The Beneficial Effect of a Healthy Dietary Pattern on Androgen Deprivation Therapy-Related Metabolic Abnormalities in Patients with Prostate Cancer: A Meta-Analysis Based on Randomized Controlled Trials and Systematic Review

Metabolic abnormalities as side effects of androgen-deprivation therapy (ADT) can accelerate progression of prostate cancer (PCa) and increase risks of cardiovascular diseases. A healthy dietary pattern (DP) plays an important role in regulating glycolipid metabolism, while evidence about DP on ADT-related metabolic abnormalities is still controversial. To explore the effect of DP on metabolic outcomes in PCa patients with ADT, PubMed, Embase, Cochrane, and CINAHL were searched from inception to 10 September 2022. Risk of biases was evaluated through Cochrane Collaboration’s Tool. If heterogeneity was low, the fixed-effects model was carried out; otherwise, the random-effects model was used. Data were determined by calculating mean difference (MD) or standardized MD (SMD) with 95% confidence intervals (CIs). Nine studies involving 421 patients were included. The results showed that healthy DP significantly improved glycated hemoglobin (MD: −0.13; 95% CI: −0.24, −0.02; p = 0.020), body mass index (MD: −1.02; 95% CI: −1.29, −0.75; p < 0.001), body fat mass (MD: −1.78; 95% CI: −2.58, −0.97; p < 0.001), triglyceride (MD: −0.28; 95% CI: −0.51, −0.04; p = 0.020), systolic blood pressure (MD: −6.30; 95% CI: −11.15, −1.44; p = 0.010), and diastolic blood pressure (MD: −2.94; 95% CI: −5.63, −0.25; p = 0.030), although its beneficial effects on other glycolipid metabolic indicators were not found. Additionally, a healthy DP also lowered the level of PSA (MD: −1.79; 95% CI: −2.25, −1.33; p < 0.001). The meta-analysis demonstrated that a healthy DP could improve ADT-related metabolic abnormalities and be worthy of being recommended for PCa patients with ADT.


Introduction
Prostate cancer (PCa) is the most common cancer of the urinary system and ranks second of all malignant tumors in men worldwide [1]. Androgen-deprivation therapy (ADT) is considered the standard treatment for advanced PCa. It is used to reduce prostate tumor growth and prolong progression-free survival [2,3]. However, ADT as a doubleedged sword, can also result in severe adverse effects, of which metabolic syndrome is one of the common side-effects [4]. About 60% of PCa patients with ADT experienced at least one metabolic abnormality [5]. ADT-related metabolic abnormalities could accelerate progression of PCa, increase insulin resistance (IR), and lead to development of dyslipidemia and sarcopenic obesity, which increase the risks of cardiovascular diseases (CVD)

Study Selection
The two authors independently reviewed the titles and abstracts of all studies to screen for eligibility. In the case of disagreement, a third author was the final arbiter. The Population, Intervention, Comparator, Outcomes, Studies (PICOS) model followed [16]. Inclusion criteria were as follows: (1) All participants had been clinically diagnosed with prostate cancer by prostate puncture and had received any form of ADT including hormone therapy or bilateral orchiectomy; (2) dietary intervention either alone or in combination with exercise; (3) at least one of metabolic-related outcome had been reported (e.g., blood pressure (BP), body mass index (BMI), weight, glucose or lipid metabolism, body composition); (4) randomized controlled trial (RCT) design; and (5) a comparison group with no dietary intervention, regardless of exercise. Exclusion criteria were as follows: (1) cross-sectional, case-control studies, prospective or protrospective cohorts, meta-analyses, reviews, comments, editorial letters, or conference abstracts, studies without results and (2) animal studies.

Data Extraction
Two authors independently extracted the data of included articles based on a preestablished Excel spread sheet. Any disagreement was resolved by a third author. Extracted data of all included studies were as follows: (1) general information (authors, year of publication, and country); (2) participant information (sample size, age, and ADT duration); (3) intervention information (intervention protocol and duration); (4) outcomes (BP, BMI, weight, glucose or lipid metabolism, body composition, prostate-specific antigen (PSA), and fatigue).

Assessment of the Risk of Biases
Quality assessment of the included studies was evaluated by two investigators using The Cochrane Collaboration's Tool for assessing risk of bias [17].Two authors independently screened the potential sources of bias that included selection bias (random sequence generation and allocation concealment), performance bias (blinding of participants and personnel), detection bias (blinding of outcome assessment), attrition bias (incomplete outcome data), and reporting bias (selective reporting). If there was any disagreement, a third investigator made the final decision. The criteria for risk assessment of bias were "low risk bias", "high risk bias", and "uncertainty".

