen's research output. Every effort has been made toensure that content in the Research Portal does not infringe any person's rights, or applicable UK laws. If you discover content in theResearch Portal that you believe breaches copyright or violates any law, please contact openaccess@qub.ac.uk.

Download date:11. Nov. 2020

METHODS

Planning a ClusterRandomized

Controlled Trial

Methodological Issues

Janice Christie 4 Peter O’Halloran 4 Mike Stevenson

b Background: The standard approach in a randomized controlled trial (RCT) is to ran-domize individuals to intervention and control groups. Yet, nursing and other health

interventions are often implemented at the levels of health service organizational unit

or geographical area. It may be more appropriate to conduct a cluster RCT. How-

ever, cluster randomization requires consideration of a number of important issues.

b Objective: The objective of this study was to show how critical issues in relation to

design and analysis can be addressed.

b Approach: Two cluster RCTs conducted by the authors are used as examples.

Guidance on the conduct and reporting of cluster RCTs is also offered.

b Results: A rationale for choosing this design was provided, and issues in relation to

study design, calculation of sample size, and statistical analysis were clarified. A

decision tree and checklist are provided to guide researchers through essential

steps in conducting a cluster RCT.

b Discussion: Cluster RCTs present special challenges in relation to design, conduct,

and analysis. Nevertheless, they are an appropriate and potentially powerful tool for

nursing research. With careful attention to the issues addressed in this article,

researchers can use this approach successfully.

b Key Words: cluster randomized trial &multilevel model & nested structure

A lthough many nurseYpatient in-teractions are one-to-one en-

counters, nursing is essentially a groupactivity. Most nurses work with groupsof colleagues, for example, as part of award team in a hospital in NorthernIreland or as a network of communitynurses in rural Romania. Similarly,nurses care for people who live orwork in groups, perhaps students froma school in Idaho or members of amountain tribe in Thailand. Conse-

quently, when carrying out nursing re-search, whether focusing on outcomesfor nurses themselves or their patients,it is important to recognize that partici-pants are often linked through mem-bership of a group, and hence, any datacollected potentially are clustered. Clus-tered data are collected from peoplewho are members of a group and whomay be presumed, by virtue of the factthat they are members of that group,to have a greater similarity to those

within the group than that to individ-uals outside it. These similarities maybe evident to the observer or unknown.Data collected from such clusters shouldbe distinguished from data that are col-lected from unrelated individuals andgrouped by a researcher on the basis ofcommon characteristics (e.g., age andgender) for the purposes of subgroupanalysis. Early researchers often did notaddress clustering effects either becauseof computation difficulties or becausesuch effects were not recognized. In re-cently conducted studies, however, it ismore likely that clustered data are con-trolled for (Bland, 2004).

In this article, the importance oftaking into account the clustered na-ture of the data obtained from groupsin planning and analyzing randomizedcontrolled trials (RCTs) is considered.A variant of randomized trials, some-times called group randomized orcommunity randomized trials but more

128 Nursing Research March/April 2009 Vol 58, No 2

Janice Christie, PhD, MA, RN, RSCPHN,PGCE, is Teaching Fellow, School of Nurs-ing and Midwifery; Peter O’Halloran, PhD,MSc, RN, RNT, is Lecturer, School of Nurs-ing and Midwifery; andMike Stevenson, BSc,FSS, PGCHET, is Senior Lecturer in Medi-cal Statistics, School of Medicine and Den-tistry, Queen’s University Belfast, Belfast,Northern Ireland, United Kingdom.

Editor’s Note

Additional information provided by the authors expanding this article ison the Editor’s Web site at http://www.nursing-research-editor.com.

9Copyright @ 200 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

commonly known as cluster random-ized trials, will be discussed. Here,groups of people rather than individu-als are randomized to intervention andcontrol groups. Also identified are themajor methodological issues that needto be addressed when planning andundertaking a cluster RCT. This is illus-trated through discussion of two trials,one an evaluation of a policy to providehip protectors (O’Halloran et al., 2004)and the other an evaluation of homevisitation schedules offered to first-time mothers by health visitors (UnitedKingdom-based specialist communitypublic health nurses; Christie, 2005).

