The ATBC Study was a randomized, double-blind, placebo-controlled, primary prevention trial conducted to determine the effects of supplementation with α-tocopherol and β-carotene on cancer incidence. In all, 29,133 white male smokers from southwestern Finland were recruited between 1985 and 1988. Enrollment criteria included age between 50 and 69 years and smoking of at least 5 cigarettes per day. Men were ineligible if they had previously had cancer, had another serious illness at enrollment, or reported current daily use of supplements containing vitamin E (>20 mg), vitamin A (>20,000 IU), or β-carotene (>6 mg). Men who were enrolled in the trial were assigned to 1 of 4 groups on the basis of a 2 × 2 factorial design: 1) α-tocopherol (dl-α-tocopherol acetate, 50 mg/day), 2) β-carotene (20 mg/day), 3) both supplements, or 4) placebo. Trial participants took the capsules for 5–8 years (median, 6.1 years), until death, or until the trial ended on April 30, 1993.
At enrollment, participants completed questionnaires about general risk factors, smoking, and medical history, as well as a validated food frequency questionnaire. Participants underwent a physical examination at baseline, during which registered nurses measured their height and weight and collected an overnight fasting blood sample. Fasting blood samples were collected again 3 years into the study. Although the trial has ended, follow-up is ongoing through the Finnish Cancer Registry and the Register of Causes of Death. As of April 30, 2006, a total of 2,041 incident prostate cancer cases occurred during 417,532 person-years of follow-up. Men were excluded from this analysis if they had missing or invalid (i.e., below the limit of detection) information on baseline serum retinol concentration (n = 29), leaving 2,041 cases among 29,104 men and 417,220 person-years for baseline analyses and 1,732 cases among 22,843 men and 349,353 person-years for analyses using the 3-year follow-up measurement.
The ATBC Study was approved by institutional review boards at both the US National Cancer Institute and the Finnish National Public Health Institute, and written informed consent was obtained from all participants.
Exposure and outcome assessment
Prostate cancer cases were identified through linkage with the Finnish Cancer Registry, which provides nearly 100% complete incident cancer ascertainment in Finland (48
). Medical records for the cases diagnosed before September 2001 were reviewed by 1 or 2 study oncologists to confirm diagnosis and staging; when available, pathologic specimens were reviewed by a pathologist. For cases diagnosed after September 2001, only the information from the Finnish Cancer Registry is available. Cases were defined as “aggressive” if they were TNM Classification of Malignant Tumors stage III or IV, American Joint Committee on Cancer stage 3 or higher, or Gleason sum 8 or higher. Stage or Gleason sum information was available for 63% of the cases.
As part of the intervention protocol to assess the influence of the supplemental β-carotene, serum retinol concentration was measured for all trial participants at enrollment and again after 3 years of follow-up. Assays were conducted over a several-year period in one dedicated laboratory at the National Public Health Institute in Helsinki, Finland, that was certified through a National Institute of Standards and Technology quality-control testing program. Retinol was measured in fasting serum samples by using isocratic high-performance liquid chromatography (49
). All samples were protected from light and stored at −70°C until they were assayed within 2–4 years of serum collection. The study samples from the cohort members were sequentially batched in the order in which they were collected and transported to the institute's central laboratory from the study centers. Quality-control samples were embedded in each batch, and the laboratory was blinded to the samples’ quality-control status. The quality-control samples included internal standards, individual quality-control duplicate aliquots, and external reference samples provided by the National Institute of Standards and Technology as part of the quality-control program for β-carotene, retinol, and other micronutrients. The between-run coefficient of variation was 2.4%, and the overall coefficient of variation was 2.2%. The intraclass coefficient for correlation between the baseline and 3-year retinol measurements was 80.1%. The limit of detection was 20 μg/L.
Cox proportional hazards modeling was used to estimate the association between quintiles of baseline serum retinol concentration (quintile 1, <483 μg/L; quintile 2, 483–546 μg/L; quintile 3, 547–606 μg/L; quintile 4, 607–684 μg/L; and quintile 5, ≥685 μg/L) and risk of total prostate cancer (n = 2,041) and aggressive disease (n = 461). Men for whom information on disease aggressiveness was not available were excluded from analyses of aggressive disease. All models were adjusted for age at baseline as a continuous variable. The following factors, which are hypothesized or known to be associated with either prostate cancer or retinol, were assessed as potentially confounding variables: α-tocopherol treatment group, β-carotene treatment group, total cholesterol level, serum α-tocopherol, serum β-carotene, number of cigarettes smoked per day, years spent smoking, family history of prostate cancer, physical activity, body mass index, height, weight, educational level, marital status, urban residence, total energy intake, total fat intake, fruit intake, vegetable intake, red meat intake, dietary retinol intake, dietary vitamin D intake, dietary calcium intake, supplemental vitamin A intake, supplemental vitamin D intake, and supplemental calcium intake. Each variable was entered into the age-adjusted model to evaluate whether the point estimates for retinol categories changed by at least 10%, and none did so. Thus, our final model was adjusted for age only. We also present our primary findings adjusted for all potentially confounding factors to show that addition of those factors did not appreciably alter the results. We confirmed the proportional hazards assumption by including in the model term for interaction between serum retinol and follow-up time and testing that term by using the Wald test (P-interaction = 0.45).
To examine the associations between prostate cancer risk and low and high extremes of serum retinol levels, we also evaluated retinol deciles. The inferences were similar using these cutpoints, so we present our results for baseline serum retinol in quintiles. We also examined the association between risk and quintiles of serum retinol on the basis of concentrations measured in blood from the 3-year follow-up visit (i.e., <494, 494–559, 560–622, 623–704, and ≥705 μg/L), as well as with quintiles of the average of the retinol concentration measured at baseline and at the follow-up visit (i.e., <498, 498–558, 559–614, 615–688, and ≥689 μg/L). We obtained identical results when we categorized the 3-year serum retinol measurement using the baseline cutpoints; thus, we retained the distinct 3-year cutpoints described above for analysis. We further examined prostate cancer risk in relation to the difference in serum retinol between the baseline and 3-year measurement, both as categories of the 3-year concentration minus baseline concentration (i.e., <−75, −75 to −11, −10 to 9, 10 to 74, and ≥75 μg/L) and through joint classification based on the baseline and 3-year measurements.
Exploratory subgroup analyses were conducted by stratifying by follow-up time (<10 years vs. ≥10 years), age at prostate cancer diagnosis (<60 years vs. ≥60 years), cigarettes smoked per day (<10, 10–19, 20−39, or ≥40), years spent smoking (<36 vs. ≥36), pack-years of smoking (<35 vs. ≥35), family history of prostate cancer, intervention arms (α-tocopherol vs. β-carotene), less than median vs. greater than or equal to median of body mass index, serum α-tocopherol, serum β-carotene, serum total cholesterol, use of supplemental vitamin A or vitamin D, and dietary intake of retinol, vitamin E, vitamin A, vitamin D, and alcohol. Statistical interaction was assessed using the likelihood ratio test. We considered α = 0.05 to be the threshold for statistical significance in all analyses.