Accounting for Intraclass Correlations and Controlling for Baseline Differences in a Cluster-Randomised Evidence-Based Practice Intervention Study

Department of Clinical Sciences-Division of Biostatistics and Simmons Comprehensive Cancer Center, The University of Texas Southwestern Medical Center, Dallas, Texas 75390, USA.
Worldviews on Evidence-Based Nursing (Impact Factor: 2.38). 02/2008; 5(2):95-101. DOI: 10.1111/j.1741-6787.2008.00125.x
Source: PubMed


Background: In health care and community-based intervention studies, cluster-randomised designs have been increasingly used because of administrative convenience, a desire to decrease treatment contamination, and the need to avoid ethical issues that might arise. While useful, cluster-randomised designs present challenges for data analysis. First, because of dependencies that exist among subjects within a cluster, methods that account for intra-class correlations have to be used. Second, on many occasions, because of unavailability of large numbers of clusters, lack of balance on baseline measures has to be carefully examined and appropriately controlled for.
Aim/Methodology: Two strategies are presented that can be used when analysing data from a cluster-randomised design; both account for baseline differences. Examples of these challenges are provided by a pain management intervention study designed to promote the adoption of evidence-based pain management practices. One approach involves use of a mixed model via SAS PROC MIXED. The other approach involves use of a marginal model: Generalised estimating equations using SAS PROC GENMOD.
Implications: In cluster-randomised design, one must adjust for intra-class correlation when evaluating the intervention effect. Although the parameter estimates and their standard errors might be comparable with both random effect and marginal strategies for certain link functions (identity link or log link only), the interpretations are quite different and the two approaches are suitable for indicating answers to different questions. If differences are present concerning baseline measures between experimental and control groups, accounting for baseline measures is important. The choice between a mixed model or marginal approach should be dictated by whether the primary interest is a population or individual.

7 Reads
  • Source
    • "Marginal models such as GEE are appropriate when the interest of the study is not the clustering effect and their variances, but the inferences about the average response over the population and when the differences among clusters are minimal [34]. GEE was proposed for correlated data by Liang and Zeger [35,36], using the quasi-likelihood approach [37]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: To analyze the regular dental care behavior and prevalence of edentulism in adult Danes, reported in sequential cross-sectional oral health surveys by the application of a marginal approach to consider the possible clustering effect of birth cohorts. Data from four sequential cross-sectional surveys of non-institutionalized Danes conducted from 1975-2005 comprising 4330 respondents aged 15+ years in 9 birth cohorts were analyzed. The key study variables were seeking dental care on an annual basis (ADC) and edentulism. For the analysis of ADC, survey year, age, gender, socio-economic status (SES) group, denture-wearing, and school dental care (SDC) during childhood were considered. For the analysis of edentulism, only respondents aged 35+ years were included. Survey year, age, gender, SES group, ADC, and SDC during childhood were considered as the independent factors. To take into account the clustering effect of birth cohorts, marginal logistic regressions with an independent correlation structure in generalized estimating equations (GEE) were carried out, with PROC GENMOD in SAS software. The overall proportion of people seeking ADC increased from 58.8% in 1975 to 86.7% in 2005, while for respondents aged 35 years or older, the overall prevalence of edentulism (35+ years) decreased from 36.4% in 1975 to 5.0% in 2005. Females, respondents in the higher SES group, in more recent survey years, with no denture, and receiving SDC in all grades during childhood were associated with higher probability of seeking ADC regularly (P < 0.05). The interaction of SDC and age (P<0.0001) was significant. The probabilities of seeking ADC were even higher among subjects with SDC in all grades and aged 45 years or older. Females, older age group, respondents in earlier survey years, not seeking ADC, lower SES group, and not receiving SDC in all grades were associated with higher probability of being edentulous (P<0.05). With the use of GEE, the potential clustering effect of birth cohorts in sequential cross-sectional oral health survey data could be appropriately considered. The success of Danish dental health policy was demonstrated by a continued increase of regular dental visiting habits and tooth retention in adults because school dental care was provided to Danes in their childhood.
    BMC Oral Health 03/2011; 11(1):9. DOI:10.1186/1472-6831-11-9 · 1.13 Impact Factor
  • Source

  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Little is known about the consequences of intensivists’ work schedules, or intensivist continuity of care. To assess the impact of weekend respite for intensivists, with consequent reduction in continuity of care, on them and their patients. In five medical intensive care units (ICUs) in four academic hospitals we performed a prospective, cluster-randomized, alternating trial of two intensivist staffing schedules. Daily coverage by a single intensivist in half-month rotations (continuous schedule) was compared with weekday coverage by a single intensivist, with weekend cross-coverage by colleagues (interrupted schedule). We studied consecutive patients admitted to study units, and the intensivists working in four of the participating units. The primary patient outcome was ICU length of stay (LOS);we also assessed hospital LOS and mortality rates. The primary intensivist outcome was physician burnout. Analysis was by multivariable regression. A total of 45 intensivists and 1,900 patients participated in the study. Continuity of care differed between schedules (patients with multiple intensivists = 28% under continuous schedule vs. 62% under interrupted scheduling; P < 0.0001). LOS and mortality were nonsignificantly higher under continuous scheduling (ΔICU LOS 0.36 d, P = 0.20; Δhospital LOS 0.34 d, P = 0.71; ICU mortality, odds ratio = 1.43, P = 0.12; hospital mortality, odds ratio = 1.17,P = 0.41). Intensivists experienced significantly higher burnout, work–home life imbalance, and job distress working under the continuous schedule. Work schedules where intensivists received weekend breaks were better for the physicians and, despite lower continuity of intensivist care, did not worsen outcomes for medical ICU patients.
    American Journal of Respiratory and Critical Care Medicine 06/2011; 184(7):803-8. DOI:10.1164/rccm.201103-0555OC · 13.00 Impact Factor