Blood concentrations of carotenoids and retinol and lung cancer risk: an update of the WCRF–AICR systematic review of published prospective studies

Abstract Carotenoids and retinol are considered biomarkers of fruits and vegetables intake, and are of much interest because of their anti‐inflammatory and antioxidant properties; however, there is inconsistent evidence regarding their protective effects against lung cancer. We conducted a meta‐analysis of prospective studies of blood concentrations of carotenoids and retinol, and lung cancer risk. We identified relevant prospective studies published up to December 2014 by searching the PubMed and several other databases. We calculated summary estimates of lung cancer risk for the highest compared with lowest carotenoid and retinol concentrations and dose–response meta‐analyses using random effects models. We used fractional polynomial models to assess potential nonlinear relationships. Seventeen prospective studies (18 publications) including 3603 cases and 458,434 participants were included in the meta‐analysis. Blood concentrations of α‐carotene, β‐carotene, total carotenoids, and retinol were significantly inversely associated with lung cancer risk or mortality. The summary relative risk were 0.66 (95% confidence interval [CI]: 0.55–0.80) per 5 μg/100 mL of α‐carotene (studies [n] = 5), 0.84 (95% CI: 0.76–0.94) per 20 μg/100 mL of β‐carotene (n = 9), 0.66 (95% CI: 0.54–0.81) per 100 μg/100 mL of total carotenoids (n = 4), and 0.81 (95% CI: 0.73–0.90) per 70 μg/100 mL of retinol (n = 8). In stratified analysis by sex, the significant inverse associations for β‐carotene and retinol were observed only in men and not in women. Nonlinear associations were observed for β‐carotene, β‐cryptoxanthin, and lycopene, with stronger associations observed at lower concentrations. There were not enough data to conduct stratified analyses by smoking. In conclusion, higher blood concentrations of several carotenoids and retinol are associated with reduced lung cancer risk. Further studies in never and former smokers are needed to rule out confounding by smoking.


Abstract
Carotenoids and retinol are considered biomarkers of fruits and vegetables intake, and are of much interest because of their anti-inflammatory and antioxidant properties; however, there is inconsistent evidence regarding their protective effects against lung cancer. We conducted a meta-analysis of prospective studies of blood concentrations of carotenoids and retinol, and lung cancer risk. We identified relevant prospective studies published up to December 2014 by searching the PubMed and several other databases. We calculated summary estimates of lung cancer risk for the highest compared with lowest carotenoid and retinol concentrations and dose-response meta-analyses using random effects models. We used fractional polynomial models to assess potential nonlinear relationships. Seventeen prospective studies (18 publications) including 3603 cases and 458,434 participants were included in the meta-analysis. Blood concentrations of αcarotene, βcarotene, total carotenoids, and retinol were significantly inversely associated with lung cancer risk or mortality. The summary relative risk were 0.66 (95% confidence interval [CI]: 0.55-0.80) per 5 μg/100 mL of αcarotene (studies [n] = 5), 0.84 (95% CI: 0.76-0.94) per 20 μg/100 mL of βcarotene (n = 9), 0.66 (95% CI: 0.54-0.81) per 100 μg/100 mL of total carotenoids (n = 4), and 0.81 (95% CI: 0.73-0.90) per 70 μg/100 mL of retinol (n = 8). In stratified analysis by sex, the significant inverse associations for βcarotene and retinol were observed only in men and not in women. Nonlinear associations were observed for βcarotene, βcryptoxanthin, and lycopene, with stronger associations observed at lower concentrations. There were not enough data to conduct stratified analyses by smoking. In conclusion, higher blood concentrations of several carotenoids and retinol are associated with reduced lung cancer risk. Further studies in never and former smokers are needed to rule out confounding by smoking.

Study selection
Included were prospective cohort, nested case-control or case-cohorts studies that reported estimates of the relative risk (RR) (e.g., hazard ratio, risk ratio, or odds ratio) and 95% confidence intervals (CIs) of specific carotenoids, total carotenoids, or retinol in blood and lung cancer incidence or mortality. In case of multiple publications of the same study, the newest publication that included the largest number of cases was selected.

