Multicenter Dose-Response Analysis — Pooled Exposure-Response Modeling Across Study Sites
Multicenter Dose-Response Analysis · Also known as: pooled dose-response analysis, multicenter exposure-response analysis, multi-site dose-response modeling, collaborative dose-response study
Multicenter dose-response analysis estimates the quantitative shape of the relationship between a graded exposure and a health outcome by pooling data or effect estimates across two or more study centers. Using flexible regression tools such as restricted cubic splines or fractional polynomials within a two-stage meta-analytic framework, it characterizes whether the relationship is linear, supra-linear, threshold-based, or J-shaped — providing far greater statistical power and generalizability than any single center could achieve alone.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use multicenter dose-response analysis when (1) the research question concerns the shape — not just the direction — of an exposure-outcome relationship, (2) data from two or more independent cohorts or clinical study centers are available, and (3) no single site has sufficient power to detect non-linear patterns across the full exposure range. It is the method of choice for large collaborative epidemiological consortia studying environmental exposures, dietary factors, or pharmacological doses. Do not use it when centers differ so drastically in outcome definition or exposure measurement that harmonization is impossible; when the exposure is purely categorical with no meaningful ordering; or when fewer than three well-characterized exposure categories are available per center, which leaves the shape of the curve unidentifiable.
Strengths & limitations
- Substantially increases statistical power to detect and characterize non-linear dose-response shapes by pooling data across sites.
- Produces results generalizable across diverse populations, settings, and exposure distributions rather than a single study context.
- Two-stage framework (estimate then pool) allows centers to protect participant-level data while still contributing to the collaborative analysis.
- Multivariate pooling preserves the covariance structure of the dose-response estimates, avoiding the bias introduced by treating category-specific estimates as independent.
- Enables rigorous heterogeneity assessment across the full exposure range, not just at a single summary point.
- Harmonizing exposure categories and confounder sets across centers requires extensive coordination and may involve irreducible measurement inconsistencies.
- The two-stage approach using aggregated data (rather than full IPD) loses efficiency and cannot adjust for individual-level confounders that were not included in each center's model.
- Spline-based shape estimation is sensitive to knot placement and reference category choice; different a priori decisions can yield visually different curves.
- Publication bias can distort pooled curves if centers with null or J-shaped findings are less likely to contribute data to the consortium.
Frequently asked
What is the difference between multicenter dose-response analysis and a standard meta-analysis?
Standard meta-analysis typically pools a single summary estimate (e.g., the RR for exposed versus unexposed) from each study. Multicenter dose-response analysis pools an entire curve — a set of correlated estimates across exposure levels — using a multivariate model that respects the covariance between category-specific estimates within each center. This allows characterization of the shape of the relationship, not just its direction or average magnitude.
Do I need individual-participant data (IPD) or can I use published summary statistics?
Both are possible. The two-stage approach of Greenland and Longnecker works from published category-specific log RRs and their variances, making it feasible when full IPD is unavailable. However, IPD-based one-stage pooling offers greater flexibility in confounder adjustment, exposure harmonization, and handling of missing data. If IPD is accessible, a one-stage analysis within a mixed-effects framework is generally preferred.
How do I choose where to place knots for the restricted cubic splines?
A common practice is to place knots at the 5th, 35th, 65th, and 95th percentiles of the pooled exposure distribution, following Harrell's recommendations. The number and placement of knots should be specified in the analysis plan before examining the data. Sensitivity analyses with alternative knot placements are expected in high-quality publications to demonstrate that conclusions do not depend on an arbitrary choice.
How should I handle centers that use different exposure categories?
Before pooling, exposure categories must be mapped to a common quantitative scale and a shared referent value. Centers with incompatible categorization may need to be excluded or analyzed separately in sensitivity analyses. If the number of categories per center is very small (fewer than three), that center's data cannot support spline modeling and must be analyzed with restricted linear terms or excluded from the shape analysis.
What software is available for multicenter dose-response pooling?
The Stata command 'glst' (generalized least squares for trend) by Orsini et al. is the most widely used tool for the two-stage approach. The R package 'dosresmeta' provides equivalent functionality and supports both fixed- and random-effects pooling. For one-stage IPD analysis, standard mixed-effects packages such as 'lme4' (R) or 'mixed' (Stata) combined with spline basis functions are appropriate.
Sources
- Greenland, S., & Longnecker, M. P. (1992). Methods for trend estimation from summarized dose-response data, with applications to meta-analysis. American Journal of Epidemiology, 135(11), 1301-1309. DOI: 10.1093/oxfordjournals.aje.a116237 ↗
- Orsini, N., Li, R., Wolk, A., Khudyakov, P., & Spiegelman, D. (2012). Meta-analysis for linear and nonlinear dose-response relations: examples, an evaluation of approximations, and software. American Journal of Epidemiology, 175(1), 66-73. DOI: 10.1093/aje/kwr265 ↗
How to cite this page
ScholarGate. (2026, June 3). Multicenter Dose-Response Analysis. ScholarGate. https://scholargate.app/en/epidemiology/multicenter-dose-response-analysis
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Dose-Response AnalysisEpidemiology↔ compare
- Multicenter cohort studyEpidemiology↔ compare