Intraclass correlation coefficients for cluster randomized trials in primary care: The cholesterol education and research trial (CEART)

https://doi.org/10.1016/j.cct.2005.01.002Get rights and content

Abstract

Cluster randomization trials are increasingly being used in primary care research. The main feature of these trials is that patients are nested within large clusters such as physician practices or communities and the intervention is applied to the cluster. This study design necessitates calculation of intraclass correlation coefficients in order to determine the required sample size. The purpose of this study is to determine intraclass correlation coefficients for a number of outcome measures at the primary care practice level. The CEART study is a randomized trial testing the effectiveness of translating ATP III guidelines into clinical practice, with primary care physician practices as the unit of randomization and patients as the unit of data collection. The intraclass correlation coefficient (ICC) was<0.02 and the design effect ranged from 1.0 to 2.3, respectively, for weight, total cholesterol, LDL, non-HDL, glucose, creatinine, and % at non-HDL goal. For smoking status, body mass index, systolic blood pressure, HDL cholesterol triglycerides, total cholesterol/HDL ratio and % at LDL goal, the ICC was 0.02–0.047 and the design effect was 2.6–4.1. The largest ICCs (0.05–0.12) and design effects (4.4–9.4) were found for height and diastolic blood pressure. These findings suggest that cluster randomization may substantially increase the sample size necessary to maintain adequate statistical power for selected outcomes such as diastolic blood pressure studies compared with simple randomization for most outcomes evaluated in this study where the design effect is small to moderate. Overall, the ICCs presented will be useful in calculating sample sizes at the primary care level.

Introduction

Cluster randomization in primary care practice intervention trials is an increasingly popular design whereby individuals are nested within larger clusters such as practices, hospitals, or communities [1], [2], [3], [4]. The practical advantages are that “whole practice populations can be studied; the organization of the trial might be simpler; many primary care providers may find this method of research less intrusive on the daily clinical practice; and the practical problems of offering an intervention to some, but not others, within a practice are overcome” [5] which is particularly important for quality improvement initiatives where most of the interactions will occur at the practice level using a team-based approach. However, to implement this study design, it necessitates special power calculations and data analysis because observations on individuals within the same cluster may be correlated [6]. Within-cluster correlation affects the power of a trial because a greater homogeneity of members in the clusters will increase the standard error of the estimate of the treatment effect resulting in a loss of power to detect a difference between the intervention and control groups [7], [8]. Therefore, primary care intervention trials may require calculation of intraclass correlation coefficients in order to determine the required sample size.

There are repeated calls for the publication of intraclass correlation coefficients to aid in the design of future cluster-based intervention studies [1], [9], [10], [11], [12]. Our objective was to provide intraclass correlation coefficients for a number of outcome measures at the primary care practice level from practices in response to this need.

Section snippets

The intraclass correlation coefficient

The intraclass correlation coefficient is the measure of variation between and within clusters of individuals and measures the clustering effect or lack of independence among individuals who make up the cluster [13]. The intraclass correlation coefficient is based on the relationship of the between-cluster to within-cluster variance and is given by ρ=σb2/(σb2+σw2), where σb2 is the between-cluster component of the variance and σw2 is the within-cluster component of the variance [9].

Factors that

Methods

The CEART study is a randomized trial testing the effectiveness of translating the ATP III guidelines into clinical practice, with primary care physician practices as the unit of randomization and patients as the unit of data collection. The prevention arm received an intervention of academic detailing, a computerized patient activation tool placed in the primary care providers office waiting room, and an interactive ATP III Cholesterol Guideline decision support tool made for a hand help

Results

The demographic characteristics of the study population are presented in Table 1. The age range of the patients was between 20 and 80 years of age and all patients were seen by their physician at least once in the past 2 years. For the total group, the mean age was 52.4 years, mean body mass index was 28.3 kg/m2, total mean cholesterol was 192.4mg/dL, mean LDL cholesterol was 115.3 mg/dL, and mean non-HDL cholesterol was 139.1 mg/dL. The percent of current smokers for the total group was 13.5%.

Discussion

This is one of a few papers to present intraclass correlation coefficients for a range of outcomes at the primary care practice level in the United States [22]. The intraclass correlation coefficients (<0.01 to 0.12) we obtained were in the range of previous published estimates [1]. In addition, to our knowledge, this is the first paper to present intraclass correlation coefficients for a number of cardiovascular measures important for primary prevention trials.

