Disentangling discordant vitamin D associations with prostate cancer incidence and fatality in a large, nested case–control study

Abstract

Background

Published analyses of prostate cancer nested case–control and survival data in the Alpha-Tocopherol, Beta-Carotene Cancer Prevention (ATBC) Study cohort suggested that men with higher baseline vitamin D [25(OH)D] concentrations have both (i) increased prostate cancer risk and (ii) decreased prostate cancer-specific fatality.

Methods

To investigate possible factors responsible for a spurious association with prostate cancer fatality, we reanalysed baseline serum vitamin D associations with prostate cancer risk and prostate cancer-specific fatality in case–control data nested within the ATBC Study (1000 controls and 1000 incident prostate cancer cases). Conditional logistic regression and Cox proportion hazard models were used, respectively, to estimate odds ratios for risk and hazard ratios for prostate cancer-specific fatality, overall and by disease aggressiveness. We replicated these case–control analyses using baseline serum measurements of alpha-tocopherol (vitamin E), beta-carotene and retinol (vitamin A), and used the entire ATBC Study cohort ($n$ = 29 085) to estimate marginal associations between these baseline vitamins and prostate cancer incidence and fatality following blood collection.

Results

Vitamin D analyses agreed closely with those originally published, with opposite risk and fatality associations. By contrast, the analyses of alpha-tocopherol, beta-carotene and retinol yielded concordant associations for prostate cancer incidence and prostate cancer-specific fatality.

Conclusions

We found evidence of neither artefacts in the nested prostate cancer case–control data set nor detection or collider biases in the fatality analyses. The present findings therefore support a valid inverse (i.e. beneficial) association between vitamin D and prostate cancer-specific survival that warrants further evaluation, including possibly in controlled trials.

Key Messages.

• The analyses support a valid inverse association between vitamin D biochemical status and prostate cancer-specific fatality in the nested case–control study within the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study.

• The study does not provide evidence of sampling artefacts, detection bias or collider bias as explanations for the opposite associations between vitamin D and prostate cancer risk and prostate cancer-specific fatality.

• Deciding whether the beneficial association between vitamin D and prostate cancer-specific survival is reproducible and clinically useful requires further investigation.

Introduction

The Alpha-Tocopherol, Beta-Carotene Cancer Prevention (ATBC) Study was a controlled trial originally designed to examine the effects of daily supplementation with alpha-tocopherol (vitamin E) and beta-carotene on the incidence of cancer in male smokers. A case–control study nested within the ATBC Study found that baseline circulating vitamin D [25(OH)D] was associated with increased prostate cancer risk.¹ Intriguingly, a subsequent survival analysis of the same prostate cancer cases found that higher baseline vitamin D concentrations were associated with decreased prostate cancer-specific fatality (i.e. improved survival).²

Our goal here was to analyse and disentangle these paradoxical findings and to try to rule out possible factors that might account for them, including artefacts and biases. One hypothesis is that ‘health-conscious’ men would have prostate cancer detected at an earlier stage through increased screening and would also be more likely to take vitamin D supplements or have increased sun exposure through outdoor activity. Such a combination of lifestyle factors could bias the association between baseline vitamin D and prostate cancer fatality following a prostate cancer diagnosis. A second hypothesis is that oversampling of aggressive cases in the ATBC Study nested case–control study or some other aspect of participant sampling biased the results. A third hypothesis is that collider bias^3–7 affects the association between baseline vitamin D and survival following prostate cancer because prostate cancer diagnosis is a potential ‘collider’ that ‘follows’ vitamin D measurements and ‘precedes’ prostate cancer fatality (see Figure 1).

Figure 1.