Statistical Analysis
Meta-analysis was performed with RevMan 5.3 (http://tech.cochrane.org/revman/ download, accessed on 12 September 2022). The assessment of heterogeneity was by means of I 2 statistic and Cochran's Q, and p < 0.10 and I 2 > 50% indicated that the heterogeneity was statistical significance [18]. If there had been no statistical heterogeneity, the fixedeffects model was carried out. Otherwise, the random-effects model was executed. Mean difference (MD) with 95% confidence intervals (CIs) was pooled for metabolic outcomes and PSA. However, for fatigue, these was converted into standardized mean difference (SMD) due to the use of different measurement scales. Subgroup analysis was performed to identify the effect of only healthy DPs on clinical outcomes. In addition, sensitivity analysis was conducted by removing one study at a time from the meta-analysis to assess the level of consistency of the results. Forest Plot was used to show the effect measures of each included study and the pooled effect measures. When there was missing data, we contacted the first author to obtain the original data. If the study only reported the median and quartile range, data were converted to mean and SD. p < 0.05 was statistical significance for the overall effect of the intervention, and the results of the meta-analysis are shown as forest plots [18].

Literature Data
The results of the literature search are illustrated in Figure 1. A total of 449 articles, which included four additional articles identified through hand-searching of reference lists, were identified in the literature search. One hundred and fifty-three articles were duplicates, 248 articles were excluded by screening titles and abstracts, and five articles were removed due to not being relevant to this topic. After full-text articles had been assessed and studied for eligibility, nine articles were deemed eligible in our study. The reasons for exclusion are shown in Figure 1. When there was missing data, we contacted the first author to obtain the original data. If the study only reported the median and quartile range, data were converted to mean and SD. p < 0.05 was statistical significance for the overall effect of the intervention, and the results of the meta-analysis are shown as forest plots [18].

Literature Data
The results of the literature search are illustrated in Figure 1. A total of 449 articles, which included four additional articles identified through hand-searching of reference lists, were identified in the literature search. One hundred and fifty-three articles were duplicates, 248 articles were excluded by screening titles and abstracts, and five articles were removed due to not being relevant to this topic. After full-text articles had been assessed and studied for eligibility, nine articles were deemed eligible in our study. The reasons for exclusion are shown in Figure 1.

Study Characteristics
The details of the nine studies that were included are presented in Table 1. There were 421 participants included in this review. Of the nine included studies, five [11,12,15,19,20] used general healthy dietary patterns, and the other four used a Mediterranean diet [13,21], low-carbohydrate diet [14], or low-glycemic index diet [10]. Dietary

Study Characteristics
The details of the nine studies that were included are presented in Table 1. There were 421 participants included in this review. Of the nine included studies, five [11,12,15,19,20] used general healthy dietary patterns, and the other four used a Mediterranean diet [13,21], low-carbohydrate diet [14], or low-glycemic index diet [10]. Dietary advice in all studies had been based on healthy foods or an appropriate ratio of food-to-food recommendation. All studies were RCTs published from 2011 to 2022. The range of age was from 64.3 to 71 years. The duration of ADT at recruitment ranged from 15.3 to 36.4 months. The intervention duration of five studies, three studies, and one study were 12 weeks, 24 weeks, and 20 weeks, respectively. Four studies were conducted in UK [10,11,15,19], three studies in US [12,14,20], and two in Australia [13,21]. The types of interventions were dietary patterns combined with supervised resistance and/or aerobic exercise [10][11][12]15,20,21] and dietary patterns only [13,14,19]. Additionally, fatigue reported in four studies [11,13,19,21], was assessed using Functional Assessment of Cancer Therapy-Fatigue (FACT-F) in three studies [11,13,21] and the Fatigue Severity Scale (FSS) in one study [19].

Risk Assessment of Bias in Included Studies
The risk of bias in the included studies is shown in Figure 2. Most studies reported appropriate random sequence generation, outcome data, selective reporting, and allocation concealment. The major sources of bias risk were in the failure to implement doubleblinding of participants and personnel in all studies [10][11][12][13][14][15][19][20][21] and a blind outcomeassessment in two studies [19,21].

