UCLA

Background: Studying risk factors of rare outcomes can be difficult as single studies tend to have few exposed cases, resulting in rather wide confidence intervals. Therefore, combination of multiple studies is usually necessary to draw a conclusion. Pooling of individual patient data (IPD) is considered the gold standard due to increased statistical power through large sample sizes and other advantages. However, present pooling is limited to studies of the same designs, which does not make full use of existing data.

Objectives: To generalize pooling to studies of different designs (including cohort study, case control study, nested case control study and matched case control study), by incorporating simulation of the association of extremely low frequency magnetic field (ELF-MF) and childhood leukemia.

Method: I first simulated large cohort and case control samples based on parameters extracted from both the literature and existing large cohort and case control datasets, which included ELF-MF exposure prevalence, childhood leukemia incidence rate, prevalence of confounders including age, gender, race and SES. Then I combined these simulated data using three different methods: two stage meta-analysis, one stage pooling and two stage meta-analysis with pooling.

Results: Estimates from three synthesis methods were close to the causal estimate and there was no obvious trend of overestimation or underestimation. One stage pooling seemed to have the worst efficiency with the widest 95%CI but the difference was not significant.

Conclusion: The performance of three synthesis methods in the study was not certain. Further simulations with varying parameters and possible mathematical derivations are needed to assess why and when these methods lead to different effect estimates.

Keywords: Individual patient data (IPD), meta-analysis, pooling, simulation with R, Extremely Low Frequency Magnetic Field (ELF-MF), childhood leukemia