Data extraction
The following data were extracted from each publication: first author's last name, publication year, country where the study was conducted, the study name, follow-up period, sample size, sex, age, number of cases, laboratory method for analysis, concentrations of carotenoids or retinol, and associated RRs and 95% CIs, and variables used in adjustment in the analysis.
The search and data extraction of articles published up to December 2005 was conducted by several reviewers at the John Hopkins University during the systematic literature review for the WCRF/AICR Second Expert Report (available online: http://www.wcrf.org/sites/default/files/SLR_lung.pdf). The search and extraction from January 2006 and up to December 2014 was conducted by the CUP team at Imperial College London.

Statistical methods
Meta-analysis of the highest compared with the lowest blood concentrations of carotenoids and retinol, and the dose-response associations with lung cancer were conducted. Random effect models were used to calculate the summary RRs and 95% CIs to take into account heterogeneity across studies [18]. Heterogeneity was determined using Q and I² statistics [19], and was explored in stratified analyses when there were eight or more studies in the analysis.
When continuous risk estimates were not provided in the articles, dose-response associations and 95% CIs were derived from categorical data using generalized least-squares for trend estimation [20], which required the RRs and CIs associated to at least three categories of blood concentrations, number of cases, and noncases or person years of follow up per category.
The mean or median values per category were used if provided in the articles, or the midpoint was calculated for studies that only reported a range of blood concentrations of carotenoids and retinol by category. When the range of the highest or lowest category of carotenoid/ retinol concentrations was open-ended, its width was assumed to be the same as the adjacent category.
If only the total number of cases or person years was reported in the articles, and the exposure was categorized in quantiles, the distribution of cases or person years was calculated by dividing the total number of cases or person years by the number of quantiles. If the results were reported for men and women separately, they were combined using a fixed effects meta-analysis before being pooled with other studies.
For studies that reported blood concentrations in μmol/L, the units were converted to μg/100 mL by dividing the concentration in μmol/L by 0.01863 for αcarotene, βcarotene, lycopene, and total carotenoids, and by 0.01809, 0.01758, and 0.03491 for βcryptoxanthin, lutein/zeaxanthin, and retinol, respectively [21].
Small-study effects, such as publication bias, were assessed using funnel plot and Egger's test [22].
A potential nonlinear dose-response association between blood concentrations of carotenoids and retinol was assessed using fractional polynomial model [19] and the best-fitting second-order fractional polynomial regression model, defined as the one with the lowest deviance was determined. A two-tailed P < 0.05 was considered statistically significant.
In all analyses, the results of each paper with the most comprehensive adjustment for confounders were included. Stata version 12 software (StataCorp, College Station, TX) was used for all analyses.
Only three studies could be included in nonlinear metaanalysis and no evidence of nonlinearity was observed, P nonlinearity = 0.11 (Fig. 2B).
There was some evidence of a nonlinear dose-response of lung cancer and blood concentrations of βcarotene (P nonlinearity = 0.05, n = 6), with the curve showing a slightly steeper slope in the low range of βcarotene concentrations (Fig. 2D), however, there was clear evidence of an inverse dose-response relationship across the range of βcarotene concentrations.
Although, the test for nonlinearity was significant (P nonlinearity = 0.03, n = 4) and there was a slightly stronger association at lower blood concentrations of βcryptoxanthin, the association was nearly linear from 5 μg/mL and above (Fig. 3B).
There was some evidence of nonlinear dose-response of lung cancer and blood concentration of lycopene (P nonlinearity = 0.01, n = 3) (Fig. 3D). The inverse doseresponse association appeared to be stronger at low blood concentrations of lycopene (approximately up to 20 μg/100 mL) with a weaker association beyond this level.

Blood total carotenoids
Four studies (693 cases) were included in the dose-response meta-analysis [4,24,26,29]. The summary RR for an increase of 100 μg/100 mL was 0.66 (95% CI: 0.54-0.81, I² = 0%, P heterogeneity = 0.43) (Fig. 5A). There was no evidence of publication bias with Egger's test (P = 0.30) but the number of studies was small. The overall RR for the high versus low analysis was 0.64 (95% CI: 0.44-0.93, L. Abar et al. Blood carotenoids and retinol and lung cancer risk I² = 23%, P heterogeneity = 0.27) in five studies (724 cases) (Fig. 5B). The nonlinear dose-response analysis was not conducted because of the small number of studies with the required data (n = 2).

