Association of obesity with prostate cancer: a case-control study within the population-based PSA testing phase of the ProtecT study

Background: Obesity has been inconsistently linked to prostate cancer, mainly with mortality rather than incidence. Few large-scale studies exist assessing obesity in relation to prostate-specific antigen (PSA)-detected prostate cancer. Methods: We used cases and stratum-matched controls from the population-based PSA-testing phase of the Prostate testing for cancer and Treatment study to examine the hypothesis that obesity as measured by body mass index (BMI), waist circumference and waist-to-hip ratio (WHR) is associated with increased prostate cancer risk, and with higher tumour stage and grade. In all, 2167 eligible cases and 11 638 randomly selected eligible controls with PSA values were recruited between 2001 and 2008. A maximum of 960 cases and 4156 controls had measurement data, and also complete data on age and family history, and were included in the final analysis. BMI was categorised as <25.0, 25.0–29.9, ⩾30.0 in kg m−2. Results: Following adjustment for age and family history of prostate cancer, we found little evidence that BMI was associated with total prostate cancer (odds ratio (OR): 0.83, 95% confidence interval (CI): 0.67, 1.03; highest vs lowest tertile; P-trend 0.1). A weak inverse association was evident for low-grade (OR: 0.76, 95% CI: 0.59, 0.97; highest vs lowest tertile; P-trend 0.045) prostate cancer. We found no association of either waist circumference (OR: 0.94, 95% CI: 0.80, 1.12; highest vs lowest tertile) or waist-to-hip ratio (WHR; OR: 0.93, 95% CI: 0.77, 1.11; highest vs lowest tertile) with total prostate cancer, and in analyses stratified by disease stage (all P-trend>0.35) or grade (all P-trend>0.16). Conclusion: General adiposity, as measured by BMI, was associated with a decreased risk of low-grade PSA-detected prostate cancer. However, effects were small and the confidence intervals had limits very close to one. Abdominal obesity (as measured by WHR/waist circumference) was not associated with PSA-detected prostate cancer.

Obesity is associated with a number of chronic diseases, including coronary artery disease, hypertension, diabetes and some cancers (Rodriguez et al, 2001;Calle et al, 2003). The association of obesity with hormone-related cancers, such as prostate cancer, has been inconsistent in epidemiological studies. Obesity has been more strongly positively associated with prostate cancer mortality rather than showing consistent associations with incidence (Andersson et al, 1997;Rodriguez et al, 2001;Calle et al, 2003;Wright et al, 2007). Studies examining obesity during prostate development in earlier life, and its possible effect on prostate cancer development later, have also produced contradictory results (Giovannucci et al, 1997;Schuurman et al, 2000). Overall, there are studies showing positive (Gronberg et al, 1996;Veierod et al, 1997;Putnam et al, 2000), null (Whittemore et al, 1995;Giovannucci et al, 1997;Nilsen and Vatten, 1999;Habel et al, 2000;Lee et al, 2001;Jonsson et al, 2003;Gallina et al, 2007) and inverse (Giovannucci et al, 2003;Wright et al, 2007) associations between body mass index (BMI) and prostate cancer risk.
The inconsistency between these studies might be attributed to a possible interaction between obesity and factors such as age, or to differential effects of obesity on low-grade and high-grade cancer implying aetiological heterogeneity for different tumour subtypes (Freedland et al, 2006). In addition, many studies do not distinguish between central and peripheral adiposity.
There is some evidence, from prospective studies, that obesity is associated with a reduction in risk of incident prostate cancer (Giovannucci et al, 2003;Wright et al, 2007). There are biological mechanisms which may explain potential protective effects against initiation of prostate cancer (Giovannucci et al, 2003). For example, there tend to be lower circulating levels of testosterone in obese compared with non-obese men (Pasquali et al, 1991;Field et al, 1994). Observed protective effects may also be an artefact of haemodilution of prostate-specific antigen (PSA) values in larger men, thereby reducing PSA threshold-based detection of true cancers by misclassifying them as non-cancers (Grubb et al, 2009).
In two case -control studies (Hsing et al, 2000;von Hafe et al, 2004), central adiposity was associated with an increased risk of prostate cancer progression. However, the risk was increased for non-advanced stage as well. Most central adiposity case -control and prospective studies did not show any associations when BMI was assessed (Hsing et al, 2000;Lee et al, 2001;von Hafe et al, 2004) and Giovannucci et al (1997), in their prospective cohort study, showed no association between either adult BMI or waistto-hip ratio (WHR) and risk of total or advanced prostate cancer (Giovannucci et al, 1997). One prospective cohort study (MacInnis et al, 2003) found no overall association with prostate cancer, but modest associations with the risk of aggressive disease, whereas other studies of the same type (Hubbard et al, 2004) report an overall increased risk with increasing WHR.
We conducted a case -control study nested within the PSA-testing phase of the Prostate testing for cancer and Treatment (ProtecT) study (Donovan et al, 2002) to examine associations of obesity with screen-detected prostate cancer. The measures of obesity included BMI, waist circumference and WHR. Associations with obesity were examined for total prostate cancer as well as its subtypes (localised, advanced; high-grade, intermediate grade and low grade).

