Empirically derived dietary patterns and prostate cancer risk: a systematic review and meta-analysis of prospective cohorts

Abstract

Prostate cancer (PCa) is the second most common cancer in men and the fourth worldwide. Diet has been suspected as a contributor to PCa onset and progression; however, evidence remains inconsistent, because many studies focus on individual foods and use heterogeneous methods to derive dietary patterns. This systematic review evaluates the association between empirically derived dietary patterns and overall PCa risk in prospective cohort studies. A literature search was conducted on PubMed, Web of Science, Embase, the Cochrane Library, and Food Science and Technology Abstracts, following preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines. Risk estimates comparing the highest versus lowest adherence categories were extracted. A random-effects meta-analysis using restricted maximum likelihood estimation pooled hazard ratios for overall PCa risk. Six eligible cohort studies were included. Two main dietary patterns were identified: a ‘western’ pattern, characterized by high intakes of red and processed meats, sweets, desserts, and a ‘prudent’ pattern characterized by high intakes of fruits, vegetables, fish, and seafood. No significant association with overall PCa was observed: prudent, hazard ratio_Pooled = 0.98; confidence interval_95% [0.91–1.06]; western, hazard ratio_Pooled = 1.07; confidence interval_95% [0.93–1.22]. Empirically derived dietary patterns were not significantly associated with overall PCa risk. Some evidence suggested an association between western dietary pattern and localized and aggressive PCa in populations with limited prior exposure, highlighting the role of dietary shifts. However, these findings were based on a limited number of studies, reflecting the scarcity of prospective cohort data. Future research should aim to standardize dietary pattern derivation and confounder selection to improve comparability across studies.

Introduction

Prostate cancer (PCa) accounts for 15% of cancers in 112 countries, and it is estimated that the number of cases will increase from 1.4 million cases to 2.9 million by 2040 (James et al, 2024). The established risk factors for PCa are age, ethnicity, and family history (American cancer society, 2023). However, lifestyle habits such as diet are also suspected to play important roles in the development and progression of the disease (Kathryn M and Lorelei A, 2019). The rationale for a potential association between diet and PCa risk stems from the disparity in incidence rates, which are highest in Australia, North America, and Western/Northern Europe and lowest in South Central Asia, North Africa, and South Asia (Culp et al, 2020). These variations may reflect differences in dietary habits across regions, although other factors, such as genetic predispositions and screening practices, may also contribute (Rebbeck, 2017). The association between diet and PCa has been investigated; however, the findings remain largely inconclusive (European Association of Urology, 2025). A meta-analysis reported no significant association between total fruit or vegetable intake and PCa risk (Meng et al, 2014), whereas a systematic review suggested a protective effect of fruit consumption (Ubago-Guisado et al, 2021). This ambiguity is likely related to classic approaches assessing foods or nutrients individually (FB, 2002). Such methods may yield conflicting results, as individuals typically consume a combination of foods rather than single food items (Schneider et al., 2019; Matsushita et al., 2020). Thus, the study of dietary patterns was introduced to address these limitations (Gherasim et al, 2020). A dietary pattern is defined as the total dietary intake of all foods consumed on a regular basis (Sánchez-Villegas and Martínez-Lapiscina, 2018). They can be studied with either ‘a priori’ or ‘a posteriori’ methods. ‘A priori’ approaches are based on a score or an index evaluating adherence to dietary guidelines (e.g. Healthy Eating Index) or predefined diet principles (e.g. healthful plant-based diet index). A posteriori approaches rely on statistical models to identify dietary patterns from empirical data, offering the advantage of exploratory analysis unconstrained by predefined food groupings (Agnoli et al, 2019). This flexibility not only supports the development of population-specific public health strategies but also allows for the identification of distinctive dietary behaviors in subpopulations, including ethnic minorities (Nature, 2025).

To our knowledge, only one systematic review and meta-analysis (2016), including 12 observational studies (three cohort studies and nine case controls), has examined the association between a posteriori dietary patterns and PCa risk (Fabiani et al, 2016). This study reported that a western diet, rich in meat and processed foods, was associated with a higher risk of PCa. However, a major limitation was the integration of findings from various study designs, resulting in substantial heterogeneity. Most of the included studies were case control in design and are susceptible to recall bias and reverse causation (Gail et al, 2019). To address these limitations, this systematic review focused on cohort studies, and to leverage population-specific data, we focused on dietary patterns identified ‘a posteriori’ in association with PCa risk. Our approach aims to provide a more nuanced understanding of how population-specific dietary patterns are associated with PCa risk while ensuring greater methodological consistency across the included studies.

Methods

This systematic review was conducted following the preferred reporting items for systematic reviews and meta-analyses (PRISMA) and registered on PROSPERO under the ID CRD42024551585 as of 11 June 2024.

Information sources and search strategies

Studies were searched in the following databases: Medline, Embase, the Cochrane Library, Web of Science and Food Science and Technology Abstracts (FSTA). The search strategies for each database are presented in Supplemental 1, Supplemental Digital Content, https://links.lww.com/EJCP/A592 (dated 13^th June 2024). Systematic reviews and meta-analyses were searched but not included, as they were secondary data.

