Telomerase Reverse Transcriptase Locus Polymorphisms and Cancer Risk

A Field Synopsis and Meta-analysis

Simone Mocellin; Daunia Verdi; Karen A. Pooley; Maria T. Landi; Kathleen M. Egan; Duncan M. Baird; Jennifer Prescott; Immaculata De Vivo; Donato Nitti


J Natl Cancer Inst. 2012;104(11):840-854. 

In This Article

Materials and Methods

Search Strategy, Eligibility Criteria, and Data Extraction

We followed the methods proposed by the Human Genome Epidemiology Network (HuGENet)[10] as well as the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)[11] and Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guidelines.[12] A two-step search strategy was adopted. First, a systematic review of original articles, reviews, and meta-analyses analyzing the association between TERT locus polymorphisms and cancer risk was performed by searching PubMed, Embase, Cancerlit, Google Scholar, and ISI Web of Knowledge databases (Supplementary Figure 1, available online). The search included the following keywords: "cancer," "tumor," "carcinoma," "melanoma," "sarcoma," "lymphoma," "leukemia," "polymorphism," "SNP," "variant," "risk," "association," "TERT," "telomerase," "locus," "5p15.33," and "gene." The 5′-end of TERT resides in a 62-kb region of high linkage disequilibrium (LD) that encompasses the upstream gene cleft lip and palate transmembrane 1-like (CLPTM1L)[9] (Supplementary Figure 2, available online). Therefore, polymorphisms not belonging to the TERT gene but localizing within this LD area might be tagging relevant TERT polymorphisms; accordingly, CLPTM1L was used as an additional search term in the second step. Finally, in the light of the growing diffusion of high-throughput technological platforms for the investigation of gene polymorphisms, the expression "genome-wide association study" and its acronym "GWAS" were also used as key words. We then performed the following to retrieve other potentially relevant data: 1) the name or identity of each polymorphism was used as a keyword to further refine the search; 2) cited references from selected articles were reviewed; 3) publicly available databases dedicated to associations between genotype and phenotype (eg, Database of Genotypes and Phenotypes [dbGaP], were searched; 4) authors were contacted whenever unreported data were potentially useful for the systematic review or to rule out overlapping data reported in different publications.

Figure 1.

Flow chart for the estimation of the joint risk of lung adenocarcinoma in the general population attributable to three TERT locus polymorphisms. Top panel. A forest plot depicting the meta-analysis of the studies that contributed to define the association between the minor alleles of three single-nucleotide polymorphisms (SNPs) and the risk of developing lung adenocarcinoma. Open squares represent odds ratios (ORs) of single studies (the width of each square is proportional to the weight of the corresponding study; the horizontal line represents the 95% confidence interval [CI] of the study OR); solid black diamonds represent summary OR for each SNP (the width of each diamond is proportional to the 95% CI of the corresponding summary OR). Bottom left panel. Only SNP showing strong cumulative evidence for association with lung adenocarcinoma were selected. Cumulative evidence was assessed as per the Venice criteria (see text for more details). OR refer to risk alleles (alleles associated with increased cancer risk). Bottom right panel. The joint PAR (population attributable risk) represents the proportion of lung adenocarcinoma cases estimated to be attributable to the three SNPs showing strong cumulative evidence of association; it depends on both the magnitude of the association (OR) and the risk allele frequency in the general population.

Studies dealing with the association between any TERT locus polymorphism and predisposition to any type of cancer in humans were considered eligible, provided that the raw or summary data necessary to calculate the risks were available. Exclusion criteria were non-English language and data published in abstract form. For each polymorphism, exclusion criteria were less than 5% minor allele frequency (MAF) in control subjects (rare polymorphisms) and violation of Hardy–Weinberg equilibrium.

The following data were extracted from eligible studies: authors' names; region or country where the study was conducted; year of publication; number of case subjects with cancer and healthy control subjects; ethnicity; allelic frequency in both case subjects and control subjects (if no raw data were available, summary data were collected; ie, odds ratios [ORs] and 95% confidence intervals [CIs]); MAF and Hardy–Weinberg equilibrium in control subjects; study design; genotyping; and statistical methods. For analysis purposes, the database, which will be updated on a yearly basis and will be publicly available on the Melanoma Molecular Map Project website [[13]], was frozen in April 2011. Data were extracted independently by the two investigators (D. Verdi and S. Mocellin) to ensure homogeneity of data collection and to rule out the effect of subjectivity in data gathering and entry. Disagreements were resolved by iteration, discussion, and consensus. To unravel potential systematic biases, a third investigator (D. Nitti) performed a concordance study by independently reviewing all eligible studies; complete concordance (100%) was reached for all variables assessed.

Statistical Analysis

Meta-analysis. Because all the investigated gene variants were biallelic polymorphisms, per-allele odds ratios and corresponding 95% confidence intervals were used to assess the strength of association between each genetic variant and cancer risk, where protective and risk alleles were associated with ORs less than or greater than 1, respectively. Per-allele ORs were calculated for each study and each polymorphism, assuming a codominant genetic model. This assumption was suggested by the following reasons: 1) for some studies (including many GWAS), neither raw nor summary genotype data were available (only per-allele ORs were reported, which only allows to explore the codominant model); 2) the codominant model is widely used as a conservative choice between the recessive and dominant models; 3) the codominant model does not require adjustment for multiple hypotheses (which is necessary when different models are tested); 4) methods that let the data dictate the genetic model (ie, model-free approach) require raw data on genotype distributions (which were not available for many identified studies).

