Research Designs for Complex Community Interventions for Childhood Obesity Prevention

1
Research Designs for Complex Community
Interventions for Childhood Obesity Prevention

• Robert W. Jeffery, Ph.D.
• Division of Epidemiology
• University of Minnesota
• School of Public
• May 4-5, 2004

2
Hypothesis
Changes in the food environment of middle schools
and high schools will reduce the prevalence of
overweight in students.
3
Design Option 2-Group Randomized Trial

• Unit of randomization is school
• Primary outcome, change in BMI, is measured on
  students.
• Two-year duration

4
School Selection Eligibility Criteria

• Co-educational
• Student ages 12-18
• At least 2-year attendance expectation
• Feasibility of intervention
• Feasibility of data collection
• Consent of school
• Satisfactory consent procedure for students

5
Recruitment Enrollment Procedure

• Random selection of schools to approach
• Possible stratification by public/private, SES,
  ethnic composition (stratification very limited)
• Age range (e.g., middle school, high school)

6
Intervention Components Food Service

• School lunch
• A la carte
• Vending machines

7
Food Service Targets

• Reduce overall availability of food at school
• Increase availability of low-energy food
• Decrease availability of high-energy food
• Increase price of high-energy food
• Decrease price of low-energy food
• Reduce food advertising on campus

8
Primary Outcome

• Change in BMI measured at baseline, 12 months,
  and 24 months
• Measured in as many students as possible who
  attend school for the 2-year period

9
Secondary Measures

• Individual level (self-report)
• Diet
• Physical activity
• Demographics
• Psychosocial measures
• Intervention exposure
• School level (direct observation)
• Intervention implementation
• Food sales

10
Pretest-Posttest Control Group Design
R O X O OR 0 0 0
11
Threats to Validity Internal

• Differential experience not related to treatment
  as a function of treatment group
• Measurement or dropout bias by treatment group
  (lack of blinding is problematic, but unavoidable)

12
Threats to Validity External

• Volunteer bias, both individual and school
• Limited school pool
• Feasibility constraints
• Overall sample size

13
Pragmatic IssuesCost

• Maintaining good fidelity of intervention
  delivery can be costly, since school food
  service, a la carte food service, and vending
  involve different business operations.
• Objective assessment of food sales can also be
  costly.
• Number of schools is a major determinant of cost.

14
Pragmatic IssuesQuality Control

• Political considerations often make intervention
  implementation incomplete.
• Food service is a dynamic activity that requires
  careful monitoring.
• Large sample size may reduce quality of
  intervention.

15
Analysis IssuesSources of Variance in Outcomes

• Individual characteristics
• Intervention variability
• School effect

16
Analysis OptionsBy School

• Maximum protection against threats to internal
  and external validity
• Including individual characteristics and
  intervention fidelity terms reduces variance
• Cost of producing enough replications to achieve
  adequate power is problematic

17
Analysis OptionsBy Individual

• Internal validity threatened by differential
  factors between schools other than
  interventionplausibility of this threat is
  estimable by historical data
• Individual level and intervention delivery
  effects may again be useful covariates
• External validity and generalizability are weak
  with this approach

18
Summary Conclusion

• The group randomized trial is the gold standard
  for evaluating community intervention trials.
• It is strong on both internal and external
  validity and thus provides a strong evidential
  basis for important decisions like public health
  policy.
• It is not necessary for establishing the
  plausibility of an intervention approach.
• Thus, it is best thought of as representing the
  culmination of a line of research in an area
  rather than a prerequisite for research funding
  related to community intervention.