Several factors could have

Acknowledgments

We would like to thank the physicians, nurses, and other primary care staff in the participating practices (Family Care Center at MHRI: Arthur Frazzano, Arnold Goldberg, Susan McGee; South Attleboro: Joshua Guttman, Heidi Brownlee, Dan Brown, Suyin Lee, Linda LeGendre FNP; Franklin Family Practice: Patrick McSweeney, Cheryl Hardenbrook; East Greenwich: John Slattery, Samuel Kagan; South County: Monica Gross, Scott Hanson). We would also like to thank research assistants: Jennifer Vancura and

References (26)

  • L. Smeeth et al.

    Intra-class correlation coefficients for cluster randomized trials in primary care: data from the MRC trial of the assessment and management of older people in the community

    Control. Clin. Trials

    (2002)
  • S.M. Eldridge et al.

    Lessons for cluster randomized trials in the twenty-first century: a systematic review of trials in primary care

    Clin. Trials

    (2004)
  • S.M. Kerry et al.

    Unequal cluster sizes for trials in English and Welsh general practice: implications for sample size calculations

    Stat. Med.

    (2001)
  • O.C. Ukoumunne et al.

    Methods in health service research. Evaluation of health interventions at area and organization level

    BMJ

    (1999)
  • R. Reading et al.

    Cluster randomised trials in maternal and child health: implications for power and sample size

    Arch. Dis. Child.

    (2000)
  • J. Cornfield

    Randomization by group: a formal analysis

    Am. J. Epidemiol.

    (1978)
  • A. Donner

    An empirical study of cluster randomization

    Int. J. Epidemiol.

    (1982)
  • R.H. Cosby et al.

    Randomizing patients by family practice: sample size estimation, intracluster correlation and data analysis

    Fam. Pract.

    (2003)
  • S.M. Kerry et al.

    The intra-cluster correlation coefficient in cluster randomization

    BMJ

    (1998)
  • D.M. Murray et al.

    Design and analysis issues in community trials

    Eval. Rev.

    (1994)
  • A. Donner et al.

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

    Int. J. Epidemiol.

    (1990)
  • M.K. Campbell et al.

    Cluster randomized trials: time for improvement

    BMJ

    (1998)
  • N.B. Baskerville et al.

    The effect of cluster randomization on sample size in prevention research

    J. Fam. Pract.

    (2001)
  • Cited by (91)

    • Imputing intracluster correlation coefficients from a posterior predictive distribution is a feasible method of dealing with unit of analysis errors in a meta-analysis of cluster RCTs

      2021, Journal of Clinical Epidemiology
      Citation Excerpt :

      Twenty-six additional sources were obtained from colleagues (or their references) and screened [14,45–69]. Based on our target population, setting, and outcome criteria, we abstracted 17 additional ICC estimates from 6 sources (1 survey [n = 5] [66], 1 database [n = 1] [45], 1 registry [n = 1] [56], and 3 CRTs [n = 10] [63–65]) yielding a completed database of 59 ICC estimates for 12 of the 13 review outcomes (range 1 to 10 per outcome; Table 2). ICC estimates varied from 0.000 to 0.129 for continuous outcomes and 0.000 to 0.330 for dichotomous outcomes and exhibited a wide range of calculated standard deviations (0.004 to 0.057), numbers of clusters (6 to 121), and patients (231 to 8808) across studies.

    • Estimating the intra-cluster correlation coefficient for evaluating an educational intervention program to improve rabies awareness and dog bite prevention among children in Sikkim, India: A pilot study

      2017, Acta Tropica
      Citation Excerpt :

      Correlation between members of a cluster, or variation between clusters is quantified using intra-cluster correlation (ICC) estimates. ICCs are used in the design phase of cluster intervention trials to increase sample size estimates to account for lack of independence in study outcomes arising from individuals within the same cluster (e.g. schools) (Parker et al., 2005). Rabies is a fatal zoonotic disease, posing a major public health risk in countries where it is endemic (Knobel et al., 2005).

    View all citing articles on Scopus

    This study was funded by Grant No. NIH-NHLBI RO1 HL70804.

    View full text