Causal diagram illustrating collider bias. W includes confounding factors for the relationship between Vitamin D and Prostate Cancer Diagnosis or Prostate Cancer Death, such as body mass index, educational status, etc. U represents uncontrolled risk factors of Prostate Cancer Diagnosis and Prostate Cancer Death. The arrowheads from Vitamin D and from U meet in Prostate Cancer Diagnosis, which is thus a collider. Restricting the prostate cancer-specific fatality analysis to prostate cancer patients (i.e. conditioning on Prostate Cancer Diagnosis) would open the path Vitamin D $\to$ Prostate Cancer $\leftarrow$ U $\to$ Prostate Cancer Death and could induce a spurious association between Vitamin D and Prostate Cancer Death

To evaluate these hypotheses, we studied the distribution of baseline serum vitamin D concentrations by disease aggressiveness and case status in the original case–control study nested within the ATBC Study. We replicated the case–control analysis substituting baseline serum measurements of alpha-tocopherol (vitamin E), beta-carotene, and retinol (measured for the entire cohort) in order to examine the possible presence of a sampling artefact that should have led to similar contradictory conclusions for these other analytes as well. The possibility of collider bias was also investigated in the entire ATBC Study cohort by studying the associations between these three analytes and prostate cancer incidence and prostate cancer-specific fatality using the age scale beginning at age at the time of baseline blood collection (i.e. not on the scale of time-on-study since prostate cancer diagnosis).

Methods

ATBC Study

The ATBC Study was a randomized, double-blind, placebo-controlled cancer prevention trial that studied the effect of daily supplementation of alpha-tocopherol and beta-carotene in 50- to 69-year-old male cigarette smokers from Finland, all of European ancestry, who were recruited between 1985 and 1988. The trial ended in April 1993 but follow-up of the $n$ = 29 133 participants continued through the Finnish Cancer Registry and Statistics Finland. Additional details of the ATBC Study design have been published elsewhere.^8–10

Prostate cancer incidence and fatality ascertainment

Incident prostate cancer cases were identified through active follow-up during the trial period and through the Finnish Cancer Registry during and after the trial. Prostate cancer cases diagnosed in Stage 3 or 4 or with a Gleason score of ≥8 were categorized as having aggressive disease, whereas cases diagnosed in Stage 1 or 2 and with a Gleason score of <8 were categorized as having a non-aggressive disease. Cases missing both stage and Gleason score were categorized as having missing disease aggressiveness information. All deaths (including prostate cancer deaths) in the nested case–control study and in the entire ATBC Study cohort were identified through linkage with Statistics Finland.

Nested case–control study of vitamin D and prostate cancer within the ATBC Study

One thousand prostate cancer cases diagnosed up to 30 April 2005 were selected from among the 1628 cases identified at that time, including all aggressive cases. The controls were randomly selected from the ATBC Study participants who were alive and cancer-free at the time of the prostate cancer diagnosis and individually matched 1:1 to cases based on age at randomization ($\pm$1 year) and date of baseline blood collection ($\pm$30 days).

Baseline serum vitamin D, alpha-tocopherol, beta-carotene and retinol concentrations

Serum 25-hydroxyvitamin D [25(OH)D] was considered the indicator of vitamin D biochemical status and measured by using a Diasorin Liaison 25(OH)D platform with a direct, competitive chemiluminescence immunoassay (Heartland Assays, Inc.) in 2010 specifically for the nested case–control study as previously reported.¹ Serum alpha-tocopherol, beta-carotene and retinol were measured in the baseline samples for the entire ATBC Study cohort between 1985 and 1988 using high-performance liquid chromatography.

Data collection

Self-reported questionnaires about smoking, diet and medical history were completed by the participants at baseline. They also underwent height and weight measurements, with body mass index (BMI) calculated as weight in kilograms/height (metres)². Information on family history of prostate cancer was obtained during follow-up. A third category (missing) was created, as family history of prostate cancer was missing for some participants.

Statistical analyses

Analyses of analyte concentrations and associations with prostate cancer risk in the nested case–control study

The distribution of the continuous season-standardized 25(OH)D concentration in the nested case–control data was studied in the six groups defined by disease aggressiveness (aggressive, non-aggressive, missing) and case status (case, control). Each control was assigned to the same disease aggressiveness category as their matched case. The season-standardized 25(OH)D values were obtained by subtracting the season-specific mean and dividing by the season-specific standard deviation, which were computed from the controls whose blood was drawn during the ‘dark season’ (November to April) or during the ‘sunny season’ (May to October).