Selection process

Only prospective observational cohort studies were included. This design allows us to adequately study, in the medium or long term, the associations between diet and health outcomes (Tapsell et al, 2016). Case–control studies were excluded because of the possibility of recall bias and reverse causation, where cases tend to make lifestyle changes upon diagnosis (Gail et al, 2019). The dietary patterns must be identified via an ‘a posteriori’ approach [principal component analysis (PCA) or factor analysis (fatty acid)]. Our main outcome is overall PCa risk ascertained with either patient record linkage or cancer registries. The study selection process was conducted by two independent reviewers (F.B.S. and R.T.). The selection was first by title and abstract and then by full text. Discrepancies between reviewers were resolved by consensus.

Data extraction and quality assessment

Information on study details, population characteristics, dietary pattern identification and description, follow-up, PCa incidence, and statistical analysis parameters were extracted. A detailed extraction guide is presented in Supplemental 2, Supplemental Digital Content, https://links.lww.com/EJCP/A592.

The risk of bias was assessed via the Newcastle–Ottawa Scale (NOS) for Quality Assessment for Cohort Studies (Wells and B. S. D. O. J. P. V. W. M. L. P. T, 2025). Each study was assessed based on the following sections: selection, comparability, and outcome. For assessing the adequacy of cohort follow up, a follow-up rate of at least 80% is considered acceptable and appropriate for large cohort studies (Wayne, 2024). The complete grid is presented in Supplemental 3, Supplemental Digital Content, https://links.lww.com/EJCP/A592.

Statistical analysis

A meta-analysis comparing the highest versus lowest categories for each dietary patterns and overall PCa risk was conducted, and the results are presented as a forest plot. Since the patterns are identified ‘a posteriori’, the names are chosen arbitrarily. The meta-analysis was performed on patterns that had the same predominant food groups across studies, and similar patterns were pooled via a fixed effect model. A logarithmic transformation was performed on the hazard ratios or relative risks (RRs) to standardize the values (Higgins et al, 2024). The meta-analysis was performed via the R package ‘metafor’ (Wolfgang Viechtbauer, Maastricht, Limburg, Netherlands), which uses a random effect model with restricted maximum likelihood accounting for the between-study variance (Viechtbauer, 2010). Dietary patterns that were population-specific were not considered in the meta-analysis, as pooling for quantitative synthesis requires that exposure be sufficiently comparable to avoid introducing bias and undermining the validity of the estimates (West et al, 2010).

Heterogeneity across studies was assessed by the I² statistic, defined as the percentage of the variability in effect estimates that is due to heterogeneity rather than sampling error. Publication bias was assessed via a funnel plot via Egger’s and Begg’s tests (Sterne and Egger, 2001).

We conducted a leave-one analysis to assess the influence of each individual study on the pooled effect estimates. We conducted sensitivity analyses excluding the study conducted in a cohort of health professionals, as well subgroup analyses to assess the variability according to the dietary assessment: self-reported food frequency questionnaire (FFQ) versus supervised FFQ. Due to the variability in covariate adjustment across studies, age being the only consistently included confounder, interaction and effect-modification were not feasible.

Results

Study selection

The study selection process is described in the flowchart in Fig. 1. From 8647 studies identified from the databases, 3822 duplicates were removed. A total of 4825 studies were screened by title and abstract. After screening, 4773 studies did not match the required eligibility criteria, and 52 studies were retrieved; six studies were considered in this systematic review.

Fig. 1.

Study selection flowchart.

The study and population characteristics are reported in Table 1. The studies were published between 2004 and 2023 and were conducted in five different countries: the USA (n = 2) (Tseng et al, 2004; Wu et al, 2006), Japan (n = 1) (Shin et al, 2018), Australia (n = 1) (Muller et al, 2009), Canada (n = 1) (Willemsen et al, 2021), and Spain (n = 1) (Castelló et al, 2023). The studies included populations between 3779 and 47 725 men aged 25–81 years. The follow-up period was a minimum of 8 years and a maximum of 22 years. In total, the studies selected allowed the follow-up of 153 925 men at risk of PCa. The ethnicities reported are mostly of European descent, except for two studies that reported a Japanese population and a Caucasian population without reference to origin (Tseng et al, 2004; Shin et al, 2018). All studies used a FFQ to collect information about dietary habits. One study used an FFQ, and a 24-hour recall validating the nutrient content and portion size of the FFQ (Tseng et al, 2004). All studies used PCA to identify dietary patterns, with five out of six applying orthogonal rotation to ensure uncorrelated components and better segregation of dietary patterns, thereby improving interpretability (Tseng et al, 2004; Wu et al, 2006; Muller et al, 2009; Shin et al, 2018; Willemsen et al. 2021). One study did not use any matrix rotation (Castelló et al, 2023). The factor loading threshold, correlating variables to factors, varied between 0.15 and 0.30 across the studies. Outcome ascertainment was performed via medical records or country-specific cancer registries. For PCa incidence, all studies used a multivariable Cox regression model and reported a relative risk (Tseng et al, 2004; Wu et al, 2006) or hazard ratio (Muller et al, 2009; Shin et al, 2018; Willemsen et al. 2021; Castelló et al, 2023).

Table 1.