Subgroup and sensitivity analyses
The subgroup analysis stratified by sex, cancer outcome, and geographic location was conducted only for blood βcarotene and retinol because of small number of studies in the other blood carotenoids investigated. It was not possible to conduct stratified analyses by smoking status or histologic type of lung cancer because of lack of such data from the studies included.
The subgroup analysis stratified by blood fasting status was conducted and there was no strong evidence of different association as the CIs mostly overlap.
In terms of geographic location, the results were significant only in studies conducted in the United States (five studies) but not in Asia (three studies) ( Table 2).
In stratified analysis by geographic location, the results were significant only in studies conducted in the Asia (three studies) but not in United States (three studies) ( Table 2).

Discussion
In this meta-analysis, there was an inverse dose-response relationship of blood concentrations of αcarotene, βcarotene, and total carotenoids, and lung cancer risk. An inverse association with blood concentrations of retinol was also observed. Subjects with the highest blood concentrations of total carotenoids and retinol had 19% and 34% lower RR of lung cancer when compared to those with the lowest blood concentrations, respectively. There was little evidence of heterogeneity in these analyses. Apart from the analysis of lycopene, there was no evidence of publication bias with the statistical tests used; however, the number of studies was limited.
To our knowledge, this is the first meta-analysis to examine a potential nonlinear association between blood concentrations of carotenoids and retinol, and lung cancer risk. The nonlinear dose-response analyses suggested inverse associations for all carotenoids, and in general, there was a stronger dose-response relationship in the lowest range of carotenoid and retinol concentrations than at the highest range. Nonlinearity was most pronounced for lycopene and retinol, for which there was a flattening of the doseresponse curve at the highest concentrations, while for most of the remaining carotenoids associations were slightly stronger at lowest compared to highest concentrations, but there was a clear inverse dose-response relationship with further reductions in risk with increasing carotenoid concentrations. These findings suggests that it might be most important to avoid low blood concentrations of lycopene and retinol, and that there is little further benefit in people with highest blood concentrations, while for alpha-carotene, beta-carotene, and beta-cryptoxanthin there might be further reductions in risk with increasing blood concentrations.
This study has several limitations which should be considered when interpreting the results. Smoking tends to be associated with lower intakes of fruit and vegetables, high intakes of fat and higher consumption of alcohol [38] and smokers have lower blood concentrations of some of carotenoids [39][40][41]. Therefore, it is possible that the observed inverse associations could have been due to residual confounding by cigarette smoking. With the exception of one study that only adjusted for age [31], all the studies included in our analysis were adjusted at least for smoking status, but there was not enough data to conduct subgroup analysis by smoking status. In the only study that showed separate results in smokers and never/former smokers [17], an inverse association with lung cancer mortality was observed for αcarotene and βcryptoxanthin only in current smokers but not in never/former smokers, however, in a previous meta-analysis of fruit and vegetable intakes (some of which are high in carotenoids) and lung cancer risk, we found similar summary RRs among never smokers as compared to current or former smokers [42], although power was more limited among never smokers as the number of cases was modest.
Given the lack of data stratified by smoking status, further studies are needed in never smokers to rule out the potential confounding by smoking. Residual confounding by other factors potentially related to the blood levels of the biomarkers investigated and to lung cancer is also a possibility. When the studies in high-risk populationshigh-risk miners, heavy smokers or people exposed to asbestos-were excluded from the meta-analysis in sensitivity analysis, the inverse association with βcarotene [25,27,28] was slightly strengthened from 16% to 19% and the inverse association with retinol [25,27,28,37] was no longer statistically significant.
