Single Nucleotide Polymorphisms in the TP53 Region and Susceptibility to Invasive Epithelial Ovarian Cancer

The p53 protein, which is a transcription factor regulating multiple cellular functions, is critical for maintenance of genomic stability. The importance of TP53 in carcinogenesis is evident in the high incidence of malignancies in the rare familial Li-Fraumeni Syndrome, which is most commonly characterized by germline mutations in the TP53 gene (1). Somatic mutations can inactivate TP53 and are found in approximately one half of human cancers, including epithelial ovarian carcinomas (2–4). Activation of p53 prevents the replication of damaged DNA until repair or apoptosis can be completed (5).

There is evidence to suggest that minor alterations in the TP53 gene may have a significant effect on its biological function. First, whereas cancer-causing mutations in most tumor suppressor genes result in truncated protein products, deleterious TP53 mutations generally are missense changes. Although these mutations affect just a single amino acid, they usually are sufficient to abrogate the transcriptional regulatory function of p53. In addition, there is evidence that the common P72R (rs1042522) nonsynonymous single nucleotide polymorphism (SNP) in TP53 affects its function (6–9). In view of the known role of TP53 inactivation in a majority of serous ovarian cancers, and the propensity for subtle genetic variations to affect its function, TP53 is an appealing candidate as an ovarian cancer susceptibility gene (2, 10).

Epidemiologic studies have evaluated ovarian cancer risks associated with TP53 variants in intron 2 (rs1642785; ref. 11), intron 3 (16bp insertion; ref. 12–15), exon 4 (P72R or rs1042522; refs. 14, 16–19), and intron 10 (rs9894946; refs. 14, 15, 20); however, the results were inconclusive.

This report was motivated by results from three independent discovery studies focusing on TP53 polymorphisms and risk of ovarian cancer: the North Carolina Ovarian Cancer Study (NCOCS; ref. 21), the Mayo Clinic Case-Control Study (MAYO; ref. 22), and the Polish Ovarian Cancer Study (POCS; ref. 23). The NCOCS and the MAYO used a tagSNP approach, whereas the POCS used resequencing data to assess common genetic variation in TP53 and two flanking genes in linkage disequilibrium (LD)²⁹

Data from three studies, NCOCS, MAYO, and POCS focusing on genetic variation in TP53 and risk of ovarian cancer were compiled and analyzed. Details of these studies have been published elsewhere and are described briefly below (21–23). An a priori decision by all three discovery analyses was made to restrict the discovery analysis to non-Hispanic White women with newly diagnosed, histologically confirmed, primary invasive epithelial ovarian cancer and to non-Hispanic White controls. Both the NCOCS and MAYO studies restricted analyses to serous cases, whereas the POCS analyses were run overall and restricted to serous cases because of the a priori belief that TP53 variants might be more closely related to serous cancers.

All ovarian cancer cases were reviewed at each site, respectively, by an expert pathologist. The NCOCS is a population-based, case-control study. Cases between the ages of 20 and 74 y were identified from the North Carolina Central Cancer Registry from January 1999 to January 2007. Control subjects were matched by age and race and were identified through random digit dialing. The discovery analysis included 391 serous invasive ovarian cases and 786 controls.

The MAYO study included participants recruited at the Mayo Clinic between January 2000 and March 2006. Participation was limited to women from six states: Minnesota, Iowa, Wisconsin, Illinois, North Dakota, and South Dakota. Clinic-based controls frequency-matched to cases on age, race, and geographic region of residence were selected from women seeking general medical evaluation. The discovery analysis included 200 serous invasive cases and 458 controls.

The POCS is a population-based, case-control study among residents of Warsaw and Lodz (Poland) who were ages 20 to 74 y. Control women were frequency matched to cases on age and study site (Warsaw and Lodz), and were randomly selected within matching strata from population lists of Warsaw and Lodz residents. Discovery analyses were restricted to 264 invasive ovarian cases of which 118 are serous and 625 controls. The study protocols for the three discovery studies were approved by respective Institutional Review Boards.

SNP selection and genotype analysis. In all, up to 23 SNPs were genotyped in the 3 discovery studies. A full list of the polymorphisms along with their chromosomal positions is given in Supplementary Table S1. The POCS genotyped 20 SNPs, of which 8 were genotyped in the NCOCS and 9 were genotyped in the MAYO study. The NCOCS and MAYO studies genotyped 10 and 9 SNPs, respectively, of which 8 were the same. Supplementary Table S2 shows the genotype frequencies of the SNPs for each of the 13 participating case-control studies.

