Fat Intake Is Not Linked to Prostate Cancer: A Systematic Review and Dose-Response Meta-Analysis

Background Since the late 1960s, the average global supply of fat has increased by 20 g per capita per day. While fat intake has been considered a potential risk factor for prostate cancer (Pca), the hypothesis from previous epidemiologic studies remained equivocal. Materials and Methods Relevant cohort studies were identified through a literature search in PubMed, ScienceDirect and Wiley Online Library up to March 1, 2015. A systematic review and dose-response meta-analysis were used to assess the relationship between fat intake and the risk for Pca. Results We identified 14 cohort studies, which included 37,349 cases and a total of 751,030 participants. We found no evidence of a non-linear association between fat intake and the risk for Pca. Overall, the summarized relative risks for every 28.35 g increment a day was 0.99 (95%CI: 0.98, 1.01; P=0.94; n=13) for total fat intake, 1.00 (95%CI: 1.00, 1.00; P=0.72; n=9) for saturated fat, 0.99 (95%CI: 0.95, 1.03; P=0.55; n=7) for polyunsaturated fat, and 1.00 (95%CI: 0.95, 1.04; P=0.85; n=8) for monounsaturated fat. Additionally, there was no link to the risk for advanced stage Pca regarding total fat intake (RR=1.02, 95%CI: 0.96, 1.08; P=0.63; n=5), saturated fat (RR=0.96, 95%CI: 0.84, 1.11; P=0.61; n=6), polyunsaturated fat (RR=0.96, 95%CI: 0.79, 1.17; P=0.68; n=6), or monounsaturated fat (RR=0.96, 95%CI: 0.86, 1.07; P=0.42; n=6). Subgroup and sensitively analyses showed consistent results. Conclusion Little evidence from published cohort studies supports the statement that total fat, saturated fat or unsaturated fat intake increases the risk for Pca or advanced stage Pca.


Introduction
Prostate cancer (Pca) is the second leading cause of cancer death among American males [1] and has a crude incidence of 38.2 per 100, 000 men for 1-year prevalence and 151.2 for 5-year prevalence worldwide [2]. According to the National Cancer Control Institute (NCCN), an estimated 233,000 men were diagnosed with Pca in 2014, which accounted for 27% of newly diagnosed cancer cases [3].
The World Health Organization reports that since the late 1960s, the average global supply of fat has increased by 20 g per capita per day [4]. In many countries, such as America, Canada, Australia, France, Finland, New Zealand the incidence of prostate cancer has increased [2]. Previous epidemiology studies have reported potential correlations between fat intake and the risk for Pca [5,6]. The mechanism is complex and unclear, one possible explanation may be the oxidative stress generated during fat metabolism [7,8]. Other proposed mechanisms, including serum testosterone level [9], free radicals [10], and insulin-like growth factor levels [11] related to fat intake. Nevertheless, the relationship between fat intake and risk of prostate cancer remains controversial.
In several previous reviews and meta-analyses [12,13,14], total fat consumption was associated with Pca risk while saturated and unsaturated fat consumption were not. In another metaanalysis, however, no association was confirmed between fat intake and risk for Pca [15]. The current data on Pca risk and fat consumption were highly heterogeneous and insufficient. The limited study numbers or less robust design also made them low reliability. Therefore, we conducted a systematic review and dose-response meta-analysis, with more available cohorts and flexible design. We aimed to investigate the relationship between the consumption of different types of fats and the risk for Pca.

Methods
We conducted our meta-analysis following the preferred reporting method for systematic reviews and meta-analyses (PRISMA) statement [16] (S1 PRISMA Checklist).