Reasons for Randomizing by ClusterThe standard approach in an RCT isto randomize individuals, and, as dis-cussed later, randomizing by clusterraises a number of important designimplications. So why use this approachto test an intervention? Such studiesmay be justified on a number of grounds(Medical Research Council, 2002). First,it may be that the intervention itself isdesigned to be delivered to groups ofpeople rather than to individualsVforexample, nurse-facilitated smoking ces-sation support groups. Alternatively,the intervention might be a treatmentthat requires health professionals tochange their behavior to impact patientoutcomesVfor example, where nursesprovide an educational intervention forpatients with diabetes. It would be dif-ficult for such nurses to avoid using theintervention techniques on patients des-ignated as controls. Randomizing bycluster may also be used if there is a riskof contamination, where individualsrandomized to receive an intervention,within a group, may influence otherswithin the group. For example, patientsattending a primary care practice whoare offered breast screening may talk toother control-group-allocated patientsattached to the practice who may thenask for screening themselves. A clusterRCT may be justified if the interventioninvolves supplying equipment to a par-ticular set of units (clinics, hospitals,etc.). In this scenario, it may be cheaperor more convenient to randomize byunit rather than to supply every unitwith the equipment.

Randomizing by cluster can alsobe appropriate for pragmatic trials inwhich the effectiveness of an interven-tion in routine clinical practice is mea-

sured. This method typically requiresthat interventions are offered to all whoare likely to benefit rather than to ran-domly assigned individuals within a par-ticular patient subgrouping. It should benoted that pragmatic trials are differentfrom explanatory trials which measurethe efficacy of an intervention underideal conditions, often using carefullyselected participants in a research envi-ronment (Roland & Torgerson, 1998).

Effects of Cluster RandomizationRandomization of individuals in a tra-ditional or nonclustered RCT is car-ried out in an attempt to ensure thatthere is a probabilistically equal dis-tribution of relevant individual char-acteristics between intervention andcontrol groups. Standard statistical testsare suitable for this design because theyare based on the assumption that thecharacteristics of individuals (and there-fore their outcomes) are independent ofone another. With a cluster RCT design,clusters (rather than individuals) arerandomized to treatment and controlgroups. Consequently, statistical inde-pendence cannot be assumed becausegeographical areas and organizationstend to contain individuals who aremore similar to each other than individ-uals within other areas or organizations.The causes of these group similaritieswill be discussed later, with examplesto help nurse researchers identify poten-tial clusters in their study population.

There are a number of reasons clus-tering effects exist within study popula-tions, including choices of individuals,the impact of factors that directly in-fluence the whole cluster, and indirecteffects of the cluster’s social environ-ment. For example, older people ortheir families may select nursing carehomes on the basis of cost, care ethos,or perceived desirability of the institu-tion. In consequence, individual choicemay result in residents within the homebeing more similar in status or aspira-tion than that of individuals in other es-tablishments. The effect of a commoncluster-level influence is evident whenma-ternalwell-beingoutcomes vary systemati-cally between the case loads of communitynurses. This is because community nursesmay have differing approaches to provid-ing care (Gomby, 1999).

Outcomes may also vary betweenclusters for reasons related to the widersocial environment. For example, com-

munity nurses may be allocated fami-lies to visit on the basis of geographicalcatchment areas. Within these areas,residents and hence potential clientsare likely to share socioeconomic sta-tus and environmental influences ontheir health status. If such patients arerandomized according to geographicalarea, the variation in health betweenclusters (area) is likely to be greaterthan the health variation within clus-ters (Ukoumunne, Guilliford, Chinn,Sterne, & Burney, 1999). Once poten-tial clusters have been identified, theresearcher will need to consider howthese clustering effects will influence thedesign and analysis of a cluster RCT.

