br Since we were able to obtain population based data
Since we were able to obtain population-based data for HBV, HCV, and H. pylori prevalence, the first formula was used for estimating PARs for HBV, HCV, and H. pylori. The PARs for the remaining infections, EBV, HPV, HHV-8 and human T-cell lymphotropic virus type 1 (HTLV-
1) were estimated with the third formula because they either demon-strate strong relationships with their associated cancers or mechanistic evidence exists for the role of the infection in cancer thus allowing for the PAR to be approximated by the prevalence in cancer cases (International Agency for Research on Cancer, 2012; Plummer et al., 2016; D'Souza et al., 2007).
2.3. Data collection and selection
The data needed to estimate PARs were identified by reviewing IARC monographs (International Agency for Research on Cancer, 2012, 1997, 2007), PAR analyses from other regions (de Martel et al., 2012; Plummer et al., 2016; Antonsson et al., 2015; Parkin, 2011), the Catalan Institute of Oncology HPV Information Centre reports for Canada and the United States (Bruni et al., 2017a; Bruni et al., 2017b), and results of our systematic literature reviews. A systematic literature search was conducted for each infection (details in Supplementary Table 1, S1) to extract data on the infection prevalence and identify meta-analyses on infection-associated cancers. Since the most recent IARC meeting that reviewed each infectious agent considered data published to the end of 2007, we searched for records published in English or French from January 1, 2008 to the search date of June 20, 2017. When data were sparse, we performed more targeted searches in PubMed and contacted experts in their respective fields. Ethics approval was granted for this project by the Health Research Ethics Board of Alberta - Cancer Com-mittee (HREBA.CC-14-0220_REN4), and McGill Univeristy exempted this study from Research Ethics Board review.
Cancers for which the infection is a necessary cause or part of the diagnostic criteria for a given cancer were: cervical cancer, extranodal natural killer T-cell lymphoma - nasal type, Kaposi sarcoma, primary effusion lymphoma, and adult T-cell leukemia/lymphoma, 100% were attributable to their associated infection and therefore inclusion criteria were not required. For all other infections and cancers, the inclusion Preventive Medicine 122 (2019) 109–117
criteria were: adult Benzenebutyric acid (defined as age 15 and older), North American study population, non-specialized population (e.g. studies performed in exclusively HIV-positive participants were excluded), 10 or more cancer cases, and use of the gold standard method to detect the infection. The inclusion criteria specific to each infection-cancer pair are noted in the tables of included studies (Supplementary Tables 2–13).
When the prevalence in cancer cases approximated the PAR (for-mula 3), the infection had to be detected in the cancer tumor, such as in a biopsy or surgical specimen. To extrapolate prevalence estimates to recent cancer incidence, rather than incorporating a latency period, the aim was to select studies conducted closer to the timeframe when cancer incidence data were collected. For this reason, studies had to be published in 1995 or later. Specifically, the prevalence of any HPV in the oropharynx has increased over time in the USA; pre-1990 HPV prevalence was 20.9% and from 2000 to 2013 it rose to 65.4% (Stein et al., 2014), further emphasizing the importance of utilizing more re-cent studies.
The prevalence of HBV and H. pylori were derived from North American population-based serosurveys, and HCV prevalence was ex-tracted from a study that modeled chronic HCV prevalence in the Canadian population (Trubnikov et al., 2014). Due to limited data on the measures of association for H. pylori associated cancers, a posteriori decision was made to consider studies conducted among European populations and studies that used the detection method that preceded the current gold standard method (we corrected to the new standard). The chosen detection method for assessing the presence of infection was crucial to the PAR estimation. Selecting studies that utilized the gold standard detection method was prioritized over other factors such as having a Canadian population or sex and age-specific results leading to sparser data.
2.4. Estimating infection prevalence in the Canadian population
Below is a brief description of how we adjusted population-based data to obtain sex- and age-specific estimates of HBV, HCV, and H. pylori prevalence for the Canadian population. The prevalence esti-mates and further details are provided in supplementary Tables S2–S5.
The Canadian Health Measures Survey (CHMS) was the first popu-lation-based survey to provide estimates of HBV and HCV prevalence for the Canadian population (Rotermann et al., 2013). Data from two cycles of the CHMS (Statistics Canada, n.d.), collected from 2007 to 2009 and 2009 to 2011, were combined for the analysis. The combined participation rate for those providing direct health measures after sample strategy adjustments was 52.8% for the two cycles (Rotermann et al., 2013). Sera from CHMS participants aged 14–79 testing positive for hepatitis B core antigen (anti-HBc) were then tested for hepatitis B surface antigen (HBsAg). Chronic HBV infection is defined as the pre-sence of HBsAg six months after a positive HBV test (National Notifiable Diseases Surveillance System, 2012). Given the cross-sectional design of the CHMS, we assumed that HBsAg positivity at one time point re-presented chronic HBV infection. Privacy restrictions limited HBsAg results to either sex or broad age groups (14–49 and 50–79), yet sex and age effect HBV prevalence. To obtain Canadian age-specific prevalence estimates, we used the HBsAg 10-year age-group prevalence from two merged cycles of the weighted National Health and Nutrition Ex-amination Survey (NHANES) (Centers for Disease Control and Prevention, 2009, 2011) to partition the CHMS estimates by 10-year age groups. The first two cycles of the CHMS were collected from 2007 to 2011, resulting in a six-year latency. This time period does not cor-respond to the prolonged latency for hepatocellular carcinoma (El-Serag, 2012), yet it is still plausible as the CHMS measured prevalent not incident HBV infection.