β-carotene supplementation for patients with low baseline levels and decreased risks of total and prostate carcinoma
The Physicians' Health Study was a randomized, double-blind, placebo-controlled trial using a 2 × 2 factorial design including supplementation with β-carotene (50 mg every other day) in the primary prevention of cancer among 22,071 U.S. male physicians ages 40–84 years at randomization. Before randomization, the authors collected baseline blood specimens to determine whether any benefit was greater among or confined to those with low baseline levels of β-carotene.
Baseline blood samples were collected from 14,916 participants. These samples were assayed, according to a nested case–control design, from 1439 men subsequently diagnosed with cancer over 12 years of follow-up (631 with prostate carcinoma) and 2204 controls matched by age and smoking habits.
Men in the lowest quartile for plasma β-carotene at baseline had a marginally significant (P = 0.07) increased risk of cancer compared with those in the highest quartile (relative risk [RR] = 1.30, 95% confidence interval [CI], 0.98–1.74). Men in the lowest quartile assigned at random to β-carotene supplementation had a possible but nonsignificant decrease in overall cancer risk (RR = 0.83, 95% CI, 0.63–1.09) compared with those assigned to placebo. This was primarily due to a significant reduction in the risk of prostate carcinoma (RR = 0.68, 95% CI, 0.46–0.99) in this group. After the first 2 years of follow-up were excluded, the results were virtually unchanged.
These prespecified subgroup analyses appeared to support the idea that β-carotene supplementation may reduce risk of prostate carcinoma among those with low baseline levels. Further follow-up of this population will help determine whether these findings are valid. [See editorial on pages 1629–31, this issue.] Cancer 1999;86:1783–92. © 1999 American Cancer Society.
Observational epidemiologic studies consistently have indicated that individuals with diets high in fruits and vegetables, particularly those rich in β-carotene, are at lower risk for cancer incidence, suggesting a role for β-carotene in the primary prevention of cancer.1, 2 Although these findings have been supported by plausible biologic mechanisms,3 case–control or cohort studies cannot distinguish the components of fruits and vegetables responsible for the observed benefit.4 Recent evidence from randomized trials have not supported a protective effect of β-carotene supplementation in cancer prevention. In the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study (ATBC)5 and the Beta-Carotene and Retinol Efficacy Trial (CARET),6 incidence of lung carcinoma was increased among those assigned to active β-carotene in populations of smokers or asbestos workers. In contrast, the Physicians' Health Study (PHS) found no significant benefit or harm of β-carotene supplementation with respect to total carcinoma or lung carcinoma incidence rates among U.S. male physicians.7 Although current smokers comprised 11% of PHS participants at baseline, there were no significant differences in cancer incidence or mortality rates with β-carotene supplementation by baseline smoking status.
Before randomization to the PHS, participants were asked to provide a baseline blood specimen that then was frozen and stored. The primary aim was to assess whether any effect of β-carotene supplementation was either larger or even limited to those with the lowest levels of baseline plasma β-carotene.8 Evidence supporting such effect modification was provided by data from the Chinese Cancer Prevention Trial,9 in which supplementation with a combination of β-carotene, vitamin E, and selenium led to a reduced risk of gastric carcinoma and total mortality in a poorly nourished population. Although data from both the ATBC10 and CARET11 studies suggest no such modification for lung carcinoma among those at high risk, the PHS is the only trial to examine this issue in a population comprised largely of nonsmokers, in which total cancer is influenced largely by prostate carcinoma. In this report, we describe the observed effects of β-carotene supplementation in the PHS on total and prostate carcinoma among those with low baseline plasma levels of β-carotene, using a prospective, nested, case–control design.