Cluster Randomization Design IssuesThe unit of analysis within a clusterrandomized trial can be either thecluster or the individuals within clus-ters. When analyzing individual-leveldata, the researcher must take intoaccount the lack of statistical indepen-dence of data collected within each clus-ter. In this case, the researcher needs tostatistically account for clustering effectsat the power calculation and data analy-sis stages of a cluster RCT. Ukoumunneet al. (1999) recommend that research-ers address a number of issues whendesigning a cluster RCT, including tak-ing the cluster design into account whencalculating the sample size, allowing forthe number and size of included clus-ters, considering the use of stratificationof clusters, and taking account of thecluster design when analyzing the re-sults of the study.

At the outset, the cluster designmust be taken into account when esti-mating the sample size. Standard sam-ple size calculations are based on theassumption that the responses of indi-viduals within clusters are indepen-dent. This assumption is unwarrantedin a cluster RCT. In comparison withan RCT randomized at the level of theindividual, a cluster RCTwith the samesample size has a reduced power to de-tect an intervention effect, thus increas-ing the risk of concluding that there isnot an effect when in reality there is (aType II error). Therefore, it is essentialthat an accurate sample size is calcu-lated before the study by allowing forthe effect of a cluster randomized design(Figure 1). A cluster randomized trialwill always require a larger sample sizethan that of a comparative noncluster

Nursing Research March/April 2009 Vol 58, No 2 Cluster Randomized Controlled Trial 129

9Copyright @ 200 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

trial (Kerry & Bland, 1998). This willhave cost and workload implications,and therefore calculation of the pro-spective sample size should be a preludeto decisions regarding the feasibility ofa cluster randomized study.

The introduction of a covariate thatexplains (and hence reduces) between-cluster variation will also increase thepower of a study (Raudenbush, 1997).For example, in the study by Christie(2005), maternal well-being variableswere measured pretest and posttest forcontrol and intervention groups, anddata regarding potential confoundingvariables such as nurses’ years of expe-rience and qualifications were collectedand analyzed. In concordance with theCONsolidated standards of ReportingTrials statement (CONSORT) (Begg et al.,1996)Va set of standards endorsed bythe International Committee of MedicalJournal Editors (2007) designed to alle-viate the problems arising from inade-quate reporting of RCTsVit is goodpractice to specify the assumptions usedwhen estimating the size of cluster andwithin-cluster samples in the trial report(Campbell, Elbourne, & Altman, 2004).

In addition to the overall samplesize, thought should be given to thenumber of clusters recruited. Considertwo studies with the same number ofparticipants; a study with a larger num-ber of clusters and fewer individualswithin clusters will be better able todistinguish intervention effects thandoes the one that has fewer clusters butlarger numbers of individuals withinclusters. It should be noted, however,that recruiting a large number of smallclusters may become counterproduc-tive, especially if there is a large amountof individual variation within the clus-ter. In this case, there will be largestandard errors within the cluster (i.e.,statistical estimates of intracluster var-iation will be unstable), leading togreater uncertainty in sample size cal-culations (Ukoumunne et al., 1999).

Researchers should also considerthe use of stratification of clusters.In stratified randomization, recruitedclusters are assigned to groups accord-ing to cluster-level characteristics thatare thought to affect individual out-comes. Clusters are then randomizedwithin strata to ensure a more even dis-tribution of cluster-level characteris-tics between intervention and controlgroups. By systematically randomizing

within relatively homogeneous strata,the amount of variation within thesample is reduced, making inferentialstatistical testing more efficient. Forexample, in the study undertaken byO’Halloran et al. (2004), homes thatagreed to participate were allocatedrandomly to either intervention or con-trol groups in a 1:2 ratio using block

randomization within strata. Stratawere determined by the organizationalcharacteristics of the homes thought toaffect the risk of hip fracture or the rateof adherence to the use of hip protec-tors (type, size, client category, andaffiliation of the home).

Ukoumunne et al. (1999) recom-mend that researchers take the cluster

FIGURE 1. Estimating sample size in a cluster randomized trial. CDC = Centers for DiseaseControl and Prevention.

