Data sources
The Oslo Ischemia Study; a cohort of working males, aged 40 to 59 years, recruited from five companies in Oslo, Norway, in the period 1972–75. They were apparently healthy, free from somatic diseases and not using drugs. Details about the selection criteria are presented elsewhere [26, 27]. In total, 2341 employees were invited, of whom 2014 (86 %) participated by completing the study protocol.
Information on cancer and cause of death were obtained through linkages between the cohort data, the Cancer Registry of Norway and the Cause of Death Registry, respectively. The data were linked through the unique national identity numbers assigned to every individual residing in Norway from 1960 onwards. The Cancer Registry of Norway has registered all cancer diagnoses nationwide from 1953 onwards. Mandatory reporting from multiple independent sources ensures the collection of complete and high quality data [28]. The Cause of Death Registry contains information on all recorded deaths of Norwegian citizens living in Norway at time of death since 1951.
Measurements
In accordance with the study protocol, a comprehensive medical history was taken from all participants in addition to an extensive physical examination, a chest X-ray, a panel of fasting blood tests, and a maximal exercise tolerance bicycle test. All participants were asked to abstain from smoking for at least 8 h, and eating for 12 h, in advance of the examination.
Total serum cholesterol was analyzed by Liebermann Burchart's no-enzymatic method [29] and divided into quartiles. Smoking status was categorized as never or ever smoker. Physical fitness was measured as work capacity (sum of work performed in the bicycle test) divided by body weight, kJ/kg, and divided into tertiles. Body mass index (BMI) was calculated based on objectively measured height and weight (body weight/height2, kg/m2), and divided into normal weight (BMI < 25) or overweight (BMI ≥ 25). Resting blood pressure was measured after the participant had been in the supine position for five minutes [30], and systolic blood pressure was divided into quartiles. Participants were encouraged to continue exercising until exhaustion [24]. Age at inclusion was divided into four groups (<45, 45–49, 50–54, 55+ years).
The cancer data consisted of information on cancer type (according to the International Classification of Disease, revision 7), date of diagnosis and stage of disease at diagnosis. Stage at diagnosis was divided into two categorises based on information on metastases at diagnosis: localized (invasive prostate cancer without any metastases) and advanced (any infiltration into surrounding structures, regional or distant metastases). Cases lacking information on metastasis were categorized as “unknown”.
The cohort of men was followed throughout 2012. Of the 2014 men included, we excluded two men with missing vital status data and 15 men due to cancer diagnosis prior to date of the examination. Thus, 1997 men remained in the cohort for analyses of prostate cancer incidence. Of these, 1520 men were followed for 20 years or more. The registrations of baseline variables used in this study were 100 % complete for all men included.
Statistical analyses
The baseline characteristics are presented as frequencies and proportions. Comparisons of proportions were tested by chi-square tests.
Standardized incidence rates (SIR) were computed to measure the relative risk of cancer (overall and the six most prevalent sites) in the cohort compared to the general male population in the two counties (Oslo and Akershus) where the men were recruited. Cancer incidences were extracted from the Cancer Registry of Norway. Reference rates were computed for 5-year age groups and 3-year calendar periods. Expected numbers were computed by applying the age- and period-specific rates to the observed person-years in the cohort. SIRs were computed by taking the ratio of the observed to expected incidence, with accompanying 95 % confidence intervals (CI).
The cohort of men was followed up longitudinally from the date of examination to the date of diagnosis of prostate cancer. Cumulative incidence of prostate cancer was estimated in competing risk models, where death (all causes) was considered as a competing event. The men were censored at the date of emigration or study end (December 31, 2012). Cumulative incidence of prostate cancer, which describes the absolute risk over the time course, was plotted for each quartile of cholesterol level. Quartile 1 was compared to quartile 4 by performing the Pepe and Mori test of equality in cumulative incidences [31].
To evaluate the effect of cholesterol on prostate cancer incidence, we performed competing risk regression analysis with adjustment for potential confounding factors; age, smoking, physical fitness, BMI, and systolic blood pressure (all categorical as given above). We used the Fine and Gray model [32], which extends the Cox proportional hazards model to competing risks data by considering the sub-distribution hazard while adjusting for the competing risk of (all cause) death. The strength of the association between each variable and the outcome was assessed using the sub-hazard ratios (SHR). Tests for trend across categories were performed by entering categorical variables as continuous in the models. The model assumptions were found to be adequately met and no significant interactions were observed.
We conducted sensitivity analyses restricted to men who were still alive and free from prostate-cancer at 20 years of follow-up, and followed them for the next 20 years. Further, we performed separate analyses for localized and advanced stage prostate cancer. In the stage-specific analysis, diagnosis at the other stage was defined as a competing event (i.e., localized stage prostate cancer was considered a competing event when studying advanced stage prostate cancer). For comparison, we conducted Cox regression analysis equivalent to the competing risk regression analysis of prostate cancer incidence. Results from the Cox analyses are presented as supplementary files.
Among men diagnosed with prostate cancer, we performed a separate competing risk analysis to evaluate the effect of cholesterol on prostate cancer death. The method of Fine and Gray was used to analyze time to prostate cancer specific death with death from other causes as a competing event.
All statistical analyses were performed using Stata [33]. In particular, we used the stcrreg and stcompet modules when analyzing competing events.