PATIENTS AND METHODS
The PHS was a randomized, double-blind, placebo-controlled trial with a 2 × 2 factorial design testing the effects of aspirin and β-carotene in the primary prevention of cardiovascular disease and cancer among 22,071 U.S. male physicians. The methods and main study results have been described previously.7 Participants were ages 40–84 years in 1982 and had no history of cancer (except nonmelanoma skin cancer), myocardial infarction, stroke, or transient cerebral ischemia; had no contraindications to aspirin use; and were not currently taking aspirin, other platelet-active medications, or supplements of vitamin A. Participants were asked to respond to a baseline questionnaire regarding medical history and health habits. This included a limited semiquantitative food frequency questionnaire containing food items high in dietary β-carotene. At baseline in 1982, 50% of participants were never, 39% were past, and 11% were current smokers.
Physicians were randomly assigned to 1 of 4 groups including active aspirin (325 mg on alternate days [Bufferin; Bristol-Myers, New York, NY]) plus β-carotene placebo, active β-carotene (50 mg on alternate days [Lurotin; BASF Corporation, Mount Olive, NJ]) plus aspirin placebo, both active agents, or both placebos. Written informed consent was obtained from all study participants, and the research protocol was reviewed and approved by the institutional review board at the Brigham and Women's Hospital in Boston, Massachusetts. The randomized aspirin component of the study was terminated early on January 25, 1988, primarily due to a statistically extreme (P < 0.001) 44% reduction in the incidence rate of first myocardial infarction among those patients assigned to active aspirin.12 The β-carotene component continued until its scheduled end in December 1995 after an average of 12 years of follow-up. At the end of 11 years of follow-up, the last year completed by all participants, 99.2% of participants still were providing morbidity information; follow-up for mortality was virtually 100% complete. At that time, 80% of participants still were taking the study pills in both the active and placebo β-carotene groups.
Prior to randomization, study participants were asked to provide baseline blood specimens for analysis of plasma β-carotene levels. Blood kits were mailed to all enrolled physicians with instructions to have their blood drawn into vacutainer tubes containing ethylenediamine tetracetic acid. Participants were asked to fractionate the blood by centrifugation and to return (by overnight, prepaid courier) the plasma in polypropylene cryopreservation vials. Each kit contained a coldpack to keep the specimens cool until they were received the following morning, when they were divided into aliquots and stored at -82 °C. During storage, precautions were taken so that no specimen was thawed or warmed substantially. Specimens were received from 14,916 of the randomized physicians (68%), with 70% of samples received in the fall, between September and November 1982. Participants who returned specimens were slightly older (age 53.4 years vs. 53.0 years; P = 0.006), weighed less (24.9 kg/m2 vs. 25.1 kg/m2; P < 0.001), exercised more (2.4 times/week vs. 2.2 times/week; P < 0.001), smoked less (10.1% current smokers vs. 13.1% current smokers; P < 0.001), and drank slightly more alcohol (3.6 drinks/week vs. 3.5 drinks/week; P = 0.04) than those who did not return specimens. They were similar with respect to β-carotene assignment and dietary β-carotene.
Selection of Cases and Controls
On report of a diagnosis of cancer, medical records were requested from hospitals and treating physicians and were reviewed by the study endpoints committee. Reported cases of cancer required confirmation by pathology report. As of October 24, 1995, 2426 cases of cancer had been confirmed. Of these, 1643 cases (68%) provided adequate baseline blood specimens. For each case, one or more control subjects were selected who had provided baseline plasma and who had no report of cancer up to the time of diagnosis of the case. Controls also were matched on the basis of smoking at baseline and age within 1 year, if possible. Due to a scarcity of controls by smoking status among the older age groups, age matching was relaxed to up to 5 years when necessary; analyses provided for additional control of age through statistical modeling. Matches could not be found for 92 subjects (6%) due to a lack of appropriate control subjects or missing smoking status. To control for laboratory variability of assays of β-carotene, only matched sets that were analyzed in the same laboratory batch were included in these analyses. This left a total of 1439 cancer cases and 2204 matched controls, including 631 cases of prostate carcinoma with 1377 matched controls. Additional controls for prostate carcinoma cases were available from previous analyses of other hypotheses13, 14; up to four controls per case were selected to increase statistical power. The majority of cases (69%) had 1 control, 10% had 2 controls, 21% had 3 controls, and 7 (< 1%) had 4 controls. Case subjects included in the analysis differed somewhat from those not included with respect to average age (59 years vs. 62 years, respectively; P < 0.001) and smoking status (14% vs. 18% current smokers, respectively; P = 0.01). These two groups were similar with respect to β-carotene assignment, body mass index (BMI), alcohol use, and exercise, but reported slightly different dietary consumption of β-carotene (6914 μg/day vs. 7386 μg/day, respectively; P = 0.03).
Frozen plasma samples were delivered to the MicroNutrient Analysis Laboratory in the Department of Nutrition at the Harvard School of Public Health. Each case and its matching control samples were assayed in random order within the same batch to minimize interassay variability, and aliquots from a pool of quality control plasma were inserted randomly. Laboratory personnel were unable to distinguish case, control, or quality control samples. Thawed samples were treated with ethanol to precipitate proteins, internal standards were added, and multiple hexane extractions were performed to remove lipid extractable analytes. The extracted samples then were dried and reconstituted with a 3:1:1 mixture of acetonitrile:ethanol:dioxane to a total volume of 250 μL. Each group of 20 unknown samples was combined with 1 blank sample, 1 internal quality control serum pool sample, 2 method blanks, and 2 internal standard blanks. The system is validated against serum samples from the National Institute of Standards and Technology two to three times annually. Measurement of β-carotene was achieved by high performance liquid chromatography (HPLC) using a Hitachi dual wavelength system (Hitachi Instruments, Danbury, CT) with detectors connected inseries and set to the appropriate wavelengths. Separation of injected samples was achieved on Beckman C18 Ultrasphere ODS reversed-phase columns (Beckman Coulter, Fullerton, CA) maintained at 28 °C. HPLC peaks were integrated using Hitachi software. The mean intraassay coefficient of variation for β-carotene based on 94 blinded quality control samples was 7.75%.
Medical Record Review
Cases of prostate carcinoma (by far the single most common site) were reviewed further to determine stage of disease. Study physicians, unaware of randomized assignment or baseline plasma β-carotene level, reviewed the medical records for each prostate carcinoma case to determine the tumor stage at diagnosis, the tumor grade, and the Gleason score. Stage was recorded according to the modified Whitmore-Jewett classification scheme.15 If multiple tissue samples were examined, the highest reported grade and Gleason score were recorded. Cases without pathologic staging were classified as indeterminate staging unless there was clinical evidence of distant metastases. Aggressive cases were defined as those diagnosed at Stage C or D (extraprostatic) plus those diagnosed at Stage A, Stage B, or indeterminate stage with either poor histologic differentiation or a Gleason score of ≥ 7. Among 574 prostate carcinoma cases reviewed, 266 (46%) were classified as having aggressive disease and 308 (54%) were classified as having nonaggressive disease.
Baseline characteristics of the selected cases and controls were compared in analyses stratified by matched set. Proportions were compared using the stratified Cochran–Mantel–Haenszel test. Means were estimated and compared with multivariate models allowing for the correlation of values within matched sets using PROC MIXED of SAS.16 The natural logarithms (ln) of baseline plasma β-carotene and of dietary β-carotene were used to normalize the distributions, and all statistical tests of mean levels for these variables are based on these ln transformations. Geometric means and an approximate standard error based on a Taylor series expansion are provided in the tables. Median levels were very similar to the geometric means. Predictors of ln baseline plasma β-carotene among the controls were assessed through multiple linear regression. Pearson correlations were used with the ln transformed values; Spearman correlation coefficients based on ranks were very similar.
Plasma β-carotene at baseline was divided into quartiles based on its distribution among controls. Mean levels of baseline characteristics by quartile among controls were estimated and tested using linear regression, controlling for age. Proportions of baseline characteristics across quartiles were compared using tests for trend stratified by 5-year age groups, and age-adjusted estimates were obtained using direct standardization to the age distribution of the sample. Baseline characteristics among controls randomized to active versus placebo β-carotene were compared similarly.
Conditional logistic regression analysis was performed using Cox proportional hazards models stratified by matched set17 for both total and prostate carcinoma. To evaluate the joint association of baseline β-carotene and randomized β-carotene assignment with cancer incidence rates, we fit models with interaction terms between quartile of baseline plasma level and randomized assignment. The highest baseline quartile with no supplementation was used as the reference group. Tests for trend in effects across quartiles were conducted using a linear term for quartile evaluated at the median plasma level within each quartile. Models also included terms for age and randomized aspirin assignment. Additional analyses controlled for baseline levels of BMI, alcohol use, exercise, and number of cigarettes per day smoked among current smokers. Analyses were repeated eliminating events occurring within 2 years of randomization and excluding subjects with cardiovascular events prior to case or control selection. Analyses also were conducted separately for aggressive and nonaggressive cases of prostate carcinoma.
Of the 1439 total cancer cases analyzed, 631 (44%) were prostate carcinoma, 169 (12%) were colon carcinoma, 93 (6%) were melanoma, 85 (6%) were lung carcinoma, 94 (6%) were lymphoma, and < 50 were of any other single type of cancer. Comparisons of baseline characteristics among the 1439 cases and their 2204 matched controls are shown in Table 1. Of the cases, 41% were never smokers, 45% were past smokers, and 14% were current smokers at baseline. The average age was 59 years. The distributions among controls within matched sets were identical. Cases and controls showed a slight but significant difference in baseline BMI, with a mean of 25.1 kg/m2 among cases and of 24.7 kg/m2 among controls. Cases and controls were similar with respect to alcohol use and exercise at baseline. The geometric means of baseline plasma levels of β-carotene were 215 ng/mL among cases and 225 ng/mL among controls (P = 0.04); median levels were very similar. There was a nonsignificant difference of 237 μg/day in reported dietary intake of β-carotene (P = 0.14). Among cases, 706 (49%) were assigned to active β-carotene supplementation as were 1121 controls (51%).
Table 1. Distribution of Baseline Characteristics among 1439 Matched Sets from a Nested Case–Control Comparison within the Physicians' Health Studya
|Age (yrs) (mean ± SE)||59.1 ± 0.2||58.9 ± 0.2||Matched|
|BMI, kg/m2 (mean ± SE)||25.1 ± 0.1||24.7 ± 0.1||0.001|
|Alcohol use (%)|
| ≥1 time/week||72.7||71.5||0.62|
|Plasma β-carotene, ng/mL (mean ± SE)b||215.1 ± 3.9||224.9 ± 3.5||0.04|
|Dietary β-carotene, μg/day (Mean ± SE)b||6913.8 ± 123.6||7150.9 ± 102.8||0.14|
|β-carotene assignment (%)||49.1||50.9||0.44|
Baseline plasma β-carotene levels were divided into quartiles based on the distribution among the controls, yielding cutpoints of ≤ 153.3 ng/mL, > 153.3–227.3 ng/mL, > 227.3–343.8 ng/mL, and > 343.8 ng/mL. The distribution of baseline characteristics according to baseline plasma level among controls is shown in Table 2. Both age and BMI showed clear trends with baseline levels, with those at lower levels being both younger and heavier. In addition, those at lower plasma β-carotene levels tended to drink more and exercise less. Smoking also was associated with baseline level, with more current smokers at lower levels. The age-adjusted geometric means of plasma levels among controls were 255 ng/mL among never smokers, 218 ng/mL among past smokers, and 172 ng/mL among current smokers. Dietary β-carotene also was associated with plasma level. The correlation between these measures was 0.17 among controls, and 0.18 among cases. In linear regression analyses of ln plasma β-carotene on these predictors among controls, all these variables remained statistically significant (P < 0.01) after simultaneous adjustment (data not shown).
Table 2. Comparison of Baseline Characteristics by Quartile of Plasma β-Carotene among 2204 Controlsa
|Age (yrs)||59.3||59.4||60.3||61.0||≤ 0.001|
|Smoking status (%)|
| Past||48.7||45.8||46.6||42.4||≤ 0.001|
|BMI (kg/m2)||25.3||24.9||24.7||24.1||≤ 0.001|
| Weekly||44.0||42.9||49.3||42.7||≤ 0.001|
| ≥ 1 time/week||67.0||70.6||73.9||73.8||0.012|
|Dietary β-caroteneb (μg/day)||6268.3||6655.3||7637.8||8204.9||≤ 0.001|
|β-carotene assignment (%)||50.1||51.0||51.7||51.0||0.73|
Among controls, there were no significant associations between randomization assignment and age, BMI, alcohol use, or dietary intake of β-carotene (Table 3). There was some difference in the frequency of exercise at baseline, and there were more current smokers in the placebo group. Randomized β-carotene assignment also was independent of baseline plasma β-carotene level among controls (Tables 2 and 3). Geometric mean plasma levels were 228 ng/mL among controls assigned to active β-carotene and 224 ng/mL among those assigned to β-carotene placebo (P = 0.49). When baseline plasma levels of β-carotene in cases and controls were examined separately by randomized assignment, the mean case–control difference was greater among those assigned to placebo (-14 ng/mL; P = 0.04) than among those assigned to active β-carotene (-5 ng/mL; P = 0.46), although a test for interaction was not significant (P = 0.37) (data not shown).
Table 3. Comparison of Baseline Characteristics by β-Carotene Assignment among 2204 Controlsa
|Smoking status (%)|
| ≥ 1 time/week||73.4||69.4||0.04|
|Plasma β-carotene (ng/mL)b||228.4||223.9||0.49|
|Dietary β-carotene (μg/day)b||7212.6||7088.9||0.54|
Results from the full PHS cohort study of 22,071 randomized physicians showed no significant effects of β-carotene supplementation on either total (relative risk [RR] = 0.98; 95% confidence interval [95% CI], 0.91–1.06) or prostate carcinoma (RR = 0.98; 95% CI, 0.87–1.11)7. Results in the nested case–control sample were similarly nonsignificant, with an estimated RR of 0.96 (95% CI, 0.83–1.10) for total cancer and 0.91 (95% CI, 0.75–1.11) for prostate carcinoma after adjustment for age, smoking, and randomized aspirin assignment. In conditional logistic regression analyses of the effects of both baseline plasma β-carotene level and β-carotene assignment on total cancer (Table 4), there was a marginally significant increase in risk among those in the lowest versus highest quartile of plasma level at baseline (RR = 1.30; 95% CI, 0.98–1.74), with a nonsignificant trend in risk over baseline quartiles (P = 0.11). The latter are the estimated effects among those randomized to β-carotene placebo. The effect of randomized assignment also was marginally different across quartiles (P value for trend = 0.09). Those in the lowest quartile of plasma β-carotene experienced a nonsignificant 17% reduction in risk with active supplementation (RR = 0.83; 95% CI, 0.63–1.09), whereas those in the highest quartile at baseline had a nonsignificant 14% increase in risk (RR = 1.14; 95% CI = 0.86–1.52). These results were nearly identical when cancers occurring within the first 2 years of randomization were eliminated or when individuals with prior cardiovascular disease were excluded. In analyses adjusting for baseline BMI, alcohol use, exercise, and number of cigarettes per day among current smokers, the RR for those in the lowest versus highest plasma level was reduced to 1.22 (95% CI, 0.90–1.63). However, the effect of supplementation was virtually unchanged in these analyses.
Table 4. Matched Analysis of Nested Case–Control Data from the Physicians' Health Study: Association of Quartile of Baseline Plasma Level of β-Carotene and Randomized β-Carotene Assignment within Quartile with Total and Prostate Carcinomaa
| 1 (lowest)||402||552||1.30||0.98–1.74||0.07||0.83||0.63–1.09||0.18|
| 4 (highest)||339 ||551 ||1.00||—||—||1.14||0.86–1.52||0.36|
| 1 (lowest)||177||353||1.45||0.98–2.15||0.06||0.68||0.46–0.99||0.04|
| 4 (highest)||162||359 ||1.00||—||—||1.33||0.91–1.96||0.14|
In analyses of prostate carcinoma (Table 4), some stronger and significant trends were observed. Those in the lowest versus highest quartile of baseline plasma β-carotene level experienced a marginally significant (P = 0.06) 45% increased risk of prostate carcinoma (RR = 1.45; 95% CI, 0.98–2.15), and the trend over plasma quartiles also was marginally significant (P = 0.09). However, β-carotene supplementation led to a significant 32% reduction in risk among those in the lowest quartile of baseline plasma β-carotene (RR = 0.68; 95% CI, 0.46–0.99), and the trend in the effect of supplementation over quartiles was statistically significant (P = 0.01). For those in the highest baseline quartile, β-carotene supplements were associated with a nonsignificant 33% increase in risk (RR = 1.33; 95% CI, 0.91–1.96). There was little change in these estimates when cases occurring within 2 years of randomization or those with prior cardiovascular disease were eliminated. Results also were nearly identical after adjusting for baseline BMI, alcohol use, exercise, and number of cigarettes smoked. In separate analyses by stage, similar results were observed for aggressive and nonaggressive prostate carcinoma (Table 5). Although no significant results were observed in subgroup analyses by stage, the effect of lowest versus highest baseline plasma level was stronger for aggressive prostate carcinoma and the effect of supplementation among those in the lowest quartile was similar for both aggressive and nonaggressive disease.
Table 5. Matched Analysis of Nested Case–Control Data from the Physicians' Health Study: Association of Quartile of Baseline Plasma Level of β-Carotene and of Randomized β-Carotene Assignment within Quartile with Type of Prostate Carcinomaa
|Aggressive prostate carcinoma|
| 1 (lowest)||84||150||1.62||0.88–2.95||0.12||0.61||0.34–1.08||0.09|
| 4 (highest)||74 ||154||1.00||—||—||1.33||0.74–2.37||0.34|
|Nonaggressive prostate carcinoma|
| 1 (lowest)||81||184||1.33||0.76–2.32||0.31||0.65||0.38–1.12||0.12|
| 4 (highest)||73 ||181||1.00||—||—||1.00||0.56–1.76||0.99|
When analyses were stratified by smoking status, the results were neither statistically significant nor different for total cancer. The estimated RRs for those in the lowest versus highest quartile of plasma β-carotene at baseline were 1.25 (95% CI, 0.79–1.97) among never smokers, 1.33 (95% CI, 0.87–2.03) among past smokers, and 2.10 (95% CI, 0.78–5.67) among current smokers. The corresponding RRs for the randomized active versus placebo β-carotene groups among those in the lowest quartile of baseline plasma β-carotene were 0.78 (95% CI, 0.48–1.27) in never smokers, 0.70 (95% CI, 0.48–1.04) in past smokers, and 1.34 (95% CI, 0.74–2.45) in current smokers. The apparent increase in the current smokers may be due to the fact that the overall RR associated with β-carotene supplementation among current smokers in the nested case–control sample was 1.32 (95% CI, 0.90–1.92), which is higher than that observed among this group in the total PHS cohort (RR = 1.05; 95% CI, 0.86–1.28).7
For prostate carcinoma, the effects of lowest versus highest baseline plasma level were similar to those for total cancer among never and past smokers, but greater among the 54 matched sets of current smokers (RR = 3.85; 95% CI, 0.72–20.66), although power was limited in this small group. The effect of randomized β-carotene assignment on prostate carcinoma among those in the lowest quartile of baseline β-carotene was null in never and current smokers (RR = 0.99; 95% CI, 0.51–1.91, and RR = 0.96; 95% CI, 0.37–2.48, respectively), but was a statistically significant reduction in risk among past smokers (RR = 0.45; 95% CI, 0.26–0.77). These relations persisted when smoking-specific quartiles of baseline plasma levels were used. Because of the small numbers within subgroups of smoking status in addition to baseline quartiles, all these analyses, and particularly those for prostate carcinoma, suffered from low power, as reflected in the wide 95% CIs.
Analyses of these prespecified hypotheses are compatible with the possibility that β-carotene supplementation may reduce the risk of prostate carcinoma among those individuals with low baseline levels. If so, this also may have implications for total cancer. In this prospective, nested case–control study, among those in the lowest quartile of plasma β-carotene at baseline there was a possible but nonsignificant 17% reduction in total cancer and a stronger and significant 32% reduction in prostate carcinoma among those randomized to β-carotene. A marginally significant increased risk with a lower baseline plasma level of β-carotene also was found for both total and prostate carcinoma among those randomized to placebo. Coupled with this, the effect of supplementation among those in the lowest quartile of baseline plasma β-carotene was to reduce the risk to that of physicians in the highest quartile at baseline. It also was apparent that those with the highest levels at baseline had a possible but nonsignificant increase in risk with supplementation.
Recent analyses of the data from the ATBC and CARET studies also have examined the issue of effect modification of β-carotene supplementation by baseline level. Neither has found any such modification for lung carcinoma. In the ATBC trial, an analysis of the β-carotene intervention by quartiles of baseline serum β-carotene showed very similar RRs of lung carcinoma by quartile (P value for trend = 0.90).10 In CARET, the RRs of lung carcinoma for those with serum levels above versus serum levels below the median value were identical among those randomized to placebo versus active β-carotene and retinyl-palmitate combined.11 Thus, the benefit of supplementation for total and prostate carcinoma among those with low baseline levels found in the PHS has not been observed for lung carcinoma in these two other trials. In a more recent analysis of 246 incident prostate carcinoma cases from the ATBC study,18 the overall effect of β-carotene on prostate carcinoma was a nonsignificant 23% increase among those assigned to active β-carotene, with no reduction observed among those with serum levels below the median at baseline. However, perhaps because it is the largest study to examine this question, the PHS found a significant 32% reduction in prostate carcinoma among those assigned to β-carotene who were in the lowest quartile of blood level at baseline.
Both the ATBC and CARET studies included only those individuals at high risk of lung carcinoma, either heavy smokers or those with an occupational exposure to asbestos. The PHS, with a small proportion of current smokers, had too little power to assess effect modification for lung carcinoma, with only 85 cases among those with plasma levels available. The data for total cancer, marginally significant in the PHS, are consistent with results observed in the Linxian study,9 which showed a 13% reduction in total cancer mortality and a 21% reduction in gastric carcinoma mortality after supplementation with a combination of β-carotene, vitamin E, and selenium in a poorly nourished population. However, reported baseline levels in Linxian were lower than those in the lowest quartile in the PHS, with a mean range of 59–68 ng/mL in Linxian versus a median of 110 ng/mL in the lowest quartile in the PHS.
Baseline β-carotene levels in the PHS were more comparable to those observed in the ATBC and CARET studies. Mean levels were 172 ng/mL among controls who were current smokers in the PHS, approximately 170 ng/mL in the ATBC study,10 and 152 ng/mL in the CARET study.11 Levels were associated strongly with smoking among controls in the PHS, with never smokers and past smokers at levels that were higher by 83 ng/mL and 46 ng/mL, respectively, after age adjustment. This relation with smoking has been observed in other studies,19, 20 and a relation with amount smoked has been confirmed in CARET.21 We also found significant independent associations of baseline β-carotene levels with age, BMI, alcohol use, and exercise among controls in our study, results that also have been observed in other studies.16–18
The observed marginally significant inverse relation of baseline β-carotene with total cancer is supported by other data. Observational studies have suggested an association of green and yellow fruits and vegetables in the diet with cancer prevention,1 and β-carotene has been hypothesized to be one of the active components.3 A possible mechanism for a protective effect of β-carotene could be its antioxidant properties, although other possibilities such as an increase in gap junctional communication22 have been proposed. Studies of dietary intake of β- carotene generally have supported this, although the result is most consistent for lung carcinoma and perhaps gastric carcinoma,2, 23 and could be due to other components of the diet. Studies of serum levels of β-carotene and total cancer have been less consistent, with some studies suggesting inverse associations24, 25 but others no relation26 or different relations in men and women.27 Studies regarding carcinomas of the lung or stomach show a more consistent association with serum levels.2 In this regard, observational analyses of both the ATBC and CARET data found an inverse association between higher serum levels of β-carotene and risk of lung carcinoma.10, 11 Neither study has reported such information for other types of cancer.
Observational analyses of β-carotene and prostate carcinoma have been inconclusive. Both cohort and case–control studies of dietary intake of green and yellow fruits and vegetables and of carotenoids have shown mixed results, with some studies finding significant inverse associations, whereas others find no relation.2, 20 Three recent studies of dietary β-carotene and prostate carcinoma have found no association,28–30 but one study did find an inverse association.31 In four studies that examined serum levels of carotenoids and prostate carcinoma23, 24, 32, 33 no significant association was reported. However, two of these studies had very small numbers with prostate carcinoma, with 11 cases23 and 37 cases,24 respectively. In another study, among 103 prostate carcinoma cases and matched controls, there was no apparent trend in risk over quartiles of baseline level.29 However, there was some fluctuation of the effect of level with years since the baseline measurement. In the most recent study, with 142 cases of prostate carcinoma accrued up to 20 years after the blood collection, no significant trend was observed, although the data were more consistent with a positive effect.30 In contrast, the observational result from the PHS, based on a total of 631 prostate carcinoma cases, indicated a marginally significant inverse association of baseline plasma β-carotene with prostate carcinoma, with a 45% increased risk among those in the lowest versus the highest quartile. In separate analyses by time since randomization, the data (not shown) suggested that this effect of baseline β-carotene level was strongest in the early years after randomization, but diminished over time. The benefits of β-carotene supplementation appeared strongest in the period between 5–9 years after randomization, although these subgroup analyses are based on fewer numbers.
Limitations of the current study include a lack of information regarding the development of lung carcinoma, which has the strongest relation with dietary β-carotene in observational data. Because of the small proportion of smokers in the PHS, the number of lung carcinoma cases was small. These results therefore may not be directly comparable to those observed in the ATBC or CARET studies, both of which were conducted among those individuals at high risk of lung carcinoma. Baseline blood specimens also were only available for 68% of the trial participants. Although this may have limited power for these analyses, it is unlikely to have influenced the estimated effects of β-carotene supplementation. Blood specimens were collected prior to randomization, and receipt of a blood sample was not related to β-carotene assignment. Baseline plasma level possibly could be different among those providing a sample and those who did not. Cases with available blood samples were similar to those without blood samples with respect to age, smoking status, BMI, alcohol use, and exercise, but did differ in dietary consumption of β-carotene. However, in multivariate analyses controlling for factors related to whether cases contributed blood samples, the results were unchanged. Strengths of this study include the large number of total and prostate carcinoma cases that have developed among a cohort of highly compliant participants randomized to β-carotene supplementation or placebo for an average of 12 years of follow-up. To our knowledge it is the largest randomized trial to date to examine the effects of β-carotene supplementation on total and prostate carcinoma and the only trial conducted largely among healthy nonsmokers.
The hypothesis that β-carotene supplementation has beneficial effects on cancer, both overall and site specific, and particularly prostate carcinoma, among those cases with low baseline levels requires further testing. Whether the differences observed among studies are due to different populations, different cancer sites, different formulations or combinations of agents, or to chance remains to be determined. Further research, including longer follow-up of participants in all randomized trials of β-carotene supplementation, is necessary to allow more definitive results to emerge.
The authors thank James Anderson, technical specialist at the vitamin analysis laboratory, for his assistance with the laboratory assays.