Pooling results from epidemiological studies using dose-response meta-analysis




Oral session: Statistical Methods (1)


Tuesday 22 October 2019 - 14:00 to 15:30


All authors in correct order:

Zeraatkar D1, Han M2, Vernooij RR3, Valli C4, Rabassa M4, El Dib R5, Bala MM6, Alonso-Coello P4, Johnston BC7, Guyatt GH1
1 Department of Health Research Methods, Evidence, and Impact, McMaster University, Canada
2 School of Medicine, Chosun University, South Korea
3 Department of Research, Netherlands Comprehensive Cancer Organisation (IKNL), The Netherlands
4 Iberoamerican Cochrane Centre Barcelona, Biomedical Research Institute San Pau (IIB Sant Pau), Spain
5 Institute of Science and Technology, Universidade Estadual Paulista, Brazil
6 Department of Hygiene and Dietetics, Jagiellonian University Medical College, Poland
7 Department of Community Health and Epidemiology, Faculty of Medicine, Dalhousie University, Canada
Presenting author and contact person

Presenting author:

Dena Zeraatkar

Contact person:

Abstract text
Background: epidemiological studies typically report on the relationship between levels of exposure to a putative disease-causing agent and risk of disease. Dose-response is considered an important criterion for evaluating the plausibility of causal inferences and is a factor considered in rating the certainty of evidence using the GRADE approach. Dose-response meta-analysis is a statistical method for pooling results from epidemiological studies while taking into account potential dose-response relationships.

Objectives: to present an application of dose-response meta-analysis using our systematic review on the association between red meat consumption and cardiometabolic outcomes as an example.

Methods: we conducted a systematic review of cohort studies reporting on the association between consumption of red meat and adverse cardiometabolic outcomes. We used methods by Greenland & Longnecker to approximate the covariance of relative effects for each study and estimated a corrected trend using generalized least-squares regression. We then pooled study-specific trend estimates using random-effects meta-analysis. We also investigated the possibility of non-linear associations using restricted cubic splines.

Results: we found 55 eligible studies, of which 32 reported sufficient information to be included in dose-response meta-analyses. We found a small positive dose-response association between consumption of red meat and adverse cardiometabolic outcomes with effects associated with decreasing consumption of red meat by three servings/week ranging from risk ratio (RR) 0.90 (95% confidence interval (CI) 0.88 to 0.92) to RR 0.94 (95% CI 0.90 to 0.98) for diabetes and stroke, respectively. We did not find evidence of non-linearity.

Conclusions: dose-response meta-analysis allows estimation of linear and non-linear dose-response relationships across epidemiological studies. Unlike the alternative approach of meta-analyzing relative effects of extreme categories of exposure, dose-response meta-analysis uses all available data without discarding intermediate levels of exposure and considers variations in levels of exposure within and across studies. However, to be eligible for inclusion in dose-response meta-analysis, besides reporting relative effects and their associated variances, studies must additionally report exposure levels and the number of events and participants or person-years across exposure levels.

Patient or healthcare consumer involvement: no patients or consumers were involved in the development or analysis of this study. The study presents an overview of an infrequently used method for meta-analysis, which may improve evidence synthesis, and subsequently improve patient or public health.