Background

Pulmonary function, which is reliably measured by spirometry, is central to the diagnosis and staging of chronic obstructive pulmonary disease (COPD). COPD, the third most common cause of death in the US, led to $29.5B in direct costs and $20B in lost productivity costs in 2010 [1]. The age-related rate of decline in the forced expiratory volume in the first second (FEV1) is a marker of mortality risk in the general population [2] and in healthy smokers [3], the rate of decline is steeper in smokers [4] and in COPD patients, although the latter association varies by disease attributes [5,6]. The forced expiratory flow at the mid-portion of forced vital capacity (FEF25–75) reflects the state of small airways, and offers a measure of lung function reflecting airflow rather than volume [7]. Attenuating lung function decline may reduce morbidity and mortality, both in healthy persons and in COPD patients. The identification of factors that affect lung function decline is important to the development of clinical or public health interventions.

Both smoking and aging accelerate the annual rate of decline in FEV1 [4,8]; the effect of other factors, including diet, is less clear [9]. Observational studies of nutrients (whole foods, micronutrients, dietary patterns, biomarkers) and lung outcomes (COPD, lung function) support the broad hypothesis that nutrients with antioxidant properties improve lung health [10-13], presumably by altering the oxidant/antioxidant balance in lung tissue. A recent critical review of the evidence for a causal relation between nutrition and lung outcomes [9] concluded there was a “limited/suggestive” role for diet, reflecting that the majority of studies are cross-sectional. However, most existing longitudinal studies report protective associations of antioxidant nutrients and lung outcomes, and there is strong evidence that the beneficial effects of diet may be limited to smokers [14]. Given that observational studies are limited by potential confounding due to lifestyle factors associated with healthful diets, and given that measures of diet based on self-report are subject to bias and have poor precision, experimental studies are needed to fully understand whether nutrition affects lung function and its decline with age, particularly in cigarette smokers.

The few randomized controlled trials considering respiratory endpoints other than lung cancer used post-hoc, secondary analyses [15-17] mainly in special populations (all heavy smokers, all with vascular disease), did not study lung function decline, and did not test selenium. Using the infrastructure of the Selenium and Vitamin E Cancer Prevention Trial (SELECT) [18], we conducted an ancillary study nested within SELECT to investigate the a priori hypothesis that supplementation with selenium and/or vitamin E, two nutrients with antioxidant potential, would attenuate the annual decline in lung function; we hypothesized a stronger effect in cigarette smokers given their higher exposure to inhaled oxidants.

Methods

Study design

The Respiratory Ancillary Study (RAS) was nested within SELECT [18], a phase 3 randomized, placebo-controlled double-blind trial of 35,533 men testing whether selenium (200 μg/d L-selenomethionine) and vitamin E (400 IU/d all rac-α-tocopheryl acetate) alone and/or in combination would prevent prostate cancer. SELECT eligibility included age ≥55 y (≥50 y in African-Americans), serum prostate-specific antigen ≤ 4 ng/mL, and no clinical evidence of prostate cancer. SELECT enrolled men in the United States and Canada between 2001–2004; use of study supplements stopped on 10/23/2008, after an interim analysis determined that there was no effect and that further intervention was unlikely to show significant reduction in prostate cancer incidence [18]. RAS used a post-randomization design, due to rapid enrollment in SELECT relative to the start of RAS; thus, we did not measure pre-randomization lung function, but we captured the rate of decline over the intervention period through repeated measurements of lung function. This design assumed that the intervention effect is reached early in the study, and is stable over time. To test the hypothesis that current smokers benefit more from intervention, RAS enrolled men from the 16 SELECT sites with the greatest number of current cigarette smokers (Additional file 1: Table S1). Based on the predicted effect of intervention on annual FEV1 decline, assuming a 7–10 year follow-up, the target sample size in RAS was 3000 men.

Participants

Eligibility requirements for registration to RAS included SELECT registration at one of the 16 SELECT sites with a high proportion of smokers and adherence to supplements (either active or placebo) at the time of RAS registration. Each SELECT site invited all eligible current smokers to RAS and, depending on the number of participants at the site, either a random sample or all eligible former and never smokers. Ultimately, men were registered at their first (5%), second (17%), third (38%) or fourth (40%) annual SELECT visit. The RAS was approved by local IRBs at each of the 16 study sites, and by the Cornell University IRB.

End point assessment

The primary endpoint was annual decline in FEV1; the secondary endpoint was annual decline in FEF25–75. Spirometry was assessed at three out of four annual visits spanning three years. Due to early termination of SELECT, not all RAS participants completed all scheduled pulmonary function tests (PFTs); the endpoint (the third and final PFT) was available for 57% of participants. We assume this is an unbiased sample given the timing of supplementation withdrawal relative to the timing of a participant’s annual visit is expected to be random and thus equal across arms.