Search Strategy
Eligible cohort studies were identified by searching PubMed, ScienceDirect, and Willey Online Library published up to March 1, 2015. Two reviewers (T.Z. Liu and Z.Y. Gao) independently searched each database and any disagreements were resolved by a methodologist (X.T. Zeng) for a final decision. A Kappa statistical test was used for measuring agreement [17]. The following search terms were used: "fat intake" OR "high-fat diet" OR "dietary fat" AND "prostate cancer" OR "prostate tumor" OR "prostate neoplasm" OR "prostate carcinoma" OR "prostate tumour". References in identified articles were also reviewed. There was no language limit.

Eligibility criteria
Because case-control studies may introduce considerable bias, particularly recall bias, only cohort or case-cohort studies were included in our meta-analysis [17,18,19]. The primary outcome of interest was any stage of Pca and the exposures analysed were total fat, saturated fat, or unsaturated fat intake. Secondary tumors from other organs were not considered. The estimated effect was either provided in the study or could be calculated from raw data. All studies included at least three quantitative categories of fat intake. Studies reporting animal fat (except for fish oil) were categorized as saturated fat. We found that most studies pooled vegetable and fish oils into total, saturated or unsaturated fat. Thus, vegetable and fish oils were not considered in this meta-analysis. Grey literature, meeting paper, and animal studies were excluded from this meta-analysis.

Data extraction
From all included studies two experienced reviewers (S. Li and F.F. Han) extracted the first author's name, publication year, country, study type, follow-up, age distribution, types of fat number of cases or person-years, serving size, adjusted or crude relative risk (RR) with 95% confidence intervals (CI), adjusted variables, and the degree of Pca using a standardized data collection sheet. When different models were used to adjust for confounders, we extracted the RR that controlled for the most factors. A third investigator (Z.Y. Gao) checked the data and corrected potential errors.

Data conversion
For studies that measured fat intake with energy percentage, the data was converted to grams by multiplying by the mean daily total energy intake and then dividing by 9 (1 gram of total fat provides 9 Kcal energy). If studies did not report the mean daily energy intake, we assumed it to be 2,418 Kcal, which is the age-specific energy value for 50-71 year old males according to the National Cancer Institute [20]. This data conversion may overestimate the amount of saturated fat consumed and underestimate unsaturated fat consumption since saturated fat contains more calories than unsaturated fat.

Statistics analyses
Relative risk (RR) was used to measure the risk. Odds ratios (OR) and hazard ratios (HR) were roughly regarded as relative risk (RR) [21]. Missing data was evaluated as described by Bekkering et al [22]. Briefly, if the number of non-cases was missing, the group sizes were assumed to be approximately equal. If the number of cases were missing, the reported RRs and non-control numbers were used to calculate the number of cases. There was no valid way to evaluate if the serving size was missing.
The dose-response meta-analysis was conducted in two steps. First, the generalized leastsquares method estimated the coefficient per unit increment of exposure within each study. Second, the regression coefficients were combined in a random-effect model with the weight calculated by inverse variance [23,24]. All effect sizes were logarithm transformed for the meta-analysis. The lowest exposure level served as the reference category in each study and the estimated log relative risk in the reference category was set to zero (log 1) [24]. Every 28.35 g (approximately 1 ounce) increment of fat intake per day was used to measure the doseresponse relationship.
We used the mean value of the lower and upper boundaries of each category as the assigned dose. For open-ended lower categories, the assigned dose was calculated by dividing the cut-off point by 1.
2. For open-ended upper categories, the cut-off point was multiplied by 1.2 [25]. The non-linear trends between total fat, saturated fat, and unsaturated fat intake and the risk for Pca were fitted by modelling both tails (left-tail and right-tail) restricted cubic splines with three knots at fixed 10th, 50th, and 90th percentiles of exposure distribution [26]. The Wald test was used to evaluate linearity or non-linearity trends by assuming the regression coefficient of the second spline equalled zero [26]. Some studies reported RRs by subtypes (such as sex or area), in our meta-analysis we pooled the subtypes using a fixed-effect model before including them in the overall analysis [19,27].
The Egger's test was used to determine publication bias, the I 2 statistic assessed heterogeneity, and subgroup and sensitivity analyses evaluated whether the results were consistent. A random-effects meta-regression was used to assess which covariates in the subgroup analysis influenced the intervention effect [17]. All the analyses were performed using the Stata SE12.0 software (Stata SE 12.0 Corp LP, College Station, Texas, USA).