130 Cluster Randomized Controlled Trial Nursing Research March/April 2009 Vol 58, No 2

9Copyright @ 200 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

design into account when analyzingthe results of the study. The limitationsthat affect sample size calculationsmust also be allowed in individual-levelanalyses. Within a cluster random-ized trial, a cluster of individuals ratherthan individuals themselves are random-ized to intervention or control groups.As individuals within such clusters sharea measure of common variance, thestatistical assumption of independencewithin the sample is violated. If a re-searcher conducts a traditional t test oran analysis of variance or analysis ofcovariance type of analysis to test forthe effect of treatment on individualswithout considering the clustered na-ture of the data, the results are likelyto be inaccurate as standard errorswould be underestimated. This is be-cause the between-cluster variancewould be averaged across the sample,and this may produce confidence in-tervals that are narrower and p valuesthat are smaller than their actual value.In consequence, because the data havebeen inappropriately analyzed, it couldbe inferred falsely that there is a differ-ence between the control and interven-tion groups when in fact there is none:a Type I error (Snijders & Bosker,1999). In this scenario, the researchermay assume that the intervention hashad an effect when in reality there hasbeen no effect. Figure 2 considers dataanalysis options for adjusting for clus-tering effects.

Planning a Cluster RCTMany of the challenges faced by re-searchers planning cluster RCTs willalso be confronted by those conduct-ing trials randomized by individual.However, there are some issues thatpertain to cluster trials in particular(Campbell et al., 2004). All of theseissues should be addressed in pub-lished reports of the trial. This willnot only allow readers to appraise theresearch critically but also be of useto those planning similar clusterRCTs in the future. Planning andreporting issues pertaining specificallyto cluster RCTs and those in commonwith all RCTs are summarized in atable on the Editor’s Web site at http://www.nursing-research-editor.com tohelp guide nurse researchers throughthe process.

Given the special challenges entailedin conducting a cluster RCT, the first

task is to provide a rationale for using acluster randomized design. A decisiontree to help researchers determinewhether a cluster design is the opti-mum approach to their trial is provided(Figure 3). As indicated in Figure 3, ifan RCT is not an appropriate method-ology for a particular research ques-tion or situation, then nurse researcherscan also address the impact of clusteringeffects when using survey and quasi-experimental approaches (Ukoumunneet al., 1999). When a cluster RCT isthe chosen approach, then it is neces-sary to specify whether interventions,objectives, hypotheses, and primaryand secondary outcome measures per-

tain to the level of the individual, thecluster, or both. Similarly, researchersshould determine whether to analyzedata at the level of the individual, thecluster, or both. For example, Christie(2005) hypothesized that more post-partum home visits would impactboth the individual mother and publichealth nurse outcomes measured atthe level of the case load for eachnurse. Thus, in this study, both indi-vidual- and cluster-level data werecollected and analyzed.

Researchers should specify howthe intraclass correlation will be calcu-lated for each primary group-leveloutcome. The intraclass correlation

FIGURE 2. Analyzing the results of a cluster randomized trial. ICC = intraclass correlation coefficient.

Nursing Research March/April 2009 Vol 58, No 2 Cluster Randomized Controlled Trial 131

9Copyright @ 200 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

coefficient (ICC) is Bthe proportion ofthe true total variation in the outcomethat can be attributed to differencesbetween clusters[ (Ukoumunne et al.,1999, p. 22). Thus, an ICC gives anindication of the shared unexplainedvariance within a cluster (see Figure1). Then, potential confounding vari-ables should be identified at both theindividual and cluster level, and themethod of measurement should bespecified. Researchers will need to esti-mate the likely size of clusters as thiswill affect the calculation of the ICCand therefore the final sample size,together with the number of clusters tobe included (studies in which fewer thanfour clusters are randomized to eachgroup are unlikely to yield conclusiveresults). It will be evident from thisdiscussion that the ICC is an estimatewith a degree of uncertainty around theresult. It is reasonable to take this intoaccount by estimating ICC confidencelimits and obtaining the correspondingconfidence limits for the estimated sam-ple size (Ukoumunne et al., 1999).