Main characteristics of studies included in the systematic review

Study details	Population characteristics	Exposure	Outcome	Measure of association
1. First author 2. Year of publication 3. Country/region 4. Study name 5. Reference	1. Age (mean) 2. Ethnicity 3. Total sample size 4. Follow-up	1. Dietary patterns name 2. Tool for collecting dietary information 3. Method to identify patterns 4. Parameters of the method	1. Outcome name 2. Case ascertainment	1. Statistical model (measure of association) 2. Description of measure 3. Sample size for analysis (N) 4. PCa cases (n)
1. Marylin Tseng 2. 2004 3. USA 4. NHANES I/NHEFSa 5.(Tseng et al., 2004)	1. 25–74 years 2. 91% Caucasians, 11% Afro Americans 3. 3779 4. 8–10 years	1. Vegetable and fruit, red meat-starch Southern 2. 105-item FFQ + 24 h food record (for validation) 3. PCA 4. Orthogonal rotation/ factor loading ≥ 0.20 /3 factors	1. Overall PCa 2. Record linkage	1. Multivariable Cox model (relative risk RR) 2. Tertiles 3. N = 3779 4. n = 136
1. Kana Wu 2. 2006 3. USA 4. HPFSb 5.(Wu et al., 2006)	1. 40–75 years 2. 72% North European, 21% South European, 4.8% other 3. 47 725 4. 14 years	1. Prudent, western 2.131 item FFQ 3. PCA 4.Orthogonal rotation/ factor loading ≥ 0.15 /2 factors	1. Overall PCa 2. Record linkage	1. Multivariable Cox model (relative risk RR) 2. Quintiles 3. N = 47 725 4. n = 3002
1. David C. Muller 2. 2009 3.Australia 4. MCCSc 5.(Muller et al., 2009)	1. 27–81 years 2. 69% (Australia); 13% (Italy) 11% (Greece); 6% (England) 3. 14 627 4. 13.6 years	1. Vegetable, meat and potatoes, mediterranean, fruit and salad 2. 121-item FFQ 3. PCA 4.Orthogonal rotation/ factor loading ≥ 0.30 /4 factors	1. Overall PCa 2. Victorian Cancer Registry	1. Multivariable Cox model (relative risk RR) 2. Quartiles 3. N = 14 627 4. n = 1018
1. Sangah Shin 2. 2018 3. Japan 4. JPHCd 5. (Shin et al., 2018)	1. 40–69 years 2. Japanese 3. 46 469 4. 19–22 years	1. Prudent, westernized, traditional 2. 138-item FFQ 3. PCA 4.Orthogonal rotation/ factor loading ≥ 0.15 /3 factors	1. Overall PCa 2. Record linkage/ hospital notification/ cancer registry	1. Multivariable Cox model (Hazard ratio HR) 2. Quintiles 3. N = 46 469 4. n = 1156
1. R.F.Willemsen 2. 2021 3. Canada 4. ATPe 5. (Willemsen et al. 2021)	1. 35–69 years 2. 92.4% European descent; 7.6% other 3. 10 109 men 4. 13.3 years	1. Western, prudent, sugar, fruit and dairy 2. 124-item FFQ 3. PCA 4. Orthogonal rotation/ factor loading ≥ 0.20 /3 factors	1. Overall PCa 2. Alberta’s cancer registry	1. Multivariable Cox model (Hazard ratio HR) 2. Quartiles 3. N = 10 109 4. n = 442
1. Adela Castelló 2. 2023 3. Spain 4. EPICf 5. (Castelló et al., 2023)	1. 29–69 years 2. Five Spanish regions: Asturias, Gipuzkoa and Navarra, Granada and Murcia 3. 15 296 men 4. 16.99 years	1. Western, prudent and mediterranean 2. 117-item FFQ 3. PCA 4. No rotation/ factor loading ≥ 0.30 /3 factors	1. Overall PCa 2. Population-based cancer registries of the five regions	1. Multivariable Cox model (Hazard ratio HR) 2. Quartiles 3. N = 15 296 4. n = 609

FFQ, Food Frequency Questionnaire; PCa, prostate cancer; PCA, principal component analysis; RRR, reduced rank regression.

^aNational Health and Nutrition Examination Survey/ Epidemiologic Follow-up Study.

^bHealth Professionals Follow-Up Study.

^cThe Melbourne Collaborative Cohort Study.

^dJapan Public Health Center-Based Prospective Study.

^eAlberta’s Tomorrow Project.

^fEuropean Prospective Investigation into Cancer and nutrition.

The adjustment variables varied across studies (Table 3). Five of the six studies adjusted for age as a continuous covariate (Wu et al, 2006; Muller et al, 2009; Shin et al, 2018; Willemsen et al. 2021), one study dichotomized age (Tseng et al, 2004), and one study used age for stratification (Castelló et al, 2023). Four studies adjusted for BMI (Wu et al, 2006; Shin et al, 2018; Willemsen et al. 2021; Castelló et al, 2023), and for total energy intake (Wu et al, 2006; Muller et al, 2009; Willemsen et al. 2021; Castelló et al, 2023). Three studies adjusted for race/ethnicity (Tseng et al, 2004; Wu et al, 2006; Muller et al, 2009) and for smoking (Wu et al, 2006; Shin et al, 2018; Willemsen et al. 2021). Only one study adjusted for PCa family history (Wu et al, 2006).

Table 3.

