Lecture 11 Cluster randomized and community trials Clusters

  • Slides: 24
Download presentation
Lecture 11: Cluster randomized and community trials • • • Clusters, groups, communities Why

Lecture 11: Cluster randomized and community trials • • • Clusters, groups, communities Why allocate clusters vs individuals? Randomized vs nonrandomized designs Methods of allocation of intervention Design issues 1

Clusters, groups, communities • Intervention directed at entire community vs individuals: – mass educational

Clusters, groups, communities • Intervention directed at entire community vs individuals: – mass educational programs – immunization campaigns • Targeting interventions to total population vs high risk group (e. g. , hypertension): – population strategy aims to shift population blood pressure distribution – high-risk strategy targets those with HBP 2

Clusters, groups, communities • Intervention directed at entire community vs individuals: – mass educational

Clusters, groups, communities • Intervention directed at entire community vs individuals: – mass educational programs – immunization campaigns • Targeting interventions to total population vs high risk group (e. g. , hypertension): – population strategy aims to shift population blood pressure distribution – high-risk strategy targets those with HBP 3

What is a community? • “. . Group of people living in a defined

What is a community? • “. . Group of people living in a defined geographic area who share a common culture, are arranged in a social structure and exhibit some awareness of their identity as a group” (Nutbeam, 1986) • “A group of individuals organized into a unit, or manifesting some underlying trait or common interest; loosely, the locality or catchment area population for which a service is provided, or more broadly, the state, nation, or body politic. ” (Last, 2001) 4

What is a cluster? • (Last) – CLUSTER/CLUSTERING: Aggregation of relatively uncommon events. .

What is a cluster? • (Last) – CLUSTER/CLUSTERING: Aggregation of relatively uncommon events. . In space and/or time … greater than expected by chance. – CLUSTER ANALYSIS: Statistical methods to group variables or observations into strongly interrelated subgroups – CLUSTER SAMPLING: Each unit selected is a group rather than individual 5

What is a cluster? • (Webster’s) – CLUSTER: a number of things growing together

What is a cluster? • (Webster’s) – CLUSTER: a number of things growing together OR of things or persons collected or grouped closely together 6

Clustering - reasons • Clustering: – individuals within clusters tend to be more similar

Clustering - reasons • Clustering: – individuals within clusters tend to be more similar to each other than to individuals in other clusters • Reasons: – selection – common exposures 7

Examples of community-level interventions • Screening or immunization programmes delivered to residents of a

Examples of community-level interventions • Screening or immunization programmes delivered to residents of a geographic area • Health promotion programmes delivered to towns, schools • Services provided to primary care practice populations 8

Examples of group or cluster interventions • Educational interventions • Group psychological interventions •

Examples of group or cluster interventions • Educational interventions • Group psychological interventions • Nutritional, environmental sanitation interventions: – delivered to household, village etc – latrines, dietary supplements 9

Rationale for community interventions • Environmental change may be easier than voluntary behavior change

Rationale for community interventions • Environmental change may be easier than voluntary behavior change (e. g, tax cigarettes vs stop smoking) • Risk behaviors are socially influenced • Some interventions are not selective (e. g. , fluoridation) 10

Reasons for carrying out evaluations at group or cluster level • More appropriate for

Reasons for carrying out evaluations at group or cluster level • More appropriate for interventions delivered to groups • Individual randomization may not be feasible because all members of group are treated same way • Individual randomization, although feasible, may result in substantial “contamination” 11

12

12

Examples • “Grass roots” intervention: – Nurse-midwife program for low-income women in Colorado –

Examples • “Grass roots” intervention: – Nurse-midwife program for low-income women in Colorado – Various needle exchange programs for IDUs • Usually not true experiments – communities not randomly allocated – quasiexperimental “non-equivalent” control group design 13

Examples • Social experiment: – COMMIT – 11 pairs of matched communities – intervention:

Examples • Social experiment: – COMMIT – 11 pairs of matched communities – intervention: multi-component smoking cessation • • media and community-wide events health care providers work-site and other organizations cessation resources 14

Community trial designs • Single community: Before-after: O X O Single (interrupted) time series:

Community trial designs • Single community: Before-after: O X O Single (interrupted) time series: O O O X O O O • One intervention and one control community O X O O O • One intervention and multiple control communities • Multiple intervention and control clusters 15

To randomize or not? • Complete randomization usually feasible only when large # clusters

To randomize or not? • Complete randomization usually feasible only when large # clusters 16

Allocation of intervention • Allocation of communities: – in pairs – stratified – matching

Allocation of intervention • Allocation of communities: – in pairs – stratified – matching or stratification factors: • known predictors/correlates of outcome • cluster size and other characteristics • matching can be ignored in analysis when matching variable is weakly correlated with outcomes 17

Study design • Serial cross-sectional surveys vs follow-up of cohort – is intervention aimed

Study design • Serial cross-sectional surveys vs follow-up of cohort – is intervention aimed at whole community of “stayers” only? – individual or community-level change? – Testing effects – attrition • Because blinding of subjects not possible, try to use objective outcome measures (e. g. , serum cotinine vs self-reported smoking) 18

Study design (cont) • Community-level vs individual-level outcomes/indicators – e. g. , tobacco sales

Study design (cont) • Community-level vs individual-level outcomes/indicators – e. g. , tobacco sales to assess smoking prevention intervention – cluster-level measures may be less biassed and less costly than individual-level measures 19

Study design (cont) • Develop causal model (hypotheses about how program should work) –

Study design (cont) • Develop causal model (hypotheses about how program should work) – measure key elements of model to understand why intervention was (or was not) successful – assess process and outcomes • Formative evaluation: – feedback of results of process evaluation to help improve intervention? • Qualitative (ethnographic) methods 20

Ethical issues in cluster randomization • Individual consent not possible prior to randomization (or

Ethical issues in cluster randomization • Individual consent not possible prior to randomization (or other method of allocation) 21

Analysis of community-level trials • Failure to account for clustering in analysis is common

Analysis of community-level trials • Failure to account for clustering in analysis is common in group-level interventions (Donner) • Analysis that accounts for clustering will yield more conservative level of statistical significance 22

10 Key Considerations (adapted from Ukoumunne et al, 1999) • Recognize the cluster as

10 Key Considerations (adapted from Ukoumunne et al, 1999) • Recognize the cluster as the unit of intervention or allocation • Justify the use of cluster as unit of intervention or allocation (these methods are not as powerful as individual designs) • Include enough clusters (at least 4 per group) • Randomize clusters when possible • Allow for clustering when computing sample size 23

10 Key Considerations (cont. ) • Consider the use of matching or stratification of

10 Key Considerations (cont. ) • Consider the use of matching or stratification of clusters where appropriate (but matching methods limit the statistical analyses that can be done) • Consider different approaches to repeated assessments in prospective evaluations: – cohort vs repeated cross-sections • Allow for clustering at time of analysis • Allow for confounding by individual and cluster characteristics • Include estimates of intracluster correlations of key outcomes, to aid in planning of future studies 24