Effect of Healthy DP on Glucose Metabolism
Two studies [10,14] investigated the effect of healthy DPs on blood-glucose-related indicators. As shown in Figure 3, a fixed-effects model was used to assess the outcomes because no significant heterogeneity was observed between the two groups _(HbAlc: p = 0.660, I 2 = 0%; homeostasis model assessment of insulin resistance (HOMA-IR): p = 0.200,

Effect of Healthy DP on Glucose Metabolism
Two studies [10,14] investigated the effect of healthy DPs on blood-glucose-related indicators. As shown in Figure 3, a fixed-effects model was used to assess the outcomes because no significant heterogeneity was observed between the two groups _(HbAlc: p = 0.660, I 2 = 0%; homeostasis model assessment of insulin resistance (HOMA-IR): p = 0.200, I 2 = 40%). Compared with that in the usual-care group, a healthy DP could significantly improve the HbAlc (MD: −0.13; 95% CI: −0.24, −0.02; p = 0.020). There was no difference in improving HOMA-IR (MD: −0.52; 95% CI: −1.04, −0.00; p = 0.050), while there was an improving trend in the healthy DP group.

BMI, BFM, and BLM
Eight studies [10,11,[13][14][15][19][20][21] including 365 participants reported the effect of healthy DP on BMI in Pca patients undergoing ADT. Due to low heterogeneity among eight studies (p = 0.24, I 2 = 24%), a fixed-effects model was performed. The pooled analysis showed that healthy DP significantly decreased BMI (MD: −1.02; 95% CI: −1.29, −0.75; p < 0.001) ( Figure 4A). By excluding any one article from the pooled analysis, the above results did not change. To exclude the effect of supervised exercise on BMI, subgroup analysis was performed. Because a significant heterogeneity of subgroup in the healthy DP only was observed (p = 0.010, I 2 = 77%), the random effects model and sensitivity analysis was performed. The results of subgroup analysis demonstrated that healthy DP only also led to a significant decrease in BMI (MD: −1.65; 95% CI: −3.11, −0.19; p = 0.030) compared with usual-care group ( Figure 4A1).
Five studies [13,14,[19][20][21] involving 189 participants assessed the effect of healthy DP on BLM. The fixed effects model was used as there was no significant heterogeneity (p = 0.14, I 2 = 42%). Compared with that of the usual-care group, healthy DP did not obviously increase BLM (MD: −0.41; 95% CI: −1.06, 0.24; p = 0.21) ( Figure 4C). We performed subgroup analysis to observe the effect of healthy DP only on BLM. Due to high heterogeneity found in the healthy DP subgroup only (p = 0.100, I 2 = 56%), the random effects model and sensitivity analysis was performed. After O'Neill et al. [19] was removed, the results suggested that healthy DP only could not significantly increase BLM (MD: −1.23; 95% CI: Eight studies [10,11,[13][14][15][19][20][21] including 365 participants reported the effect of healthy DP on BMI in Pca patients undergoing ADT. Due to low heterogeneity among eight studies (p = 0.24, I 2 = 24%), a fixed-effects model was performed. The pooled analysis showed that healthy DP significantly decreased BMI (MD: −1.02; 95% CI: −1.29, −0.75; p < 0.001) ( Figure 4A). By excluding any one article from the pooled analysis, the above results did not change. To exclude the effect of supervised exercise on BMI, subgroup analysis was performed. Because a significant heterogeneity of subgroup in the healthy DP only was observed (p = 0.010, I 2 = 77%), the random effects model and sensitivity analysis was performed. The results of subgroup analysis demonstrated that healthy DP only also led to a significant decrease in BMI (MD: −1.65; 95% CI: −3.11, −0.19; p = 0.030) compared with usual-care group ( Figure 4A1
Five studies [13,14,[19][20][21] involving 189 participants assessed the effect of healthy DP on BLM. The fixed effects model was used as there was no significant heterogeneity (p = 0.14, I 2 = 42%). Compared with that of the usual-care group, healthy DP did not obviously increase BLM (MD: −0.41; 95% CI: −1.06, 0.24; p = 0.21) ( Figure 4C). We performed subgroup analysis to observe the effect of healthy DP only on BLM. Due to high heterogeneity found in the healthy DP subgroup only (p = 0.100, I 2 = 56%), the random effects model and sensitivity analysis was performed. After O'Neill et al. [19] was removed, the results suggested that healthy DP only could not significantly increase BLM (MD: −1.23; 95% CI: −2.47, 0.01; p = 0.050) ( Figure 4C1), while there was a trend of decreased BLM.

Effect of Healthy DP on Fatigue
Four studies [11,13,19,21] including 216 participants assessed the effect of healthy DP on fatigue. The random-effects model was used because heterogeneity was high (p = 0.02, I 2 = 70%). The pooled analysis showed healthy DP could not significantly improve their fatigue (SMD: 0.43; 95% CI: −0.13, 0.99; p = 0.13) ( Figure 6). By excluding any one article from the pooled analysis, the above results did not change.

