The effect of β-carotene on the mortality of male smokers is modified by smoking and by vitamins C and E: evidence against a uniform effect of nutrient

Abstract

A previous analysis of the Alpha-Tocopherol Beta-Carotene (ATBC) Study on male smokers found that β-carotene supplementation increased the risk of pneumonia 4-fold in those who started smoking at the age of ≥21 years and smoked ≥21 cigarettes/d (a subgroup of 7 % of the study population). The present study hypothesised that β-carotene increases mortality in the same subgroup. The ATBC Study (1985-1993) recruited 29 133 Finnish male smokers (≥5 cigarettes/d) aged 50-69 years. Cox regression models were constructed to estimate the effect of β-carotene supplementation in subgroups. β-Carotene increased mortality (risk ratio 1·56; 95 % CI 1·06, 2·3) in those who started to smoke at ≥21 years and smoked ≥21 cigarettes/d. Within this subgroup, there was strong evidence of further heterogeneity. The effect of β-carotene supplementation was further modified by dietary vitamin C intake, fruit and vegetable intake (P = 0·0004), and by vitamin E supplementation (P = 0·011). Thus, harm from β-carotene was not uniform within the study population. Interactions between β-carotene and vitamins C and E were seen only within a subgroup of 7 % of the ATBC participants, and therefore should not be extrapolated to the general population. Heterogeneity of the β-carotene effect on mortality challenges the validity of previous meta-analyses that have pooled many diverse antioxidants for one single estimate of effect using the assumption that a single estimate equally applies to all antioxidants and all people. Trial registration: ClinicalTrials.gov NCT00342992.

Keywords: AT, all rac-α-tocopheryl acetate; ATBC, Alpha-Tocopherol Beta-Carotene; Antioxidants; BC, β-carotene; Cohort studies; Dietary supplements; Effect modifiers; Oxidative stress; Population characteristics; RR, risk ratio; Randomised controlled trials.

Fig. 1.

Mortality by β-carotene (BC) supplementation: participants who initiated smoking at ≥21 years and smoked ≥21 cigarettes/d (Alpha-Tocopherol Beta-Carotene Study 1985–1993). The Kaplan–Meier survival curves for the BC and placebo arms are shown. Each step indicates one death. (a) Mortality after randomisation. For the difference between the two curves, P = 0·024. (b) Mortality by follow-up age.

Fig. 2.

Effect of vitamin C intake, and fruit and vegetable intake on the mortality caused by β-carotene (BC) supplementation: participants who initiated smoking at ≥21 years and smoked ≥21 cigarettes/d in male smokers (Alpha-Tocopherol Beta-Carotene Study 1985–1993). ○, Deaths in the placebo arm; ●, deaths in the BC arm. Assuming that BC had no effect on mortality, we would expect a similar distribution of deaths in the placebo and BC arms. The cut-off limits used in the statistical analysis are shown: 90 mg/d vitamin C and 275 g/d fruit and vegetables. The subgroup labels A, B and C are as used in Table 3.

Fig. 3.

Mortality caused by β-carotene (BC) supplementation in subgroups A to C of Fig. 2 in male smokers (Alpha-Tocopherol Beta-Carotene Study 1985–1993). The Kaplan–Meier survival curves are shown for the BC and placebo arms. Each step indicates one death. (a) Subgroup A: vitamin C < 90 mg/d. For the difference between the two arms, P = 0·004. (b) Subgroup B: vitamin C ≥90 mg/d and fruit and vegetables <275 g/d, P = 0·049. (c) Subgroup C: vitamin C ≥90 mg/d and fruit and vegetable intake ≥275 g/d, P = 0·004.

Fig. 4.

Meta-analysis to examine heterogeneity in the effects of β-carotene (BC) on mortality within the Alpha-Tocopherol Beta-Carotene Study. This meta-analysis is restricted to the 14 564 no-vitamin E participants. Subgroups A, B and C are described in Fig. 2 and Table 3. The participants who smoked between 5 and 19 cigarettes (Cig)/d (median 12/d), are separated into their own subgroup. ‘The rest’ means all the remaining participants. The percentage after the study label indicates the proportion of all participants in the particular subgroup. The horizontal lines indicate the 95 % CI for the BC effect and the squares in the middle of the horizontal lines indicate the point estimates of the effect in the particular subgroup. The diamond shape indicates the pooled effect and its 95 % confidence interval. RR, risk ratio.