Conditional logistic regression was used to estimate odds ratios (ORs) and 95% CIs of prostate cancer diagnosis by season-specific categories of baseline 25(OH)D based on season-specific quintiles of the distributions of baseline 25(OH)D in the controls ($\le$16.3, >16.3 to $\le$23.8, >23.8 to $\le$33.3, >33.3 to $\le$45.6, >45.6 nmol/L for the dark season and $\le$25.9, >25.9 to $\le$35.7, >35.7 to $\le$48.3, >48.3 to $\le$59.9, >59.9 nmol/L for the sunny season). The logistic regression model was conditioned on the matched pairs and adjusted for age at blood collection, family history of prostate cancer (yes, no, missing), BMI (<25 or $\ge$25 kg/m²), height, marital status (married, not married/divorced/widowed), educational status (did not graduate from high school/secondary school/college, high-school graduate or higher), number of cigarettes smoked per day and number of years smoked. The analysis was then replicated in strata defined by disease aggressiveness. Trends across 25(OH)D categories were tested by creating an ordinal variable that was treated as continuous in the conditional logistic regression models.

The unstratified analysis was replicated in the same cases and controls to investigate associations between prostate cancer risk and categories of baseline alpha-tocopherol, beta-carotene and retinol concentrations. The categories were based on the quintiles of the baseline distribution of each analyte in the entire ATBC Study ($\le$9.3, >9.3 to $\le$10.8, >10.8 to $\le$12.2, >12.2 to $\le$14.2, >14.2 mg/L for alpha-tocopherol; $\le$98, >98 to $\le$145, >145 to $\le$200, >200 to $\le$289, >289 μg/L for beta-carotene; $\le$483, >483 to $\le$547, >547 to $\le$607, >607 to $\le$685, >685 $\mathrm{\mu }$g/L for retinol).

Prostate cancer fatality in the 1000 cases included in the nested case–control study

A Cox proportional-hazards model was used to estimate hazard ratios (HRs) and CIs of prostate cancer-specific fatality from date of diagnosis in the nested case–control study ($n$ = 1000). HRs were computed for season-specific categories of baseline 25(OH)D based on the season-specific quintiles in the controls. Time of follow-up was computed from the prostate cancer diagnosis date to the earliest of date of prostate cancer death, death from other causes or censoring at end of follow-up (31 December 2016). Deaths from causes other than prostate cancer were treated as independent censoring events in the analysis. The Cox proportional-hazards model used the time since prostate cancer diagnosis scale and included the covariable factors used for the risk analysis. The analysis was also performed within strata defined by disease aggressiveness.

The unstratified analysis was replicated to investigate the associations between prostate cancer-specific fatality and categories of baseline alpha-tocopherol, beta-carotene and retinol concentrations. The categories were based on the quintiles of the baseline distribution of each analyte in the entire ATBC Study cohort. As for 25(OH)D, the analyses of HRs were on the time since prostate cancer diagnosis scale and adjusted for the same covariable factors.

Vitamin analyses of prostate cancer incidence and fatality in the entire ATBC Study cohort

Cox proportional-hazards models were used to estimate HRs and CIs of prostate cancer incidence by categories of baseline alpha-tocopherol, beta-carotene and retinol concentrations in the entire ATBC Study ($n$ = 29 085). The categories were based on the quintiles of the baseline distribution of each analyte in the entire ATBC Study. Entry age was defined as age at blood collection and exit age as the earliest of age at prostate cancer diagnosis, death or censoring at end of follow-up in 2016. Deaths were treated as independent censoring events. The Cox proportional-hazards models were on the age scale and adjusted for family history of prostate cancer, BMI, height, marital status, educational status, number of cigarettes smoked per day and number of years smoked.

Cox proportional-hazards models on the age scale were also used to estimate HRs for prostate cancer fatality from age at blood collection. The same procedures and adjustment as for prostate cancer incidence were used, with exit age defined as the earliest of age at prostate cancer death, death from other causes or censoring at end of follow-up in 2016. Deaths from non-prostate cancer causes were treated as independent censoring.

Results

Baseline characteristics of the cases and controls in the nested set and of the 1000 cases by season-specific quintiles of baseline 25(OH)D are shown in Supplementary Tables 1 and 2 (available as Supplementary data at IJE online), respectively.