Dietary patterns identified by PCA and main results

Author Country/state/study name	Main findingsa				Confounding variables for adjustment
	Name and description of the identified dietary patterna		HR/RR [CI95%]	P trend
Tseng et al. USA (NHANES I/NHEFS)b	Vegetable and fruit: fruits, vegetables, fish and seafood.	T1 T2 T3	Ref 1.5 [0.9–2.3] 1.2 [0.7–2.0]	Ptrend = 0.64	Age (dichotomized <65 vs. ≥65), race, poverty census enumeration district, family income, USA region, education, recreational sun exposure, smoking status, total energy intake
	Red meat-starch: red meats, potatoes, salty snacks, cheese, sweets and desserts.	T1 T2 T3	Ref 0.7 [0.5–1.2] 0.8 [0.4–1.4]	Ptrend = 0.37
	Southern: beans, rice, and traditional southern foods like cornbread, grits, sweet potatoes, and okra.	T1 T2 T3	Ref 0.9 [0.6–1.4] 0.6 [0.4–1.1]	Ptrend = 0.08
Wu et al. USA HPFSc	Prudent: vegetables, legumes, fruits, fish and poultry, whole grains.	Q1 Q2 Q3 Q4 Q5	Ref 0.97 [0.86–1.10] 1.10 [0.98–1.24] 1.01[0.90–1.14] 0.95 [0.84–1.07]	Ptrend = 0.37	Age, race, height, smoking status, family history of prostate cancer history of vasectomy, vigorous exercise, BMI alcohol intake, total energy intake
	Western: red meat, processed meat, refined grains, sweets and desserts, fries and high-fat dairy products.	Q1 Q2 Q3 Q4 Q5	Ref 0.98 [0.88–1.10] 1.03[0.92–1.16] 1.00 [0.89–1.13] 1.02[0.91–1.15]	Ptrend = 0.62
Muller et al. Australia MCCSd	Vegetable: vegetables, cooked dried legumes, wholemeal bread, fruits, yogurt.	Q1 Q2 Q3 Q4	Ref 1.11 [0.90–1.36] 1.02 [0.83–1.27] 1.12 [0.90–1.40]	Ptrend = 0.5	Age, total energy intake, ethnicity BMI, physical activity, smoking, alcohol intake, education
	Mediterranean: vegetables, olive oil, cooked dried legumes, steamed fish, ricotta cheese, boiled chicken, fruits.	Q1 Q2 Q3 Q4	Ref 0.93 [0.79–1.11] 1.14 [0.95–1.37] 0.93 [0.74–1.18]	Ptrend = 0.9
	Fruit and salad: fruits (apricot, peach, watermelon) and vegetables (cucumber, salad greens, celery).	Q1 Q2 Q3 Q4	Ref 1.14 [0.96–1.36] 1.10 [0.92–1.32] 1.00 [0.81–1.23]	Ptrend = 0.6
	Meat and potatoes: red meats (beef, lamp, pork), potato cooked in fat, savory pastries, White bread.	Q1 Q2 Q3 Q4	Ref 1.00 [0.84–1.20] 1.04 [0.87–1.24] 0.87 [0.71–1.08]	Ptrend = 0.2
Shin et al. Japon JPHCe	Prudent: vegetables, fruits, seaweeds, potatoes, mushrooms, soy products, noodle, seafood and fish.	Q1 Q2 Q3 Q4 Q5	Overall PCa Ref 0.91 [0.74–1.12] 1.05 [0.87–1.27] 1.01 [0.84–1.23] 0.93 [0.76–1.12]	Ptrend = 0.7155	Age at baseline (5-years groups), study area (ten public health centers) BMI, smoking status, vitamin supplement use, marital status
		Q1 Q2 Q3 Q4 Q5	Localized PCa Ref 0.98 [0.77–1.26] 1.19 [0.94–1.50] 1.15 [0.91–1.45] 1.00 [0.79–1.27]	Ptrend = 0.555
		Q1 Q2 Q3 Q4 Q5	Advanced PCa Ref 0.82 [0.56–1.21] 0.80 [0.55–1.16] 0.76 [0.52–1.10] 0.78 [0.55–1.12]	Ptrend = 0.315
	Westernized: meat and processed meat, soup, dairy products, confectionaries, fruit juice, fish products, coffee, tea, soft drink, sauces, and alcohol.	Q1 Q2 Q3 Q4 Q5	Overall PCa Ref 1.06 [0.89–1.26] 1.04 [0.86–1.24] 1.22 [1.02–1.47] 1.22 [1.00–1.49]	Ptrend = 0.02
		Q1 Q2 Q3 Q4 Q5	Localized PCa Ref 1.03[0.83–1.29] 1.03 [0.83–1.27] 1.18 [0.95–1.47] 1.24 [0.971.57]	Ptrend = 0.045
		Q1 Q2 Q3 Q4 Q5	Advanced PCa Ref 1.12 [0.80–1.57] 1.09 [0.76–1.56] 1.25 [0.87–1.81] 1.23 [0.82–1.84]	Ptrend = 0.23
	Traditional: seafood and fish (oily, salty, and lean fish, and salmon), chicken, liver, tea (green/Oolong), and sake	Q1 Q2 Q3 Q4 Q5	Overall PCa Ref 1.17 [0.95–1.45] 1.10 [0.87–1.39] 1.10 [0.86–1.40] 1.01 [0.79–1.29]	Ptrend = 0.89
		Q1 Q2 Q3 Q4 Q5	Localized PCa Ref 1.08 [0.83–1.40] 1.06 [0.80–1.41] 1.10 [0.82–1.48] 0.97 [0.72–1.31]	Ptrend = 0.869
		Q1 Q2 Q3 Q4 Q5	Advanced PCa Ref 1.36 [0.89–2.07] 1.16[0.73–1.84] 1.05 [0.65–1.71] 1.16 [0.72–1.88]	Ptrend = 0.830
Willemsen et al. Canada ATPf	Prudent: vegetables, fruits, dry bean and pea, and lean meat from fish and other sea food.	Q1 Q2 Q3 Q4	Ref 0.82 [0.63–1.06] 0.86 [0.66–1.11] 0.76 [0.57–1.00]	Ptrend = 0.077	Age, sex, BMI, total energy intake smoking status, physical activity
	Western: grain servings, especially nonwhole grains, vegetables, white potatoes, cheese, meats from lamb, pork, beef and luncheon meats (red and processed meats), discretionary fats, and teaspoons of added sugar.	Q1 Q2 Q3 Q4	Ref 1.17 [0.90–1.52] 1.18 [0.89–1.57] 1.22 [0.84–1.76]	Ptrend = 0.26
	Sugar, fruit and dairy: grain servings, especially whole grains, fruits, dairy, and teaspoons of added sugar.	Q1 Q2 Q3 Q4	Ref 1.06 [0.82–1.36] 1.01 [0.78–1.31] 1.00 [0.76–1.31]	Ptrend = 0.94
Castelló et al. Spain Spanish cohort (EPIC)g	Prudent: juices, low-fat dairy products, whole grains, vegetables, fruits.	Q1 Q2 Q3 Q4	Overall PCa Ref 0.89 [0.71–1.11] 0.90 [0.71–1.14] 1.08 [0.84–1.38]	Ptrend = 0.60	Age (stratification) total energy intake, BMI, height, educational level at recruitment
		Q1 Q2 Q3 Q4	Gleason >6 PCa Ref 0.93 [0.63–1.35] 1.01 [0.68–1.50] 1.05 [0.68–1.61]	Ptrend = 0.73	-
		Q1 Q2 Q3 Q4	ISUPh = 3 + 4 + 5 Ref 0.95 [0.59–1.53] 0.86 [0.51–1.46] 0.80 [0.44–1.46]	Ptrend = 0.43	-
	Western: caloric drinks, high-fat dairy, convenience food and sauces, refined grains, processed meats, sweets.	Q1 Q2 Q3 Q4	Overall PCa Ref 1.11 [0.85–1.44] 1.21 [0.92–1.60] 1.29 [0.96–1.72]	Ptrend = 0.07	-
		Q1 Q2 Q3 Q4	Gleason >6 PCa Ref 1.38 [0.87–2.19] 1.61 [1.00–2.59] 1.60 [0.96–2.67]	Ptrend = 0.08	-
		Q1 Q2 Q3 Q4	ISUP = 3 + 4 + 5 Ref 1.97 [0.98–3.93] 2.72 [1.35–5.51] 2.29 [1.07–4.92]	Ptrend = 0.04	-
	Mediterranean: vegetables, oily fish, potatoes, seafood and fish, olive oil, fruits	Q1 Q2 Q3 Q4	Overall PCa Ref 0.89 [0.67–1.19] 1.02 [0.78–1.33] 1.05 [0.79–1.38]	Ptrend = 0.42	-
		Q1 Q2 Q3 Q4	Gleason >6 PCa Ref 0.88 [0.55–1.43] 1.02 [0.65–1.61] 1.17 [0.74–1.87]	Ptrend = 0.28	-
		Q1 Q2 Q3 Q4	ISUP = 3 + 4 + 5 Ref 0.66 [0.34–1.26] 1.02 [0.57–1.82] 1.13 [0.62–2.06]	Ptrend = 0.29	-

