Impact of Foods and Dietary Supplements Containing Hydroxycinnamic Acids on Cardiometabolic Biomarkers: A Systematic Review to Explore Inter-Individual Variability

Plant-based diets rich in bioactive compounds such as polyphenols have been shown to positively modulate the risk of cardiometabolic (CM) diseases. The inter-individual variability in the response to these bioactives may affect the findings. This systematic review aimed to summarize findings from existing randomized clinical trials (RCTs) evaluating the effect of hydroxycinnamic acids (HCAs) on markers of CM health in humans. Literature searches were performed in PubMed and the Web of Science. RCTs on acute and chronic supplementation of HCA-rich foods/extracts on CM biomarkers were included. Forty-four RCTs (21 acute and 23 chronic) met inclusion criteria. Comparisons were made between RCTs, including assessments based on population health status. Of the 44 RCTs, only seven performed analyses on a factor exploring inter-individual response to HCA consumption. Results demonstrated that health status is a potentially important effect modifier as RCTs with higher baseline cholesterol, blood pressure and glycaemia demonstrated greater overall effectiveness, which was also found in studies where specific subgroup analyses were performed. Thus, the effect of HCAs on CM risk factors may be greater in individuals at higher CM risk, although future studies in these populations are needed, including those on other potential determinants of inter-individual variability. PROSPERO, registration number CRD42016050790.

related to specific classes of phenolic compounds, namely the class of flavonoids, including flavanols and flavonols, suggesting that these bioactives might be more effective in specific subgroups [18,19].
However, the inter-individual variability in the efficacy of the main dietary phenolic acids, namely HCAs, which are consumed in abundance in coffee and cereals, has not been fully explored. Therefore, the aim of this systematic review was to provide an overview of the randomized controlled trials (RCTs) evaluating the effect of HCAs on markers of CM health, in order to increase the knowledge about the impact of the inter-individual variability in the responsiveness to the consumption of this class of phenolic compounds.

Search Strategy and Study Selection
This systematic review was conducted following the Cochrane Handbook for Systematic Reviews of Interventions [20] and the Centre for Reviews and Dissemination's guidance for reviews in health care [21] and was reported in line with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement guidelines (Supplementary Table S1) [22]. The review protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO, registration number CRD42016050790) [23].
The systematic literature search was conducted using PubMed (http://www.ncbi.nlm.nih.gov/ pubmed) and the Web of Science (http://apps.webofknowledge.com) databases on December 2016 (updated December 2018), using the syntaxes reported in Supplementary Table S2. Electronic searches were supplemented with manual searches of references from included studies and reviews on similar topics. Studies were included in the present systematic review, based on the PICOS process (Supplementary Table S3), if (i) they were RCTs investigating the effect of the consumption of HCA-rich foods (artichoke, coffee, potato and cereal-based foods such as rye) or HCA extracts (i.e., containing HCAs extracted from one of the above-mentioned HCA-rich foods) in humans, where this was compared to a control, which was not high in HCAs; (ii) provided a quantitative characterization of the HCA content and (iii) reported on one or more of the following CM markers: Body mass index (BMI), WC, systolic and diastolic BP, TG, high-density lipoprotein cholesterol (HDL), low-density lipoprotein cholesterol (LDL), flow-mediated dilation (FMD), blood glucose, blood insulin, glycated hemoglobin (HbA1c), platelet aggregation or exercise capacity.
Exclusion criteria included (i) the presence of a co-intervention (e.g., physical activity) or other confounding factors; and (ii) not reported in a European language. No restrictions for the characteristics of study participants (e.g., age, sex and health condition) were applied.

Data Extraction
Two author-pairs independently assessed the studies for their inclusion. Disagreement between authors was resolved through consultation with a third author (PM) who performed an independent assessment to reach a consensus. Data were extracted from each identified study using a standardized form and the following information was collected: Name of first author; year of publication; study location; number of participants/controls; characteristics of intervention and placebo/control arms; characteristics of test food (e.g., HCA content); potential factors (e.g., sex, pathophysiological status) influencing the heterogeneity in the responses to the supplementation with HCA-rich food/extracts [24]; outcomes (BMI, WC, systolic and diastolic BP, TG, HDL, LDL, FMD, blood glucose, blood insulin and HbA1c); main findings. Studies were also assessed for whether there were any assessments made to explain inter-individual variability, including subgroup analyses (e.g., for age, pathophysiological status, sex, dietary patterns or specific polymorphisms). For these studies, the above-mentioned data were collected for each subgroup.
Data were then summarized by a qualitative assessment based on the results reported in individual studies. For each outcome assessed, studies were grouped by those significant or not and baseline levels of the outcome were used to compare studies. For studies that explored inter-individual variability in their own analyses, the results for each subgroup were assessed and summarized.

