Confidence interval construction for effect measures arising from cluster randomization trials

doi:10.1016/0895-4356(93)90050-B

Journal of Clinical Epidemiology

Volume 46, Issue 2, February 1993, Pages 123-131

https://doi.org/10.1016/0895-4356(93)90050-B Get rights and content

Abstract

Methods of confidence interval construction are provided for summary measures of treatment effect arising from designs randomizing clusters to one of two treatment groups. Three basic designs are considered for the case of continuous and dichotomous variables: completely randomized, pair-matched and stratified.

References (33)

A. Sommer et al.
Impact of vitamin A supplementation on childhood mortality: a randomized controlled community trial
Lancet
(1986)
M.H. Gail et al.
Aspects of statistical design for the Community Intervention Trial for Smoking Cessation (COMMIT)
Controlled Clin Trials
(1992)
L. Tabár et al.
Reduction in mortality from breast cancer after mass screening with mammography: randomized trial from the breast cancer screening working group of the Swedish National Board of Health and Welfare
Lancet
(1985)
A. Donner et al.
A methodologic review of non-therapeutic intervention trials employing cluster randomization, 1979–1989
Int J Epidemiol
(1990)
M.J. Gardner et al.
Confidence intervals rather than P values: estimation rather than hypothesis testing
Br Med J
(1986)
D.M. Wilson et al.
A randomized trial of a family physician intervention for smoking cessation
JAMA
(1988)
A. Donald et al.
Adjustments to the Mantel-Haenszel chi-square statistic and odds ratio variance estimator when the data are clustered
Stat Med
(1987)
B. Woolf
On estimating the relation between blood groups and disease
Ann Hum Genet
(1955)
N.E. Breslow et al.
A. Donner
Statistical methodology for paired cluster designs
Am J Epidemiol
(1987)

D. Byar

The design of cancer prevention trials

Recent Results Cancer Res

(1988)

M.J. Shipley et al.

Calculation of power for matched pair studies where randomization is by group

Int J Epidemiol

(1989)

M.J. Bass et al.

Do family physicians need nurse assistants to detect and manage hypertension—a randomized trial

Can Med Assoc J

(1986)

European collaborative trial of multifactorial prevention of coronary heart disease: final report on the 6-year results

Lancet

(1986)

W.W. Hauck

Odds ratio inference from stratified samples

Commun Stat Theory and Methods

(1989)

K.Y. Liang

Odds ratio inference with dependent data

Biometrika

(1985)

Cited by (79)

