Ravinder Sandhu1, Lawrence Mbuagbaw1,2, Jean-Eric Tarride1,3, Vanessa De Rubeis1, Sarah Carsley4, Laura N Anderson5,6,7. 1. Department of Health Research Methods, Evidence, and Impact, McMaster University, 1280 Main St W, CRL-221, Hamilton, ON, L8S 4L8, Canada. 2. Biostatistics Unit, St Joseph's Healthcare Hamilton, 50 Charlton Ave E, Hamilton, ON, Canada. 3. Centre for Health Economics and Policy Analysis, McMaster University, 1280 Main St W, Hamilton, Ontario, Canada. 4. Department of Health Promotion, Chronic Disease and Injury Prevention, Public Health Ontario, 480 University Ave #300, Toronto, ON, Canada. 5. Department of Health Research Methods, Evidence, and Impact, McMaster University, 1280 Main St W, CRL-221, Hamilton, ON, L8S 4L8, Canada. ln.anderson@mcmaster.ca. 6. Centre for Health Economics and Policy Analysis, McMaster University, 1280 Main St W, Hamilton, Ontario, Canada. ln.anderson@mcmaster.ca. 7. Division of Child Health Evaluative Sciences, The Hospital for Sick Children Research Institute, 686 Bay St, Toronto, Ontario, Canada. ln.anderson@mcmaster.ca.
Abstract
OBJECTIVES: Interventions for child obesity prevention are needed and it is unclear whether evidence from nonrandomized intervention studies is adequate. The objective of this research was to review the methods for the design, analysis and reporting of nonrandomized intervention studies for child obesity prevention and to assess potential for bias. METHODS: We conducted a review of nonrandomized intervention studies, including population health interventions, quasi-experimental studies and natural experiments, published from 2013 to 2017 that were identified in a recent systematic review. Data on study design, intervention and control groups, outcome measures, and statistical analyses, were extracted. Risk of bias was evaluated using the Risk of Bias in Non-Randomized Studies of Interventions (ROBINS-I) tool. RESULTS: All identified studies (n = 23) included a school or community-based intervention and had a concurrent control group. Participants were 3-18 years and sample sizes were 100 to > 1 million. Study designs were described inconsistently, and interventions ranged from 14 weeks to 5 years. Obesity was compared between control and intervention groups using logistic or linear regression, analysis of variance and mixed effects regression. Only 48% of studies accounted for clustering, and methods to control for confounding and repeated measures varied substantially. Overall risk of bias was moderate to serious for all studies. CONCLUSION: There are substantial opportunities to improve the methods for nonrandomized intervention studies and reduce bias. Future studies should use advanced statistical and causal epidemiology methods, including better control for confounding and clustering, to generate higher quality evidence and certainty regarding which obesity prevention interventions are effective.
OBJECTIVES: Interventions for childobesity prevention are needed and it is unclear whether evidence from nonrandomized intervention studies is adequate. The objective of this research was to review the methods for the design, analysis and reporting of nonrandomized intervention studies for childobesity prevention and to assess potential for bias. METHODS: We conducted a review of nonrandomized intervention studies, including population health interventions, quasi-experimental studies and natural experiments, published from 2013 to 2017 that were identified in a recent systematic review. Data on study design, intervention and control groups, outcome measures, and statistical analyses, were extracted. Risk of bias was evaluated using the Risk of Bias in Non-Randomized Studies of Interventions (ROBINS-I) tool. RESULTS: All identified studies (n = 23) included a school or community-based intervention and had a concurrent control group. Participants were 3-18 years and sample sizes were 100 to > 1 million. Study designs were described inconsistently, and interventions ranged from 14 weeks to 5 years. Obesity was compared between control and intervention groups using logistic or linear regression, analysis of variance and mixed effects regression. Only 48% of studies accounted for clustering, and methods to control for confounding and repeated measures varied substantially. Overall risk of bias was moderate to serious for all studies. CONCLUSION: There are substantial opportunities to improve the methods for nonrandomized intervention studies and reduce bias. Future studies should use advanced statistical and causal epidemiology methods, including better control for confounding and clustering, to generate higher quality evidence and certainty regarding which obesity prevention interventions are effective.
Authors: J P Ioannidis; A B Haidich; M Pappa; N Pantazis; S I Kokori; M G Tektonidou; D G Contopoulos-Ioannidis; J Lau Journal: JAMA Date: 2001-08-15 Impact factor: 56.272