Analyses of analyte concentrations and associations with prostate cancer risk in the nested case–control study

Aggressive cases had only slightly lower concentrations of season-standardized baseline 25(OH)D as compared with non-aggressive cases (Table 1), which is contrary to the hypothesis that health-conscious men are more likely to have both higher vitamin D status and prostate cancer diagnosed at earlier, less aggressive stages of disease.

Table 1.

Distribution of ‘continuous’ season-standardized^a 25(OH)D in the six groups defined by disease aggressiveness and case status in the case–control study nested within the ATBC Study cohort

Disease aggressiveness	Case status	n	Mean	SD	25th percentile	Median	75th percentile
Aggressive	Case	512b	0.02	1.01	–0.75	–0.18	0.60
Non-aggressive	Case	444	0.10	0.97	–0.62	–0.05	0.70
Missing	Case	43	0.21	1.07	–0.70	–0.08	0.78
Aggressivec	Control	513	–0.04	1.05	–0.80	–0.26	0.52
Non-aggressivec	Control	444	0.07	0.95	–0.64	–0.06	0.67
Missingc	Control	43	–0.21	0.83	–0.91	–0.07	0.27

25(OH)D, vitamin D; ATBC, Alpha-Tocopherol, Beta-Carotene Cancer Prevention; n, number of individuals (cases and controls); SD, standard deviation.

^aStandardization used the season-specific mean and SD values obtained from the controls whose blood was drawn during the ‘dark season’ or ‘sunny season’.

^bOne prostate cancer case did not have a 25(OH)D measurement.

^cEach control was assigned to the same disease aggressiveness category as their matched case.

Prostate cancer risk was positively associated with the highest quintile category of season-specific 25(OH)D in the unstratified analysis (OR = 1.44 for Q5 vs Q1; 95% CI: 1.06, 1.96; P_trend = 0.07; Table 2). The ORs for prostate cancer diagnosis by season-specific categories of baseline 25(OH)D appeared slightly stronger for aggressive prostate cancer than for non-aggressive prostate cancer.

Table 2.

Adjusted odds ratios of prostate cancer risk in the case–control study nested within the ATBC Study cohort by baseline 25(OH)D quintile categories

	Stratified by disease aggressiveness
	Aggressive	Non-aggressive	Unstratified
25(OH)D	OR e  (95% CI)	OR e  (95% CI)	OR e  (95% CI)
n (cases/controls)	512/512a,b	444/444a,b	999/999a,c,d
Quintile categoryf
Q1	Reference	Reference	Reference
Q2	1.20 (0.80–1.81)	1.00 (0.61–1.64)	1.27 (0.94–1.71)
Q3	1.10 (0.73–1.66)	1.43 (0.90–2.30)	1.31 (0.98–1.77)
Q4	1.28 (0.82–1.98)	0.95 (0.59–1.55)	1.17 (0.86–1.59)
Q5	1.45 (0.95–2.23)	1.13 (0.70–1.84)	1.44 (1.06–1.96)
P trend g	0.10	0.68	0.07

25(OH)D, vitamin D; ATBC, Alpha-Tocopherol, Beta-Carotene Cancer Prevention; BMI, body mass index; n, number of individuals (prostate cancer cases and controls); OR, odds ratio.

^aSome prostate cancer cases had missing stage and/or Gleason score.

^bEach control was assigned to the same disease aggressiveness category as their matched case.

^cOne prostate cancer case did not have a 25(OH)D measurement.

^dOne prostate cancer case was missing height and BMI measurements, with those values being imputed to the respective variable medians of the other cases.

^eLogistic regression models conditioned on matching factors and adjusted for age at blood collection + BMI + height + marital status + educational status + number of cigarettes smoked per day + number of years smoked + family history of prostate cancer.

^fQuintile categories based on quintiles of season-specific 25(OH)D in the controls.

^gTrends across categories assessed by assigning an ordinal variable to the quintile categories and treating it as a continuous variable in the models.

Men in the highest quintile category of baseline alpha-tocopherol were less likely to develop prostate cancer during follow-up than men in the lowest quintile category (OR = 0.66 for Q5 vs Q1; 95% CI: 0.49, 0.91; P_trend = 0.04; Table 3). There was no association between high baseline beta-carotene and increased or reduced prostate cancer risk. The ORs for retinol suggested a moderate positive association with prostate cancer risk (OR = 1.22 for Q5 vs Q1; 95% CI: 0.91, 1.63; P_trend = 0.10; Table 3).

Table 3.

Adjusted odds ratios of prostate cancer risk in the case–control study nested within the ATBC Study cohort by baseline alpha-tocopherol, beta-carotene and retinol quintile categories

Vitamin	OR b  (95% CI)	P trend c
n (cases/controls)	1000/1000a
Alpha-tocopherol quintile categoryd
Q1	Reference
Q2	0.83 (0.61–1.12)
Q3	0.80 (0.60–1.07)
Q4	0.90 (0.67–1.20)
Q5	0.66 (0.49–0.91)	0.04
Beta-carotene quintile categoryd
Q1	Reference
Q2	1.20 (0.88–1.64)
Q3	0.97 (0.72–1.30)
Q4	1.09 (0.81–1.46)
Q5	0.92 (0.69–1.23)	0.40
Retinol quintile categoryd
Q1	Reference
Q2	0.98 (0.73–1.31)
Q3	1.31 (0.97–1.76)
Q4	1.16 (0.86–1.56)
Q5	1.22 (0.91–1.63)	0.10

ATBC, Alpha-Tocopherol, Beta-Carotene Cancer Prevention; BMI, body mass index; n, number of individuals (prostate cancer cases and controls); OR, odds ratio.

^aOne prostate cancer case was missing height and BMI measurements, with those values being imputed to the respective variable medians of the other cases.

^bLogistic regression models conditioned on matching factors and adjusted for age at blood collection + BMI + height + marital status + educational status + number of cigarettes smoked per day + number of years smoked + family history of prostate cancer.

^cTrends across categories assessed by assigning an ordinal variable to the quintile categories and treating it as a continuous variable in the models.

^dVitamin quintile categories based on the ATBC Study cohort.

Prostate cancer fatality among the 1000 cases included in the nested case–control study

Higher baseline 25(OH)D was associated with reduced prostate cancer-specific fatality from time of diagnosis (HR = 0.87 for Q4 vs Q1; 95% CI: 0.41, 0.83 and HR = 0.73 for Q5 vs Q1; 95% CI: 0.53, 1.01; P_trend = 0.01; Table 4). The HRs in the analysis of all 1000 cases were closer to the HRs in the analysis restricted to the aggressive prostate cancer cases than to those of the non-aggressive cases, probably in part because there were somewhat more deaths among the aggressive prostate cancer cases. In the stratified analysis, the overall inverse association was present for both aggressive and non-aggressive prostate cancer but appeared stronger for the latter cases (HRs = 0.90 and 0.31 for Q5 vs Q1; 95% CIs: 0.63, 1.30 and 0.13, 0.73; P_trend = 0.22 and 0.02, respectively).

Table 4.

Adjusted hazard ratios of prostate cancer-specific fatality from time of diagnosis in men diagnosed with prostate cancer in the case–control study nested within the ATBC Study cohort by baseline 25(OH)D quintile categories

	Stratified by disease aggressiveness
	Aggressive	Non-aggressive	Unstratified
25(OH)D	HR d  (95% CI)	HR d  (95% CI)	HR d  (95% CI)
n (deaths/cases)	299/512a	61/444a	369/999a,b,c
Quintile categorye
Q1	Reference	Reference	Reference
Q2	0.86 (0.60–1.22)	0.69 (0.30–1.59)	0.85 (0.62–1.18)
Q3	0.91 (0.63–1.32)	0.47 (0.21–1.04)	0.67 (0.48–0.94)
Q4	0.60 (0.40–0.89)	0.75 (0.34–1.63)	0.58 (0.41–0.83)
Q5	0.90 (0.63–1.30)	0.31 (0.13–0.73)	0.73 (0.53–1.01)
P trend f	0.22	0.02	0.01

25(OH)D, vitamin D; ATBC, Alpha-Tocopherol, Beta-Carotene Cancer Prevention; BMI, body mass index; HR, hazard ratio; n, number of individuals (prostate cancer cases and prostate cancer deaths).

^aSome prostate cancer cases were missing stage and/or Gleason score.

^bOne prostate cancer case did not have a 25(OH)D measurement.

^cOne prostate cancer case was missing height and BMI measurements, with those values being imputed to the respective variable medians of the other cases.

^dCox proportional hazard regression models using the time-on-study scale⁺ and adjusted for age at diagnosis + age at diagnosis² + BMI + height + marital status + educational status + number of cigarettes smoked per day + number of years smoked + family history of prostate cancer. (⁺ Time of follow-up computed from the date of diagnosis of prostate cancer to the earliest of date of prostate cancer death, death from other causes or censoring at end of follow-up in 2016. Deaths from causes other than prostate cancer were treated as independent censoring events in the analysis.)

^eQuintile categories based on quintiles of season-specific 25(OH)D in the controls.

^fTrends across categories assessed by assigning an ordinal variable to the quintile categories and treating it as a continuous variable in the models.

There was no evidence of an association between high baseline alpha-tocopherol, beta-carotene or retinol and prostate cancer fatality, although the HRs for retinol suggested that men in the highest quintile category were more likely to die from prostate cancer (Table 5).

Table 5.

Adjusted hazard ratios of prostate cancer-specific fatality from time of diagnosis in men diagnosed with prostate cancer in the case–control study nested within the ATBC Study cohort by baseline alpha-tocopherol, beta-carotene and retinol quintile categories

Vitamin	HR b  (95% CI)	P trend c
n (deaths/cases)	370/1000a
Alpha-tocopherol quintile categoryd
Q1	Reference
Q2	1.12 (0.82–1.54)
Q3	0.80 (0.57–1.13)
Q4	0.98 (0.70–1.36)
Q5	1.17 (0.84–1.64)	0.71
Beta-carotene quintile categoryd
Q1	Reference
Q2	1.07 (0.76–1.50)
Q3	0.93 (0.66–1.31)
Q4	0.86 (0.62–1.21)
Q5	1.00 (0.72–1.40)	0.63
Retinol quintile categoryd
Q1	Reference
Q2	1.08 (0.76–1.54)
Q3	1.52 (1.09–2.12)
Q4	1.16 (0.82–1.64)
Q5	1.22 (0.87–1.72)	0.25

ATBC, Alpha-Tocopherol, Beta-Carotene Cancer Prevention; BMI, body mass index; HR, hazard ratio; n, number of individuals (prostate cancer cases and prostate cancer deaths).

^aOne prostate cancer case was missing height and BMI measurements, with those values being imputed to the respective variable medians of the other cases.

^bCox proportional hazard regression models using the time-on-study scale⁺ and adjusted for age at diagnosis + age at diagnosis² + BMI + height + marital status + educational status + number of cigarettes smoked per day + number of years smoked + family history of prostate cancer. (⁺ Time of follow-up computed from the date of diagnosis of prostate cancer to the earliest of date of prostate cancer death, death from other causes or censoring at end of follow-up in 2016. Deaths from causes other than prostate cancer were treated as independent censoring events in the analysis.)

^cTrends across categories assessed by assigning an ordinal variable to the quintile categories and treating it as a continuous variable in the models.

^dVitamin quintile categories based on the ATBC Study cohort.

Vitamin analyses of prostate cancer incidence and fatality in the entire ATBC Study cohort

Men in the highest quintile categories of baseline retinol were more likely to develop prostate cancer during follow-up (HR = 1.25 for Q5 vs Q1; 95% CI: 1.11, 1.41; P_trend < 0.001; Table 6). There were no associations between prostate cancer incidence for either baseline alpha-tocopherol or beta-carotene.

Table 6.

Adjusted hazard ratios of prostate cancer incidence and prostate cancer-specific fatality from age at blood collection in the ATBC Study cohort by baseline alpha-tocopherol, beta-carotene and retinol quintile categories

	Incidence		Fatality
Vitamin	HR d  (95% CI)	P trend f	HR e  (95% CI)	P trend f
Alpha-tocopherol
n/totalg	2923/29 102b,c		704/29 102b,c
Quintile categorya
Q1	Reference		Reference
Q2	0.98 (0.87–1.10)		1.20 (0.95–1.52)
Q3	1.05 (0.94–1.18)		0.96 (0.75–1.23)
Q4	0.99 (0.88–1.11)		1.00 (0.78–1.27)
Q5	0.90 (0.80–1.02)	0.15	0.99 (0.77–1.26)	0.37
Beta-carotene
n/totalg	2923/29 103b,c		704/29 103b,c
Quintile categorya
Q1	Reference		Reference
Q2	0.98 (0.86–1.11)		0.95 (0.75–1.22)
Q3	0.96 (0.85–1.08)		0.86 (0.67–1.11)
Q4	1.04 (0.92–1.17)		0.85 (0.67–1.10)
Q5	0.99 (0.87–1.11)	0.74	0.92 (0.73–1.18)	0.39
Retinol
n/totalg	2923/29 104b,c		704/29 104b,c
Quintile categorya
Q1	Reference		Reference
Q2	1.08 (0.96–1.22)		1.12 (0.87–1.44)
Q3	1.08 (0.96–1.22)		1.25 (0.98–1.60)
Q4	1.11 (0.98–1.25)		1.20 (0.94–1.54)
Q5	1.25 (1.11–1.41)	<0.001	1.38 (1.08–1.76)	0.01

ATBC, Alpha-Tocopherol, Beta-Carotene Cancer Prevention; BMI, body mass index; HR, hazard ratio; n, number of individuals (prostate cancer cases, prostate cancer deaths and entire cohort included in the analysis).

^aVitamin quintile categories based on the ATBC Study cohort.

^bThirty-one, 30 and 29 men in the ATBC Study cohort were missing alpha-tocopherol, beta-carotene and retinol measurements, respectively.

^cNineteen men in the ATBC Study cohort were missing height and BMI measurements, with those values being imputed to the respective variable medians of the other cases.

^dCox proportional hazard regression models for prostate cancer diagnosis on the age scale⁺ and adjusted for BMI + height + marital status + educational status + number of cigarettes smoked per day + number of years smoked + family history of prostate cancer. (⁺Entry age defined as age at blood collection and exit age as the earliest of age at prostate cancer diagnosis, death or censoring at end of follow-up in 2016. Deaths were treated as independent censoring events in the analysis.)

^eCox proportional hazard regression models for prostate cancer death on the age scale⁺⁺ and adjusted for BMI + height + marital status + educational status + number of cigarettes smoked per day + number of years smoked + family history of prostate cancer. (⁺⁺Entry age defined as age at blood collection and exit age as the earliest of age of prostate cancer death, death from other causes or censoring at end of follow-up in 2016. Deaths from causes other than prostate cancer were treated as independent censoring events in the analysis.)

^fTrends across categories assessed by assigning an ordinal variable to the quintile categories and treating it as a continuous variable in the models.

^g n is cases/total for incidence and deaths/total for fatalities.

A positive association was also found between prostate cancer-specific fatality from age at blood collection and the highest quintile category of serum retinol (HR = 1.38 for Q5 vs Q1; 95% CI: 1.08, 1.76; P_trend = 0.01). Neither alpha-tocopherol nor beta-carotene at baseline had demonstrable associations with prostate cancer fatality from age at blood collection to age at death.

Discussion

We reran previously published analyses and conducted additional analyses in the ATBC Study nested case–control data to investigate biochemical vitamin D associations with prostate cancer risk and prostate cancer-specific fatality, using time-on-study from prostate cancer diagnosis as the scale for the latter. When we examined the distribution of the continuous, season-standardized 25(OH)D baseline serum concentrations by disease aggressiveness and case status, we observed no differences that would provide evidence to support a detection bias relevant to prostate cancer diagnosis or fatality. Indeed, contrary to such a hypothesis, aggressive cases only had slightly ‘lower’ baseline 25(OH)D concentrations compared with non-aggressive cases. In contrast to 25(OH)D, none of the associations of baseline serum alpha-tocopherol, beta-carotene or retinol was discordant for prostate cancer risk and fatality in the nested case–control study. These findings suggest that the opposite associations for 25(OH)D were not an artefact of case–control sampling. In addition, the directions of the associations that we found for alpha-tocopherol, beta-carotene and retinol were consistent with those from previous analyses of the ATBC Study and of the 1891 prostate cancer cases diagnosed.^11–13 Analyses of the entire ATBC Study cohort to investigate associations between these three analytes and prostate cancer incidence and fatality from age at blood collection also did not lead to discordant results, with all the associations having similar directionality in the cohort and case–control analyses. These findings argue against the presence of collider bias in prostate cancer-specific fatality associations in the nested case–control data. We were unable to replicate the full cohort analyses with 25(OH)D, however, as the measurements were only collected from individuals in the nested case–control sample.

The opposite 25(OH)D associations with prostate cancer risk and fatality are paradoxical and intriguing. Our present findings agree qualitatively with previous publications from the ATBC Study.^1,2 Similar associations have also been found more recently in prospective studies for fatality, in a large cohort¹⁴ and for overall risk in a large multinational pooling project of ∼13 000 case–control sets,¹⁵ even though stratification by disease aggressiveness suggested some heterogeneity between the ATBC Study and the pooling project as well as within each analysis. Notably, the nested case–control study in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening (PLCO) trial was less susceptible to detection bias, as men followed a fixed protocol for prostate cancer screening during the trial period,¹⁴ in contrast to the ATBC Study, which had none. The present findings therefore suggest an inverse association between vitamin D and prostate cancer-specific fatality that may have potential therapeutic implications. Further observational studies and perhaps intervention trials are warranted to explore this possibility.

Intervention trials could directly address whether vitamin D supplementation in recently diagnosed men improves prostate cancer-specific survival compared with placebo. Such controlled trials would provide estimates of the effects of specific vitamin D treatment regimens and would avoid the inherent weaknesses of observational studies. These include, for example, the fact that our baseline vitamin D measurements were based on blood samples taken long before the prostate cancer diagnoses, that the ATBC Study recruited only White participants of European ancestry and the possibility of unmeasured confounding. Two trials have looked at the effects of vitamin D on prostate-specific antigen and/or prostate cancer progression^16,17 and suggested that vitamin D supplementation may benefit patients with non-aggressive prostate cancer. Another trial of vitamin D₃ supplementation for low-risk prostate cancers is ongoing.¹⁸ A convincing controlled trial of vitamin D supplementation and prostate cancer-specific fatality would need to be large, long or both in order to demonstrate protective effects as indicated by the relative hazards in Table 4, because the association appears to be of modest magnitude and the absolute hazard of death from prostate cancer is small, except for those with advanced and very aggressive disease. For example, under the assumption that a vitamin D supplement regimen raised all cases into the fifth quintile group of Table 4 and that a placebo group had a hazard that corresponded to the third quintile category of the present study, one would need to detect a hazard ratio of 0.751/0.816 = 0.920. To have power 0.9 or 0.8 at a two-sided significance alpha-level of 0.05 would require a trial with 6100 or 4552 prostate cancer-specific deaths (respectively).

In the ATBC Study, previously published analyses studying the association between vitamin D and prostate cancer fatality were restricted to patients diagnosed with prostate cancer, with vitamin D measurements obtained prior to diagnosis (see Figure 1). Our hypothesis was therefore that such an analysis could induce collider bias and lead to the paradoxical vitamin D associations with prostate cancer incidence and fatality. In addition, although we did not find evidence to support such bias, other colliders could have also been at play because of competing events.^7,19 Indeed, death from other causes can precede prostate cancer diagnosis and prostate cancer death, and bias could also be present in the analyses of prostate cancer risk. This raises the possibility of a separate study analysis that is focused on these complex issues that are beyond the scope of the present work.

Ethics approval

The ATBC Study was approved by the institutional ethics review committees of the US National Cancer Institute and the National Public Health Institute of Finland. The trial was registered as Clinical Trials.gov number NCT00342992 (ClinicalTrials.gov).

Data availability

Data are maintained by the National Cancer Institute, Division of Cancer Epidemiology and Genetics, and are available to bona fide researchers upon submission and approval of a research proposal and subsequent completion of a Data Transfer Agreement.

Supplementary data

Supplementary data are available at IJE online.

Author contributions

Conceptualization, L.E., D.A. and M.G.; writing- original draft preparation, L.E.; writing- review and editing L.E., D.A. and M.G. All authors have read and agreed to the published version of the manuscript.

Funding

This research was supported by the Intramural Research Program of the Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, USA.

Conflict of interest

None declared.