Study characteristics and quality assessment
Among the 14 studies, there were a total of 751,030 participants of which 37,349 cases developed Pca. The mean age of Pca cases was about 60.60 years and the mean follow-up period was 9.2 years. Four studies [29,30,33,37] measured fat intake with energy percentages and the remaining 10 measured grams [8,32,[34][35][36][38][39][40][41]. All of the studies were conducted in America or European countries and America (America and Canada) contributed 84.71% of total cases. Twelve studies controlled for the main confounders and two studies [32,41] adjusted for age only (Tables 1, 2, and 3).
The Newcastle-Ottawa Scale [42,43], which contains 9 terms with 1 term account for 1 score, was used by two reviewers to assess the quality of the included studies. A third author deal with any disagreements. For each study, we obtained a mean score of 8.07 of all the included studies (Kappa = 0.32) (S1 Table).

Saturated fat intake and Pca risk
Nine studies [29-34, 37, 38-41] reported an association between saturated fat intake and risk of Pca. No evidence of a non-linear relationship between saturated fat intake and the risk for Pca (P for non-linearity was 0.25; Fig 3A) was found. The combined RR was 1.00 (95%CI: 1.00, 1.00; P = 0.72; I 2 = 14.3%) for every 28.35 g (1 ounce) increment of saturated fat consumed per day ( Fig 3B).

Fat intake and the risk for advanced or high grade Pca
Seven studies [8, 29-31, 33, 34, 37] investigated the association between fat intake and the risk for advanced or high grade Pca (Fig 5). The RRs for every 28.35 g/day increment was 1.02 (95%CI: 0.96, 1.08; P = 0.63; I 2 = 48.6%; n = 5) for total fat, 0.96 (95%CI: 0.84, 1.11; P = 0.61; I 2 = 70.4%; n = 6) for saturated fat, 0.96 (95%CI: 0.79, 1.17; P = 0.68; I 2 = 55.9%; n = 6) for polyunsaturated fat, and 0.96 (95%CI: 0.86, 1.07; P = 0.42; I 2 = 37.3%; n = 6) for monounsaturated fat intake. Subgroup, meta-regression, and sensitivity analyses We conducted a subgroup analysis using primary measurement units (e.g. grams or energy), study area, and adjustment status (adjusted/not) for BMI for potential divergences within subgroups, which showed no substantial change ( Table 4). The multivariable meta-regression showed that primary measurement units, study area, adjustment status for BMI were not associated with the risk of Pca. The Knapp-Hartung adjustment P-value was 0.32 for total fat, 0.71 for saturated fat, 0.91 for polyunsaturated fat, and 0.65 for monounsaturated fat consumption. Sensitivity analysis using the random-effect model showed robust results after omitting one study at a time from each analysis. It also showed consistent results when two studies [30,33] with most weight in most cases were omitted (S2 Table).