Risk of Bias
Risk of bias of the individual studies was assessed independently by two authors following the Cochrane Risk of Bias Tool [20]. The following categories were assessed: 1. Sequence generation and allocation sequence concealment (selection bias), 2. blinding of participants and personnel (performance bias), 3. blinding of outcome assessment (detection bias), 4. incomplete outcome data (attrition bias) and 5. selective outcome reporting (reporting bias). For each study, each category was assessed as either "Low risk of bias", "High risk of bias" or "Unclear risk of bias".

Study Selection
The study selection process is shown in Figure 1. A total of 811 records were identified through the database search. After removing 185 duplicate articles, 626 studies were screened and 540 were excluded based on the title or abstract. A total of 86 eligible records went under the full text screening process, after which 41 records were excluded ( Figure 1). Forty-five publications met eligibility criteria, providing data on a total of 44 unique RCTs, which were assessed in the qualitative analysis. The difference between the number of publications and the number of RCTs was due to the fact that three publications were on the same RCT but reported on different CM risk factors and, in one case, one publication reported results on two different RCTs.

Risk of Bias
Risk of bias of the individual studies was assessed independently by two authors following the Cochrane Risk of Bias Tool [20]. The following categories were assessed: 1. Sequence generation and allocation sequence concealment (selection bias), 2. blinding of participants and personnel (performance bias), 3. blinding of outcome assessment (detection bias), 4. incomplete outcome data (attrition bias) and 5. selective outcome reporting (reporting bias). For each study, each category was assessed as either "Low risk of bias", "High risk of bias" or "Unclear risk of bias".

Study Selection
The study selection process is shown in Figure 1. A total of 811 records were identified through the database search. After removing 185 duplicate articles, 626 studies were screened and 540 were excluded based on the title or abstract. A total of 86 eligible records went under the full text screening process, after which 41 records were excluded ( Figure 1). Forty-five publications met eligibility criteria, providing data on a total of 44 unique RCTs, which were assessed in the qualitative analysis. The difference between the number of publications and the number of RCTs was due to the fact that three publications were on the same RCT but reported on different CM risk factors and, in one case, one publication reported results on two different RCTs.

Characteristics and Risk of Bias of the Included Studies
The main characteristics of the included studies are reported in Tables 1 and 2. Out of the 45 included publications, 20 of them investigated the acute (i.e., single dose) effects of HCAs, which provided data on 21 RCTs (Mills et al. [25] included two different RCTs in the same publication). The remaining 25 publications investigated the chronic (2-16 weeks) effects of HCAs, which provided data on 23 RCTs (Martínez-López et al. 2018, Sarriá et al. 2016 andSarriá et al. 2018 [26-28] provided data on different outcomes from the same RCT).
Most of the RCTs were performed in Japan (n = 11), followed by the United States of America and the United Kingdom (n = 5), Australia and Iran (n = 4), Italy (n = 3) and Germany, Denmark, Norway, Sweden, Finland, Greece, Thailand, Colombia, Mexico, Switzerland, Spain and Austria (n = 1).
For either acute or chronic studies, coffee was the most commonly investigated product, used as the HCA-source in 30 RCTs either as a beverage (n = 27), coffee polyphenols (n = 2) or as a green coffee extract (GCE; n = 1). Conversely, artichokes (n = 3), cereals (i.e., rye and wheat, n = 3) and potatoes (n = 2) were the least used sources of HCA. Pure compounds were also tested (n = 6), including both 3and 5-CQA, ferulic acid or a mix of CGAs.
Risks of bias across the studies and within individual studies are shown in Supplementary Figure S1 and Supplementary Figure S2, respectively.
Studies often lacked details useful to make a judgment, mainly no mention of whether there was allocation concealment or blinding of outcome assessment, resulting in the majority of studies with domains predominantly assessed as unclear risk of bias. Conversely, there was very little high risk of bias observed where 0/21 acute studies and 4/23 chronic studies had only 1/5 domains (one study had two domains) rated as high. Table 1 reports the main results obtained from the 21 acute studies evaluating the impact of HCA-rich foods or dietary supplements (pure extracts) on CM health outcomes. As mentioned above, coffee was the main HCA-rich source considered (n = 11), while only few studies were focused on phenolic-rich extracts from coffee (n = 4), as well as on rye (n = 2), potatoes (n = 1) and pure CGAs (n = 3). Total doses of CGAs ranged from~3 mg [29] to 600 mg [30,31] for the phenolic-rich foods and from 400 [32] to 900 mg [33] for pure compounds. The number of study participants ranged from n = 9 [34] to n = 41 [35], while mean age ranged from 23 [25,29,36] to 59 years [33,37]. Overall, studies were conducted on healthy individuals, except for two studies focused on overweight individuals [38] and on individuals with borderline (systolic BP (SBP): 130-139 mmHg, or diastolic BP (DBP): 85-89 mmHg) or stage 1 hypertension (systolic BP: 140-159 mmHg, or diastolic BP: 90-99 mmHg) [39] and one study on individuals who had self-reported gastrointestinal symptoms after eating cereal, particularly rye [40].