^aMain findings are presented for Hazard ratio (HR) or relative risk (RR) [CI_95%] of full/multivariate models and for overall risk of PCa.

^bNational Health and Nutrition Examination Survey/Epidemiologic Follow-up Study.

^cHealth Professionals Follow-Up Study.

^dThe Melbourne Collaborative Cohort Study.

^eJapan Public Health Center-Based Prospective Study.

^fAlberta’s Tomorrow Project.

^gEuropean Prospective Investigation into Cancer and nutrition.

^hISUP: International Society of Urological Pathology; ISUP 1 = Gleason 3 + 3; ISUP 2 = Gleason 3 + 4; ISUP 3 = Gleason 4 + 3; ISUP 4 = Gleason 4 + 4, 3 + 5, 5 + 3; ISUP 5 = Gleason 4 + 5, 5 + 4.

Quality assessment: risk of bias in studies

The assessment of the risk of bias by the NOS for each study is reported in Table 2. Among the six studies, two studies received the highest selection score because exposure was assessed via structured interviews and the FFQ was administered under the supervision of a qualified professional (Tseng et al, 2004; Castelló et al, 2023). For the remaining studies, the FFQ was self-reported by the participants. For comparability, all studies were rewarded with two stars since they all controlled for age and other additional factors. For the outcome, four studies received three stars because the follow-up rate was higher than 80% (Wu et al, 2006; Muller et al, 2009; Willemsen et al. 2021; Castelló et al, 2023). All the studies selected scored a minimum of 7 stars and were deemed ‘good quality’ according to the NOS.