Effect of Healthy DP on Fatigue
Four studies [11,13,19,21] including 216 participants assessed the effect of healthy DP on fatigue. The random-effects model was used because heterogeneity was high (p = 0.02, I 2 = 70%). The pooled analysis showed healthy DP could not significantly improve their fatigue (SMD: 0.43; 95% CI: −0.13, 0.99; p = 0.13) ( Figure 6). By excluding any one article from the pooled analysis, the above results did not change.

Discussion
Dietary interventions have been proposed as a way to mitigate ADT-related side- Figure 6. Fatigue: forest plot of effect of healthy DP on fatigue. DP, dietary pattern; the diamond, effect size and 95%CIs; green color, weight [11,13,19,21].

Effect of Healthy DP on Fatigue
Four studies [11,13,19,21] including 216 participants assessed the effect of healthy DP on fatigue. The random-effects model was used because heterogeneity was high (p = 0.02, I 2 = 70%). The pooled analysis showed healthy DP could not significantly improve their fatigue (SMD: 0.43; 95% CI: −0.13, 0.99; p = 0.13) ( Figure 6). By excluding any one article from the pooled analysis, the above results did not change.

Discussion
Dietary interventions have been proposed as a way to mitigate ADT-related sideeffects, but the evidence is still limited. Thus, this meta-analysis focused on the effect of

Discussion
Dietary interventions have been proposed as a way to mitigate ADT-related sideeffects, but the evidence is still limited. Thus, this meta-analysis focused on the effect of healthy DP interventions on decreasing the adverse effects of ADT in men with PCa. There were nine RCTs that reported the effect of healthy DP interventions on ADT-related metabolic abnormalities. The results suggested that although there were only minor effects in mitigating TC, LDL-C, HDL-C, BLM, and fatigue, the healthy DP interventions greatly decreased HbA1c, BMI, BFM, TG, and BP, and, more importantly, healthy DP significantly lowered PSA.

Effect of Healthy DP on Glycolipid Metabolism
Long-term ADT therapy results in abnormalities of fat and glucose metabolism and abdominal obesity including reduction of BLM and accumulation of body fat, especially visceral fat [11,14,15], which can increase risk of CVD. The possible mechanisms are that the expression of insulin receptors of insulin target tissues and glucose oxidation are depressed [22], increasing triacylglycerol uptake and lipoprotein lipase activity leading to leptin resistance and decreasing insulin sensitivity [23] and interaction with proinflammatory factors that promote IR in PCa patients with ADT [24].

Glucose Metabolism
Diabetes is a common ADT-related metabolic abnormality [25]. At present, therapeutic options of diabetes were composed of diet control, exercise therapy, drug therapy, blood-glucose monitoring, and health education to reduce the rates of diabetes. Healthy DP could significantly reduce glucose [26,27]. The results of this study showed that HbAlc in the healthy DP group was significantly decreased (p = 0.02) and HOMA-IR had a decreasing trend (p = 0.05). The benefits of a healthy DP on hyperglycemia may be contributed to by weight loss and improving IR [28]. Reduced risk of diabetes has been linked to high consumption of vegetables and fruits including high levels of DF and flavonoids [29][30][31]. Asgary et al. [32] reported that DF played a role in weight-loss by regulating the expression of obesity genes and changing the proliferation signal transduction pathway of adipose tissue. Meanwhile, DF could reduce hepatic gluconeogenesis and hepatic glucose output, improving insulin sensitivity and lowering blood glucose [33]. Flavonoids may improve glucose metabolism due to improving inflammation and oxidative stress [34,35]. Furthermore, high-quality protein including red meat, fish, shrimp, soy bean, or milk can provide 10-15% of the energy for human tissues, increasing insulin-mediated glucose uptake by skeletal muscle cells and improving insulin sensitivity [36,37], and thus, glucose metabolism.