Eligibility criteria must be definednot only for individuals but also forclusters. For example, in the clusterRCT by O’Halloran et al. (2004),where the clusters were nursing homes,Ball homes registered with the Regis-tration and Inspection Unit (RIU) ofthe Eastern Health and Social ServicesBoard (EHSSB), Northern Ireland, tooffer residential or nursing care to theold and infirm (O&I), and the elderlymentally infirm (EMI)[ were eligiblefor the study (p. 583).

When randomizing to interventionand control groups, the researcher alsoneeds to consider if they will use sim-ple cluster randomization, block ran-domization, stratification, or matchingof clusters. Although both the partici-pants and researchers ideally shouldbe unaware (blind to) who is allocatedto treatment or control conditions,this is hard to achieve in cluster ran-domized trials where often behavioralinterventions are being studied (Donner& Klar, 2000).

Researchers naturally will planwhich statistical methods will be usedto compare groups for primary out-comes and for prespecified additionalanalyses, such as subgroup analyses.However, purchase of statistical softwareand associated statistical training mayalso be required. Although multilevel mod-

els can be estimated through many typesof statistical software packages, somededicated programs are available suchas MLwiN (Rashbash, Steele, Browne,&Prosser, 2005) and Hierarchical Linearand Nonlinear Modeling (HLM;Raudenbush, Bryk, & Congdon, 2008).

As noted earlier, a cluster RCT is auseful approach to testing the effec-

tiveness of an intervention in routineclinical practice (a pragmatic trial). Withthis in mind, consideration should begiven to intention-to-treat analysis.Withthis approach, all participants are in-cluded in the arm to which they wereallocated whether or not they received(or completed) the intervention given tothat arm. This gives an estimate of theeffect of treatment unbiased by the lossof participants as it includes in theanalysis patients who did not get theintended treatment or who deviatedfrom the trial protocol. Thus, it moreclosely reflects a real-world situation(Heritier, Gebski, & Keech, 2003).

A first step to achieve an intention-to-treat analysis is to obtain completedata on all randomized participants(Lachin, 2000). Where possible, sys-tems should be considered that willfollow up nonresponders. In the studyby Christie (2005), all nonrespondentsreceived follow-up by telephone andletter. From a group of 295 motherswho agreed to take part, 3 participantswithdrew and 12 did not respond tofollow-up. Maternal outcomes were

FIGURE 3. Cluster randomized trial decision tree.

Planning and reporting

issues pertaining

specifically to cluster

RCTs and those in

common with all RCTs

are summarized in a table

on the Editor’s Web site.

qqq

132 Cluster Randomized Controlled Trial Nursing Research March/April 2009 Vol 58, No 2

9Copyright @ 200 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

analyzed according to which publichealth nurse group (intervention orcontrol) the parents’ health visitorwas allocated to initially, even if themothers did not actually get the intendednumber of home visits (treatment).

As with all RCTs, trial registrationshould be considered. Currently, theInternational Committee of MedicalJournal Editors has a policy of requir-ing researchers to record trial details onan accepted trial registry before patientrecruitment into an RCT (Laine et al.,2007). Failure to register a trial may re-sult in the study being rejected by cer-tain journals (De Angelis et al., 2004).

Ethical Issues for Cluster RCTsCluster RCTs present special challengesin relation to consent and representa-tion (Medical Research Council, 2002).Because cluster RCTs occur at the levelof the group or community, it can bedifficult to provide for individual choicewithin the group when the interven-tion is implemented. For example, if ahealth promotion initiative is imple-mented at the community level, itwould be impractical to obtain indi-vidual consent from all members ofthe community and impossible to guar-antee that a nonconsenting member ofthe community would not be exposedto the intervention. When planning acluster RCT, the first task is to deter-mine whether the intervention is re-ceived by a whole cluster together,without the possibility of individualchoice, or one where participants candecide individually (Donner & Klar,2000). If individual consent cannot beobtained, then the researcher mustmake special efforts to ensure thatthe interests of individual participantsare monitored and guarded. Gener-ally, this means identifying and gain-ing the approval of the guardians ofthe participants’ interests and the gate-keepers of access to patient groups.

