Meta-Analysis of Lung Cancer in Asphalt Roofing and Paving Workers with External Adjustment for Confounding by Coal Tar

Abstract

The study's objectives were to update Partanen's and Boffetta's 1994 meta-analysis of lung cancer among roofing and paving asphalt workers and explore the role of coal tar in explaining the statistical heterogeneity among these studies. Information retrieval strategies and eligibility criteria were defined for identifying the epidemiologic studies to be included in the analysis. The relative risk ratio (RR) for lung cancer was selected as the effect measure of interest. Coal tar bias factors were developed and used to externally adjust each eligible study's published RR for confounding by coal tar. Confidence intervals on adjusted RRs were estimated by the Monte Carlo methods outlined by Steenland and Greenland (2004). The meta-Relative Risk (meta-RR) and its variance were estimated by general variance-based methods. Heterogeneity of the RRs was assessed by heterogeneity chi-square and I ² tests. Also explored were the influences of study design; differences between internally and externally adjusted RRs; and the potential impact of coal tar misclassification. The results from this update were similar to those in Partanen's and Boffetta's original meta-analysis. The unadjusted meta-RR (random effects model) for the roofing studies was elevated (meta-RR = 1.67, 95% CI = 1.39–2.02, I ² = 80%) and significantly greater than the meta-RR for the paving studies (meta-RR = 0.98, 95% CI = 0.81–1.18, I ² = 59%). After adjustment for confounding by coal tar, however, the corrected roofers' meta-RR dropped to 1.10 (meta-RR = 1.10, 95% CI = 0.91–1.33, I ² = 60%), while the corrected pavers' meta-RR was nearly unchanged at 0.96 (meta-RR = 0.96, 95% CI = 0.80–1.16, I ² = 56%). Although the meta-RRs for the roofers and the pavers were no longer statistically significantly different from one another, significant heterogeneity remained within each of the coal tar-adjusted sectors. Stratifying the roofing and paving studies by study design revealed that statistical heterogeneity was concentrated in the weaker designs such as those relying on death certificates for work history. When the weaker studies were excluded, the remaining roofing and paving studies had a combined meta-RR close to 1.0 (meta-RR = 1.01, 95% CI = 0.88–1.17) with no significant heterogeneity (I ² = 0%). Meta-analysis of non-experimental epidemiologic studies is subject to significant uncertainties as is externally correcting studies for confounding. Given these uncertainties, the specific quantitative estimates in this (or any similar) analysis must be viewed with caution. Nevertheless, this analysis provides support for the hypothesis proposed by several major reviewers that confounding by coal tar-related PAH exposures may explain most or all of the lung cancer risks found in the epidemiologic literature on asphalt roofing and paving workers.

Keywords:

• meta-analysis
• confounding
• asphalt
• coal tar
• epidemiology
• lung cancer

ACKNOWLEDGMENTS

I would like to acknowledge Drs. Paolo Boffetta, Diane Mundt, and Edwin Wijngaarden for early reviews of the concept for this article; and Drs. Gary Marsh and John Gamble for helpful suggestions on early and later drafts of the manuscript.

Notes

^A GM = Geometric Mean.

^A Coal Tar Bias Factor: derived from (a) % of lagged bitumen person-years in the top 3 cumulative coal tar exposure quartiles (lagged); (b) % of referent population with coal tar exposure (3%, 95% CI = 1%–16%); and (c) coal tar RR = 1.87 (95% CI = 1.57–2.22) for roofing workers and 1.37 (1.25–1.49) for paving workers.

^B Prevalence of coal tar exposure.

^A Mundt's calculations for the Coal Tar Bias Factor by decade assume that the coal tar prevalence in asphalt roofers is as shown; the coal tar prevalence in the general population is 1–5% over all decades; and coal tar's Relative Risk for lung cancer is 1.8–2.5.

^A DC-WH (Death-Certificate-Based Work History); PMR (Proportional Mortality Study); CS-PS (Cross-Sectional/Public Census); MOR (Mortality Odds Ratio).

^A DC-WH (Death-Certificate-Based Work History); PMR (Proportional Mortality Study); CS-PS (Cross-Sectional/Public Census); MOR (Mortality Odds Ratio).

^A DC-WH (Death-Certificate-Based Work History); PMR (Proportional Mortality Study); CS-PS (Cross-Sectional/Public Census); MOR (Mortality Odds Ratio).