The napkin-ring sign indicates advanced atherosclerotic lesions in coronary CT angiography
2012, JACC: Cardiovascular Imaging
This study sought to determine the accuracy of plaque pattern assessment by coronary computed tomography angiography (CCTA) to differentiate between early and advanced atherosclerotic lesions as defined by histology.
A ringlike attenuation pattern of coronary atherosclerotic plaques termed as napkin-ring sign (NRS) was described in CCTA of patients who had acute coronary syndrome.
All procedures were performed in accordance with local and federal regulations and the Declaration of Helsinki. Approval of the local ethics committees was obtained. We investigated 21 coronary arteries of 7 donor hearts. Overall, 611 histological sections were obtained and coregistered with CCTA images. The CCTA cross sections were read in random order for conventional plaque categories (noncalcified [NCP], mixed [MP], calcified [CP]) and plaque patterns (homogenous, heterogeneous with no napkin-ring sign [non-NRS], and heterogeneous with NRS).
No plaque was detected in 134 (21.9%), NCP in 254 (41.6%), MP in 191 (31.3%), and CP in 32 (5.2%) CCTA cross sections. The NCP and MP were further classified into homogenous plaques (n = 207, 46.5%), non-NRS plaques (n = 200, 44.9%), and NRS plaques (n = 38, 8.6%). The specificities of NCP and MP to identify advanced lesions were moderate (57.9%, 95% confidence interval [CI]: 50.1% to 65.6%, and 72.1%, 95% CI: 64.7% to 79.4%, respectively), which were similar to the homogenous and heterogeneous plaques (62.6%, 95% CI: 54.8% to 70.3%, and 67.3%, 95% CI: 58.6% to 76.1%, respectively). In contrast, the specificity of the NRS to identify advanced lesions was excellent (98.9%, 95% CI: 97.6% to 100%). The diagnostic performance of the pattern-based scheme to identify advanced lesions was significantly better than that of the conventional plaque scheme (area under the curve: 0.761 vs. 0.678, respectively; p = 0.001).
The assessment of the plaque pattern improves diagnostic accuracy of CCTA to identify advanced atherosclerotic lesions. The CCTA finding of NRS has a high specificity and high positive predictive value for the presence of advanced lesions.
Chronic care model and shared care in diabetes: Randomized trial of an electronic decision support system
2008, Mayo Clinic Proceedings
Citation Excerpt :
Specifically, Part A billed charges were adjusted by using hospital department cost-to-charge ratios and wage indexes, and Part B physician service costs were proxied by 2000 Medicare reimbursement rates. All analyses used hierarchical models, in which the unit of analysis was the unit of randomization, ie, the physician.51,52 To evaluate the success of our randomization process in terms of the prognosis of the patients (given that the physicians were randomized), we used generalized linear models to test the hypothesis of no difference at baseline in patient characteristics across trial arms.
To assess the effect of a specialist telemedicine intervention for improving diabetes care using the chronic care model (CCM).
As part of the CCM, 97 primary care physicians at 6 primary care practices in Rochester, MN, referred 639 patients to an on-site diabetes educator between July 1, 2001, and December 31, 2003. On first referral, physicians were centrally randomized to receive a telemedicine intervention (specialty advice and evidence-based messages regarding medication management for cardiovascular risk) or no intervention, keeping outcome assessors and data analysts blinded to group assignment. After each subsequent clinical encounter, endocrinologists reviewed an abstract from the patient's electronic medical record and provided management recommendations and supporting evidence to intervention physicians via e-mail. Control physicians received e-mail with periodic generic information about cardiovascular risk reduction in diabetes. Outcome measures included diabetes care processes (diabetes test completion), outcomes (metabolic and cardiovascular risk factors, estimated coronary artery disease risk), and patient costs (payer perspective).
During the intervention, 951 (70%) of the 1361 endocrinology reviews detected performance gaps and resulted in a message; primary care physicians reported using 49% of messages in patient care. With a mean of 21 months' follow-up, the intervention, compared with control, did not significantly enhance metabolic outcomes or reduce estimated risk of coronary artery disease (adjusted mean difference, -1%; 95% confidence interval, -19% to 17%). The intervention group incurred lower costs (P=.02) but not in diabetes-related costs.
Specialty telemedicine did not significantly enhance the value of CCM in primary care.
Trial Registration: clinicaltrials.gov identifier: NCT00421850
Evaluation of an epilepsy education program for Grade 5 students: A cluster randomized trial
2007, Epilepsy and Behavior
Citation Excerpt :
For ease of interpretation, knowledge and attitude difference scores were calculated (posttest knowledge/attitudes minus pretest knowledge/attitudes as the outcome variables) and compared by intervention status. The intervention group experienced, on average, an 11.8-point increase in knowledge following the program (range −4.0–24.0, 95% CI = 11.4–12.5) [22]. This is compared with the control group which, on average, experienced only a 2.2-point increase in knowledge (range −10.0–14.0, 95% CI = 1.8–2.6).
Epilepsy is a common and often highly stigmatized disorder. Several international organizations indicate a need to assess the stigma that exists and to develop and evaluate interventions to dispel myths about epilepsy.
A stratified cluster randomized controlled trial evaluated whether an epilepsy education program (intervention) increases knowledge of and positive attitudes about epilepsy in Grade 5 students (ages 9–11). The study also investigated characteristics of the individuals (gender, language spoken at home, familiarity with epilepsy) and schools (Catholic vs public, size of school, and school level socioeconomic status) that affect epilepsy knowledge and attitudes. We assessed epilepsy knowledge and attitudes at baseline and 1 month following the program using a 33-item questionnaire.