Both the NCOCS and the MAYO study identified sets of tagSNPs for the TP53 region using releases 19 and 20, respectively, of the International HapMap Project's³⁰

For the NCOCS, 1 within-plate and 1 across-plate duplicate sample was included on each 96-well DNA plate to evaluate consistency in genotyping. The concordance rate for duplicates was 99.9%. In addition, six CEPH-Utah trios from the Coriell Institute, Camden, NJ, were distributed across six plates. Any SNP where >1% of samples within a batch failed genotyping on an Illumina BeadStudio file was disregarded, and only the remaining SNPs were carried forward for statistical analysis. There was 95.8% concordance between HapMap calls on the CEPH data and the genotype calls for 9 of the SNPs evaluated. One SNP, rs8079544, showed 50% discordance and these data were excluded from additional analyses. A χ², 1 degree of freedom, test found only one SNP (rs2287499) in the TP53 region that showed evidence (P < 0.05) for departure from Hardy-Weinberg Equilibrium among controls.

For the MAYO study, 8 replicates of a CEPH family trio (mother, father, child) from the Coriell Institute, and replicates of an additional 5 standard DNAs were included in each 96-well plate (n = 10 plates). The replicate and inheritance data were used to review and refine clustering. In addition, 2 samples per 96-well plate were blindly duplicated (n = 20). Samples and SNPs with a call rate of <95% were excluded: of the remainder the mean per-SNP call rate was 98.64, the mean per-sample call rate was 99.72, and reproducibility was 99.9%. All 7 TP53 tagSNPs that were attempted were successfully genotyped (mean call rate = 99.57). Genotyping details are provided elsewhere (22).

The POCS study selected SNPs based on resequence analysis of over 7200 bp including the flanking 5, 3, conserved regions, and all exons in TP53 in 94 Norwegian women and 102 individuals in the SNP500Cancer panel.^29,31

Statistical analyses. Plots of LD that were drawn using Haploview version 4.1 (26) are in Supplementary Figs. S1 and S2, for those of European and African ancestry, respectively. SNP data for all three studies were independently analyzed, fitting unconditional logistic regression models adjusted for age, assuming a log-additive genetic model to estimate per allele odds ratios (OR) and corresponding 95% confidence intervals (95% CI) for associations between genotype and case status. SNPs with the lowest P values were nominated for evaluation in the replication study. The decision to move forward on the results of the NCOCS and MAYO was an a priori decision to achieve homogeneity likely depicted by serous histology especially in the context of the TP53 gene where p53 mutations are the common among serous cases. The POCS analyses were run overall and restricted to serous cases because of the a priori belief that TP53 variants might be more closely related to serous cases. In the PCOS study, the selection of SNPs for replication was based on SNPs with significant associations for all tumors or for serous tumors only.

Study participants. Ten additional sites contributed data for the TP53 replication study: the Australian Ovarian Cancer Study (AOCS) and the Australian Cancer Study presented together as AUS (27), the Family Registry for Ovarian Cancer (presented as STA; ref. 28), the Hawaiian Ovarian Cancer Study (HAW; ref. 29), the Malignant Ovarian Cancer Study Denmark (MALOVA; ref. 30), the New England Case-Control Study (NEC; ref. 31), the Nurses' Health Study (NHS; ref. 32), SEARCH Cambridge (SEA; ref. 33), the Los Angeles County Case-Control Study of Ovarian Cancer (presented here as USC; ref. 34), the University of California at Irvine study (UCI; ref. 35), and the United Kingdom Ovarian Cancer Population Study (presented here as UKO; ref. 36). Characteristics of the study populations are shown in Supplementary Table S3. Of the 10 replication case-control studies, 9 used population-based ascertainment for cases and controls and 1 was a nested case-control study (NHS). All studies received ethical committee approval and all study subjects provided informed consent. Key clinical and questionnaire data on study participants were merged into a common data set, including case-control status, ethnicity, race, tumor behavior, histologic subtype, age at diagnosis/interview, and history of prior cancers. Among the epithelial ovarian cancer cases across all OCAC sites, 293 had missing data on histologic subtype and were omitted. Race/ethnicity was missing for nine cancer cases. There was no missing data for any of the other variables. Centralized pathology review was conducted for HAW, NEC, and a sample of STA and AUS cases. The remaining sites used pathology reports only.