Once the trial is underway, theremust be an identified representativeindividual or body with the authorityand ability to monitor and representthe interests of the clusters. Institutionalreview boards will expect informed in-dividual consent where possible, evenin cluster RCTs; however, the Office forHuman Research Protections guidanceto institutional review boards indicatescircumstances where studies may be ap-proved, although they alter standard

consenting processes (Department ofHealth and Human Services, 2005).These special circumstances includeminimal risk to the well-being, rights,or welfare of participants and whereit is impractical to carry out theresearch using standard consentingprocedures. The statement furtherrecommends that information shouldbe given to research participants afterthe research.

Summary

Nursing interventions are often imple-mented across groups of people withinhealthcare organizations or geographi-cal areas. Consequently, RCTs evalu-ating nursing interventions may be moreappropriately randomized at the levelof the group rather than at the level ofthe individual. However, cluster RCTsare typically more complex and expen-sive than nonclustered designs.

Researchersmust consider a numberof methodological issues when design-ing a cluster RCT and should determinewhether interventions, objectives, hy-potheses, and primary and secondaryoutcome measures pertain to the levelof the individual, the cluster, or bothand analyze accordingly. They shouldalso specify the potential confoundingvariables at both individual and clusterlevel, eligibility criteria for clusters andtheir likely size, the approach to ran-domization, and whether the analysiswill be by intention to treat. Multilevelmodeling is a suitable approach tostatistical analysis for cluster RCTs asit can be used to estimate treatmenteffects and, in addition, outcome varia-tion between and within clusters.

Cluster RCTs make many demandson research design, conduct, and anal-ysis. The decision tree, checklist, andworked examples provided will helpguide nurses who are planning a clusterRCT. With careful attention to the is-sues addressed in this article, research-ers can avoid methodological pitfallsand use this approach successfully. q

Accepted for publication July 10, 2008.The authors thank Professor Jean Orr andMrs. Marianne Moutray, head and associatehead of the School of Nursing and Midwifery,Queen’s University Belfast, respectively, fortheir help and support in undertaking thisarticle. Acknowledgement is also due to theResearch and Development Office, Depart-

ment of Health, Social Services and PublicSafety in Northern Ireland which supportedand funded the original research upon whichthis article is based.Corresponding author: Janice Christie, PhD,MA, RN, RSCPHN, PGCE, School of Nurs-ing and Midwifery, MBC Building, Queen’sUniversity Belfast, 97 Lisburn Road, BelfastBT9 7BL, Northern Ireland, United Kingdom(e-mail: j.christie@qub.ac.uk).

ReferencesBegg, C., Cho,M., Eastwood, S., Horton, R.,

Moher, D., Olkin, I., et al. (1996). Im-proving the quality of reporting of random-ised controlled trials. The CONSORTstatement. Journal of the American Medi-cal Association, 276(8), 637Y639.

Bland, M. (2004). Cluster randomised trialsin the medical literature: Two biblio-metric surveys. BMC Medical ResearchMethodology, 4(21).

Campbell, M. K., Elbourne, D. R., &Altman, D. G. (2004). CONSORT state-ment: Extension to cluster randomisedtrials. British Medical Journal, 328,702Y708.

CDC. (1999). Epi info (Version 6). Re-trieved August 24, 2007, from http://www.cdc.gov/epiinfo/epiinfo.htm

Christie, J. (2005). An evaluation of healthvisitors’ postpartum home visits to first-time mothers. A mixed method study,including a cluster randomised trialevaluating the effect of health visitorpostpartum home visits to first-timemothers. Unpublished doctoral thesis,University of Ulster, Jordanstown,Northern Ireland, United Kingdom.

De Angelis, C., Drazen, J. M., Frizelle, F. A.,Haug, C., Hoey, J., Horton, R., et al.(2004). Clinical trial registration: A state-ment from the International Committeeof Medical Journal Editors. Annals ofInternal Medicine, 141(6), 477Y478.