Lipid Metabolism
Obesity accelerated the progression and recurrence of PCa and increased the incidence of CVD that were closely related to non-neoplastic mortality [38]. Appropriate diet contributed to weight control in PCa patients undergoing ADT [5]. The result of this study showed that BMI levels in the healthy DP group decreased significantly (p < 0.01) compared to that in usual-care group. This result is consistent with that of Freedland et al. [14]. To exclude the effect of supervised exercise on BMI, subgroup analysis still suggested that healthy DP only could lower BMI (p < 0.05).
Deterioration in body composition including accumulation of BFM and loss of BLM was common metabolic alterations in PCa patients with long-term ADT therapy [39]. Loss of BLM was strongly associated with cancer-related fatigue (CRF) in PCa patients [40,41], while the accumulation of BFM, especially visceral fat was a source of various metabolic diseases [42]. Diet-induced weight loss decreased body mass without adversely affecting muscle strength [43]. However, this study showed that while a healthy DP significantly reduced BFM (p < 0.01), it also indicated a trend of reducing BLM (p = 0.05), which is likely a reflection of the greater weight loss in the DP group.
Hyperlipidemia, diabetes mellitus, and obesity may be complementary, with diabetes and obesity exacerbating occurrence and development of hyperlipidemia [44]. Through improving blood glucose and decreasing BMI, healthy diet plays an important role in preventing hyperlipidemia. The results of this study showed that TG in the healthy DP group was significantly decreased (p = 0.02). Consistent with our research, Freedland et al. [14] reported there was significant difference in TG in the LCD group.
In this meta-analysis, most dietary advice interventions focused on energy restriction, increased DF, and reduced fatty acids and cholesterol. Energy restriction could cause a decrease in fat accumulation [45]; dietary fiber not only did not yield energy which induces a catabolic-state and reduced liver dirty lipid deposition [33], but it also increased a feeling of satiety and lead to a strong dietary compensation effect [46] which reduces food intake and limits energy intake, which, thus, can lead to reduced weight and BMI. In addition, studies indicated DF could increase the abundance of the gut microbiota, such as Bacteroides, Bifidobacterium, and Lactobacillus, and low Firmicutes to Bacteroidetes ratios, which was associated with reduced weight [47,48]. Reduced saturated fatty acids and cholesterol in food interfere with bile-acid metabolism and improved liver-lipid metabolism [49]; unsaturated fatty acids, especially omega-3 fatty acids, lowered plasma TG by increasing fatty acid oxidation, which suppressed hepatic lipogenesis [50].

Effect of Healthy DP on Blood Pressure
Low androgen status during ADT therapy might activate nuclear factor kappa-lightchain enhancer of activated B cells, increase tumor necrosis factor-α (TNF-α)-induced expression of vascular cell adhesion molecule−1, and promote leukocyte adhesion to the arterial intima and atheromatous plaque formation [51], which could result in an increase in BP, and thus increase CVD morbidity and mortality [52]. Unsaturated fatty acids and DF in this healthy DP may reduce BP by modulating the inflammatory response and improving endothelial function [17,53,54]. This meta-analysis suggested that a healthy DP significantly lowered SBP and DBP (p < 0.05) which might contribute to an improvement in long-term CVD morbidity and mortality [55].

Effect of Healthy DP on Cancer Related Fatigue
Cancer-related fatigue (CRF) is one of the most frequent side-effects of cancer and its treatment [56]. PCa patients receiving ADT often experienced CRF which may be associated with the disturbance of hormone levels, resulting in the imbalance of inflammatory regulatory factors. Excessive inflammatory factors acted on the nervous-endocrine system, which led to the occurrence of fatigue [57]. Improving dietary quality and increasing nutrients may be effective ways to relieve CRF. Zick et al. [58]. found that there were more whole grains and vegetables consumed by patients with non-fatigue, compared with fatigue patients. Another study suggested that breast-cancer patients consuming <25 g/d DF had significant fatigue, compared with those consuming ≥25 g/d [59]. However, our findings were that a healthy DP did not significantly decrease the score of CRF, which might be associated with a decreased trend of BLM. Future work can be focused on developing a dietary regimen which is beneficial in lowering BFM while maintaining BLM, which may be helpful in improving CRF.

Effect of Healthy DP on PSA
Importantly, we have paid attention to the side-effects of dietary interventions. PSA is the best first-step serum marker as a screening test for PCa, and it plays an important role in the diagnosis, staging, and prognosis evaluation of PCa [60]. Therefore, PSA was used as a secondary indicator to evaluate the effect of the dietary interventions in this meta-analysis. Our findings showed that the healthy DP intervention lowered the level of PSA, indicating healthy DP intervention increased sensitivity to ADT therapy for PCa patients.

Conclusions
The meta-analysis demonstrated that a healthy DP could improve ADT-related metabolic abnormalities and be worthy of being recommended for PCa patients with ADT.

Limitations
Some limitations in this meta-analysis must be taken seriously. First, the study designs of only three studies were strictly dietary-intervention programs, which may have resulted in implementation bias. Second, only two or three studies, with limited sample sizes, were included in the meta-analysis for blood lipid markers, PSA, fatigue, and glucose markers, which might limit the broader application of our findings. Third, the results of sensitivity analysis for BLM and fatigue were unstable, and need to be further explored.