PARTICIPANTS AND METHODS
The methodology of the ProtecT study has been described previously (Donovan et al, 2002). ProtecT is an ongoing randomized controlled trial that will compare the effectiveness and acceptability of treatments for localised prostate cancer in men aged 50 -69 years. Between 2001 and 2008, over 110 000 men aged 50 -69 years, from B300 primary care centres (general practises) across the United Kingdom, attended prostate check clinics, where histologically confirmed prostate cancer cases were identified through a combination PSA testing, digital rectal examination (DRE) and 10-core transrectal ultrasound-guided biopsy (the latter two investigations only apply to those with PSA X3). Repeat biopsies were offered to men with a normal initial biopsy, in whom there was a high index of clinical suspicion (evidence of high-grade prostatic intraepithelial neoplasia or suspicious features on initial biopsy) or in whom PSA concentration was persistently elevated. Tumours were staged using the TNM staging system. A central pathology review is also conducted.

Selection of cases and controls
Cases were men aged 50 -69 years, who underwent PSA measurement and had a histological diagnosis of primary prostate cancer. We defined localised cancer as T1 -T2, NX or NO, MX or MO and advanced cancer as T3 -T4 or N1 or M1; there were 1894 localised cases and 257 advanced cases (not taking account of the availability or not of anthropometric measurements). We defined highgrade cancer as Gleason grades 8 -9; intermediate grade cancer as Gleason grade 7; and low-grade cancer as Gleason grade p6.
All participants with no evidence of prostate cancer after PSA testing, DRE and/or biopsy were eligible to be controls. Controls were stratum matched to cases by age (5-year bands), and the primary care centres from which they were recruited. The index date for controls was the date of the prostate check clinic. Such matching automatically matches for calendar time, as prostate check clinics were completed sequentially. Detailed descriptions of ProtecT and the protocol for nested case -control selection are published elsewhere (Zuccolo et al, 2008).

Exposure assessment
Obesity indicators were measured as well as self-reported. The measured weight value was taken at the prostate check clinic appointment by clinical staff. The weight was recorded, and it was explained to the participant that general health measurements were taken in order to examine the links with prostate cancer. The participant was weighed to the nearest 0.1 kg, and the measurement was noted by the clinical staff.
Participants were given a self-completion diet, health and lifestyle questionnaire at the clinic; this included questions on body size and weight in stones/pounds, height in feet/inches, and inside leg measurement in inches. They were also provided with a tape measure with which to take waist and hip measurements themselves and return it together with the questionnaire. Participants were instructed regarding the use and placement of the tape measure in order to make the measurements. Each tape measurement (in inches) was marked with a single line on the tape measure, labelled with the relevant letter (W or H) (measured value) and also recorded in the relevant questionnaire box by the participant (self-reported measurement).
We used a metric based on both available measurements, derived from the measured value if that was available and the selfreported value otherwise. BMI was derived from weight and height measurements as kg m À2 ; WHR was computed as waist circumference divided by hip circumference.
In November 2008, there were 2167 prostate cancer cases and 11 638 controls, randomly selected from B100 000 men not diagnosed with prostate cancer. The data composition of the stage and grade analyses is present in Table 1a. Data on BMI could be derived for 4769 controls and 1025 cases; data on WHR could be derived for 4917 controls and 1075 cases, whereas waist circumference data were available for 5020 controls and 1089 cases (Table 1a). Subjects included in the final BMI analyses had complete data on BMI, age and family history, and comprised 3931 controls and 919 cases. For the final analyses for waist circumference, 4156 controls and 960 cases had complete data on waist circumference, age and family history. For the final WHR analyses there were 4069 controls and 948 cases with complete data on WHR, age and family history (analyses presented in Table 2).
The study received ethical approval from Trent Multicentre Research and Ethics Committee and all participants provided written informed consent.
Conditional logistic regression was used to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for associations of obesity with total prostate cancer. We computed two models. First, the basic conditional logistic regression model, in which the stratum matching of cases to controls by age and recruitment centre is taken into account. The second multivariable model was additionally adjusted for exact age at prostate check clinic and family history, as these are the established risk factors for prostate cancer.
The ORs for associations with advanced and localised cancer vs controls and for low, intermediate and high-grade disease vs controls were compared using a multinomial logistic regression model. This model provides a statistical test for heterogeneity in ORs comparing associations of the obesity indicators with localised vs advanced prostate cancers, but it is unconditional; it was, therefore, adjusted for exact age at prostate check clinic and the study centre where the recruiting general practice was based (nine-level variable).
The maximum number (1089 cases and 5020 controls) of subjects used in the analysis, before adjusting for other factors, was compared with the eligible participants not included in the analysis (1078 prostate cancer cases and 6618 randomly selected controls) using t-tests for continuous variables and the w 2 -test for binary response variables.

