Inclusion and exclusion criteria for this paper were set as follows. We would include a paper for analysis if the paper was published between 1995 (January) and 2016, if the language of publication was English, if the paper was published in a peer reviewed journal, and the research would have to be based on individual data, as opposed to aggregated data. If a paper was an editorial, or if the paper did not include individual level data analysis, or if the paper did not fulfil any of the above criteria, we would not include this paper in our analyses. We also searched for reference section of the articles we selected for meta analysis and identified additional articles this way. We also excluded studies if they were conducted in animal models. We only considered cross-sectional surveys, case control studies, meta-analysis of observational epidemiologic studies, and cohort studies for inclusion in this paper. We have excluded ecological studies.
Initially, we read the abstracts and full texts of all the retrieved articles. We applied the inclusion and exclusion criteria as detailed above to select or reject the studies. We then created a spreadsheet where we stored the following information we abstracted from each article:
- A study ID (last name of the first author and the year)
- Year
- Study Population
- Cancer Type
- Risk Estimate
- Study Type
- Standard Error
- 95% Confidence Interval
- Participants
- Cases
- Controls
- Exposed
- Non-exposed
This is a master database and we conducted separate meta analyses based on whether the studies were cohort studies or case control studies. For each class of studies, we would examine subgroup analyses based on the population subgroup where the study was conducted. The summary estimates for each population subgroup would provide us with the information around variability of the estimates. We estimated and conducted subgroup analyses based on the population to identify the association between arsenic exposure either as total arsenic exposure or as arsenic in the form of MMA3 and estimated the risk ratios or odds ratios from random and fixed effects meta analyses and compared the rates. We also calculated the PAR%s from the data based on prevalence of arsenic exposure levels and we compared the PAR% of the individual populations. This was done to ascertain what would happen if the arsenic were to be removed completely or by what proportion would bladder cancer be reduced if that were to happen.