The combined data set (discovery and replication) comprised 5,206 White, non-Hispanic invasive epithelial ovarian cancer cases, of which 2,829 were classified as invasive serous ovarian cancer, and 8,790 White non-Hispanic controls. Analyses were restricted to serous invasive ovarian cancer cases to achieve a homogenous subgroup likely to have a common etiology.

SNPs found to be associated with risk of serous ovarian cancer in the replication study also were analyzed in endometrioid, mucinous, and clear cell invasive ovarian cancer. Additionally, we conducted analyses on potentially associated SNPs in 73 invasive serous ovarian cancer cases and 189 controls who self-reported themselves to be of African descent from 9 centers.

Genotype analyses. Eight of the 10 replicating OCAC sites used the 5′ nuclease TaqMan allelic discrimination assay (Taqman; Applied Biosystems) according to manufacturer's instructions. Samples from the Australian Ovarian Cancer Study and AUS in Australia were genotyped using the iPlex Sequenom MassArray system (Seqeunom, Inc.). The MAYO study additionally used the TaqMan assay for the follow-up of rs9894946 and rs2287498. To ensure quality control across laboratories, the TP53 SNPs were genotyped at each site using the HAPMAPPT01 panel of CEPH-Utah trios-standard plate by Coriell.³²

Statistical analyses. For the replication analysis, we used a mixed effects logistic regression, fit using Bayesian methods to appropriately handle heterogeneity in risks across studies. Each model associating case-control status to a SNP was adjusted for study site, reference age, and personal history of breast cancer. Study site, age, and SNP were treated as random effects with study site–specific distributions allowing for heterogeneity in effects across sites. In particular, the site-specific log ORs for each SNP were independently normally distributed with mean μ_TP53 and variance σ²_TP53, with the overall population level log OR, μ_TP53, assigned a proper but relatively noninformative normal distribution with mean 0 and variance 10. For the variance component σ²_TP53, the precision 1/σ²_TP53 was given a γ prior distribution with shape variable 1 and rate 0.05. This choice was based on assuming a range for the OR of between 0.5 and 2, leading to a prior expectation for the precision of ∼1/0.05. The shape variable of 1 corresponds to adding two sites in terms of degrees of freedom a priori. The same hierarchical distributions were used for the other random effects. The variable “previous breast cancer” was treated as a fixed effect with a common coefficient across all sites, which was a priori normally distributed with mean 0 and variance 10. Posterior inference was based on running 50,000 iterations of a Markov chain Monte Carlo algorithm using WinBUGS version 1.4.3 (37). We computed Bayesian 95% posterior probability intervals (PI) by taking the 2.5th and 97.5th percentiles of the sampled values from the Markov chain Monte Carlo output. Point estimates of ORs were estimated by the median of the sampled values. We summarized strength of association by computing the posterior probability that the OR was greater than one, P(OR > 1.0 | Data). This quantity was estimated by the frequency of ORs >1.0 in the Markov chain Monte Carlo samples. We also computed Bayes Factors (BF) as a measure of the strength of associations for each SNP in the replication study. This BF is the ratio of the posterior probability for an OR > 1.0 to the posterior probability for an OR of <1.0. Based on Jeffreys' scale of evidence (40, 41), BFs between 3.2 and 10 provide substantial evidence of a positive association, BFs between 10 and 100 provide strong positive evidence for an association, and BFs of >100 provide decisive evidence of a positive association.