Pulmonary function testing followed American Thoracic Society guidelines [19] and used the EasyOne handheld, flow-sensing spirometer, which has excellent validity and reliability, and significantly simpler field implementation in comparison to desktop devices [20]. Only PFTs meeting criteria for acceptable start and end of test and for reliability were included in analyses; further details are provided in Additional file 1: Table S2).

Statistical analysis

There were three pre-specified main effect comparisons between each active treatment arm and placebo, with a P-value threshold of 0.018 to account for the three tests with a common placebo group. All analyses were intent-to-treat, and effects were estimated using a linear mixed-effects regression model incorporating the repeated measurements of pulmonary function (either FEV1 or FEF25–75) as the outcome. The model included random intercept and slope, and the following fixed effects: time (time between each PFT and baseline), treatment arm and its interaction with time (treatment-by-time), age, height, race and smoking. The treatment-by-time coefficient estimated the effect of treatment on annual rate of decline in lung function. All RAS men with ≥1 (n = 2920) PFT were included in the model and contributed to estimates of effects of age, height, race and smoking, but only men with ≥2 PFTs (n = 1677) were informative for the estimate of the time-by-treatment coefficient. Missing data were assumed to be missing at random, given very low drop-out rates. To test whether treatment effects differed by smoking status, models were extended to include the treatment-by-time-by-smoking interaction terms, and the significance threshold for the interaction effects was a nominal P-value of 0.05.

Results

Participants

RAS enrolled 2,920 men (Figure 1) at 16 SELECT sites between 7/2/2004 and 4/30/2007. RAS eligibility and enrollment were blind to intervention arm, and indeed the number of participants across the four arms was balanced (Figure 1). RAS experienced minimal attrition, with only 2-3% of men in each intervention group refusing further RAS and/or SELECT follow-up at some time point after registration. All participants studied herein had at least one acceptable PFT, and 56 to 60% of participants in each arm had repeated PFTs, confirming that repeated measurements for the endpoint assessment were similar by arm. The mean number of PFTs per participant was 2.3 (SD 1.1; median 3) with a mean of 36.1 months (SD 4.2) between first and last PFT (median 35.8 months; range 24 to 52). Spirometry quality control scores ranged from 3.2 to 3.7 out of maximum score of 4 (Additional file 1: Table S2).

Figure 1
figure 1

Flow of participants included in FEV 1 analysis by intervention group. 1Participants lost to follow-up defined as deaths and participants not in contact for 2 years prior to the end of supplemented time period (March 1, 2009 was used in analyses to proxy the end of supplement). 2Refusals defined as withdrawals from SELECT and/or RAS.

Full size image

RAS participants had similar distributions of age, race/ethnicity, education, smoking history and height across intervention arms (Table 1), confirming that the post-randomization design yielded four groups balanced on characteristics. The participants with one PFT were similar on all baseline characteristics to participants with repeated PFTs (data not shown); this is consistent with our expectation because the lack of repeated PFT data was a function of the date the intervention was withdrawn, and was not driven by participant choice or participant characteristics. Thus, we expect the estimate of the effect of treatment on lung function decline to be unbiased. Among participants who completed the final PFT, the mean time from SELECT registration to a participant’s last PFT was 60.4 months (SD 10.8; median 59.8), thus results reflect intervention effects of about 5 years duration.

Table 1 Baseline characteristics of study participants †
Full size table

Adherence among RAS participants (Table 2), determined using pill count, compares well to all SELECT participants [18]. Across the four arms, 87 to 92% of RAS men were adherent to the selenium supplement (or matching placebo) in year 1, and 80 to 84% were adherent in year 5. Similarly, 87 to 92% of RAS men were adherent to the vitamin E supplement (or matching placebo) in year 1, and 79 to 81% were adherent in year 5. Across all arms, for the full study period, self-supplementation with non-study vitamin E and selenium (drop-in rate, assessed by self-reported use of either supplement) was reported by ≤2.3% and ≤1.2% of participants, respectively.

Table 2 Adherence* to study supplements by pill counts
Full size table

Rate of decline in pulmonary function

Overall, the distribution of rate of decline in FEV1 was consistent with expectations of decline, and the mean annual change in FEV1 was −37.5 mL (SD 12.5; Additional file 1: Figure S1). Compared to never smokers, FEV1 was 363 mL lower in current smokers and annual decline in FEV1 was 6.9 mL/y steeper. In unadjusted analyses of raw data, compared with the placebo group (Table 3), participants randomized to intervention experienced an attenuation of between 3 and 6 mL/y in rate of change in FEV1, but there were no statistically significant differences between arms. Similarly, for rate of decline in FEF25–75 (mL/second/y), there were no statistically significant differences between arms (Table 4).