Table 2.

Newcastle–Ottawa Quality Assessment Scale (NOS) for cohort studies

References	Selection				Comparability	Outcome			Final score	aAssessment
	Representativeness of the exposed cohort	Selection of the nonexposed cohort	Ascertainment of exposure	Demonstration that outcome of interest was not present at start of study	Comparability of cohorts based on the design or analysis	Assessment of outcome	Was follow-up long enough for outcomes to occur	Adequacy of follow-up of cohortsb
Tseng et al. (2004)	*	*	*	*	**	*	*		8	Good quality
Wu et al. (2006)		*		*	**	*	*	*	7	Good quality
Muller et al. (2009)	*	*		*	**	*	*	*	8	Good quality
Shin et al. (2018)	*	*		*	**	*	*		7	Good quality
Willemsen et al. (2021)	*	*		*	**	*	*	*	8	Good quality
Castelló et al. (2023)	*	*	*	*	**	*	*	*	9	Good quality

^aA study can be awarded a maximum of one star for each numbered item within the selection and exposure categories. A maximum of two stars can be given for comparability. Good quality: 3 or 4 stars in selection domain AND 1 or 2 stars in comparability domain AND 2 or 3 stars in outcome/exposure domain; fair quality: 2 stars in selection domain AND 1 or 2 stars in comparability domain AND 2 or 3 stars in outcome/exposure domain; poor quality: 0 or 1 star in selection domain OR 0 stars in comparability domain OR 0 or 1 stars in outcome/exposure domain.

^bA follow-up rate ≥80% is deemed adequate for large cohort studies.

Systematic review: dietary patterns and their associations with prostate cancer risk

The main results are presented in Table 3. Each study identified at least two dietary patterns. Similar patterns emerged across studies. A ‘western’ pattern, also referred to as ‘westernized’ (Shin et al, 2018), meat and potatoes (Muller et al, 2009) and ‘red meat and starch’ (Tseng et al, 2004), is characterized by a high intake of red meat and processed meats, refined grains, high-fat dairy products and ultra-processed foods. A ‘prudent’ pattern, also called ‘vegetable and fruits’ (Tseng et al, 2004) or ‘fruit and salad’(Muller et al, 2009), is characterized by a high intake of fruits and vegetables, fish and poultry. A detailed list of all food items contributing to each pattern is provided in Supplemental 4, Supplemental Digital Content, https://links.lww.com/EJCP/A592.

Other patterns observed were population-specific, such as the ‘southern’ pattern in the USA, which is rich in traditional southern food such as cornbread, grit, sweet potatoes, and okra (Tseng et al, 2004). A ‘traditional’ pattern was observed in a Japanese population, characterized by high intakes of fish and seafood, chicken, liver, tea, and sake (Shin et al, 2018). A ‘sweet, fruits and dairy’ pattern has been identified in a Canadian population that is rich in whole grains, fruits, dairy products and added sugar. Two studies identified a pattern named ‘mediterranean’; one study was conducted in Australia, and one study was conducted in Spain (Muller et al, 2009; Castelló et al, 2023).

For the association with overall PCa risk, two studies reported no association (Wu et al, 2006; Muller et al, 2009). Two studies reported a negative trend for the ‘southern’ pattern (Tseng et al, 2004) [RR_Q3vs.Q1 = 0.6; confidence interval (CI)_95% (0.4–1.1); P_trend = 0.08] and the ‘prudent’ pattern (Willemsen et al. 2021) [hazard ratio_Q4vs.Q1 = 0.76; CI_95% (0.57–1.00); P_trend = 0.077]. One study reported a significant positive association for the ‘westernized’ pattern in a Japanese population [hazard ratio_Q5vs.Q1 = 1.22; CI_95% (1.00–1.49); P_trend = 0.021] (Shin et al, 2018). A positive trend was reported for the ‘western’ diet in a Spanish population [hazard ratio_Q4vs.Q1 = 1.29; CI_95% (0.96–1.72); P_trend = 0.074] (Castelló et al, 2023).

Meta-analysis

A meta-analysis was conducted on the ‘western’ and ‘prudent’ patterns identified in the six studies, comparing the highest level of exposure with the lowest. The pooled hazard ratio for the ‘prudent’ patterns (Fig. 2a) reported no association with the overall risk of PCa [hazard ratio_Pooled = 0.98; CI_95% (0.91–1.06); P_value = 0.572], with no heterogeneity detected between studies (I² = 0%). Similarly, no association was reported for the ‘western’ patterns (Fig. 2b) [hazard ratio_Pooled = 1.07; CI_95% (0.93–1.22); P_value = 0.337], in the presence of moderate heterogeneity (I² = 46.55%, Q = 8.77, τ² = 0.01). For western patterns, the leave-one out analysis (Supplemental 5, Supplemental Digital Content, https://links.lww.com/EJCP/A592) indicated that no single study drove the observed heterogeneity.

Fig. 2.

Forest plot for predominant dietary patterns. (a) Forest plot for ‘prudent’ dietary patterns. (b) Forest plot for ‘western’ dietary patterns.

Publication bias

The funnel plot for the ‘prudent’ and ‘western’ patterns revealed no evidence of asymmetry (Supplemental 6 and 7, Supplemental Digital Content, https://links.lww.com/EJCP/A592), Egger’s test, P_value = 0.87 and P_value = 0.96; Begg’s test, P_value = 0.71 and P_value = 0.71, respectively.