This Bayesian mixed effects model is comparable with a mixed effects logistic regression and may be viewed as a limiting case of the meta-analytic approach of DerSimonian and Laird (40). Both approaches are more appropriate for a multicenter study evaluation than a fixed effects analysis as each allows for center-to-center heterogeneity in effects. We also present the results from comparable frequentist mixed effects model fits using SAS PROC GLIMMIX.³³

Using the log-additive model, analyses of the three discovery studies led to the selection of five SNPs for replication (rs2287498, rs12951053, rs1042522, rs16258595, and rs9894946). Two SNPs, rs2287498 and rs12951053, showed the strongest association with serous invasive tumors in the NCOCS and showed similar associations in either POCS (rs12951053) or MAYO (rs2078486, which was in high LD with rs2287498; r² = 0.98, MAYO and POCS data). The other three SNPs were selected based only on analyses from the POCS (NCOCS and MAYO were not available at the time of selection), which showed significant associations (P < 0.05) with all invasive cancers and/or serous cancers in the POCS. Other SNPs, including rs2078486 in the NCOCS and MAYO and rs2909430 and rs1642785 in the POCS, which also showed association in these discovery studies, were not genotyped because they had high correlations in the control populations with the SNPs that were selected for replication. In the POCS controls, rs1042522 was in strong LD with rs1642785 (r² = 0.94) and rs2909430 was in strong LD with rs1625895 (r² = 0.92).

A mixed effects SNP-at-a-time analysis of 5 SNPs (rs2287498, rs12951053, rs1042522, rs1625895, rs9894946,) was performed in up to 10 additional OCAC studies (Table 1). Eight studies that included a total of 1,859 serous invasive ovarian cancer cases and 5,927 controls participated in the follow-up of SNP rs2287498. The results of the analyses, including the discovery studies, showed a significant overall posterior median per allele OR of 1.30 (95% PI, 1.08–1.57;P (OR > 1 | Data) = 0.99; BF, 165.7). When the discovery data sets were omitted, the per allele overall posterior median OR was 1.23 (95% PI, 0.94–1.58), with a posterior probability P(OR > 1 | Data) of 0.94. The BF for this analysis was 16.3, indicating that the evidence is still in favor of a positive association between the SNP and case-control status. Using Proc GLIMMIX, we found the overall per allele OR for the association between serous invasive ovarian cancer and rs2287498 using all sites was 1.30 [95% confidence interval (CI), 1.10–1.53; P = 0.0059]. Omitting the discovery data sets, the per allele OR was 1.23 (95% CI, 0.99–1.54; P = 0.0573). Figure 1 is a forest plot of study specific ORs and CIs generated using a mixed effects model and fixed effects model, illustrating the shrinkage of the point estimates.

Follow-up analysis of rs12951053 included nine studies in the final analysis (Table 1). The results showed a significant association between SNP rs12951053 with a posterior median per allele OR of 1.19 (95% PI, 1.01–1.38; P(OR > 1 | Data) 0.98; BF, 47.8). When the data from the three discovery studies were omitted, the posterior median per allele OR was 1.14 (95% PI, 0.93–1.36; P (OR > 1 | Data), 0.90). The BF for this analysis was 9.2 indicating that the evidence is still in favor of the hypothesis that the OR of >1.0 when the discovery data sets were omitted. Using Proc GLIMMIX, the overall per allele OR for rs12951053 for all sites combined from the frequentist mixed effects analysis was 1.20 (95% CI, 1.06–1.36; P = 0.0081). Omitting the three discovery data sets, the per allele OR was 1.16 (95% CI, 1.00–1.35; P = 0.056). These per allele MLEs of ORs are comparable with Bayesian median ORs (differing by <0.02). Figure 2 is a forest plot of study specific ORs and CIs generated using a mixed effects model and fixed effects models, illustrating the shrinkage of the point estimates.

We found no evidence of association for rs9894846, rs1625859, or rs1042522 (Table 1). We also analyzed the combined data for five additional SNPs included in two or more data sets (see Supplementary Table S4). A significant finding in these analyses was for the association between ovarian cancer and SNP rs2078486, which was highly correlated with SNP rs2287498 among controls in the combined OCAC data (r² = 0.99) and therefore not a surprising finding. The per allele posterior median OR for rs2078486 was 1.49 (95% PI, 1.04–2.15; P (OR >1 | Data), 0.98).

In Table 2, we provide results of combined analyses of rs2287498 and rs12951053 with other histologic subtypes of invasive epithelial ovarian cancer. Analysis of rs2284798 found an overall per allele OR of 1.25 (95% PI, 0.94–1.65; P(OR > 1 | Data), 0.93; BF, 14.0) with a positive association for endometrioid invasive cancers. Evidence for a decisive association between SNP rs12951053 and endometrioid cancers was detected, with an overall per allele median OR of 1.31 (95% PI, 1.05–1.62; P(OR > 1 | Data), 0.99; BF, 109.6). No relationship was observed with mucinous or clear cell invasive cancers and either of these SNPs.