The PSA detection bias
Some previous studies have shown a decreased PSA level with increasing adiposity (Baillargeon et al, 2005;Werny et al, 2007). This suggests that there could be differential prostate cancer detection with respect to obesity, particularly for screen-detected case finding by PSA-based thresholds for biopsy. Therefore, to assess the potential for PSA detection bias, associations of the adiposity measures with serum PSA concentration amongst controls were investigated using linear regression performed on log-transformed PSA concentrations, adjusted for exact age and recruitment centre.
Analyses were conducted using Stata/IC 10.1 for Windows (StataCorp, 2007, College Station, TX, USA). Table 1b presents the baseline characteristics of the maximum number of cases and controls with adiposity data (1089 cases and    (2167) with missing BMI information was 52.7%; the respective percentage for potentially eligible controls (11 638) was 59%. The percentages for missing data on waist measurements were 49.8% for cases and 56.9% for controls. For 50.4% of cases and 57.8% of controls information on WHR could not be derived. The maximum number (1089 cases and 5020 controls) of subjects used in the analysis, before adjusting for other factors, was similar to the eligible participants not included in the analysis (1078 prostate cancer cases and 6618 randomly selected controls) in terms of family history of prostate cancer in 1st degree relatives (w 2 P-value ¼ 0.100), but not in terms of age (P-value from t-test o0.001). Those included in the analysis were on average a year older than those not included. Table 2 presents the association of the adiposity measures BMI, waist measurement and WHR with total prostate cancer. There was no evidence of any important relationship between any of the measures of adiposity and total prostate cancer.

RESULTS
To examine the association of the obesity measures with stage of prostate cancer, we calculated the ORs and CIs for the categories of BMI/waist circumference/WHR. These are shown in Table 3.   Table 4 are ORs for the categories of BMI, waist circumference and WHR, in order to examine the association of obesity with grade of prostate cancer. For the fully adjusted model for BMI, we observed a weak association with prostate cancer grade, limited to the low-grade group (P-trend 0.045; highest vs lowest tertile OR: 0.76, 95% CI: 0.59, 0.97), with the effect being small and the respective CI close to 1. Results for the basic model were largely similar and are not presented.
We did not observe any associations with either waist circumference or WHR after adjusting for age, family history and study centre for disease stage (all P40.35) or grade (all P40.16).
In a sensitivity analysis, we compared different sources of measurement for weight, waist and hip -as well as the derived WHR. Comparison was between measured and self-reported values. Measured and self-reported weight had a Spearman correlation coefficient of 0.98 (significance level 0.01). The correlation coefficient for measured and self-reported waist circumference was 0.99 (significance level 0.01). For measured and self-reported hip measurements the correlation coefficient was 0.98 (significance level 0.01). Therefore, it is unlikely that using different sources of measurement introduces bias.