Sensitivity analysis

One study was conducted on a cohort of health professionals (Wu et al, 2006), and a sensitivity analysis was performed to assess its impact. The pooled hazard ratio for the five remaining studies was hazard ratio_Pooled = 1.09; CI_95% [0.91–1.31]; P_value = 0.363 for the ‘western’ patterns and hazard ratio_Pooled = 0.99; CI_95% [0.88–1.11]; P_value = 0.902 for the ‘prudent’ patterns. Similarly, the pooled estimate was nonsignificant for the principal analysis. No significant difference was observed in the subgroup analysis comparing studies with self-reported FFQ (Wu et al, 2006; Muller et al, 2009; Shin et al, 2018; Willemsen et al. 2021) and supervised FFQ (Tseng et al, 2004; Castelló et al, 2023) for the western and prudent patterns (Supplemental 8, Supplemental Digital Content, https://links.lww.com/EJCP/A592).

Discussion

Main findings

This systematic review identified six cohort studies that explored dietary patterns identified ‘a posteriori’ and their associations with the risk of PCa. Two main common patterns emerged: a ‘western’ pattern and a ‘prudent’ pattern. The meta-analysis revealed no significant association with PCa risk for either pattern, with low heterogeneity for the ‘prudent’ pattern (I² = 0%) and moderate heterogeneity for the ‘western’ pattern (I² = 46.55%). This heterogeneity likely reflects clinical differences, primarily geographical differences across study settings and the use of population-specific FFQs.

These results are partly in accordance with the only other systematic review published on the association between an ‘a posteriori’ dietary pattern and PCa risk (Fabiani et al, 2016). The latter reported no association with the ‘prudent’ pattern but an overall significant association with the ‘western’ pattern. This association was driven by the inclusion of case–control studies in the meta-analysis, which had substantial heterogeneity for the pooled results ($I_{{}_{{}_{\mathrm{pooled results}}}}^2$ = 74.63%; P = 0.0001), moderate heterogeneity for the case–control studies ($I_{{}_{{}_{\mathrm{c}\mathrm{a}\mathrm{s}\mathrm{e}–\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}}}^2$ = 54.61%; P = 0.02) and low heterogeneity for the cohort studies ($I_{{}_{{}_{\mathrm{c}\mathrm{o}\mathrm{h}\mathrm{o}\mathrm{r}\mathrm{t}}}}^2$ = 4.14%; P = 0.35) (Deeks et al, 2025). Case control studies are typically subjected to recall bias and potential reverse causation, as the study design does not ensure the temporality between exposure and outcome (Tenny et al, 2023; Kopec and Esdaile, 2025).

Population-specific patterns emerged in each study reflecting geographical diversity. While pooling estimates for these patterns in not valid as quantitative synthesis requires sufficiently comparable exposures (West et al, 2010), they remain valuable for characterizing and identifying local dietary patterns. Notably, the ‘traditional’ pattern in the Japanese population and the ‘mediterranean’ pattern in the Spanish population, were not associated with PCa risk (Shin et al, 2018; Castelló et al, 2023), although prior studies reported that both patterns were associated with a lower risk of PCa (Castelló et al, 2018; Gregg et al, 2021). These discrepancies may stem from differences in study design, population characteristics possibly rendering these dietary patterns more uniform, limited statistical power and variations in the adjustment for confounders. In contrast, the westernized dietary pattern was associated with an increased risk of PCa in the Japanese population. The most associated factors were high intake of processed meats and dairy products, core components of the western diet and primary sources of saturated fatty acids among Japanese men (Sasaki et al, 2003). The incidence of PCa in Japan has risen rapidly over the past decades from 7.1 in 1975 to 50.1 per 100 000 in 2022; this increase has potentially been attributed to the introduction of prostate-specific antigen testing (Kageyama et al, 2022). However, a 13-year (2003–2015) trend analysis also reported the continuous westernization of the Japanese diet, suggesting that it may also be associated to the observed increase in PCa incidence (Murakami et al, 2018).

Similarly, in a study conducted in Spain (Castelló et al, 2023), a positive trend was observed for the ‘western’ pattern for overall PCa risk, and a significant association was observed for tumors with International Society of Urological Pathology (ISUP) grades 3–5. The differential effect of tumor grade may result from PCa screening, which was introduced in 1994 to detect low-grade tumors, leading to distinct risk groups that explain the observed associations (Schröder et al, 2009). This finding also suggests a tumor grade-specific association between adherence to the ‘western’ diet and PCa risk. A plausible explanation for this difference may be that aggressive PCa exhibits greater sensitivity to dietary modulation (Labbé et al, 2015). An example of this may be the more pronounced inverse association reported between fish/long-chain omega-3 fatty acids and aggressive PCa tumors compared with earlier-stage tumors, which are often detected by screening (Reese et al, 2009). Notably, it has also been reported that the Spanish population is gradually shifting away from the traditional mediterranean diet and increasingly adopting a more westernized dietary pattern (León-Muñoz et al, 2012). This shift might also explain the observed association, although further studies are needed to confirm these findings. The shift toward a ‘western’ diet has been documented mainly in lower and middle-income countries (Popkin, 2004). In China, for example, this transition has been partially modulated by household income (Howard et al, 2021). However, to the best of our knowledge, this shift has not been thoroughly investigated in high-income countries, where dietary changes may be driven by different factors, such as inflation, food availability, or broader nutritional awareness.