To further elucidate the relationship between genetic variation in TP53 and ovarian cancer risk, we examined the association between rs2287498 and rs12951053 and ovarian cancer among a small number of subjects of African ancestry using mixed effects logistic regression. For rs12951053 the analysis included 73 serous invasive ovarian cancer cases and 189 controls enrolled in nine participating OCAC sites. The results are in Table 3. For rs2287498, 69 serous invasive cases and 173 controls among 6 OCAC sites were included. The association between rs12951053 in intron 7 of TP53 and serous ovarian cancer was stronger in magnitude and consistent in direction with the corresponding association in non-Hispanic Whites, whereas this was not true for rs2287498 in exon 2 of the neighboring gene WDR79. In particular, the median per allele OR for rs12951053 was 1.79 (95% PI, 0.96–3.25;P (OR > 1 | Data), 0.97; BF, 29.4), whereas the median per allele OR for rs2287498 was 0.71 (95% PI, 0.40–1.27;P (OR < 1 | Data), 0.88; BF, 0.1). These markers were essentially uncorrelated in these samples (r² = 0.01), whereas they were in LD in the non-Hispanic White samples (r² = 0.62). This suggests that there may be a single functional locus that it is closer to SNP rs12951053 than SNP rs2287498.

This large investigation of 13 individual studies identified regions in or near TP53 associated with increased risk of serous ovarian cancer in non-Hispanic White women. The strongest association was seen with rs2287498, a synonymous amino acid change in exon 2 (F150F) of the neighboring gene WDR79 [median per allele OR, 1.30 (95% PI, 1.07–1.57)]. The posterior probability and BF indicate that the evidence for an association is strong. SNP rs12951053 also showed evidence of association in non-Hispanic White women [median per allele OR, 1.19 (95% PI, 1.01–1.38)]. Similar associations were found for endometrioid tumors with weak or no evidence for the other histologic subtypes. Of note, endometrioid cancers are often a mixture of endometrioid and serous, which may account for the consistent findings in only these two subtypes.

The r² between rs12951053 and rs2287498 is 0.62, making determination of the location of the association, assuming there is a single risk variant, difficult. To shed light on this, we examined evidence for association separately in OCAC samples of African descent. Although our power was limited, it showed an increased risk with rs12951053, located in intron 7 of TP53, and not with rs2287498 or any other marker in LD with rs12951053 in samples for non-Hispanic Whites. The contrasting results between women of African decent and non-Hispanic Whites are consistent with a single risk-associated marker at or near rs12951053. However, the lack of evidence for an association with rs2287498 may be a chance finding due to the small sample of women with African ancestry.

Of the SNPs evaluated in the replication study, only rs1042522 (P72R) has been previously assessed in relation to ovarian cancer risk. Several small studies have suggested that there may be an association between this common amino acid–changing polymorphism and ovarian cancer risk (11, 14, 19, 41). A higher frequency of the variant R allele among cases compared with controls has been reported; however, these differences were not statistically significant. Two additional small studies, 1 of 45 cases in South Africa among women with BRCA1 or BRCA2 mutations (18) and the other of 51 cases in Greece (16), did not find a greater frequency of the R allele among cases than controls. Likewise, we also were not able to confirm an association between the P72R polymorphism and ovarian cancer.

Few other TP53 polymorphisms have been previously investigated in relation to ovarian cancer risk. The intron 2 polymorphism (rs1642785) was evaluated in a study of 184 cases in Denmark with no evidence for a higher frequency of the variant allele in ovarian cancer cases than in controls (11). In the current study, rs1042522 tagged rs1642785 (r² = 0.94, in POCS) but was not confirmed in the replication phase.

Two case-control studies of 310 cases in Germany (15) and 225 cases in the United Kingdom (20) found an ∼2-fold increase in ovarian cancer risk associated with the intron 10 variant rs9894946. We were not able to replicate this finding, although a 16 bp insertion in intron 3, not measured in our study, was also related to risk and strongly correlated with rs9894946 in the German study population (15). This finding was not replicated in two other studies (12, 13).

