Association of low plasma antioxidant levels with all-cause mortality and coronary events in healthy middle-aged men from France and Northern Ireland in the PRIME study

The PRIME Study is a multi-centre, prospective cohort composed of men aged 50–59 years when recruited from the general population of Belfast in NI and Lille, Strasbourg and Toulouse in France, between 1991 and 1993 [20, 21]. A detailed medical, lifestyle and food frequency questionnaire (FFQ) was completed at baseline examination which included information on lifetime smoking, physical activity [22], alcohol consumption [23] and diet [24]. A cardiovascular screening examination was also conducted at baseline, which included a detailed history of previous CVD. Follow-up included annual questionnaires for 10 years and possible cardiovascular events were confirmed by a medical committee [21, 25]. Of the 10,600 men recruited, 317 (3.0%) were lost to follow-up after 10 years, 215 (2%) withdrew their consent for participation within the study, and 653 (6.2%) died during this time. Cardiovascular events included validated myocardial infarction and/or stroke (fatal and non-fatal). Death certificates were obtained for all men who died and causes of death were categorised using the International Classification of Diseases (ICD) Ninth Revision.

Fasting venous blood samples were drawn at baseline and centrifuged within 4 h of collection. Plasma samples were aliquoted for long-term storage at − 80 °C. Plasma concentrations of retinol, α-tocopherol, γ-tocopherol, and six carotenoids (α-carotene, β-carotene, β-cryptoxanthin, lutein, lycopene, zeaxanthin) were determined by high performance liquid chromatography (HPLC) with diode array detection [26]. Results were analysed using Chromquest 4.2 software (Thermo Fisher Scientific, Massachusetts, USA). Standards used in the assay were supplied by either Sigma (Sigma-Aldrich, Poole, Dorset, UK) or Chemos (Chemos GmbH, Regensburg, Germany). Vitamin C was measured in plasma which had been stabilised on collection on a BMG FLUOstar Optima plate reader adapted from the method by Vuilleumier and Keck [27]. These assays were validated against the National Institute of Standards and Technology (NIST) Standard Reference Material 968d for fat-soluble vitamins, carotenoids and cholesterol and NIST Standard Reference Material 970 for ascorbic acid in human serum, which were externally quality assured three times per year using samples supplied by the French Society for Vitamins and Biofactors. In-house quality control samples were also included in every run. The inter- and intra-assay coefficients of variation (CV) for vitamin A, E and carotenoid assays were < 15% and < 7.5% for vitamin C. A mean standardised antioxidant score (SAS) was estimated from the mean standardised Z scores for the ten logarithmically transformed antioxidants.

Body mass index (BMI) was calculated as the body weight/height² (kg/m²). Lifetime smoking was expressed in five categories: never smoked, smoked other than cigarettes, smoked < 15 cigarette pack years, smoked ≥ 15 but < 30 cigarette pack years and smoked ≥ 30 cigarette pack years. Alcohol intake was also expressed as five categories of consumption; none, 1–128, 129–265, 266–461 and ≥ 462 mL/week. Physical activity (in metabolic equivalent of task (METs) scores/week) was positively skewed and was therefore summarised using median and interquartile range. Diabetes was defined by self-report. Socioeconomic status was categorised into high, middle and low using a composite measure of material conditions based on the type of living accommodation (rented or owned, mortgaged), number of cars/vans/motorcycles in the household and the number of baths and/or showers and toilets in the home [22].

The daily fruit, vegetable and juice (FVJ) score included all fruit and vegetable variables from the PRIME FFQ. This involved eight categories: ‘salad’, ‘carrot’, ‘tomato’, ‘natural’ and ‘baked vegetables’, ‘citrus fruits’, ‘other fruits’ and a ‘juice’ variable. Potatoes were not included as they were included under the ‘Bread, cereals and potatoes’ food group. Frequencies of each variable were transformed into daily consumption values (the ‘juice’ variable was limited to a maximum of one daily portion in-line with UK dietary guidelines) and these new variables were summed for each participant to give total daily FVJ consumption. This variable was positively skewed and was therefore summarised using median and interquartile range.