Acute Studies
Among the outcomes, markers of blood glucose metabolism (i.e., postprandial glucose and insulin) were most often considered (n = 16), followed by BP (n = 8) and markers of endothelial function (i.e., FMD, n = 8). There were no studies focused on platelet aggregation or exercise capacity. For acute glucose metabolism, 16 studies reported a measure of postprandial glucose, 13 of which calculated an incremental area under the curve (IAUC) [29][30][31][34][35][36]38,[40][41][42][43][44][45], while three of them evaluated the effect at various postprandial time points [37,39,46]. Of the 13 reporting an IAUC, only two reported a significant reduction in IAUC for glucose after a HCA-rich intervention compared to a control [36,44], whereas one study demonstrated an increase [43]. One of these two studies [36], conducted on 12 healthy young adults, found a significantly lower 2 h IAUC for glucose after consumption of 25 g sucrose with coffee enriched with GCE and containing~432 mg CGAs, compared to 25 g sucrose in water. However, there was no effect when sucrose was consumed with normal or decaffeinated instant coffee containing~220 mg CGAs [36]. The second study [44] was conducted in 12 healthy young adults who consumed one of three potato chips (from red, purple or white potatoes) containing up tõ 360 mg CGAs and providing 50 g of available carbohydrates and compared to the consumption of 50 g available carbohydrates from plain salted wheat crackers. The 2 h IAUC for glucose was significantly lower after each of the potato chips compared to the cracker control. Nevertheless, it is possible that the observed differences here may be confounded by the greater fat or fiber content of the potato chips tested compared to the crackers. The one study that reported a significant increase in 2 h IAUC for glucose was that by Robertson et al. [43], in which they only found the increasing effect on IAUC glucose when caffeine was added to the test decaffeinated coffees in overweight men, using water as the control drink. No effect on IAUC glucose was found between the test drinks and control when the same amount of caffeine was added to each beverage. The study by Schubert et al. in 12 healthy adults, however, did not find an effect of decaffeinated coffee vs. water in either case when consumed without caffeine or when caffeine capsules were consumed [45]. However, Schubert et al. [45] used a standard breakfast with the test beverages, which included pancakes, butter and jam as a means to provide 48 g of carbohydrate as opposed to glucose load. Thus, the effects of caffeine on the glycemic response may differ when consumed with carbohydrates, which may affect its metabolism [47].
Of the three studies that only reported differences in postprandial glucose at various time points, one demonstrated a reduction at the peak glucose (30 min) after the ingestion of a coffee polyphenol extract providing 355 mg CQAs consumed with a meal compared to when the meal was consumed without the polyphenol extract in healthy men [46]. The other two studies found no differences between test and control groups in postprandial glucose response. However, one of these [39] in 19 middle-aged adults with borderline hypertension found a significant increase in postprandial glucose at 1 h and 2 h after the consumption of either the roasted coffee with CGAs or the coffee without CGAs. This study differs from the other two in that the test and control beverages were consumed alone, without a glucose load or meal as it was used in the other two studies, which may limit comparability as well as applicability.
Twelve studies reported a measure of acute postprandial insulin, ten of which calculated an IAUC [29][30][31]34,35,38,[40][41][42][43] and two assessed the effect only at various postprandial time points [39,46]. Of the ten, one study reported a reduction in 3 h IAUC and one study an increase, while the others found no significant effects. The one study that found the reduction in IAUC was conducted by Rosen et al., which assessed five different varieties of rye made into breads and compared to a white wheat bread in 14 healthy adults [29]. Interestingly, there were only two rye bread varieties, which significantly reduced 3 h IAUC insulin compared to the white wheat control bread. These two breads differed from the other three