The strengths of this study include a comprehensive tagging approach for identifying potentially relevant polymorphisms, the large sample size, and stringent quality control requirements that included a high proportion of blinded samples and high concordance and genotyping rates. Additionally, restricting the analysis to non-Hispanic Whites reduces the likelihood of significant population-stratification or confounding bias by race or ethnicity. Our ability to characterize the SNP association within the histologic subtypes of ovarian cancer is an additional strength of this study, although the power to detect associations in the less frequent subtypes was limited.

We used a staged design for the discovery and replication of a subset of the most promising findings to reduce genotyping costs. The consequence of this may have been loss of power to localize the putative SNP(s) compared with an approach where all SNPs are genotyped in all samples. A more powerful approach would have been for all centers to genotype all variants to more fully characterize the variation in TP53 associated with ovarian cancer risk.

Overall, this evaluation of common genetic variation in TP53 by an international ovarian cancer consortium indicates that certain SNPs in this gene are likely to be related to an increase in risk of serous and endometrioid invasive ovarian cancer. Although the removal of the three discovery data sets for the analysis of rs12951053 and rs2287498 represents a very conservative approach, evidence for an association with both SNPs remained evident. As described by Jeffrey's (38), BFs of >10 indicate “strong confidence that a result would survive further investigation.” Although it is possible that common variants not selected by the replication strategies we used are related to disease, these are likely to be captured by the tag SNPs. Fine mapping studies around the regions of interest will be needed to address this question.

No potential conflicts of interest were disclosed.

Grant support: Genotyping for the replication of findings was supported by a grant from the Ovarian Cancer Research Fund provided by the family and friends of Kathryn Sladek Smith. Additional support was provided by USPHS grant CA58598 and contracts N01-CN-55424 and N01-67001 from the National Cancer Institute, NIH; Department of Health and Human Services (Hawaii); the Roswell Park Alliance and the National Cancer Institute CA71966 and Core Grant CA16056 (Family Registry for Ovarian Cancer), Intramural Funds from the National Cancer Institute, NIH, Bethesda, MD (POCS); the California Cancer Research Program grants 00-01389V-20170 and 2110200; USPHS grants CA122443 (MAYO), CA76016 (NCOCS), NCA76016, CA14089, CA17054, CA61132, CA63464, N01-PC-67010, HL090559 and R03-CA113148; Department of Defense grant DAMD17-02-1-0666, and California Department of Health Services subcontract 050-E8709 as part of its statewide cancer reporting program (USC), R01 CA 58598 and N01 PC 35137 (Hawaii), CA54419,CA105009(NECC,) CA49449, and CA087969(NHS). H. Song was funded by a grant from WellBeing. The Australian Ovarian Cancer Study Management Group (D. Bowtell, G. Chenevix-Trench, A. deFazio, D. Gertig, A. Green, P. Webb) thanks the contribution of all the clinical and scientific collaborators. Australian Ovarian Cancer Study and the Australian Cancer Study Management Group (A. Green, P. Parsons, N. Hayward, P. Webb, D. Whiteman) also thank all of the project staff, collaborating institutions, and study participants. Financial support was provided by U.S. Army Medical Research and Materiel Command under DAMD17-01-1-0729, the Cancer Council Tasmania and Cancer Foundation of Western Australia (Australian Ovarian Cancer Study); The National Health and Medical Research Council of Australia (199600; Australian Cancer Study); G. Chenevix-Trench and P. Webb are supported by the National Health and Medical Research Council. SEARCH is funded through a program grant from Cancer Research UK. Dr. Paul Pharoah is a CRUK Senior Clinical Research Fellow.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

The Polish Breast Cancer study thanks Meredith Yeager from the Core Genotyping Facility, both at the National Cancer Institute (USA), Szeszenia-Dabrowska of the Nofer Institute of Occupational Medicine and W. Zatonski of the Department of Cancer Epidemiology and Prevention, Cancer Center and M. Sklodowska-Curie Institute of Oncology, 02-781 Warsaw, Poland, for their contribution to the POCS. Anita Soni (Westat, Rockville, MD) for her work on study management; Pei Chao (IMS, Silver Spring, MD) for her work on data and sample management; Douglas Richesson for statistical analyses; Meredith Yeager and Robert Welch for genotyping; and Neonila Szeszenia-Dabrowska and Witold Zatonski for their work during study design and data collection. Finally, we thank all the study participants who contributed to this research.