In total, 24 schools (783 individuals) were cluster randomized. Those in the intervention group had an average 11.8-point increase (95% confidence interval (CI) = 11.4–12.5) in knowledge 1 month following the program, compared with the control group increase of 2.2 points (95% CI = 1.8–2.6) out of a total of 57 points. For attitudes, the intervention group had a mean increase of 8.15 points (95% CI = 4.70–11.60), compared with the control group increase of 1.64 points (95% CI = −0.84–4.42) out of a total of 50 points. The intervention was responsible for 63% of the variation in postprogram epilepsy knowledge and 28% of the variation in postprogram attitudes about epilepsy.
The epilepsy education program was associated with a significant increase in epilepsy knowledge and positive attitudes in the intervention group 1 month later compared with the control group.
Effect of community-based newborn-care intervention package implemented through two service-delivery strategies in Sylhet district, Bangladesh: a cluster-randomised controlled trial
2008, The Lancet
Citation Excerpt :
For pregnancy-care indicators, the denominator was the total number of births; for newborn-care indicators, the denominator was the total number of livebirths that occurred at home. We used a t test to compare differences in neonatal mortality rate between intervention and comparison clusters.14,15 We estimated the adjusted relative risk (RR) of neonatal mortality for each intervention arm relative to the comparison arm and constructed 95% CIs for RR with Taylor series approximated variance16 and log-transformed RR, controlling for mother's age and years in education, sex and birth order of the index child, and wealth index.
Neonatal mortality accounts for a high proportion of deaths in children under the age of 5 years in Bangladesh. Therefore the project for advancing the health of newborns and mothers (Projahnmo) implemented a community-based intervention package through government and non-government organisation infrastructures to reduce neonatal mortality.
In Sylhet district, 24 clusters (with a population of about 20 000 each) were randomly assigned in equal numbers to one of two intervention arms or to the comparison arm. Because of the study design, masking was not feasible. All married women of reproductive age (15–49 years) were eligible to participate. In the home-care arm, female community health workers (one per 4000 population) identified pregnant women, made two antenatal home visits to promote birth and newborn-care preparedness, made postnatal home visits to assess newborns on the first, third, and seventh days of birth, and referred or treated sick neonates. In the community-care arm, birth and newborn-care preparedness and careseeking from qualified providers were promoted solely through group sessions held by female and male community mobilisers. The primary outcome was reduction in neonatal mortality. Analysis was by intention to treat. The study is registered with ClinicalTrials.gov, number 00198705.
The number of clusters per arm was eight. The number of participants was 36059, 40159, and 37598 in the home-care, community-care, and comparison arms, respectively, with 14 769, 16 325, and 15 350 livebirths, respectively. In the last 6 months of the 30-month intervention, neonatal mortality rates were 29·2 per 1000, 45·2 per 1000, and 43·5 per 1000 in the home-care, community-care, and comparison arms, respectively. Neonatal mortality was reduced in the home-care arm by 34% (adjusted relative risk 0·66; 95% CI 0·47–0·93) during the last 6 months versus that in the comparison arm. No mortality reduction was noted in the community-care arm (0·95; 0·69–1·31).
A home-care strategy to promote an integrated package of preventive and curative newborn care is effective in reducing neonatal mortality in communities with a weak health system, low health-care use, and high neonatal mortality.
United States Agency for International Development and saving newborn lives programme by Save the Children (US) with a grant from Bill and Melinda Gates Foundation.
Conditional versus non-conditional incentives to maximise return of participant completed questionnaires in clinical trials: a cluster randomised study within a trial
2023, Trials
Impact of a phone call with a medical student/general practitioner team on morbidity of chronic patients during the first French COVID-19 lockdown (COVIQuest): a cluster randomised trial
2022, BMJ Open

View all citing articles on Scopus

View full text

Original articleConfidence interval construction for effect measures arising from cluster randomization trials

Abstract

Lancet

Controlled Clin Trials

Lancet

A methodologic review of non-therapeutic intervention trials employing cluster randomization, 1979–1989

Int J Epidemiol

Confidence intervals rather than P values: estimation rather than hypothesis testing

Br Med J

A randomized trial of a family physician intervention for smoking cessation

JAMA

Adjustments to the Mantel-Haenszel chi-square statistic and odds ratio variance estimator when the data are clustered

Stat Med

On estimating the relation between blood groups and disease

Ann Hum Genet

Statistical methodology for paired cluster designs

Am J Epidemiol

The design of cancer prevention trials

Recent Results Cancer Res

Calculation of power for matched pair studies where randomization is by group

Int J Epidemiol

Do family physicians need nurse assistants to detect and manage hypertension—a randomized trial

Can Med Assoc J

European collaborative trial of multifactorial prevention of coronary heart disease: final report on the 6-year results

Lancet

Odds ratio inference from stratified samples

Commun Stat Theory and Methods

Odds ratio inference with dependent data

Biometrika

Original article
Confidence interval construction for effect measures arising from cluster randomization trials