Alcohol intake and gastric cancer: Meta-analyses of published data versus individual participant data pooled analyses (StoP Project)

Publication date: June 2018

Source: Cancer Epidemiology, Volume 54

Author(s): Ana Ferro, Samantha Morais, Matteo Rota, Claudio Pelucchi, Paola Bertuccio, Rossella Bonzi, Carlotta Galeone, Zuo-Feng Zhang, Keitaro Matsuo, Hidemi Ito, Jinfu Hu, Kenneth C. Johnson, Guo-Pei Yu, Domenico Palli, Monica Ferraroni, Joshua Muscat, Reza Malekzadeh, Weimin Ye, Huan Song, David Zaridze

Abstract

Background

Individual participant data pooled analyses allow access to non-published data and statistical reanalyses based on more homogeneous criteria than meta-analyses based on systematic reviews. We quantified the impact of publication-related biases and heterogeneity in data analysis and presentation in summary estimates of the association between alcohol drinking and gastric cancer.

Methods

We compared estimates obtained from conventional meta-analyses, using only data available in published reports from studies that take part in the Stomach Cancer Pooling (StoP) Project, with individual participant data pooled analyses including the same studies.

Results

A total of 22 studies from the StoP Project assessed the relation between alcohol intake and gastric cancer, 19 had specific data for levels of consumption and 18 according to cancer location; published reports addressing these associations were available from 18, 5 and 5 studies, respectively. The summary odds ratios [OR, (95%CI)] estimate obtained with published data for drinkers vs. non-drinkers was 10% higher than the one obtained with individual StoP data [18 vs. 22 studies: 1.21 (1.07–1.36) vs. 1.10 (0.99–1.23)] and more heterogeneous (I²: 63.6% vs 54.4%). In general, published data yielded less precise summary estimates (standard errors up to 2.6 times higher). Funnel plot analysis suggested publication bias.

Conclusion

Meta-analyses of the association between alcohol drinking and gastric cancer tended to overestimate the magnitude of the effects, possibly due to publication bias. Additionally, individual participant data pooled analyses yielded more precise estimates for different levels of exposure or cancer subtypes.

https://ift.tt/2u8NrHR