Ethical approval for baseline examination and follow-up was obtained at each study centre; all participants provided informed consent and the study was compliant with the Declaration of Helsinki guidelines.

Data for antioxidant levels were not normally distributed and were therefore logarithmically transformed before analysis. Results were summarised using the geometric mean and inter-quartile range. Correlation coefficients between fruit and vegetable consumption and antioxidant measures were assessed. Antioxidant levels were compared between men with and without a CVD outcome and all-cause mortality using independent samples t-tests. Lifestyle behaviours were compared across socioeconomic variables either by chi-square test for categorical variables or by Kruskal–Wallis one way ANOVA for continuous variables.

Cox proportional hazards models were used for analysis of antioxidant status for cardiovascular outcomes (first cardiovascular event) and separately for all-cause mortality, during a 10-year follow-up period. In the analysis of cardiovascular events, individuals were followed from baseline assessment to the first CVD event or death or end of follow-up. Each antioxidant was split into fourths using quartiles to categorise participants for analysis. The lowest quarter of the distribution of each antioxidant was used as the reference category for the calculation of hazard ratios. Firstly, models were run with minimal adjustment for potential confounders which included age and country, followed by a multivariate analysis that also included systolic blood pressure (SBP), diabetes, BMI, total cholesterol, high density lipoprotein (HDL) cholesterol, height, social class (low, medium and high), alcohol consumption (none, 1–128, 129–265, 266–461 and ≥ 462 mL/week), smoking status (never smoked, smoked other than cigarettes, smoked < 15 cigarette pack years, smoked ≥ 15 but < 30 cigarette pack years and smoked ≥ 30 cigarette pack years), and total physical activity (total physical activity was square root transformed, before inclusion in the model). Hazard ratios were compared across quarters of antioxidant distribution by a test for trend, and a likelihood ratio test was used to check for non-linearity by comparing the trend model with a model specifying a different hazard for each quarter. Additional sensitivity analyses considered the exclusion of variables that could potentially contribute to the causal pathway of CVD or all-cause mortality. Tests of the hazard proportionality assumption in the Cox model were performed using Schoenfeld residuals. All tests were conducted at the p < 0.05 significance level using SPSS version 25 (IBM Corp, Armonk, NY, USA) and Stata release 14 (StatCorp, College Station, TX, USA). To check for possible effect modification, interactions between country and antioxidants were added to the Cox proportional hazards models. A more stringent level of significance was set at p ≤ 0.005 to allow for multiple comparisons of potential interactions in the absence of any prior hypotheses.

Of the 10,600 PRIME study participants, 891 had a previous history of CVD at study entry (355 from NI and 536 from France) and were excluded from the analysis. The remaining 9709 men were followed up for 10 years (mean = 3506.81 days) which equated to 93,281 person years. Follow-up was incomplete for 470 men (the number who were not known to have died and were not followed for the full 10 years), so the lost to follow-up rate was 4.8%. Of 9709 men, there were 538 deaths from any cause and 440 fatal and non-fatal cardiovascular events after 10 years of follow-up. Population characteristics for those 9709 men are summarised according to CVD outcome and all-cause mortality (Table 1). Individuals who experienced a CVD outcome were on average older than men who had not (55.5 years versus 54.8 years), were more likely to be from Northern Ireland (38.4% versus 24.0%), to have diabetes (7.3% versus 3.1%) and higher mean SBP (141 mmHg versus 133 mmHg). In addition, those who experienced a CVD outcome had higher mean BMI (27.3 kg/m² versus 26.5 kg/m²), higher mean total cholesterol (2.30 g/L versus 2.21 g/L) and median triglycerides (1.44 g/L versus 1.21 g/L) but lower mean HDL cholesterol (0.45 g/L versus 0.49 g/L), and height (172.1 cm versus 172.8 cm) and be less physically active (median METs 76.0 versus 83.2). Furthermore, they were more likely to be from the lowest social class (31.0% versus 23.8%), consume fewer daily FVJ portions (median 2.21 versus 2.43), be heavy smokers (31.3% versus 21.5%) and refrain from alcohol consumption (25.9% versus 16.3%).