For each allelic contrast (ie, data regarding a specific polymorphism and a given tumor type), summary per-allele ORs (meta-risks) were calculated by performing random-effects meta-analysis as per Der Simonian and Laird (ie, using the inverse variance method to weight the studies), a Z test being used to formally prove the statistical significance. The choice of the random-effects model was suggested by three main reasons: 1) the variety of histological cancer types, which was far from being fully represented for each genetic variant; 2) because the Q test for between-study heterogeneity is characterized by low statistical power, which is especially relevant when few studies are available; 3) in general, the random-effects model is a more conservative choice when heterogeneity is present, whereas it reduces to the fixed effect model when heterogeneity is absent.

For each polymorphism (whose allele of interest is indicated by the corresponding nucleotide letter in squared brackets), a meta-analysis was performed if at least two data sources were available. Stratification by ethnicity and histological subtype was done if data permitted. When a publication reported the data in an ancestry-specific way, each ancestral subset was considered as a distinct study. Similarly, with regard to GWAS, if data were available, discovery (hypothesis testing) and replication (validation) phases were considered as separate studies.

Meta-analysis also included evaluation of between-study heterogeneity, sensitivity analysis, and examination for bias. Heterogeneity (true variance of effect size across studies) was formally investigated using a Q test (to assess whether observed variance exceeds expected variance) and I2 statistic (it indicates the percentage of the variability in effect estimates because of true heterogeneity rather than sampling error).[14] The Q test for heterogeneity was also used to formally compare effects (ie, meta-risk) between groups of interest (eg, different ethnicities).

The extent to which the combined risk estimate might be affected by individual studies was assessed by consecutively omitting every study from the meta-analysis (leave-one-out sensitivity analysis). This approach would also capture the effect of the oldest or first positive study (first study effect).

Funnel plot was used to detect the so-called "small study effect." Publication and selection biases in meta-analysis are more likely to affect small studies, which also tend to be of lower methodological quality.[15] Funnel plot asymmetry was formally investigated with the Egger linear regression approach (Egger test) and the Begg rank correlation test (Begg test). The impact of small study effect bias on the summary effects was formally assessed by means of the trim and fill method described by Duval and Tweedie.[16] The excess of statistically significant findings (potentially indicating the so-called "chasing bias") was evaluated by the test proposed by Ioannidis and Trikalinos (Ioannidis and Trikalinos test).[17]

The population attributable risk (PAR) was calculated using the following formula:

where Pr is the proportion of control subjects exposed to the allele of interest and the relative risk (RR) was estimated using the summary estimates (ORs) calculated by the meta-analysis. The joint PAR for combinations of polymorphisms was calculated as follows:

where PAR i corresponds to the individual PAR of the ith polymorphism and n is the number of polymorphisms considered.[18] Because PAR is a relative measure of effect and thus it does not account for the absolute risk of disease, we also calculated the attributable community risk, according to the following formula:

where Ic is the crude risk in the general population (probability of developing the disease during the entire lifetime) and I0 is the risk of disease in nonexposed (ie, in people carrying the protective allele).[19]

P values less than .05 were considered statistically significant for all tests, except for Q test for heterogeneity, Egger test, Begg test, and Ioannidis and Trikalinos test, for which a less stringent 10% alpha level of statistical significance was applied. The latter three tests were performed if at least four studies were available. All tests were two-sided. The Bonferroni method was used for P value adjustment in case of multiple testing. Statistical analyses were performed with STATA 11.0/SE software (College Station, TX).

Assessment of Cumulative Evidence. We used the Venice criteria[10,20] to evaluate the epidemiological credibility of each statistically significant association identified by meta-analysis. Briefly, credibility was defined as "strong," "moderate," or "weak" based on grades A, B, or C in three categories: 1) amount of evidence, 2) replication of the association, and 3) protection from bias. Amount of evidence (which roughly depends upon the study sample size) was graded by the sum of test allele among case and control subjects in the meta-analysis: grades A, B, or C were assigned for values greater than 1000, 100–1000, and less than 100, respectively. Replication was graded by the heterogeneity statistic: grades A, B, and C were assigned for I 2 less than 25%, 25–50%, and greater than 50%, respectively. Protection from bias was graded as grade A if there was no observable bias (bias was unlikely to explain the presence of the association), grade B if bias could be present, or grade C if bias was evident. Assessment of protection from bias also considered the magnitude of association; grade "C" was assigned to an association with a summary odds ratio less than 1.15, or greater than 0.85 in case of protective effect.

Overall, the cumulative epidemiological evidence for statistically significant associations upon meta-analysis were considered to be "strong" if all three grades were A, "weak" if any grade was C, and "moderate" in all other cases. In case of no statistically significant association upon meta-analysis, the minimum detectable risk was calculated considering a hypothetical study with sample size equal to the combined sample sizes of the studies reporting on a given polymorphism (error alpha and power were set at 5% and 90%, respectively). If no heterogeneity was found (I 2 <25% and Q test P >.10) and the detectable alternative included non-negligible associations (ie, OR ≥1.15 or ≤0.87), we considered the cumulative evidence sufficient to rule out any meaningful relationship between that polymorphism and cancer risk (strong evidence); in the other cases, the evidence for lack of association was considered weak (ie, more data are necessary before the association can be ruled out).


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: