Serum Retinol and Risk of Overall and Site-Specific Cancer in the ATBC Study

Abstract

Retinol, the most biologically active form of vitamin A, might influence cancer-related biological pathways. However, results from observational studies of serum retinol and cancer risk have been mixed. We prospectively examined serum retinol and risk of overall and site-specific cancer in the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study (n = 29,104 men), conducted in 1985–1993, with follow-up through 2012. Serum retinol concentration was measured using reverse-phase high-performance liquid chromatography. Cox proportional hazards models estimated the association between baseline serum retinol quintile and overall and site-specific cancer risk in 10,789 cases. After multivariable adjustment, higher serum retinol was not associated with overall cancer risk (highest vs. lowest quintile: hazard ratio (HR) = 0.97, 95% confidence interval (CI): 0.91, 1.03; P for trend = 0.43). Higher retinol concentrations were, however, associated with increased risk of prostate cancer (highest vs. lowest quintile: HR = 1.28, 95% CI: 1.13, 1.45; P for trend < 0.0001) and lower risk of both liver and lung cancers (highest vs. lowest quintile: for liver, HR = 0.62, 95% CI: 0.42, 0.91; P for trend = 0.004; and for lung, HR = 0.80, 95% CI: 0.72, 0.88; P for trend < 0.0001). No associations with other cancers were observed. Understanding the mechanisms that underlie these associations might provide insight into the role of vitamin A in cancer etiology.

cancer, cohort study, prospective study, serum retinol, vitamin A

Abbreviations

 
• ATBC

  Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study

 
• CI

  confidence interval

 
• HR

  hazard ratio

 
• ICD-9

  International Classification of Diseases, Ninth Revision

Retinoids are a class of synthetic and biological molecules that have chemical structures similar to vitamin A, of which retinol is the most biologically active in humans (1). Retinol is found in some foods, but much of the body’s retinol is derived from ingestion of provitamin A carotenoids which are subsequently converted to retinol. These compounds have been shown to have potentially anticarcinogenic properties such as induction of apoptosis and cellular differentiation, inhibition of proliferation, antioxidant/free radical quenching activities, and enhancement of immune surveillance (2). However, there is also evidence that retinoids might enhance tumor growth at some sites (3, 4). Thus, the multifaceted role of retinol in cancer remains unclear.

Previous studies, examining the association between vitamin A and cancer at various sites by measuring dietary and supplemental intake of vitamin A and provitamin A carotenoids using food frequency questionnaires, have had inconsistent results. The recent World Cancer Research Fund Second Expert Report on diet and cancer judged that there was probable evidence that foods containing carotenoids protect against cancers of the head and neck and of the lung, and that foods containing β-carotene protect against cancer of the esophagus (5). However, studies have demonstrated poor correlation between dietary or supplemental intake and circulating retinol levels (6), likely due to dietary measurement error and because retinol concentrations are contributed to not only by dietary and supplemental intake but also by factors related to absorption, cleavage of provitamin A compounds, and transport to and from retinol stores in the liver (7). Thus, circulating retinol concentration is a better measurement of retinol status than self-reported intake.

Several epidemiologic studies have examined circulating retinol concentrations and risk of cancer, with inconsistent results. For example, an inverse association between retinol and cancer has been reported at several sites such as oral (8), liver (9–11), prostate (12, 13), lung (14–17), and stomach (18), while other studies have reported no association between retinol and cancers of the cervix (19), colon (20), prostate (21), breast (22), and liver (23). Meanwhile, a positive association between retinol and prostate cancer has been reported by several groups (24–27). In some of these studies, the inconsistency in the findings might be due to small sample sizes (i.e., n < 100), which limit the statistical power to detect true associations (12, 23). Furthermore, differences in study design, screening prevalence, dietary and lifestyle factors, and laboratory methods for measuring retinol could also contribute to the differences in the findings (8, 18). Therefore, to comprehensively evaluate the role of retinol across cancer sites within the same cohort, we investigated the association between serum retinol and risk of cancer at multiple sites within the Alpha-Tocopherol, Beta-Carotene Cancer Prevention (ATBC) Study.

METHODS

Study design and population

The ATBC Study was a randomized, double blind, placebo-controlled study conducted between 1985–1993. The objective of the study was to determine the effect of supplemental α-tocopherol and β-carotene on the incidence of lung and other cancers. Men from southwestern Finland, aged 50–69 years, who smoked at least 5 cigarettes per day, were recruited for the study. Informed consent was obtained from the participants, and the institutional review boards of both US National Cancer Institute and the National Public Health Institute of Finland approved the study. A total of 29,133 participants were randomized to one of four groups: 1) α-tocopherol (50 mg/day), 2) β-carotene (20 mg/day), 3) both supplements, or 4) placebo (28). The participants took the supplements until death or until April 30, 1993, when the trial ended.

At the first baseline visit, the participants completed detailed questionnaires about their medical, smoking, and dietary history, and their height and weight were measured by registered nurses. Overnight fasting blood samples were collected and stored at −70°C, protected from light.

Serum retinol measurement

Baseline serum samples from all participants were analyzed for α-tocopherol, β-carotene, retinol, and total and high-density lipoprotein cholesterol. Retinol was measured using reverse-phase high-performance liquid chromatography (29). All the assays were conducted at a central laboratory at the National Public Health Institute in Helsinki, Finland. Of 29,133 participants, 29 were excluded due to missing retinol values, leaving a total analytical cohort of 29,104 men.

Case identification

All cases were identified by the Finnish Cancer Registry, which has been shown to correctly identify and classify nearly all cancer cases for this cohort (30). For cases that were diagnosed before September 2001, medical records were reviewed by 1 or 2 oncologists to confirm diagnosis. For cases that were diagnosed after September 2001, information was obtained from the Finnish Cancer Registry and Register of Causes of Death (28). Included in this report are all of the following cancers, diagnosed through December 31, 2012 (n = 10,798): cancers of the biliary tract (International Classification of Diseases, Ninth Revision (ICD-9), code 156), bladder (ICD-9 code 188), brain/central nervous system (ICD-9 code 191), colorectum (ICD-9 codes 153 and 154, excluding cancers of appendix and anus), upper gastrointestinal tract (including esophageal squamous cell carcinoma (ICD-9 code 150), esophagogastric junctional adenocarcinoma (ICD-9 codes 150 and 151.0), and gastric noncardia adenocarcinoma (ICD-9 codes 151.1–151.9)), hematologic (ICD-9 codes 200–208), kidney (ICD-9 codes 189.0 and 189.1), larynx (ICD-9 code 161, including only squamous cancers), liver (including intrahepatic bile duct (ICD-9 code 155)), lung (ICD-9 code 162), melanoma (ICD-9 code 172), oropharynx (ICD-9 codes 140–149, including only squamous cancers), pancreas (ICD-9 code 157, excluding 157.4), and prostate (ICD-9 code 185). With the exception of two sites with small numbers (small bowel cancer, n = 36, and cancer of the pleura, n = 71), we included all site-specific cancers for which we received data from the Finnish Cancer Registry.

Statistical analysis

Cox proportional hazards regression was used to estimate the association between quintiles of baseline serum retinol concentration and overall as well as site-specific cancer incidence. We confirmed the proportional hazards assumption for all individual cancer sites examined (all P > 0.12) by including in the model a term for interaction between serum retinol and follow-up time and by evaluating its statistical significance using the Wald test. All models included age as a continuous variable. Risk factors known or hypothesized to be associated with different cancers were assessed as potential confounders by entering each factor in the age-adjusted model for overall cancer. The variables considered were: α-tocopherol treatment group; β-carotene treatment group; height; weight; body mass index; number of cigarettes smoked per day; years of smoking; marital status; education and training; physical activity; urban residence; intake of fruit, vegetables, red meat, dietary fat, cholesterol, alcohol, retinol, vitamin D, and calcium; and baseline concentrations of serum total and high-density lipoprotein cholesterol, α-tocopherol, and β-carotene. Although none of these variables met the definition of a confounder (i.e., their inclusion did not change the retinol point estimates by >10%), we selected the following covariables for inclusion in our multivariable models: α-tocopherol treatment group (yes/no), β-carotene treatment group (yes/no), alcohol consumption (up to vs. at least the median), age, body mass index, number of cigarettes smoked per day, years of smoking, serum α-tocopherol, serum β-carotene, and serum cholesterol (all continuous).

To evaluate potential effect modification, models were stratified on the following variables: α-tocopherol treatment group, β-carotene treatment group, follow-up time (up to 10 years vs. at least 10 years), number of cigarettes smoked per day, number of years smoked regularly, body mass index, alcohol consumption, serum α-tocopherol, serum β-carotene, serum cholesterol, and age (all up to vs. at least the median). Stratified analyses were performed for overall cancer incidence, as well as for those cancers where a main association with serum retinol was observed (i.e., liver, prostate, lung). Statistical interaction was assessed using the likelihood ratio test by comparing models with and without an interaction term. All reported P values are 2 tailed, and α = 0.05 is considered to be the threshold for statistical significance for most analyses, with the exception of the exploratory interaction analyses where a Bonferroni correction was used to account for multiple testing (α = 0.00125 based on 44 tests). All analyses were performed using SAS, version 9.4 (SAS Institute, Inc., Cary, North Carolina).

RESULTS

Characteristics of the cohort according to baseline serum retinol concentration are shown in Table 1. Men with higher retinol status had higher average body mass index, serum total and high-density lipoprotein cholesterol, and serum α-tocopherol, and were more highly educated. Men with higher serum retinol also had higher dietary intakes of total vegetables, red meat, alcohol, and retinol, and were more likely to take calcium, vitamin A, vitamin D, and vitamin E supplements.

Table 1

Selected Baseline Characteristics According to Serum Retinol Quintiles in the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study, Finland, 1985–2012

Characteristic	Quintile of Baseline Serum Retinola
	1(n = 5,883)			2(n = 5,806)			3(n = 5,841)			4(n = 5,792)			5(n = 5,782)
	No.	%	Median (10th–90th Percentile)	No.	%	Median (10th–90th Percentile)	No.	%	Median (10th–90th Percentile)	No.	%	Median (10th–90th Percentile)	No.	%	Median (10th–90th Percentile)
α-tocopherol treatment group	2,946	49.9		2,904	50.1		2,930	50.3		2,876	49.6		2,891	50.1
β-carotene treatment group	2,955	50.1		2,896	49.9		2,897	49.7		2,925	50.4		2,875	49.9
Age, years			58 (51–65)			57 (51–65)			57 (51–65)			56 (51–64)			56 (51–64)
Height, cm			173 (164–180)			173 (166–181)			174 (166–189)			174 (166–182)			174 (166–182)
Weight, kg			75.6 (60.7–93.7)			77.7 (63.5–95.2)			78.4 (64.2–96.2)			79.1 (65.5–96.0)			80.4 (66.0–97.3)
Body mass indexb			25.3 (20.8–30.6)			25.8 (21.6–31.1)			26 (21.9–31.2)			26.2 (22.2–31.2)			26.6 (31.5–22.5)
Education and training combined above eighth grade	3,690	62.7		3,684	63.4		3,852	65.9		3,913	67.5		4,084	70.6
Married	4,603	78.2		4,700	81		4,743	81.2		4,706	81.2		4,592	79.4
Urban residence	3,578	60.8		3,362	57.9		3,465	59.3		3,336	57.6		3,456	59.7
Physically active	3,307	56.2		3,392	58.4		3,455	59.2		3,459	59.7		3,310	57.3
No. of cigarettes smoked per day			20 (10–30)			20 (10–30)			20 (10–30)			20 (10–30)			20 (10–30)
No. of years smoked regularly			39 (28–47)			37 (27–45)			36 (25–45)			36 (25–45)			35 (25–45)
Serum biomarkers
Retinol, mg/L			438 (357–476)			517 (491–542)			577 (553–601)			642 (614–675)			755 (697–900)
Total cholesterol, mmol/L			5.8 (4.5–7.3)			6.04 (4.81–7.53)			6.17 (4.90–7.73)			6.3 (5.0–7.9)			6.5 (5.1–8.0)
HDL cholesterol, mmol/L			1.12 (0.83–1.56)			1.14 (0.84–1.59)			1.14 (0.85–1.6)			1.15 (1.63–0.85)			1.17 (0.85–1.71)
α-tocopherol, mg/L			10.6 (7.5–14.3)			11.1 (8.2–15.1)			11.6 (8.5–15.6)			11.9 (8.7–16.3)			12.4 (8.8–17.6)
β-carotene, μg/L			171 (70–385)			177 (76–386)			180 (79–395)			175 (72–390)			151 (59–357)
Daily dietary intake
Cholesterol, mg			529 (316–877)			535 (319–891)			539 (331–882)			546 (328–901)			542 (318–881)
Total fat, g			118 (76.5–176.1)			119 (77.2–178.8)			119 (78–177)			117 (77–175)			115 (73–173)
Retinol, mg			1,172 (519–2,634)			1,213 (545–2,748)			1,249 (542–2,746)			1,273 (571–2,809)			1,326 (552–2,872)
Vitamin D, mg			4.56 (2.08–9.33)			4.66 (2.21–9.21)			4.62 (2.18–9.20)			4.81 (2.25–9.41)			4.87 (2.33–9.52)
Vitamin E, mg			10.7 (6.3–20.2)			10.7 (6.4–19.7)			10.6 (6.4–19.2)			10.8 (6.4–19.5)			10.6 (6.3–18.9)
Calcium, mg			1,303 (717–2061)			1,338 (752–2,113)			1,334 (746–2,113)			1,347 (766–2,126)			1,325 (721–2,095)
Total energy, kcal			2,579 (2,123–3,664)			2,616 (1,820–3,666)			2,610 (1,847–3,657)			2,613 (1,854–3,676)			2,585 (2,142–3,659)
Alcohol, g			6.6 (0–34.9)			8.3 (0–35.7)			10.6 (0–40.6)			13.2 (0.4–46.4)			19.4 (1.3–56.3)
Total fruits, g			107 (25–250)			109 (27–250)			108 (25–252)			109 (27–253)			107 (23–258)
Total vegetables, g			87 (32.3–190)			91 (35–187)			95 (38–197)			99 (37–202)			98 (38–207)
Total red meat, g			129 (72–235)			132 (73–236)			132 (75–231)			134 (76–235)			133 (73.6–233)
Supplement use
Vitamin A	504	16.8		519	17.3		586	19.6		631	21.1		747	25.0
Vitamin D	316	16.1		335	17.0		385	19.5		433	22.0		499	25.3
Calcium	535	16.7		546	17.1		613	19.2		681	21.3		814	25.5
Vitamin E	472	16.1		524	17.8		582	19.8		630	21.4		725	24.7

^a Baseline serum retinol in mg/L. Quintile 1: ≤483 (n = 5,883); quintile 2: 483.1–547 (n = 5,806); quintile 3: 547.1–607 (n = 5,841); quintile 4: 607.1–685 (n = 5,792); quintile 5: >685.1 (n = 5,782).

^b Weight (kg)/height (m)².

Overall cancer

After multivariable adjustment for several potential risk factors, serum retinol was not associated with overall cancer risk (for quintile 5 vs. 1, hazard ratio (HR) = 0.97, 95% confidence interval (CI) = 0.91, 1.03; P for trend = 0.43) (Table 2, Web Table 1). This finding was unchanged when cases diagnosed within 2 years of blood collection were excluded (for quintile 5 vs. 1, HR = 0.98, 95% CI: 0.92, 1.05; P for trend = 0.71) (Web Table 2). No statistically significant interaction was observed between serum retinol and any of the factors examined for overall cancer (Table 3).

Table 2

Association Between Baseline Serum Retinol and Overall and Site-Specific Cancer in the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study, Finland, 1985–2012

Cancer Site and Modelb	No. of Cases	Quintile of Baseline Serum Retinola								P for Trend
		2		3		4		5
		HR	95% CI	HR	95% CI	HR	95% CI	HR	95% CI
Overall Cancer	10,798
Model 1		0.94	0.88, 0.99	0.91	0.86, 0.97	0.92	0.86, 0.97	0.93	0.88, 0.99	0.39
Model 2		0.95	0.90, 1.01	0.93	0.88, 0.99	0.94	0.89, 1.01	0.97	0.91, 1.03	0.43
Biliary Tract	86
Model 1		1.01	0.50, 2.05	0.75	0.35, 1.60	1.59	0.83, 3.03	1.21	0.60, 2.41	0.28
Model 2		1.02	0.50, 2.07	0.76	0.35, 1.64	1.59	0.82, 3.06	1.19	0.59, 2.42	0.33
Bladder	789
Model 1		0.95	0.76, 1.18	0.96	0.76, 1.20	0.97	0.78, 1.22	1.01	0.81, 1.26	0.81
Model 2		0.94	0.75, 1.18	0.95	0.76, 1.18	0.96	0.77, 1.20	0.99	0.78, 1.24	0.97
Brain/CNS	78
Model 1		1.36	0.68, 2.69	1.00	0.48, 2.07	1.08	0.52, 2.21	0.90	0.42, 1.93	0.58
Model 2			0.72, 2.85	1.08	0.51, 2.26	1.18	0.57, 2.47	1.03	0.47, 2.25	0.85
Colorectum	878
Model 1		0.89	0.71, 1.11	1.05	0.85, 1.30	1.13	0.92, 1.40	1.03	0.83, 1.28	0.28
Model 2		0.89	0.71, 1.11	1.05	0.84, 1.30	1.12	0.91, 1.39	1.01	0.80, 1.26	0.40
ESCC	96
Model 1		0.89	0.46, 1.70	0.87	0.45, 1.66	0.88	0.46, 1.68	1.17	0.63, 2.16	0.55
Model 2		0.93	0.48, 1.77	0.92	0.48, 1.77	0.92	0.47, 1.78	1.16	0.61, 2.19	0.61
EGJA	151
Model 1		0.93	0.57, 1.51	0.89	0.55, 1.46	0.94	0.58, 1.53	0.71	0.41, 1.21	0.25
Model 2		0.93	0.57, 1.52	0.90	0.55, 1.47	0.94	0.57, 1.55	0.71	0.41, 1.24	0.28
GNCA	332
Model 1		1.06	0.75, 1.49	1.04	0.74, 1.47	1.16	0.83, 1.63	0.89	0.62, 1.28	0.69
Model 2		1.06	0.75, 1.50	1.05	0.74, 1.49	1.16	0.82, 1.63	0.89	0.61, 1.30	0.67
Hematologic	602
Model 1		1.11	0.86, 1.42	0.93	0.72, 1.21	1.00	0.78, 1.30	0.96	0.73, 1.24	0.55
Model 2		1.12	0.87, 1.43	0.93	0.72, 1.21	1.01	0.78, 1.31	0.97	0.74, 1.27	0.61
Kidney	413
Model 1		1.06	0.78, 1.44	0.96	0.70, 1.31	0.95	0.69, 1.31	1.15	0.85, 1.56	0.48
Model 2		1.06	0.77, 1.44	0.96	0.70, 1.32	0.96	0.69, 1.32	1.15	0.84, 1.58	0.48
Larynx	193
Model 1		1.12	0.70, 1.79	1.09	0.68, 1.75	1.36	0.87, 2.14	1.13	0.71, 1.82	0.44
Model 2		1.15	0.72, 1.85	1.15	0.71, 1.85	1.42	0.90, 2.25	1.12	0.69, 1.84	0.49
Liver	233
Model 1		0.62	0.42, 0.90	0.57	0.39, 0.84	0.35	0.23, 0.56	0.66	0.45, 0.96	0.006
Model 2		0.63	0.43, 0.92	0.59	0.40, 0.86	0.36	0.23, 0.57	0.62	0.42, 0.91	0.004
HCC, model 2	150	0.67	0.41, 1.09	0.78	0.49, 1.24	0.36	0.20, 0.65	0.65	0.40, 1.07	0.03
Lung	3,940
Model 1		0.90	0.81, 0.97	0.8	0.73, 0.88	0.75	0.68, 0.82	0.73	0.66, 0.81	<0.0001
Model 2		0.92	0.84, 1.01	0.85	0.77, 0.94	0.80	0.72, 0.88	0.80	0.72, 0.88	<0.0001
Melanoma	136
Model 1		1.24	0.70, 2.22	1.07	0.59, 1.94	1.30	0.73, 2.29	1.60	0.96, 2.88	0.06
Model 2		1.20	0.67, 2.16	1.02	0.56, 1.86	1.22	0.68, 2.18	1.52	0.87, 2.67	0.13
Oropharynx	239
Model 1		0.92	0.60, 1.40	1.12	0.75, 1.67	0.99	0.65, 1.49	1.07	0.71, 1.61	0.66
Model 2		0.98	0.64, 1.50	1.21	0.81, 1.82	1.07	0.70, 1.64	1.14	0.74, 1.74	0.49
Pancreas	454
Model 1		0.73	0.55, 0.98	0.69	0.52, 0.94	0.90	0.68, 1.18	0.89	0.67, 1.18	0.93
Model 2		0.72	0.54, 0.97	0.68	0.50, 0.92	0.87	0.66, 1.16	0.86	0.64, 1.15	0.77
Prostate	2,724
Model 1		1.08	0.95, 1.22	1.07	0.95, 1.21	1.12	0.99, 1.27	1.25	1.10, 1.41	0.0002
Model 2		1.09	0.96, 1.23	1.08	0.96, 1.23	1.14	1.00, 1.29	1.28	1.13, 1.45	<0.0001

Abbreviations: CI, confidence interval; CNS, central nervous system; EGJA, esophagogastric junctional adenocarcinoma; ESCC, esophageal squamous cell carcinoma; GNCA, gastric noncardia adenocarcinoma; HCC, hepatocellular carcinoma; HR, hazard ratio.

^a Baseline serum retinol in mg/L. Quintile 1 (referent): ≤483 (n = 5,883); quintile 2: 483.1–547 (n = 5,806); quintile 3: 547.1–607 (n = 5,841); quintile 4: 607.1–685 (n = 5,792); quintile 5: >685.1 (n = 5,782).

^b Model 1 adjusted for age. Model 2 adjusted for age, α-tocopherol treatment group, β-carotene treatment group, number of cigarettes smoked per day, years of smoking, body mass index, alcohol consumption, serum α-tocopherol, serum β-carotene, and serum cholesterol. Quintile 1 is the referent.

Table 3

Association Between Baseline Serum Retinol and Overall Cancer, Stratified by Potential Effect Modifiers, in the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study, Finland, 1985–2012

Subgroupa	No. of Cases	Quintile of Baseline Serum Retinolb,c								P for Interaction
		2		3		4		5
		HR	95% CI	HR	95% CI	HR	95% CI	HR	95% CI
Trial supplementation group
β-carotene										0.62
Yes	5,441	0.91	0.84, 0.99	0.92	0.85, 1.00	0.89	0.82, 0.97	0.93	0.86, 1.01
No	5,386	0.97	0.89, 1.06	0.92	0.84, 1.00	0.95	0.87, 1.04	0.94	0.86, 1.03
α-tocopherol										0.13
Yes	5,386	0.90	0.83, 0.98	0.92	0.85, 1.00	0.87	0.80, 0.94	0.93	0.85, 1.01
No	5,441	0.98	0.90, 1.07	0.91	0.84, 0.99	0.98	0.90, 1.06	0.94	0.86, 1.03
No. of cigarettes smoked    per day										0.36
<20	3,694	0.89	0.80, 0.98	0.92	0.83, 1.01	0.92	0.83, 1.01	0.96	0.86, 1.06
≥20	7,133	0.98	0.91, 1.05	0.92	0.86, 0.99	0.92	0.86, 0.99	0.93	0.86, 1.00
No. of years smoked    regularly										0.08
<36	4,461	0.94	0.85, 1.04	0.97	0.88, 1.07	0.98	0.89, 1.08	1.02	0.93, 1.12
≥36	6,366	0.95	0.88, 1.02	0.9	0.83, 0.97	0.9	0.83, 0.97	0.9	0.83, 0.97
Body mass indexd										0.21
<26	5,649	1.00	0.92, 1.08	0.96	0.88, 1.03	0.98	0.90, 1.06	0.98	0.90, 1.07
≥26	5,178	0.88	0.80, 0.96	0.87	0.80, 0.96	0.86	0.79, 0.94	0.89	0.81, 0.97
Alcohol consumption, g										0.32
<11	5,793	0.98	0.90, 1.05	0.92	0.85, 0.99	0.96	0.89, 1.04	0.95	0.87, 1.04
≥11	5,034	0.88	0.80, 0.97	0.89	0.81, 0.97	0.84	0.77, 0.93	0.87	0.79, 0.95
Serum α-tocopherol, mg/L										0.88
<11.5	5,499	0.93	0.86, 1.00	0.92	0.85, 1.00	0.94	0.87, 1.02	0.94	0.86, 1.03
≥11.5	5,328	0.98	0.89, 1.08	0.94	0.86, 1.03	0.94	0.86, 1.03	0.98	0.89, 1.07
Serum β-carotene, μg/L										0.15
<170	5,283	0.91	0.84, 0.99	0.85	0.78, 0.92	0.9	0.83, 0.98	0.9	0.83, 0.98
≥170	5,544	0.97	0.90, 1.06	0.99	0.91, 1.07	0.94	0.86, 1.02	0.94	0.86, 1.03
Serum total cholesterol,    mmol/L										0.11
<6.14	5,449	0.89	0.82, 0.96	0.88	0.81, 0.95	0.91	0.85, 1.00	0.94	0.86, 1.02
≥6.14	5,378	1.03	0.94, 1.13	0.98	0.90, 1.08	0.96	0.87, 1.05	0.97	0.89, 1.06
Age, years										0.39
<57	4,897	0.95	0.86, 1.04	0.93	0.85, 1.03	0.95	0.87, 1.05	0.99	0.91, 1.09
≥57	5,930	0.94	0.87, 1.02	0.91	0.85, 0.99	0.9	0.83, 0.98	0.89	0.82, 0.97
Follow-up time, years										0.16
<10	4,001	0.97	0.88, 1.06	1.04	0.94, 1.14	1.03	0.93, 1.13	0.95	0.86, 1.04
≥10	6,826	1.00	0.93, 1.08	0.95	0.88, 1.02	0.98	0.91, 1.06	0.99	0.91, 1.07

Abbreviations: CI, confidence interval; HR, hazard ratio.

^a Subgroups were based on median values unless otherwise noted.

^b Baseline serum retinol in mg/L. Quintile 1 (referent): ≤483 (n = 5,883); quintile 2: 483.1–547 (n = 5,806); quintile 3: 547.1–607 (n = 5,841); quintile 4: 607.1–685 (n = 5,792); quintile 5: >685.1 (n = 5,782).

^c Adjusted for age. Quintile 1 is the referent.

^d Weight (kg)/height (m)².

Site-specific cancer

When individual cancer sites were examined, statistically significant associations were observed between serum retinol and liver, prostate, and lung cancers. Higher serum retinol was associated with higher risk of melanoma, but this was not statistically significant. Serum retinol was not associated with the other cancer sites examined.

Liver cancer

Higher baseline serum retinol was associated with a reduced risk of liver cancer (for quintile 5 vs. 1, multivariable-adjusted HR = 0.62, 95% CI: 0.42, 0.91; P for trend = 0.004) (Table 2). This finding was unchanged when cases diagnosed within 2 years of blood collection were excluded (for quintile 5 vs. 1, HR = 0.66, 95% CI: 0.44, 0.98; P for trend = 0.01) (Web Table 2) or when limited to hepatocellular carcinoma (for quintile 5 vs. 1, HR = 0.65, 95% CI: 0.40, 1.07; P for trend = 0.03). No meaningful interactions were observed between serum retinol and any of the factors examined except follow up-time (<10 years, for quintile 5 vs. 1, HR = 0.45, 95% CI: 0.24, 0.84; ≥10 years, for quintile 5 vs. 1, HR = 0.91, 95% CI: 0.55, 1.48; P for trend = 0.003) (Web Table 3).

Prostate cancer

Higher baseline serum retinol was significantly associated with increased risk of prostate cancer (for quintile 5 vs. 1, multivariable-adjusted HR = 1.28, 95% CI: 1.13, 1.45; P for trend < 0.0001) (Table 2, Web Table 1). This finding remained even when cases diagnosed within 2 years of blood collection were excluded (for quintile 5 vs. 1, HR = 1.30, 95% CI: 1.15, 1.48; P for trend < 0.0001) (Web Table 2). There were no statistically significant interactions observed with any of the factors examined (Web Table 4).

Lung cancer

Higher baseline serum retinol was associated with lower risk of lung cancer (for quintile 5 vs. 1, multivariable-adjusted HR = 0.80, 95% CI: 0.72, 0.88; P for trend < 0.0001) (Table 2, Web Table 1). This finding remained even when cases diagnosed within 2 years of blood collection were excluded (for quintile 5 vs. 1, HR = 0.80, 95% CI: 0.72, 0.89; P for trend < 0.0001) (Web Table 2). There were no statistically significant interactions observed (Web Table 5) and no differences in the association across the histological subtypes (Web Table 6).

DISCUSSION

In this study, we prospectively evaluated the association between serum retinol concentrations and overall and site-specific cancers. Although retinol status was not associated with overall cancer after adjustment for potential confounders, men with higher serum retinol were at increased risk of prostate cancer and decreased risk of lung and liver cancers.

The inverse retinol–liver cancer association is consistent with a previously published report from this cohort based on 208 cases diagnosed through 2009 (9). Two other prospective studies evaluating prediagnostic serum retinol in participants with chronic hepatitis B infection have shown elevated retinol to be associated with reduced risk of liver cancer (10, 11). Because retinol is stored in and released into circulation from the liver, low serum concentrations could result from liver abnormalities or undiagnosed liver cancer. However, we observed no change in the association between serum retinol and liver cancer risk when cases diagnosed within 2 or 5 years of blood collection were excluded, suggesting that reverse causation was not a major issue in our prospective study.

We also demonstrated an inverse risk association between serum retinol and lung cancer, consistent with a recent systematic review of prospective analyses (16). For example, an inverse association has been demonstrated in several studies (14–17, 31) including two large β-carotene chemoprevention trials, the ATBC Study (1,644 lung cancer cases diagnosed through December 31, 1998, compared with the present study with 3,924 cases) and the Beta-Carotene and Retinol Efficacy Trial (291 male and 132 female lung cancer cases). These trials showed that higher presupplementation serum retinol was associated with a lower risk of lung cancer incidence (17, 31), even though they demonstrated higher lung cancer incidence in smokers randomized to receive high-dose β-carotene or β-carotene plus vitamin A supplements (32, 33).

In contrast to lung and liver cancer, we observed a positive association between serum retinol and prostate cancer. Our findings are consistent with a pooled analysis of 15 studies (25) and with a previously-published report from the ATBC cohort based on 2,041 cases that were diagnosed through April 30, 2006 (compared with 2,724 prostate cancer cases in the current analysis) (24). A positive association between retinol and prostate cancer has also been reported in the Prostate Cancer Prevention Trial (27). One recent study of prediagnostic circulating retinol and gene fusion–positive prostate cancer (which has a poor prognosis) was null (34), and another demonstrated an inverse association (13).

Few prospective studies have examined the association between serum retinol and the other cancer sites we examined. Circulating retinol has been inversely associated with non-Hodgkin lymphoma (35), bladder cancer (36), gastric cancer (18), and kidney cancer (23) but not with risk of melanoma (37), esophageal cancer (18), or colorectal cancer (38). Further studies are needed to clarify these associations.

Studies have shown that vitamin A derivatives have both anticarcinogenic and tumorigenic properties. Cancer cells derive their energy from glycolysis rather than from oxidative phosphorylation under normal oxygen concentration. Thus, cancer cells have sufficient energy to proliferate rapidly. This phenomenon, the Warburg effect, is regarded as a hallmark of cancer cells (39). Experimental studies have shown that retinoic acid has anticancer effects through AMP-activated protein kinase suppression of glycolysis (40). Furthermore, retinoids might improve insulin resistance by activating leptin signaling pathways in hepatocellular carcinoma (41). These pathways might partly help us to understand the inverse association between serum retinol and liver cancer.

Several mechanisms have been hypothesized to explain the inconsistent role of retinoids in lung cancer. In general, it is accepted that while β-carotene and retinol at physiologic doses derived from fruit and vegetable consumption are protective, high-dose supplementation, particularly in the context of cigarette smoke exposure, is harmful. High levels of β-carotene activate cytochrome P450 enzymes, leading to increased activation of tobacco smoke precarcinogens and the formation of alternative, harmful β-carotene and retinol metabolites (42).

Our observation of a positive association between serum retinol and prostate cancer contrasts with findings from experimental studies of retinoids (43). For example, retinoic acid activates Rb, a tumor suppressor gene, in lymph node carcinoma of the prostate (LNCaP) cell culture, which reduces androgen receptor protein expression and increases apoptosis (44). It has also been proposed that retinoic acid inhibits the proliferation of prostate cancer cells by increasing p27 expression through cyclin-dependent kinase-5 up-regulation (45). While such experiments suggest anticarcinogenic properties, others indicate a role of retinol in tumor progression through increased cell proliferation and dedifferentiation (46). Furthermore, our study assumes that serum retinol is representative of tissue retinol concentrations; however, the correlation between serum retinol and intraprostatic retinol has been shown to be low and inverse (44). It is unclear whether the correlation between circulating and tissue retinol concentrations differs by organ site and whether this could partly explain the different associations observed for liver and lung cancers compared with prostate cancer.

Strengths of our study include the large cohort size and number of incident cancers, population-based case ascertainment through national registries, and a long follow-up period of more than 2 decades. The present study is, to our knowledge, the largest to examine the serum retinol association with numerous cancer sites in a single analysis. In addition, information on many potential confounders and effect modifiers was available for the analysis. Serum retinol was measured for all study participants in a single dedicated laboratory in Finland. One limitation of our study is that all the participants were male smokers, so the results might not be generalizable to other populations such as nonsmokers or women. However, smoking intensity or duration did not significantly modify or confound the association between serum retinol and cancer, and thus the results might be applicable to nonsmokers.

Our results suggest that the role of retinol in cancer risk might differ by organ site, with beneficial associations for lung and liver cancers, a harmful association for prostate cancer, and no association for cancer at other sites. Future studies exploring the role of retinoids in cancer at various sites would be useful in understanding the underlying mechanisms and clarifying the potential role of vitamin A in cancer etiology and prevention.

ACKNOWLEDGMENTS

Author affiliations: Department of Epidemiology, University of Michigan School of Public Health, Ann Arbor, Michigan (Manila Hada, Alison M. Mondul); and Metabolic Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, Maryland (Manila Hada, Stephanie J. Weinstein, Demetrius Albanes).

The ATBC Study is supported by the Intramural Research Program of the US National Cancer Institute, National Institutes of Health, Department of Health and Human Services.

Conflict of interest: none declared.