Height and lung cancer risk: A meta-analysis of observational studies

Background The association between height and lung cancer risk has been investigated by epidemiological studies but the results are inconsistent. This meta-analysis was to evaluate whether the height is associated with lung cancer. Methods We identified relevant articles by searching the MEDLINE and EMBASE databases, and reviewed the reference lists of selected papers. A random effect model was used to calculate summary odds ratios (OR) and relative risk (RR) with 95% confidence intervals (95% CI). Publication bias was estimated using Egger’s regression asymmetry test. Results We included a total 16 studies (15 prospective studies and one case–control study) on adult height and lung cancer risk in the meta-analysis. Overall, per 10-cm height increases were associated with increased risk of lung cancer (RR 1.06; 95% CI 1.03–1.09, I2 = 43.6%). Conclusions In this meta-analysis, high adult height is related to increased lung cancer risk. Well-designed, large prospective studies are required to obtain a better indication of the relationship.


Introduction
Lung cancer is one of the most common cancers worldwide in terms of both incidence and mortality [1]. Although cigarette smoking and specific occupational exposure (e.g., indoor radon, household coal smoke) are major known risk factors for lung cancer, the etiology of lung cancer remains largely elusive [2].
Adult height, which is determined by both genetic and environmental factors [3], is considered a biomarker that reflects the interplay of genetic endowment and various early-life experiences PLOS  and exposures (e.g., fetal, dietary, social, and psychological circumstances) [4][5][6][7][8]. As the study of height can provide insights into patterns of shared and differing early determinants of major diseases of later life, it would be informative to compare the associations of adult height with subsequent risk of a wide range of diseases. Previous epidemiologic studies and meta-analyses have reported positive associations between height and risk of all cancers combined and several specific cancers, including cancer of the breast [9], prostate [10], colorectal [11], kidney [12], ovary [13], pancreas [14], testis [15], and the endometrium [16], and malignant melanoma [9,17,18] and lymphohematopoietic malignancies [19], and have reported negative associations between height and risk of all-cause, cardiovascular, and respiratory disease [20][21][22][23][24][25]. Numerous studies have examined the relation between height and lung cancer [9,26]; however, the results have been inconsistent. Many factors, including selection bias and confounding, can lead to inconsistencies in such studies. However, consensus has not been reached on whether height is a risk factor for lung cancer in both women and men. To evaluate the association between height and risk of lung cancer comprehensively, we conducted a systematic review and meta-analysis of observational studies.

Search strategy
A systematic literature search with no language restrictions was conducted in MEDLINE and EMBASE for studies on the association between height and lung cancer incidence in humans. We searched all studies published before November 20, 2016. We used the following combinations of search terms: ("anthropometry" or "body size" or "height") and ("lung" or "pulmonary") and ("cancer" or "neoplasm" or "carcinoma"). In addition, we searched the reference lists of the retrieved papers for relevant articles.

Study selection criteria
We included studies that met the following criteria: (1) case-control or cohort study investigating the association between height and lung cancer; (2) the outcome was lung cancer incidence or mortality; (3) the exposure of interest was height; and (4) reported relative risk (RR) or odds ratio (OR) estimates with their corresponding 95% confidence intervals (CIs) (or sufficient data to calculate of these effect measures). We included studies in which height had been self-reported and in which it had been directly measured. When several articles reported results from the same study population, only the most recent or comprehensive study was included.

Data extraction
The following information was extracted from the included articles: first author's last name, publication year, study name or source, country where the study was conducted, study period, study duration; follow-up period (cohort studies) or data collection (case-control studies), sample size(study participants, number of cases), sex, age, height assessment method (selfreported or measured), comparison of exposure level, and RRs or ORs and 95% CIs for the highest versus lowest height or per unit increase in height. If a study reported several adjustment models for potential confounding variables, we extracted the risk estimate of the most fully adjusted model. The quality of the included studies was assessed using the 9-star Newcastle Ottawa scale [27].