Discussion
The present meta-analysis confirmed no obvious associations between total, saturated or unsaturated fat consumption and the risk for Pca. Our further analysis also confirmed a lack of association between fat intake and advanced or high grade Pca. Although obvious asymmetry was observed in the Egger's test of saturated fat intake, the trim and fill method showed consistent results and suggest that asymmetry may be not be caused by publication bias.
Obesity is often linked to high fat consumption [44] while BMI is closely related to Pca [14]. There is a complex relationship between BMI and Pca, but whether BMI increases the risk of Pca remains controversial [45,46,47]. For this reason, we conducted a subgroup analysis based upon different statuses of BMI adjustments. Our subgroup analysis showed consistent results in BMI-adjusted and non-adjusted groups, which suggested that our results may not be influenced by BMI. We also conducted a subgroup analysis using primary measurement units and study areas and found no substantial change in the results.
The sensitivity analysis showed robust results in our meta-analysis. Two studies [29,33] accounted for most of the weight, and in order to determine their potential influence on overall results, both two were omitted in each analysis. The results were consistent and supported the credibility of our meta-analysis.
A previous meta-analysis [12], of 29 observational studies with 5 cohort studies, found that only total fat consumption was associated with an increased risk for Pca (RR = 1.2). Consuming 45g of total fat per day (5 studies, combined RR = 1.12, 95%CI: 1.01, 1.25) or saturated fat (4 studies, combined RR = 1.38, 95%CI: 1.13, 1.70) increased the risk for advanced stage Pca. The meta-analysis was well-designed, but most of the studies included were case-controlled with considerable heterogeneity, which may account for the low grade of evidence. Another systematic review [48] that contained only 5 studies (including one cohort study) claimed that saturated fat consumption was associated with advanced Pca. However, their limited study numbers and sample size may explain the low statistical power of their results. The present meta-analysis is based on large numbers of cohort studies and we found no association between fat intake and the risk for Pca. Our results are similar to a meta-analysis by Chua et al [15]. Our meta-analysis included more high quality cohorts and prepared with more flexible design, may be credible. There were also reviews on this topic [13,14], but the lack of systematic statistical analysis and less rigorous design may lead to a loss of credibility.

Potential mechanism of fat intake and Pca risk
There are a few known risks and benefits to consuming dietary fat. Bioactive components in dietary fats such as N-3 polyunsaturated fat acid (n-3 PUFA), may protect against prostate cancer and other types of cancer [49] by altering COX-2 expression and prostaglandin production [50]. Fat-soluble vitamins such as vitamin D and E, are increasingly absorbed as fat is consumed may protect against prostate cancer [51,52]. In addition to potential carcinogenic pathways being linked to an increase risk Pca, oxidative stress generated during fat metabolism has been reported to increase the risk for Pca [7] though IGF-1 up regulation and increased cell growth [53,54]. Androgen signaling has been regarded an important factor for Pca progression [14]. Simultaneously, androgen was found to up regulate the IGF-1R expression [55], which may also promote Pca development. Free radicals and proinflammatory fatty acids produced by dietary fat were also considered to promote tumour growth [11]. Since we found no association between fat intake and risk of Pca, the above mentioned factors may generate an offset effect.
Physical activity status may influence our results since regular physical activity is considered to protect against Pca [56]. Two studies [8,33] in our meta-analysis controlled for the influence of physical activity on their results. We did not conduct any additional analysis (such as subgroup analysis) for these two studies because the numbers were small and would result in a low statistical power. Other confounding factors such as age, energy, family history, fruit and vegetable intake, and serum fatty acid levels may have influenced our results. However, sensitivity analysis did not reveal substantial change, which suggests that the influences mentioned above may have little impact on our results.

Strengths and limitations
To ensure that our results are reliable, we used a dose-response meta-analysis to evaluate high quality cohort studies and to find potential non-linear or linear relationships between fat intake and the risk for Pca. We used subgroup and sensitivity analyses to distinguish the effect among the subgroups and generated consistent results. All of this makes our results more reliable.
There were a few limitations in our meta-analysis as well. First, the limited number of studies included and the considerable amount of heterogeneity detected in our analysis of advanced or high grade Pca and fat consumption may have influenced the accuracy of our results. Second, all studies were conducted in American or European countries. Thus, a selection bias was introduced making this meta-analysis applicable to Americans and Europeans only. Third, we have limit data on addressing the above potential bias in our study that may also influence our results.

Conclusion
Current published cohort studies suggest no association between total fat, saturated fat, or unsaturated fat intake and the risk for Pca. More studies on the association between fat intake and high grade or advanced stage Pca are needed.