Department of Health and Human Ser-vices. (2005). Title 45Vpublic welfare,Part 46Vprotection of human subjects.Retrieved April 18, 2008, from http://www.hhs.gov/ohrp/humansubjects/guidance/45cfr46.htm#46.116

Donner, A., & Klar, N. (2000). Designand analysis of cluster randomised trialsin health care. London: Arnold.

Ekman, A., Mallmin, H., Michaelsson, K.,& Ljunghall, S. (1997). External hipprotectors to prevent osteoporotic hipfractures. Lancet, 350(9077), 563Y564.

Fleiss, J. L. (1981). Statistical methods forrates and proportions. New York: Wiley.

Gomby, D. S. (1999). Understanding eval-uations of home visitation programs.Future of Children, 9(1), 27Y43.

Heritier, S. R., Gebski, V. J., & Keech, A. C.(2003). Inclusion of patients in clini-cal trial analysis: The intention to treat

Nursing Research March/April 2009 Vol 58, No 2 Cluster Randomized Controlled Trial 133

9Copyright @ 200 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

principle. Medical Journal of Australia,179, 438Y440.

International Committee ofMedical JournalEditors. (2007).Uniform requirements formanuscripts submitted to biomedicaljournals: Writing and editing for biomedi-cal publication. Retrieved April 15, 2008,from http://www.icmje.org/index.html

Kerry, S. M., & Bland, M. (1998). Statis-tics notes: Sample size in cluster ran-domisation. British Medical Journal,316(549), 1455Y1460.

Kreft, I., & De Leeuw, J. (1998). Introduc-ing multilevel modeling. London: Sage.

Lachin, J. M. (2000). Statistical consider-ations in the intent-to-treat principle.Controlled Clinical Trials, 21, 167Y189.

Laine, C., Horton, R., DeAngelis, C. D.,Drazen, J. M., Frizelle, F. A., Godlee, F.,et al. (2007). Clinical trial registration:Looking back and moving ahead.JAMA, 298(1), 93Y94.

Lauritzen, J. B., Petersen,M.M., & Lund, B.(1993). Effect of external hip protectors

on hip fractures. Lancet, 341(8836),11Y13.

Medical Research Council. (2002). Clusterrandomised trials methodological andethical considerations. London: Author.

O’Halloran, P. D., Cran, G. W., Beringer,T. R. O., Kernohan, G. W., O’Neill, C.,Orr, J., et al. (2004). A cluster random-ised controlled trial to evaluate a policyof making hip protectors available toresidents of nursing homes. Age andAgeing, 33, 582Y588.

Park, S., & Lake, E. T. (2005). Multi-level modeling of a clustered continu-ous outcome. Nursing Research, 54(6),406Y413.

Rashbash, J., Steele, F., Browne, W., &Prosser, B. (2005). MLwiN (Version 2).Retrieved October 31, 2008, from http://www.cmm.bristol.ac.uk

Raudenbush, S., Bryk, A., & Congdon, R.(2008). HLM (Version 6). RetrievedOctober 31, 2008, from http://www.ssicentral.com/hlm/index.html

Raudenbush, S. W. (1997). Statistical ana-lysis and optimal design for clusterrandomised trials. Psychological Meth-ods, 2(2), 173Y185.

Reading, R., Harvey, I., & Mclean, M.(2000). Cluster randomised trials in ma-ternal and child health: Implications forpower and sample size. Archives of Dis-ease in Childhood, 82, 79Y83.

Roland, M., & Torgerson, D. J. (1998).Understanding controlled trials: Whatare pragmatic trials? British MedicalJournal, 7127(316), 285.

Snijders, T., & Bosker, R. (1999). Mul-tilevel analysis, an introduction to basic andadvanced multilevel modeling. London:Sage.

Ukoumunne, O. C., Guilliford, M. C.,Chinn, S., Sterne, J. A. C., & Burney,P. G. J. (1999). Methods for evaluatingarea-wide and organisation-based inter-ventions in health and health care: Asystematic review, Health TechnologyAssessment (Winchester), 3, iiiY92.

134 Cluster Randomized Controlled Trial Nursing Research March/April 2009 Vol 58, No 2

9Copyright @ 200 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.