For the population-specific patterns, only the southern American pattern, rich in typical southern foods such as beans, cornbread, sweet potatoes, and okra, showed a negative correlation with PCa risk (P_value = 0.08) (Tseng et al, 2004). This trend was not attributed to any specific food but to the pattern itself. This may also be related to sun exposure, as the incidence of PCa in the U.S. has been found to be inversely related to geographic levels of Ultraviolet radiation, with lower rates in the south (Mandal et al, 2009). This is an important observation, as it highlights the potential synergistic effects of dietary patterns when combined with other lifestyle factors rather than acting in isolation (Gherasim et al, 2020).

Strengths and limitations

The main strength of this systematic review lies in its exclusive inclusion of cohort studies, which minimize recall bias and establish temporality, as the exposure precedes the outcome (Freudenheim, 1999). Additionally, these studies feature large population samples and relatively long follow-up periods (minimum of eight years). Focusing exclusively on a posteriori dietary pattern offers the advantage of yielding more interpretable and realistic representations of actual dietary behaviors (Zhao et al, 2021). We note, however, that the number of eligible studies was limited to six which constrains the robustness of the findings.

Although the bias assessment tool indicated a lower risk of bias overall, it is important to note that in four of the six studies included, the FFQ was self-reported. This can introduce recall bias, response bias, social desirability bias and misclassification (Popoola et al, 2023; Naska et al, 2017). The FFQs employed in the studies encompassed a diverse range of foods tailored to specific populations and geographical contexts which may limit generalizability of the results to other settings. However, this consideration echoes Gene V. Glass’s original ‘apples and oranges’ metaphor, which he endorsed comparisons provided that the inferences remained at the broader categorical level of ‘fruits’ (Glass, 1978). Similarly, variations in the specific food items included in the FFQs do not constitute a substantial limitation, provided that the conclusions pertain to overall dietary patterns.

Although all studies employed PCA to derive dietary patterns, the method’s implementation involved subjective decisions. The number of retained factors ranged from 2 to 4, and the thresholds for factor loadings, indicating the correlation between a variable and a factor, varied from 0.15 to 0.30. Moreover, one study (Castelló et al, 2023) did not perform a matrix rotation, which is essential to ensure the interpretability and independence of the derived dietary patterns (Varraso et al, 2012). These methodological differences could influence the interpretation and consistency of the identified patterns across studies. There were also differences in the adjustment variables used in the Cox regression models assessing the associations between dietary patterns and PCa risk. This raises the possibility of residual confounding due to the omission of established confounders, such as family history (Rawla, 2019), which was accounted for in only one study (Wu et al, 2006), or ethnicity, which was accounted for in only three studies (Tseng et al, 2004; Wu et al, 2006; Muller et al, 2009). Failure to account for these confounders may introduce bias, leading to either an overestimation or underestimation of the associations and contributing to uncertainty in the findings. These differences also add an additional layer of complexity when interpreting population-specific patterns as the association may further be influenced by differences in adjustment. All six studies provided little to no information on how confounders were selected, and to our knowledge, no study has used directed acyclic graphs, which may lead to biased results.

From a clinical and public health perspective, our findings report null associations, which may be explained by methodological limitations such as differences in confounders adjustment, variation in PCA parameters used to derive dietary patterns, and limited number of available studies. These results should not, however, be interpreted as dismissing the possibility of an association. Notably, while null associations were observed for data-driven dietary patterns, inverse associations have been reported for ‘a priori’ dietary indices such as plant-based diet indices (Loeb et al, 2022) and Mediterranean Diet Score (Kenfield et al, 2013). This suggests that dietary patterns based on predefined scoring systems may currently offer more actionable insights for clinical practice, given the limited evidence available on a posteriori dietary patterns.

Conclusions

In this meta-analysis, we observed no significant association between a posteriori dietary patterns and PCa risk. While the systematic evaluation suggested a possible increased risk of localized and aggressive PCa associated with western dietary patterns in populations, where this pattern diverges from more traditional dietary patterns, these observations should be interpreted with caution due to the limited number of studies and methodological differences among them. Future research should prioritize the development of harmonized methodologies for both the identification of a posteriori dietary patterns and the selection of confounders in studies examining the association between diet and PCa risk.

Acknowledgements

The author thank Dr. Hervé Tchala Vignon Zomahoun for his contribution to the methodology and Dr. Karine Robitaille for the critical revision of this manuscript.

This study was funded by Cancer Research Society of Canada, ministère de l’Enseignement supérieur, de la recherche, de la science, et de la technologie du Québec, and Fonds de Recherche du Québec – Santé (FRQ-S) through the GRePEC program, as well as the CHU de Québec-Université Laval. A career award from the FRQ-S supported by Dr. Vincent Fradet. JPDC is a research-scholar of the FRQ-S.

Data described in the manuscript is a systematic review of previously published data. All data is publicly available from the original sources cited in the reference list.

PROSPERO registration: CRD42024551585.

Conflicts of interest

There are no conflicts of interest.

Supplementary Material

ejcp-35-570-s001.docx ^{(602.5KB, docx)}