Table 3 Model-based estimated mean annual decline in FEV 1 (mL/year) by treatment group in the full sample, and stratified by cigarette smoking status (Additional file 1 : Tables S3 and S4 for model coefficients)
Full size table
Table 4 Model-based estimated annual decline in FEF 25–75 (mL/second/year) by treatment group in the full sample, and stratified by cigarette smoking status (Additional file 1 : Tables S3 and S4 for model coefficients)
Full size table

In linear mixed-effects regression models adjusted for covariates, the main effect of intervention on rate of decline in lung function was similar to the estimates based on raw data, considering markers of both volume (FEV1; Table 3) and flow (FEF25–75; Table 4). Thus, in intervention groups the annual rate of decline in FEV1 was lower by between 1 and 6 mL/y versus placebo, but effects were not statistically significant (Table 3). Similarly, there were no statistically significant main effects of any intervention on the annual rate of decline in FEF25–75 (Table 4). Sensitivity analyses that considered study site or the registration contact at which men started RAS (reflecting the length of time on study intervention) showed similar results.

Effect modification by cigarette smoking

The hypothesis that smoking modifies the effect of supplementation was pre-specified, and models were extended to estimate intervention effects within categories of cigarette smoking; categories included current, former (quit prior to trial), and never (lifetime never smoker). In the placebo arm, expected differences in rate of decline in FEV1 were confirmed such that the annual decline in current smokers was 11 to 16 mL/y steeper compared to former and never smokers (Table 3). Similarly, the annual rate of decline in FEF25–75 in current smokers on placebo was more than two-fold that in never smokers (P = 0.0189; Table 4).

There was no evidence that smoking modified the effect of intervention on rate of decline in FEV1, and all P values exceeded the threshold of 0.05. However, for the FEF25–75 outcome, compared to placebo the annual rate of decline in FEF25–75 was attenuated in current cigarette smokers in the selenium arm (P = 0.0219) and in the combined arm (P = 0.0236). Further models testing any selenium (selenium alone and selenium + vitamin E, combined) vs. placebo showed that FEF25–75 rate of decline was decreased by more than half in current smokers on any selenium compared to current smokers in the placebo group (P = 0.0095).

Discussion

This is the first randomized trial of selenium and/or vitamin E intervention that studies the rate of decline in pulmonary function as the endpoint. This study is important because it contributes new information about whether interventions that presumably affect the antioxidant/oxidant balance in lung tissue can ameliorate or attenuate a functional outcome reflecting lung health. Neither supplementation with selenium nor vitamin E had statistically significant main effects on rate of decline in FEV1 or FEF25–75. Following our a priori hypothesis that effects are stronger in and/or limited to current cigarette smokers, there was evidence for a differential effect of selenium in current smokers for the flow-related endpoint such that smokers supplemented with any selenium, either alone or in combination with vitamin E, had an attenuated rate of decline in FEF25–75.

This randomized trial evidence for an effect of selenium on annual rate of decline in lung function in smokers is consistent with prior cross-sectional studies that reported strong positive associations of serum selenium with lung function [14]. An analysis of baseline bloods collected on a subset of SELECT participants [18] found that men were rarely low on serum selenium, where low selenium was defined as ≤ 121.6 ng/mL consistent with prior studies of cancer outcomes [21]. While this suggests the potential-to-benefit from selenium intervention in the overall study may be low, the potential to benefit in smokers is likely to be greater given prior evidence that selenium concentrations are lower in smokers [22], and, indeed, this is supported by our findings.