DISCUSSION
We found no evidence that our measures of central adiposity were associated with PSA-detected prostate cancer. There was only weak evidence that general adiposity was associated with decreased risk of low-grade prostate cancer.
Most studies use BMI as a measure of obesity, although BMI cannot distinguish between lean and fat mass. This can be problematic for measurement and might account for contradictory results from several retrospective and prospective cohort studies using BMI to evaluate adiposity (Andersson et al, 1997;Nilsen and Vatten, 1999;Rodriguez et al, 2001;Calle et al, 2003;Giovannucci et al, 2003;Jonsson et al, 2003). Therefore, we used additional indicators, that is, waist measurements and WHR as estimates of central adiposity and excess abdominal fat (Arner, 1997). Measures of central adiposity are also better suited for measurements in middle-aged men. Abdominal obesity has been linked to several chronic conditions through mechanisms involving hormonal and metabolic changes (Kaaks and Stattin, 2010). However, very few studies (Giovannucci et al, 1997;Hsing et al, 2000;Lee et al, 2001;MacInnis et al, 2003;Hubbard et al, 2004;von Hafe et al, 2004;Wallstrom et al, 2009) have examined prostate cancer associations with abdominal adiposity; prospective cohort and case -control studies have shown increased risk with increasing WHR but no association with BMI (Hsing et al, 2000;von Hafe et al, 2004). This might imply that abdominal fat rather than general obesity may be associated with prostate cancer risk.
Associations reported in individual studies and for certain subgroups of men have not been seen consistently across studies. Obesity (as defined by BMI 430 kg m À2 ) has been linked with increased risk in some cohort studies (Lew and Garfinkel, 1979;Snowdon et al, 1984;Chyou et al, 1994), but not in some other cohort and case -control studies (Kolonel et al, 1988;Mills et al, 1989;Nomura and Kolonel, 1991;Kolonel, 1996;Andersson et al, 1997;Nomura, 2001;Friedenreich et al, 2004). Giovannucci et al (2003), in their prospective cohort study, reported inverse associations between BMI and prostate cancer risk in younger men or those with a family history. Rohrmann et al (2003) observed a reduced risk of high-grade disease in those with a family history, but an increased risk of high-grade disease in obese men o50 years old, in a case -control study. Two cohort studies of general (Engeland et al, 2003) and central (Wallstrom et al, 2009) obesity found increased risks in younger obese men.
Some hormonal and metabolic alterations that occur in obesity, such as decreased testosterone, may decrease prostate cancer risk; however, other alterations such as high insulin, insulin-like growth factor 1 (IGF-1) and/or leptin levels (Frystyk et al, 1995) have mitogenic effects (McKeehan et al, 1984) and can potentially increase risk by promoting prostate cancer progression (Yu and Rohan, 2000;Chan et al, 2002;Jenks, 2010). Testosterone is involved in prostatic growth and low levels prevent proliferation in the prostate; therefore, decreased testosterone in obese men may explain a protective effect of obesity against incident prostate cancer. At the same time, high testosterone helps maintain differentiation in prostatic epithelium, thus, preventing tumour progression, which means that the low testosterone levels in obese men increase the risk of tumours that are poorly differentiated/ high grade. The IGF-1 is involved in androgen-independent progression of prostate cancer and leptin induces migration in prostate cancer cells, thus obesity-induced hormonal changes may promote tumour progression Freedland et al, 2004). Leptin and insulin/IGF-1 are high in obese men (Hoda et al, 2010) and suppress androgen levels, moreover leptin has been implicated in advanced and high-grade prostate cancer in two case -control studies (Saglam et al, 2003;Ribeiro et al, 2006). Therefore, the simultaneous presence of high insulin and leptin levels and low testosterone levels explains the reduced risk of incident tumours, but the increased risk of progression in existing prostate cancer tumours.
One of the reasons why obesity has been hypothesised to be associated with greater risk of prostate cancer progression in case -control, cohort and prospective studies (Andersson et al, 1997;Putnam et al, 2000;MacInnis et al, 2003;Dal Maso et al, 2004;Baillargeon et al, 2005;Freedland et al, 2005;Wright et al, 2007;Gross et al, 2009), as opposed to initiation (Giovannucci et al, 2003), is that obese men tend to have higher oestradiol and lower testosterone levels, which have been associated with more advanced and poorly differentiated tumours (Hsing et al, 2002;Massengill et al, 2003;Platz et al, 2005) as explained above. This implies that different associations of obesity might be observed for prostate cancers of different stage or grade. Evidence from prospective studies is beginning to show that energy intake in excess of expenditure, captured by a higher BMI, may affect prostate carcinogenesis and, in particular, risk of advanced disease, and that energy imbalance may function late in the carcinogenic pathway, therefore, facilitating progression rather than initiation of prostate tumours (Rodriguez et al, 2007).
The differences in risk trends in previous studies, mentioned above, are possibly because of different methods used for assessing Association of obesity with prostate cancer P Dimitropoulou et al obesity, and to different disease mechanisms acting at different prostate cancer stages and grades (Gong et al, 2006;MacInnis and English, 2006;Littman et al, 2007;Rodriguez et al, 2007). Prostate cancer has a long natural history, and it might be that BMI in earlier life is more important for the development of prostate cancer than adult BMI, explaining the absence of strong associations with BMI in most studies including the present study.
We do not consider differences in PSA values by BMI to be an important source of bias. For BMI, as well as the other adiposity measures, there was weak evidence that PSA levels were linked to obesity, as the absolute difference between the geometric mean PSA values in the highest and lowest categories did not exceed 0.16 ng ml À1 , and the maximum percent difference in means between BMI categories was 14.4%.
The strengths of the study are its large sample size and a wellcharacterised population. In addition, the collection of both selfreported and clinically measured values for the assessment of the adiposity indicators used allows assessment of bias from different sources of measurement. We found no such bias in this study. However, BMI values could be derived from only 47% of cases and 42% of the total population. Waist circumference and WHR measurements had similarly low percentages. For all obesity variables, the percentage of cases with available information was higher than the respective percentage of controls. The reductions in sample size may have affected the power of the study to detect effects, and it is theoretically possible that associations may have differed in those included in the analysis compared to those who could not be included (selection bias). These are two study limitations, although the study remains large. Our results do not support the hypothesis that obesity is involved in prostate cancer progression, although our study is limited to PSA-detected disease.