Statistical analysis
We used random effect models to calculate summary RRs and 95% CIs for the highest versus lowest height and for the quantitative analyses [28]. The average natural logarithm of the RRs was estimated, and the RR from each study was weighted by the inverse of its variance. Twotailed P 0.05 was accepted as statistically significant. In studies using height as a categorical variable, we standardized all reported RRs into comparison of the risk of the higher group with that in the lower group. Therefore, when the lowest group was not referent, we used the method proposed by Hamling et al [29] to recalculate the RRs using the lowest one as reference.
In the quantitative analyses, twelve studies [9,23,[30][31][32][33][34][35][36][37][38][39] had directly provided RRs for per unit increase in height. For four studies [40][41][42][43] that did not provide estimate for per unit, we compute study-specific slopes (linear trends) and 95% CIs from the natural logs of the RRs and CIs across categories of height using the method by Greenland and Longnecker [44]. The method requires that the number of cases and person-years or non-cases and the relative risks with the variance estimates are known for at least three quantitative categories of use. We estimated the distribution of cases or person-years in studies that did not report these but reported the total number of cases/person-years. For example, if the total number of person-years was provided and the exposure variable was categorized by quintiles, we divided the number of person-years by five. The median or mean level of height in each category was assigned to the corresponding relative risk for each study. If a study reported height expressed as a range, we estimated the mid-point in each category by calculating the average lower and upper bound. When the highest or lowest category was open-ended, we assumed that they were of the same length as the adjacent interval.
We estimated heterogeneity among studies using the Q test and the I 2 statistic [45]. I 2 takes values between 0% and 100%, and I 2 > 50% is considered a measure of high heterogeneity [46]. To examine whether the results could have been influenced by a single study or a study with an extreme result, we performed sensitivity analysis by removing one study at a time. Sources of heterogeneity were explored using subgroup analyses and random-effects meta-regression analysis, according to sex, outcome (incidence and mortality), height assessment (measured and self-measured), geographic location, and adjustment for confounding factors such as smoking, alcohol, and body mass index.
We evaluated potential publication bias using funnel plots, Egger's regression test [47], and Begg's rank correlation test [48]. P < 0.1 was considered to indicate statistically significant publication bias. All statistical analyses were performed using Stata 12.0 (StataCorp, College Station, TX, USA).
In the sensitivity analysis, the influence of each study on the pooled RR was examined by repeating the meta-analysis while omitting one study at a time. The 16 study-specific RRs ranged from1.07 (95%CI1.04-1.10) when the National Institutes of Health-AARP Diet and Health Study was excluded to1.05 (95% CI 1.02-1.08) when the Korean Adult Population Study was excluded, but were in general similar. This procedure proved that our results are reliable and robust.
No indication of publication bias was observed form either with the Egger's test (P = 0.358) or Begg's test (P = 0.673) (Fig 4).

Discussion
We found a similar, weakly positive association between height and lung cancer risk among men and women, although it was statistically significant only among men. Lung cancer risk increased6% for a 10-cm increase in height and a 15% increased risk for high versus low height. These associations were evident even after adjustment for smoking, alcohol, and body mass index, suggesting that height represents a robust and independent factor of increased lung cancer risk.
It has been confirmed that taller people are at higher risk for breast and colorectal cancer [9], and taller height is a possible risk factor for several other cancers; however, the potential biological mechanisms that underlie the association between greater height and lung cancer risk are unclear. It is believed that a combination of genetics, childhood and adolescent dietary factors, and infections determine adult height [8]. Previous epidemiological studies have indicated that insulin-like growth factor 1 (IGF-1) might play an important role in the development of breast [49], colorectal [50], and lung cancer [51]. IGF-1 can stimulate cell proliferation, adhesion, and migration and inhibit apoptosis, which could ultimately result in cancer. However, it is unlikely that the IGF1 gene alone would explain the observed increased lung cancer risk associated with adult height. Other genes also revealed recently to influence adult height, such as the genes for p53, c-Myc, and estrogen receptor α (ERα), are thought to be crucial for tumorigenesis [52].
The overall analysis revealed that there was moderate heterogeneity. To investigate the potential source of heterogeneity, we carried out subgroup analysis according to study design, sex, geographical area, height assessment method, and adjustment for confounding. However, we did not find an explanation for the heterogeneity, as it persisted in most subgroup analyses. Such heterogeneity may be due other reasons not included in our subgroups, such as different adjustments for confounding factors.
Our meta-analysis has several strengths. First, it included a larger sample size (33,824 cases among 4,907,101participants) and summarized statistics which provided sufficient power to detect the association between height and lung cancer risk. Second, we used two methods to investigate the association between height and lung cancer risk, the meta-analysis by categories of height and quantitative analysis. Third, we conducted several subgroup analyses according to study characteristics, study quality scores, and adjustment for a wide range of potential confounding variables, and our findings were generally robust. Moreover, we used the Newcastle-Ottawa scale to evaluate the quality of the eligible studies, and studies included in our metaanalysis were deemed high quality because their total scores ranged 7-9. Nevertheless, our study also has several limitations. First, although the included studies controlled for various known risk factors for lung cancer, we cannot rule out the possibility of unknown or residual confounding by dietary, behavioral, or physical activity factors. Second, the results may have been influenced by smoking, as cigarette smoke exposure has been established as an independent risk factor for developing lung cancer, but when we adjusted for smoking and age, the results did not change. Third, various height assessments were used in our analysis. Some studies used self-reporting to assess height, which may have led to overestimation of the participants' true heights. However, the subgroup analysis revealed no substantial change in the analysis that included only the studies that relied on measured height. Finally, we cannot preclude the possibility that we might have overlooked other unpublished studies, despite our extensive literature search. The potential publication bias may have been because studies with null effects are less easily published than those reporting positive effects, therefore it was difficult for us to obtain such studies, although Egger's test or Begg's test did not reveal the presence of publication bias.
In conclusion, the present data suggest a positive association between height and lung cancer risk. The mechanisms involved are likely to be complex. Additional studies are warranted to extend our findings and to clarify the unknown mechanisms. In addition, given the unexplained heterogeneity, further studies are needed before a conclusion can be drawn.