The pattern of the RAS findings, including the effect of selenium on flow (FEF25–75) but not volume parameters and the magnitude of the effect sizes, are similar to the effects of air pollution on lung function reported in the SAPALDIA study. Based on 11 years of follow-up, SAPALDIA reported mean annual rate of decline in FEV1 and FEF25–75 of 35 mL/y (SD 30) and 71 mL/second/y (SD 65), respectively [23]. In the RAS placebo arm, average annual rates of decline were very similar, although the RAS estimates are more variable given the shorter duration of follow-up and more closely spaced PFTs. SAPALDIA reported that reductions in particulate matter ≤ 10 microns in diameter (PM10) were associated with the rate of change in both FEV1 and FEF25–75, but the strength of the association and the level of statistical significance were greater for the FEF25–75 outcome [23], similar to the findings reported herein for supplement effects in RAS. In current smokers the selenium intervention effect size for annual decline in FEV1 was similar to the effect size for reducing PM10 exposure by 10 μ/m3 in SAPALDIA (attenuated FEV1 decline by 4 mL/y). Greater effect sizes were seen for FEF25–75 in both studies: in SAPALDIA, reducing PM10 attenuated FEF25–75 decline by 11 mL/second/y, in RAS selenium supplementation attenuated FEF25–75 decline by 59 mL/second/y. While FEV1 is less variable than FEF25–75 in cross-sectional studies [19], longitudinal declines in both endpoints are of interest and FEF25–75 findings may be salient in smokers given that changes in flow rates may signal early changes in small airways function [24]. Although such changes may not be predictive at the individual level, they may be informative in the comparison of treatment groups. In addition, a longitudinal endpoint, which leverages repeated measurements per participant, and uses all available spirometry data on each participant (an approach that is consistent with two prior studies of longitudinal change [5,25]), is less affected by variability in comparison to cross-sectional studies.

Although the vitamin E effect sizes were clinically meaningful and consistent in effect direction (attenuated rate of decline) across two pulmonary function measures, with stronger effects of intervention in current smokers, the effects did not meet pre-set criteria for statistical significance. These findings reflect either a true lack of effect of vitamin E on rate of decline in lung function, the possibility that baseline vitamin E levels were high (or at minimum, not deficient) and thus there was limited potential to benefit from supplementation, or the possibility that attenuation in decline might occur only with a longer period of supplementation. While several past observational studies reported that associations of vitamin E were limited to smokers [26], such studies are more likely to be affected by confounding than the randomized trial findings reported herein.

The Respiratory Ancillary Study (RAS) to SELECT used a post-randomization design, and participants were registered to the RAS after active supplementation began. The primary endpoint was rate of decline in lung function; absolute differences in lung function due to the supplements over a fixed period of time were not calculated because first measurements were obtained after the participants started taking their supplements. The design assumes that effects are achieved quickly and are stable over the supplemented period, which is reasonable given the hypothesis of support for antioxidant function provided by the supplements.

This study measured pre-bronchodilator spirometry, but the lack of post-bronchodilator spirometry is not a serious weakness given the primary outcome is rate of decline, which relies on within-person repeated measurements. A recent study shows similar associations of pre- and post-bronchodilator spirometry with mortality [3], which is welcome news given the added participant burden of conducting post-bronchodilator spirometry in healthy population studies.

The strengths of RAS include the enrichment of the study sample with current cigarette smokers, which was part of the a priori intent to test effect modification, and the inclusion of a diverse sample (24% African Americans), which supports inference to broader population groups. An additional strength was the extensive infrastructure provided by SELECT, which allowed RAS to be conducted with efficiencies of cost and effort. SELECT infrastructure included an online data collection tool, which allowed incorporation of web-based uploading of spirometry data on a weekly basis, and bi-annual meetings of study personnel, which allowed for optimal training and refresher courses on spirometry methodology.

A few limitations are worth noting. In the post-randomization design, we cannot directly estimate whether the intervention increased FEV1 early in the supplementation period in smokers on the active study supplements. This question is important given that prior studies among individuals with COPD show that some clinical treatments increase FEV1, but have no effect on FEV1 rate of decline [27], and in light of the Lung Health Study, which showed that smoking cessation led to a small but significant initial rebound in FEV1, followed by an attenuation in the rate of decline [28]. Our study was conducted in male participants in the SELECT prostate cancer prevention trial, thus whether findings apply to women requires further study. In addition, the design of SELECT did not vary dose and/or formulation, and did not consider whether genetic variation might influence nutrient requirements for optimal health. Finally, the premature termination of supplements meant that final pulmonary function test on some participants was collected well after supplements had been discontinued, and thus analyses were based on fewer participants than originally planned.

Conclusions

While smoking cessation is the key public health intervention to prevent smoking-related health effects, about 20% of the population continues to smoke [1]. This study investigated the role of nutritional supplementation in lung function decline to identify possible intervention strategies to mitigate lung effects in continuing smokers. This randomized controlled trial found statistically significant protective effects of selenium, specifically 200 μg/d L-selenomethionine, on rate of decline in FEF25–75 in current cigarette smokers. Supplementation with selenium attenuated the annual rate of decline in FEF25–75 in current cigarette smokers, but neither vitamin E nor selenium had effects on rate of decline in FEV1. Further studies are needed to understand whether intervention effects are modified by baseline selenium nutriture and/or selenium-related genetic variation [29].