Antioxidant status was available for 6961 participants and antioxidant quartile boundaries are shown in Supplementary Table 1. As expected, there was a high correlation between fruit and vegetable consumption and antioxidant measures (Supplementary Table 2). Over a third of participants (37.6%) had plasma vitamin C concentrations considered to be optimum for health (> 50 μmol/L), over three quarters (77.6%) had plasma vitamin C concentrations above the marginal deficiency level (> 22.7 μmol/L), and 9.2% of participants had concentrations considered extremely deficient (< 11 μmol/L) [28‐30] (data not shown). For α-tocopherol, severe overt deficiency has been reported at 11.6 μmol/L [28] and levels > 27.5 μmol/L have been considered optimum to reduce the risk associated with ischaemic heart disease (IHD) [29]. More than two fifths of participants (40.5%) had plasma alpha tocopherol concentrations recommended for optimum health (> 27.5 μmol/L). Almost all participants (99.7%) had alpha tocopherol concentrations in excess of the overt deficiency value of 11.6 μmol/L (data not shown).

Optimum recommended concentrations to reduce IHD risk have previously been proposed for retinol (≥ 2.2 μmol/L) and β-carotene (> 0.4 μmol/L) [28‐30]. Over half of the participants (54.5%) had plasma retinol concentrations ≥ 2.2 μmol/L and over one third (36.2%) had a β-carotene concentration > 0.4 μmol/L.

With the exception of both α- and γ-tocopherol, lower levels of all antioxidants were detected in men who experienced a CVD outcome, although this was not statistically significant for zeaxanthin and retinol (Table 2). Hazard ratios in a model minimally adjusted for age and country, showed a significant reduction in risk across quarters for alpha and beta carotene, vitamin C, lycopene, α-tocopherol and mean SAS (p < 0.05; Table 3). Following further adjustment for SBP, diabetes, BMI, cholesterol, HDL cholesterol, triglycerides, height, total physical activity, alcohol consumption, smoking and socioeconomic status, significant reductions across quarters of cardiovascular event risk were only observed for retinol, with α- and β-carotene falling just short of the significance threshold (p < 0.05; Table 4, Fig. 1). The mean SAS divided in quarters showed no significant association with CVD outcomes in the fully adjusted model (Table 4). Associations with CVD outcomes were little altered by the omission of total cholesterol, HDL cholesterol, triglycerides, and systolic blood pressure with the exception of the attenuated effect on retinol that no longer remained significant (Supplementary Tables 3a and 3b).

There were lower levels of all antioxidants associated with all-cause mortality, and only that of γ-tocopherol was not significant (Table 2). Similarly fully adjusted hazard ratios showed statistically significant reductions for all antioxidants in association with all-cause mortality, with the exception of γ-tocopherol (Fig. 1; Tables 3 and 4). Daily FVJ portions were initially significant when fitted with other confounders within the adjusted model but was no longer significant when the mean Z score was included. The mean SAS showed significantly reduced risk across quarters for all-cause mortality in minimally and fully adjusted models (Tables 3 and 4; p < 0.001). Lack of linearity was observed for the associations between mortality and α-tocopherol, β-cryptoxanthin, zeaxanthin and SAS. Similar adjusted Cox proportional HR by quartile for smoking-associated cancer and all neoplasm-related mortality according to ICD Ninth Revision codes are presented in Supplementary Table 4.

No significant deviations from the tests of the hazard proportionality assumption in the Cox model using Schoenfeld residuals were detected, for either CVD events or all-cause mortality. There was no evidence of significant interaction between country and antioxidants in any of the models for CVD events or all-cause mortality following correction for multiple testing.