Although there was a large number of studies that could be included in the dose-response analyses of βcarotene (n = 9) and retinol (n = 8), fewer studies reported on the other carotenoids (n = 4-6). The inverse associations were observed in men but not in women, and whether this is due to residual confounding, low number of cases in the analyses in women or gender differences is unclear and needs further study. Furthermore, blood concentrations of carotenoids and retinol may not only reflect dietary intake, but can be influenced by the lipid content of the diet, metabolism and absorption, and genetic variability [7,39,40]. As carotenoids and retinol are fat-soluble, the lipid content of the diet increases the absorption. Some carotenoids including α and βcarotene, and βcryptoxanthin can be partially metabolized to retinol, particularly in people with depleted vitamin A concentrations [40]. The absorption and hence the bioavailability of carotenoids can be modulated by the fat content of the diet, competition with other carotenoids, degree of colon fermentation, and hormonal factors [40].
The results of this meta-analysis provide further support that high blood concentrations of carotenoids and retinol, as biomarkers of fruits and vegetable intake, are associated with reduced lung cancer risk. Carotenoids are found in many different types of fruit and vegetables, and it has been shown in epidemiological studies that dietary intakes of green and raw vegetables, carrots and broccoli are correlated with blood concentrations of αcarotene, βcarotene, and lutein/zeaxanthin [43], and fruits and root vegetables, carrots and tomato products are good predictors of βcryptoxanthin, αcarotene and lycopene in plasma [44].
In contrast to the results of many observational studies and the current meta-analysis, two large randomized controlled trials (RCT's), the ATBC and CARET, showed an increased risk of lung cancer with high-dose supplemental βcarotene among smokers [14][15][16]. The increased risk at high doses may be related to the prooxidant activity of βcarotene when administered as a supplement in high doses (5-10 times greater than normal dietary intake) to heavy smokers [6,45,46]. In addition, it is possible that the difference in results between the RCTs and the observational studies may be because high blood concentrations of carotenoids and retinol simply may be markers of a high fruit and vegetable intake, but may not themselves be the constituent(s) responsible for the beneficial effect. Fruits and vegetables are not only good sources of carotenoids but also contain many other vitamins, minerals, fiber, antioxidants, and numerous phytochemicals [45] that could have a potential protective effect against lung cancer, and it is possible that a number of constituents may act synergistically [47].
Strength of this meta-analysis is the inclusion of prospective cohort studies which avoids potential recall biases and that are less prone to selection biases than case-control studies. Some analyses included a large number of cases and had statistical power to detect relatively small associations but for some micronutrients the power may have been insufficient. Most studies, as mentioned previously, were adjusted for main confounders including smoking status, intensity, duration of smoking, and other smoking variables. Most of the studies measured the carotenoids and retinol blood concentrations using high-performance liquid chromatography (HPLC). The cancer outcome in the included studies was identified through cancer registries, death certificates and hospital records, and loss to follow-up was very low.
In conclusion, higher blood concentrations of total carotenoids, αcarotene, βcarotene, lycopene, and retinol were inversely associated with lung cancer risk. However, because of the lack of data in never smokers, further large scale studies stratified by smoking status are needed to rule out residual confounding by smoking.

Acknowledgment
The authors' responsibilities were as follows-L. A., A. R. V.: performed the updated literature search and the updated data extraction; L. A.: conducted statistical analyses, wrote the first draft of the original manuscript, had primary responsibility for the final content of the manuscript, and took responsibility for the integrity of data and accuracy of the data analysis; C. S.: was database manager for the project; and all authors, D. A., S. V., D. A. N. R. and D. C. contributed to the revision of the manuscript and had full access to all data in the study. D. C. G.: advised on and contributed to statistical analyses. T. N. is the principal investigator of the Continuous Update Project at Imperial College. All authors commented on drafts of the paper and approved the final version. None of the authors reported a conflict of interest related to the study. The views expressed in this review are the opinions of the authors. The views may not represent the views of World Cancer Research Fund International/ American Institute for Cancer Research and may differ from those in future updates of the evidence related to food, nutrition, physical activity, and cancer risk. The sponsor of this study had no role in the decisions about the analysis or interpretation of the data; or preparation, review, or approval of the manuscript.

Supporting Information
Additional supporting information may be found in the online version of this article: