Longitudinal biobanks-based study on the joint effects of infections, nutrition and hormones on risk of prostate cancer

Abstract

Background To evaluate the individual and combined effects of enterolactone, vitamin D, free testosterone, Chlamydia trachomatis and HPV-18 on the risk of prostate cancer in a large population-based biochemical material that combined three Nordic serum sample banks. Material and methods A joint cohort of 209 000 healthy men was followed using cancer registry linkages. From this cohort altogether 699 incident cases of prostate cancer were identified. Four controls were selected by incidence density sampling and matching for country, age and date of the blood sampling. Complete data for all investigated exposures was available for 483 eligible cases and 1055 eligible controls. Multivariate regression analyses were performed to investigate the solitary and combined effects. Results The solitary effects were small. Significantly increased risk [rate ratio 1.6 (95% CI 1.0–2.5)] was found in those seronegative for C. trachomatis infection. The joint effect in risk levels of enterolactone and vitamin D was antagonistic [observed rate ratio (RR) 1.4 (1.0–2.1), expected RR 2.0 (1.0–4.1)] as well as that of HPV-18 and C. trachomatis [observed RR 1.9 (0.8–4.5), expected RR 9.9 (1.1–87.0)]. Conclusion A large follow-up study combining data from several previously investigated exposures to investigate joint effects found no evidence that exposure to two risk factors would increase the risk of prostate cancer from that expected on basis of exposure to one risk factor. If anything, the results were consistent with antagonistic interactions.

Causes of prostate cancer pose many challenges to the epidemiological research. A large hereditary component has been identified [1]. However, there is also a substantial increase in the risk of the disease over time [2]. The diagnosis of occult lesions in asymptomatic men by the increased use of PSA-test [3] may explain part of the increasing incidence in prostate cancer over time; however, environmental factors may also play a role.

The environmental causes of prostate cancer are not well known. Dietary factors including vitamins, trace elements and plant estrogens are extensively studied, but their interactions are less known [4]. Prostate cancer is commonly regarded a hormonal dependent disease [5]. Infectious causes of cancerous diseases is an area with many breakthroughs in cancer etiology, such as oncogenic HPVs and cervical cancer as well as and Helicobacter pylori and stomach cancer [6]. We have previously analyzed separately the effects of serum biomarkers on prostate cancer risk, including 25(OH)-vitamin D [7], enterolactone [8], free testosterone [9], and infections HPV-18 [10] and Chlamydia trachomatis [11], but without considering any confounding or joint effects between these exposures. If the exposures causing the changes in the human genome are independent the joint effect (of them) on the risk of cancer is likely to be multiplicative [12]. Therefore, in the present analysis the agreement of the observed and expected joint effect assuming multiplicative interaction was considered.

Large serum sample banks allow prospective studies including evaluation of the individual and joint effects of risk factors. The Nordic countries have a long tradition of molecular epidemiological studies based on linking large scale biobanks and comprehensive, population-based cancer registries. A major advantage with coordinated biobanks-based studies is that it is possible to perform joint analysis on several individual serum biomarkers to analyze whether they may confound or modify the individual associations with prostate cancer incidence. Here, we report the results based on a Nordic study that identified 483 prospectively occurring prostate cancer cases and 1055 controls.

Material and methods

In our study, we used follow-up data from population-based cohorts of 209 000 men who had donated blood samples stored in biobanks in Finland, Norway and Sweden.

Cohorts

In Finland, the cohort consisted of approximately 19 000 men who attended the first screening visit within the Helsinki Heart Study. In this clinical trial the effect of a drug modulating lipid levels was investigated [13]. The participants, middle-aged employees in two governmental agencies and five industrial companies, were recruited in 1981 and 1982. A blood sample was drawn from the participants in the morning and serum samples were stored at −20 °C.

The Janus Serum Bank in Norway was established in 1973 and enrolled a population-based sample of about 165 000 men [14]. Most donors (91%) were invited to participate in nationwide health examinations mostly for cardiovascular diseases. A minority (9%) of donors were recruited from the Red Cross Blood Donor Center in Oslo. The average age at enrollment for donors was 41 years in the health examination group and 34 years in the Red Cross blood donor group. The blood collection took place during office hours and serum samples were stored at −25 °C.

In Sweden, men in The Northern Sweden Health and Disease Cohorts were recruited through the Västerbotten Intervention Project (VIP) and the Northern Sweden part of WHO MONICA study [15]. The VIP started in 1985. Each year all residents of Västerbotten county at the ages 40, 50, and 60 years were invited by letter to participate in a health promoting project with the aim of reducing cardiovascular disease and cancer. In the present study participants from the surveys performed in 1990 and 1994 were used. Together these two projects have recruited about 30 000 men used in the present study as a representative population sample from the counties of Västerbotten and Norrbotten. For the large majority of participants, blood collection took place in the morning and heparin plasma samples were stored at −80°C.

All participants signed an informed consent form and the studies were approved by each respective local Research Ethical Committee in Finland and Sweden. During the Janus Serumbank’s first years of operation the donors gave their broad consent for their samples to be used for ‘cancer research’. Samples donated in 1997 and onwards are collected based on an explicit informed consent.

Case ascertainment and control selection

All incident cases of prostate cancer and all cases of death were identified through linkage with national cancer and mortality registers in 1997, using a nationwide individual identification number as the identity link. If several samples were available for the same case subject, the first sample was chosen. Four control subjects for each case subject were selected within each cohort from all members alive and free of cancer at the time of diagnosis of the case, by matching on age (±2 years) and date (±2 months) of the blood sampling and county in Norway. Finnish samples were also matched for thawing, as some of the frozen samples had been accidentally thawed during storage.

Controls were randomly selected within each group of eligible subjects if more than four subjects matched to the case were identified. If less than four control subjects were found, the date of sampling was widened up to six months. If no suitable control subjects were found in this procedure, less than four control subjects were accepted.

In this joint analysis, the case or control subject was excluded if any of the risk factors under study remained not analyzed or unknown. Consequently, a total of 135 case subjects were available for analysis in the Finnish cohort. Four controls were available for 92 case subjects, three for 26, two for 16, and one control for one case subject. In Norway, 480 case subjects were available for analysis. Four controls were available for 115 case subjects, three for 177, two for 144 and one for 44 case subjects. In the Swedish cohort, 84 case subjects were available for analysis; four controls were available for 78 case subjects and three for six case subjects. Thus, 699 case subjects and 2132 matched control subjects were available for analysis in the whole study group.

Biochemical assays

Coded samples were transported on dry ice to a coordinating center at the Karolinska Institute, Stockholm, where the samples were thawed, aliquoted, refrozen, and transported on dry ice to the collaborating laboratories. Throughout the study, and still to this day, the laboratory personnel were blinded as to the case control status of the samples. Samples pertaining to matched study subjects were always analyzed together in the same batch (i.e. on the same day and within the same run).

Enterolactone

The enterolactone samples were analyzed in the Institute for Preventive Medicine, Nutrition and Cancer, Folkhälsan Research Center, Helsinki. The concentration of enterolactone was measured with time-resolved fluoroimmunoassay (TR-FIA), a method that has been described in detail earlier with regard to sensitivity, specificity, accuracy and precision [16]. The working range of the assay was 8.9–3218 pg/20 μl, corresponding to plasma levels of 1.5–540 nmol/l. All the batches of samples included three quality control samples. The inter-assay coefficients of variation (CV) were 11.1% (21 nmol/l), 11.0% (47 nmol/l) and 9.8% (101 nmol/l). The average of the intra-assay CVs was 10.3%.

Vitamin D

Serum concentrations of 25 hydroxyvitamin D₃/D₂ were analyzed by radioimmunoassay [Incstar, Stillwater, MN, USA] at the University of Tampere, Department of Biomedical Sciences. The coefficients of intra- and inter-assay variations for the 25(OH)-vitamin D assay were 8.5% and 16%, respectively.

Testosterone and sex hormone binding globulin assays

The samples were analyzed in the Department of Clinical Chemistry, Helsinki University Hospital. Testosterone was quantitated by a time-resolved fluoroimmunoassay [DELFIA®, Wallac, Turku, Finland]. The detection limit of the assay was 0.3 nmol/l. Intra- and inter-assay CV at 2.4 nmol/l were 6.0% and 12.9%, at 10 nmol/l 5.5% and 6.8% and at 28 nmol/l 5.6% and 5.5%, respectively. Sex hormone binding globulin (SHBG) was quantitated by a time-resolved immunofluorometric assay [AutoDELFIA™, Wallac, Turku, Finland]. Samples were diluted 1:100 prior to assay. The detection limit of the assay was 0.5 nmol/l. Intra- and inter-assay CV at 20 nmol/l were 1.4% and 8.2%, at 57 nmol/l 1.3% and 5.3%, and at 13 nmol/l 1.8% and 10.1%, respectively. An index of free testosterone was calculated using a mass action equation based on the affinity of SHBG and albumin for testosterone, assuming a constant serum albumin concentration for all men [17]. Samples pertaining to matched study subjects were always analyzed together in the same batch (i.e. on the same day and within the same run).

Serology for Chlamydia trachomatis and HPV-18

The samples were analyzed at the University of Oulu, Department of Medical Microbiology, for C. trachomatis-IgG antibodies by the MIF method, which is considered a gold standard for chlamydial serology [18]. Elementary bodies of pooled serovars BED (B-group), CJHI (C-group), and GFK (intermediate group) of C. trachomatis, (Washington Research Foundation, Seattle, WA, USA) were used as antigens. Fluorescein isothiocyanate (FITC)-conjugated antihuman IgG (Kallestad, Chaska, MO, USA) was used as conjugate. Titers of ≥16 were considered positive. The antibody determinations of each case and the individual controls were always done simultaneously in the same titration series in a blinded fashion. All the sera were tested by one laboratory technician and fluorescence pattern was interpreted by one experienced reader.

Seropositivity for HPV-18 was determined at Lund University by the standard ELISA assay, using baculovirus-expressed disrupted capsids of bovine papillomavirus as negative control. The capsids were obtained from Dr. Martin Sapp, University of Mainz, Germany. Cut-off level of 0.100 absorbance units were used [10].

Statistical analyses

First, we analyzed the data among the total 699 cases and 2132 controls. Next, we used two exclusion criteria for the restricted material. In the Finnish cohort 52 cases and 197 controls were accidentally thawed and they were excluded from the analyses using restricted material. Due to large seasonal variation of levels of vitamin D, we applied stringent criteria for matching cases and controls. We required that the blood sample of the control was taken during the same season as the blood sample of the case. Seasons were classified as winter (December, January, February), spring (March, April, May), summer (June, July, August), and fall (September, October, November). However, the same stringent rules were applied with all explanatory variables, not only vitamin D, when using restricted material.

To obtain rate ratios (RR) of prostate cancer by explanatory variables, we used conditional logistic regression analysis taking the case control matching strategy into account. We estimated risks by two levels of these variables. For each of the risk factors we elected the group with lowest risk as the reference group. If the estimated RR in the reference groups (low risk) was zero, the high risk group was elected as the reference group to avoid infinite RR. Further, the large random variation with any zero number of observations makes the RRs non-informative.

For enterolactone, as a cut-off point we used the first tertile concentrations in all controls, 5.56 nmol/l. A previous study using the same dataset showed a U-shaped prostate cancer risk for vitamin D [7]. Thus, the reference level for vitamin D in this study was based on the second tertile of the vitamin D concentration of the total original cohort which were rounded to 40 and 60 nmol/l. We combined the risk levels (<40 and ≥60 nmol/l) as the comparison group. Cut-off point for free testosterone was the second tertile in all controls (337.67 pmol/l). The group with lower risk of prostate cancer served as a reference group for the RRs. Univariate RRs for the two levels of explanatory variables were calculated for the total as well as for the restricted material. We call these RRs crude even if they were matched for age, date of blood sampling and county in Norway.

Using multivariable regression analyses, we also mutually adjusted the RRs by other explanatory variables included in this study. These adjusted analyzes were performed for the restricted material (i.e. with exclusion of accidentally thawed samples and requirement of stringent criteria for season). To explore the reverse causality, the material was split into two by the time (10 years) between drawing of the blood sample and diagnosis of the case. We call this difference as lag and use 10 years as the cut-off point for long and short lag.

Finally, we estimated the individual and the joint effects of explanatory variables for the prostate cancer risk. This was done for two explanatory variables at a time, using the same groups and cut-off points as in multivariable analysis. The RRs were adjusted for other variables used in this study. The expected multiplicative joint effect, RR_m, of two exposures, X and Y, was calculated as a product of RRs for the solitary effects: RR_m(X,Y)=RR_X–Y RR_–XY, in which –Y denotes those not exposed to Y. The confidence limits for the expected multiplicative RR were approximated by the delta method on the logarithms of the RR estimates.

The RRs and their 95% confidence intervals were estimated using the SAS program package, version 9.3 (SAS Institute, Cary, NC, USA). If the RR estimate was zero, its confidence interval was obtained by exact conditional logistic regression with LogXact version 10 (Cytel Inc., Cambridge, MA, USA).

Results

Our results were based on a total of 699 prospectively occurring prostate cancer cases and 2132 controls from the cohorts and on the restricted material of 483 eligible cases and 1055 eligible controls. The majority of the case and controls came from the Janus Bank in Norway (64% of cases and controls), with the longest follow-up at the same time. More than 80% of the men in Norway were followed over 10 years whereas all donors from Sweden had less than 11 years of follow-up (Table 1).

Table 1. Nordic biobank prostate cancer study. Number of cases by study characteristics and country.

	All cohorts		Norway		Finland		Sweden
Variable Age at blood collection (years)	Restrictions		Restrictions		Restrictions		Restrictions
	Without	With^1,2	Without	With^1	Without	With^1,2	Without	With^1
<45	178	122	166	113	11	8	1	1
45–50	338	223	293	197	37	18	8	8
51–60	175	130	21	14	87	50	67	66
>60	8	8	0	0	0	0	8	8
Lagtime from blood collection to diagnosis (years)
≤10	214	170	74	53	56	34	84	83
11–19	352	227	273	185	79	42	0	0
≥20	133	86	133	86	0	0	0	0
Calendar year of blood collection
Before 1985	593	384	458	308	135	76	0	0
1985–1989	30	24	21	15	0	0	9	9
After 1989	76	75	1	1	0	0	75	74
Tumor stage^3
Localized (CT1, T2)	464	321	341	225	65	38	58	58
Non-localized (other)	172	128	116	87	36	22	20	19
Not determined	63	34	23	12	34	16	6	6
Number of cases	699	483	480	324	135	76	84	83

¹Cases without a seasonally matched control; ²Accidentally thawed samples; ³Tumor stage classified according to UICC (46), localized tumors defined as T1, T2 tumors without known/regional lymph node or distant metastasis.

Our data showed that those with average levels of vitamin D had low crude relative risk of prostate cancer compared to the other levels with RR 1.2 (95% CI 1.0–1.4). High levels of enterolactone were associated with somewhat increased risk RR 1.2 (1.0–1.5). C. trachomatis infection was associated with decrease in the risk RR 1.5 (1.1–2.1). Risk of prostate cancer was independent of the levels of free testosterone of HPV-18 (Table 2).

Table 2. Nordic biobank prostate cancer study, total and restricted material. Crude rate ratios (RR) with 95% confidence intervals (CI) for the development of prostate cancer by levels of variables (vitamin D, enterolactone, free testosterone, Chlamydia trachomatis, and HPV-18) and material restrictions.

		Total material				Restricted material*
		Number of				Number of
Variable (unit)	Level	cases	controls	RR	95% CI	cases	controls	RR	95% CI
25(OH)-Vitamin D	<40 or ≥60	434	1240	1.2	1.0–1.4	287	595	1.2	0.9–1.5
(nmol/l)	40–59.9	265	892	1	(ref.)	196	460	1	(ref.)
Enterolactone	1st tertile, <5.56	209	665	1	(ref.)	152	299	1	(ref.)
(nmol/l)	2nd–3rd tertiles, ≥5.56	490	1467	1.2	1.0–1.5	331	756	1.1	0.8–1.4
Free testosterone	1st–2nd tertiles, <337.67	482	1423	1.1	0.9–1.4	329	671	1.2	0.9–1.5
(pmol/l)	3rd tertile, ≥337.67	217	709	1	(ref.)	154	384	1	(ref.)
C. trachomatis	Seronegative, <16	648	1908	1.5	1.1–2.1	453	959	1.6	1.0–2.5
(Titers)	Seropositive, ≥16	51	224	1	(ref.)	30	96	1	(ref.)
HPV-18	Seronegative, <0.1	682	2065	1	(ref.)	471	1033	1	(ref.)
(Absorbance units)	Seropositive, ≥0.1	17	67	0.8	0.5–1.4	12	22	1.2	0.6–2.4

*Accidentally thawed samples (52 cases and 197 controls), 157 cases without a seasonally matched control, and 840 controls not matching seasonally were excluded. Significant RRs in bold.

The material was restricted to meet the stringent criteria for matching for the season, and those Finnish samples accidentally thawed were excluded. For the solitary effects, there were only small differences in the relative risks compared to the total material. Due to the smaller sample size the confidence intervals widened and only exposure to C. trachomatis remained statistically significant (Table 2).

Confounding was evaluated by multivariable analysis in the material with exclusions (Table 3). The model with all the exposures considered had only marginal effect on the relative risk estimates. When analyzed by time from blood drawn to diagnosis of the cancer, the tendency for increased risk in those with HPV-18 infection was seen only with short lag.

Table 3. Nordic biobank prostate cancer study, restricted material*. Mutually adjusted rate ratios (RR) with 95% confidence interval (CI) for the development of prostate cancer by levels of variables (vitamin D, enterolactone, free testosterone, Chlamydia trachomatis, and HPV-18) and lag since serum sampling until diagnosis.

Variable		Lag ≤10 years		Lag >10 years		No lag exclusion
(unit)	Level	RR	95% CI	RR	95% CI	RR	95% CI
25(OH)-Vitamin D	<40 or ≥60	1.1	0.8–1.7	1.2	0.9–1.6	1.2	0.9–1.5
(nmol/l)	40–59.9	1	(ref.)	1	(ref.)	1	(ref.)
Enterolactone	1st tertile, <5.56	1	(ref.)	1	(ref.)	1	(ref.)
(nmol/l)	2nd–3rd tertiles, ≥5.56	0.9	0.6–1.4	1.2	0.9–1.6	1.1	0.8–1.4
Free testosterone	1st–2nd tertiles, <337.67	1.3	0.8–1.9	1.1	0.8–1.5	1.2	0.9–1.5
(pmol/l)	3rd tertile, ≥337.67	1	(ref.)	1	(ref.)	1	(ref.)
C. trachomatis	Seronegative, <16	2.1	0.9–5.3	1.5	0.9–2.5	1.6	1.0–2.5
(Titers)	Seropositive, ≥16	1		1		1
HPV-18	Seronegative, <0.1	1	(ref.)	1	(ref.)	1	(ref.)
(Absorbance units)	Seropositive, ≥0.1	1.7	0.5–6.0	1.1	0.4–2.7	1.2	0.6–2.6

*Accidentally thawed samples (52 cases and 197 controls), 157 cases without a seasonally matched control, and 840 controls not matching seasonally were excluded.

Significant RRs in bold.

In the analysis of first order interactions based on the restricted material some deviations from expected (assuming multiplicative effects) values were seen (Table 4). All the joint effects with vitamin D were antagonistic when assessed with multiplicative assumption. In those with long lag time the joint effect with vitamin D was RR 1.6 (1.0–2.6) in those exposed both to high levels of enterolactone and high or low levels of vitamin D. This was, if anything, indicating an antagonistic interaction: the interaction was not significantly lower than the expected RR 3.1. In those with short lag, C. trachomatis positive men had a low risk of prostate cancer and those positive for HPV-18 had a somewhat increased risk. This indicated a high expected risk (expected RR 27.0), in those negative for C. trachomatis and positive for HPV-18. The observed risk, however, was less: RR 3.1 (0.6–16.7) which was consistent with an antagonistic interaction.

Table 4. Nordic biobank prostate cancer study, restricted material*. Mutually adjusted solitary and joint effects of variables (enterolactone, Chlamydia trachomatis, free testosterone, HPV-18, and vitamin D) for the development of prostate cancer by lag since serum sampling until diagnosis.

Variable (unit)	Variable (unit)		Lag ≤10 years		Lag >10 years		No lag exclusion
level	level		RR	95% CI	RR	95% CI	RR	95% CI
Enterolactone	Vitamin D
(nmol/l)	(nmol/l)
<5.56	40–59.9		1 (ref.)		1 (ref.)		1 (ref.)
≥5.56	40–59.5		0.8	0.4–1.5	1.7	1.1–2.7	1.3	0.9–2.0
<5.56	<40 or ≥60		1.0	0.5–2.0	1.8	1.1–3.0	1.5	1.0–2.3
≥5.56	<40 or ≥60	OBS.	1.0	0.5–1.8	1.6	1.0–2.6	1.4	1.0–2.1
		EXP.	0.8	0.2–2.7	3.1	1.3–7.4	2.0	1.0–4.1
C. trachomatis	Vitamin D
(Titers)	(nmol/l)
Seronegat., <16	<40 or ≥60		1 (ref.)		1 (ref.)		1 (ref.)
Seronegat., <16	40–59.9		0.9	0.6–1.4	0.8	0.6–1.1	0.9	0.7–1.1
Seroposit., ≥16	<40 or ≥60		1.1	0.4–3.0	0.7	0.4–1.4	0.8	0.5–1.4
Seroposit., ≥16	40–59.9	OBS.	0.1^a	0.0–0.5	0.5	0.2–1.2	0.4	0.2–0.8
		EXP.	1.1	0.3–3.3	0.6	0.3–1.3	0.7	0.4–1.4
Free testosterone	C. trachomatis
(pmol/l)	(Titers)
≥337.67	Seroposit., ≥16		1 (ref.)		1 (ref.)		1 (ref.)
<337.67	Seroposit., ≥16		1.8	0.3–12.0	0.9	0.3–2.7	1.1	0.4–2.9
≥337.67	Seronegat., <16		2.7	0.6–12.2	1.3	0.5–3.3	1.6	0.7–3.5
<337.67	Seronegat., <16	OBS.	3.3	0.7–15.3	1.4	0.5–3.7	1.9	0.9–4.1
		EXP.	4.8	0.2–120.0	1.2	0.2–8.0	1.8	0.4–9.2
HPV-18	C. trachomatis
(Absorbance units)	(Titers)
Seronegat., <0.1	Seroposit., ≥16		1 (ref.)		1 (ref.)		1 (ref.)
Seronegat., <0.1	Seronegat., <16		2.5	0.9–6.7	1.5	0.9–2.6	1.7	1.1–2.7
Seroposit., ≥0.1	Seroposit., ≥16		10.8	0.6–205.0	3.8	0.2–65.0	5.7	0.7–43.9
Seroposit., ≥0.1	Seronegat., <16	OBS.	3.1	0.6–16.7	1.6	0.6–4.6	1.9	0.8–4.5
		EXP.	27.0	0.9–810.0	5.8	0.3–110.0	9.9	1.1–87.0

*Accidentally thawed samples (52 cases and 197 controls), 157 cases without a seasonally matched control, and 840 controls not matching seasonally were excluded.

EXP., expected multiplicative RR; OBS., RR for observed joint effect.

^aunadjusted.

Significant RRs in bold.

Discussion

In cancer epidemiology a frequent problem is whether biological measurement indicates a cause or a consequence of the disease. Potential cause is called exposure whereas the consequence is called tumor marker with a clinical application, e.g. potential in diagnostics. Prostate cancer is an insidious disease; the length of the detectable preclinical phase may be of the order of 10 years [19]. Therefore, the associations based even on relatively long follow-up may indicate causal relationship either from the biological measurement to cancer or reverse causality from cancer to the levels of the measurement. Here, we report the risk of prostatic cancer by levels of vitamin D, enterolactone, free testosterone, C. trachomatis and HPV-18 based on a large Nordic study linking biobanks and cancer registries. We especially focused on the joint effects of these exposures. In most cases multiplicative interactions could be ruled out and the joint effects were consistent with antagonism. Our unusually long follow-up based on old serum samples and long history of cancer registration enabled us to study the causal direction by excluding cancers diagnosed within 10 years from drawing of the blood sample.

We applied stringent criteria for matching; especially the close matching for seasonal variation reduced the samples available. We also excluded some Finnish samples because of one more thawing than for the rest of the Finnish samples. The extra thawing in the excluded samples could otherwise have resulted in potentially biased levels of selected substances, e.g. the values of vitamins [20]. However, the same exclusions and stringent rules were applied throughout. This was because of simultaneous consideration of all the measured variables. Less stringent rules of exclusion would have affected also the relationship between those variables that are more stable for handling and other factors of incomparability because of adjusting for confounders by multivariate methods. The exclusions had no major effects on the relative risks (data not shown) even if the reduction in numbers of observations increased the statistical variations. The Swedish men contributed only to the analyses with short lag. Therefore, there is less country-wise heterogeneity in the results with long lag. However, the long lag results may suffer from less generalizability because they stem mainly from the Norwegian men. All men were to be incorporated in order to reduce the random variability with larger numbers of observations and to increase the variation in the exposure variable investigated in the present study.

There is some evidence that average levels of vitamin D are related to low risk of prostate cancer [7] whereas both low and high values indicate an increased risk. In our study the incidence of prostate cancer was high at both ends of the spectrum of vitamin D levels. The relative risk at the high end (RR 1.2) was statistically significant and it was not materially affected by the adjustment or exclusion procedures. Some of the exposures are non-linear in effect. The mechanism for the non-linearity in the effect may be an etiological one or, alternatively, it may be composed from a linear etiological effect and from a linear tumor marker effect. If so, risk after a long lag will be indicative of an etiological effect and any short-term risk can be due to either effect (because it is not known how long was the duration of exposure before drawing of the blood sample). The risks were somewhat higher with long lag and especially in the joint effects with enterolactone. This is consistent with the etiological role of vitamin D, deviations from the average levels may increase the risk of prostate cancer.

We confirmed also the decreased risk of prostate cancer in those serologically positive for C. trachomatis [11]. In our previous study [10], there was a tendency of decreased risk of prostate cancer in those positive for HPV-18. This we could not confirm in the present analysis. The exclusion criteria were not the same, and in the present study we adjusted for the other determinants analyzed from the biobanks.

Consistent with our previous results [9] and with a subsequent meta-analysis [5] free testosterone was not associated with the risk of prostate cancer when considered alone. However, somewhat increased risk was found in those with low level of free testosterone after allowing for the interaction with C. trachomatis infection. The increase appeared only in those with less than 10 year lag between drawing of the sample and diagnosis of the cancer. A possible interpretation is that the low hormonal levels are a marker of (preclinical) prostate cancer and serve as a tumor marker. However, assuming that hormonal levels are associated with sexual habits and that sexual habits are determinants of chlamydia infection our results are consistent with the hypothesis that active sexual habits prevent occurrence of prostate cancer. Studies on this issue have been inconsistent [9,21–24]. This may be because events that are determinants of the prostate cancer risk take place relatively early in the man’s life, e.g. survey on early sexual habits may more record the memorizing of the desire than objective event like frequency of intercourse.

The mean age at diagnosis was 63 years with substantial variation with minimum age 44 and maximum age 79. The background incidence of prostate cancer was low in Finland and at the same level in Sweden and Norway in 1950s. The risk in Sweden increased more than in Norway and the rank was Sweden, Norway, Finland up to mid 1990s. Since then there was a substantial increase in the risk and differences disappeared as a consequence of frequent PSA testing. The biobanks were linked with cancer registries in 1997. Therefore, the PSA testing was not likely to affect or cause incomparability in the results. Increased variation in potential exposure and in the cancer risk by age and by country was aimed at increasing the potential to identify risk factors of prostate cancer.

The long storage time from drawing of the sample to the analysis of the biochemical substances as well as the handling itself was likely to affect the levels of substances [25,26]. In order to cause bias the degradation should have been differential, systematically different between the cases and controls. It was unlikely, however [20]. Some regression towards the mean during the storage years is possible and its effect was to reduce the RR values from the true one.

Our total material consisted of 699 cases and 2132 controls. However, after exclusions relevant for the simultaneous analysis the numbers reduced to 483 and 1055. Many of the relative risk estimates were, therefore, subjected to large random variation. Since the analysis of the serum samples in 1997, many more prostate cancer patients have been diagnosed [27]. The possibility to extend the study would assume substantial resources, both human and economical, not available at present.

We studied the mutually adjusted solitary and joint effects of several risk factors of prostate cancer. The solitary effects were not materially affected by the confounding by the other risk factors. We found no evidence that exposure to two risk factors would increase the risk of prostate cancer from that expected on basis of exposure to one risk factor. If anything, the results were consistent with antagonistic interactions.

Acknowledgments

This study was supported by Academy of Finland, EU consortia ‘Evaluation of the role of infections in cancer using biological specimen banks’ (ERICBSB), Finnish Cancer Foundation, Nordic Academy of Advanced Studies (NORFA), Nordic Cancer Union, Swedish Cancer Society, and University Hospital of Tampere.

Disclosure statement

The authors declare that they have no conflict of interest.