The major strengths of our study were the large number of well-characterised participants with prospective follow-up over 10 years and clinically validated cardiovascular events. Many observational studies are restricted by small sample size and/or short follow-up duration, limiting the reliability of the associations reported. Associations were based only on those men free from cardiovascular events at study entry. In addition, the micronutrient plasma measurements were quantified using standardised and validated laboratory methods, and so provided an objective determinant, independent of dietary intake estimates recalled over a period of time. Assessment of antioxidant concentrations as continuous variables, identified similar associations to the evaluation of antioxidants by quartiles but the latter is more amenable to the identification of non-linear relationships, where they exist. Alpha-carotene and lutein were the only antioxidants that demonstrated a dose–response relationship with all-cause mortality, with higher concentrations associated with lower mortality. Retinol was the only antioxidant significantly associated with CVD events demonstrating a linear relationship with reduced risk with higher concentrations. A mean SAS offers an opportunity to assess the cumulative effects of antioxidant concentrations on CVD outcomes and all-cause mortality. Although a combined total antioxidant score was more strongly associated with all-cause mortality than individual antioxidants, the effects observed were not substantially different than for individual antioxidants alone.

Our study has several limitations. We tested multiple hypotheses and, as such, our results must be interpreted with caution as we made no correction for multiple testing in the analyses presented in Tables 2, 3 and 4, although several of the reported associations would remain significant even if tested at the p < 0.001 level. Despite controlling for the most important covariates, additional residual confounding may exist. Furthermore, additional dietary factors not considered within the analyses may be associated with antioxidant levels. Moreover, the sensitivity of nutritional biomarkers is somewhat limited by the influences of individual genetic variation associated with dietary intake and nutrient metabolism, a complex multistage process that includes ingestion, digestion, absorption and transportation in the blood [48, 49]. The categorical analyses by quartiles depends on certain parametric modelling assumptions (proportional hazards, linearity, etc.), although analyses of antioxidant levels as continuous variables demonstrated similar levels of significance (data not shown). In addition, while we evaluated daily FVJ consumption, the use of FFQ data may lead to an overestimation of the intakes of some healthy foods, which may impact the interpretation of the associations observed, although this had little effect on the associated hazard ratio estimates. Furthermore, we used a single measure of plasma antioxidant levels which may not necessarily reflect long-term nutritional status against CVD and all-cause mortality outcomes. Previous analysis of The Third National Health and Nutrition Examination Survey (NHANES III) investigated 4225 deaths in 16,008 participants over a median 14.2 years follow-up period and reported that a single serum measurement likely represents a reliable estimate of long-term antioxidant status [15]. Additionally, associations between micronutrient status and fruit and vegetable intake were strong, supporting the plasma concentration as an appropriate representative measure. Although we examined the independent and cumulative effects of ten antioxidants on cardiovascular events and all-cause mortality, potential associations with other nutrients such as omega-3-fatty acids, vitamin D, the B-vitamins and selenium were beyond the scope of this study. Finally, our study results apply to predominantly white European middle-aged men who were initially free from CVD at the point of recruitment.

In summary, we have shown that higher plasma concentrations of vitamin C, retinol, α-tocopherol and six carotenoids (α-carotene, β-carotene, β-cryptoxanthin, lutein, lycopene, zeaxanthin) were associated with a lower risk of all-cause mortality in middle-aged men. Furthermore, higher levels of retinol were associated with reduced risk of CVD events. Our findings are of major public health significance as the search for improved understanding of the modifiable risk factors associated with cardiovascular events and all-cause mortality continues. Identifying individuals at future increased risk of developing CVD and detecting adverse risk at the earliest stages, may be most effective in delaying disease progression and preventing premature death. Our findings suggest that, at a population level, small, achievable shifts in micronutrient status could have a considerable impact on mortality and CVD risk. However, there is still insufficient evidence to establish a causal relationship between low antioxidant levels and all-cause mortality. Given the potential implications of nutritional status in the aetiology of multiple chronic conditions, dietary intervention warrants further investigation and may prove beneficial in reducing premature mortality, especially given the largely null findings from randomised control trails of antioxidant supplementation. Insufficient consideration for individual participant baseline nutrition levels within study inclusion criteria, may represent a major concern for randomised trial design. Our findings suggest individuals deficient in specific micronutrients may derive greater benefit from dietary or supplement interventions than sufficient individuals.