Statistics from Altmetric.com
A meta-analysis that was recently published in thisJournal 1 suggested an association between excess pancreatic cancer risk and exposure to nickel and nickel compounds (meta-risk ratio 1.9, 95% confidence interval (95% CI) 1.2 to 3.2, based on four studies). Through correspondence with the authors (Ojajärvi et al), I learned that their analysis excluded the many epidemiological studies that had been conducted on workers in the nickel refining and alloy production industries. Although most of these studies could not contribute to the meta-analysis due to a failure to specifically examine risk of pancreatic cancer, I found two studies of nickel workers that provide relevant data.
I think that one of these studies, which examined mortality in 11 500 nickel mining and smelting workers,2 should have been included in the meta-analysis by Ojajärvi et al,1 based on the criteria used for study selection. Another study of more than 30 000 workers exposed to nickel and nickel compounds in the production of nickel alloys3was published a few months after the May 1998 cut off that Ojajärviet al used to establish the data base for their meta-analysis. The results from these studies2 3add substantially to data used in the analysis by Ojajärviet al 1.
Combining the data from all of these studies with the meta-analysis random effects model used by Ojajärvi et al 1 produces a meta-risk ratio (MRR) of 1.3 (95% CI 0.9 to 1.9). Interestingly, the two studies designed specifically to detect excess cancer risks associated with occupational exposure to nickel2 3 show the lowest relative risks for pancreatic cancer and differ substantially from the MRR for nickel exposure calculated by Ojajärvi et al (1.9). Moreover, the estimated relative risk (0.9) from the study of nickel alloy workers3 is significantly smaller (p<0.05) than even the lower 95% confidence limit (1.2) for the MRR of Ojajärviet al.1
The fact that the MRR of Ojajärvi et al 1 for nickel related pancreatic cancer significantly overestimates the risk found in a large cohort of nickel workers indicates that their meta-analysis risk estimates should be viewed with an appropriate degree of caution. These results of the meta-analysis may be considerably biased because of limitations of the studies on which they are based. In studies that relate to nickel, the potential for misclassification bias is strong because of the complete confounding of nickel exposure with known carcinogenic hazards such as cadmium,5 or asbestos, polycyclic aromatic hydrocarbons, chromium, beryllium, polychlorinated biphenyls, and hydrazine.4 Similarly, in the meta-analysis of Ojavarviet al 1 the case-control study7 that contributed the most substantial evidence of a risk of pancreatic cancer related to nickel provides equally strong statistical evidence of associations between excess pancreatic cancer and exposures to 10 other substances, some of which are likely to be correlated with occupational exposure to nickel.
Although Ojajärvi et al 1 are to be congratulated on their investigation of the aetiology of pancreatic cancer, it is my opinion that their results are most appropriately viewed as hypotheses that require further investigation, rather than compelling evidence that links substances to the induction of pancreatic cancer. As Ojajärvi et al 1 correctly suggest, research to test these hypotheses requires large studies and more refined measures of exposure. With respect to nickel and nickel compounds, data from large studies that were not included in the analysis of Ojajärviet al 1 call into question the veracity of a hypothesis that links nickel exposure to increased risk of pancreatic cancer.
Ojajärvi and Partanen reply
We thank Seilkop for his comment and have, in essence, not much to add to it. The study by Shannon et al 1-1 had obviously been overlooked and the study by Arena et al 1-2 was published after our deadline for the inclusion of studies.
Seilkop's table has errors for the study by Anderssonet al.1-3 The number of pacreatic cancer deaths should be 2; relative risk should be 1.2; and 95% confidence interval should be 0.1 to 4.5.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.