rye breads in that one (Rekrut) had the highest soluble fiber content and the other (Amilo) had the highest insoluble fiber content, as well as higher levels of certain bioactives including caffeic and sinapic acid. Dietary fiber content was related to the early (0-60 min) lower levels of glucose and insulin and it has been previously demonstrated that breads made with endosperm rye, which are rich in soluble fibers and bioactives, reduce glycaemia whereas wheat breads enriched in rye do not [29]. Therefore, the bioprocessing may influence the metabolic effects of rye breads, as has been previously reported [48]. On the other hand, the study that reported an increase in IAUC for insulin by Rakvagg et al. [42], in 11 healthy young adults, only found a significant effect of the dark roast coffee (66.6 mg CQA) vs. control (water), but not of the light roast coffee (~400 mg CQA), so the increasing effect of the dark roast coffee may not be due to the CQA content but to some other confounding factors. This study was also the only one to administer the interventions 30 min before the glucose load, which may affect the interpretation and comparison of results. Of the two studies, which only reported the effect of HCAs on insulin at various postprandial time points, no significant effects were found [39,46].
Nine acute studies reported effects of HCA on BP [25,[30][31][32][33]39,49], however only one demonstrated a significant reduction and one a significant increase while the remaining seven studies did not find significant effects. The study that demonstrated a reduction was performed in 23 healthy adults (mean age 52 y) [32], found a significant reduction in SBP and DBP over 3 h after consumption of 400 mg CGAs in 200 mL low nitrate water compared to plain water. It was not clear whether the control water was matched for nitrate content and thus, since nitric oxide (which nitrate is a precursor to) is a vasodilator, it may confound the effect on BP. The one study that found an increase was the one performed by Ioakeimidis et al. [49] in 24 healthy adults (mean age 33 y), who showed a significant increase in DBP and SBP over 2.5 h after consumption of a decaffeinated or caffeinated coffee (both with 81 mg of 3-CQA and 156 mg of 5-CQA) or caffeine tablets compared to a control of hot water. Interestingly, six of these nine studies assessing BP tested the effects of HCA when consumed alone (without a meal or glucose load) [25,32,33,39,49]. Only one study was done in a population with borderline or stage 1 hypertension (mean baseline SBP, 128 mmHg) [39], with the remaining in healthy adults (all with mean baseline SBP < 125 mmHg, except one study with 130.5 mmHg). Therefore, the applicability and generalizability of these results are limited.
Eight studies reported effects of HCAs on FMD with 50% (four studies) demonstrating a significant increase in FMD compared to control [25,30,46], and two studies demonstrating a significant increase in FMD within the HCA treatment [33,37], although not statistically significant compared to the control. Interestingly, the two studies that did not demonstrate a significant effect only assessed FMD at 2 h post-ingestion [32,39], whereas five of the six studies that found some significant effect assessed FMD between 4-6 h post-ingestion, which may be of importance since most of the HCAs are not metabolized until in the large intestine [5,6,11]. All studies were conducted in those who were healthy, other than one in borderline or stage 1 hypertension [39], with mean BMI in the normal range (21.8-25.6 kg/m 2 ) and an average age of 45 y. Of the six studies that found some significant effects, four were conducted solely in men, with the other two having 67% and 38% men, in contrast to the two studies without significant effects, which had 58% and 17% men. Thus, these results may be indicative of a possible effect in men, although more studies assessing this are needed, specifically including women, with longer follow up for assessments (>2 h), and in higher risk populations.
In general, the acute studies demonstrated no effect of HCAs on glycaemia or BP, with some potential effects on FMD. Limitations of the comparability of these studies include the wide variability in terms of study design and methods of outcome assessment. Due to the vast majority of studies being conducted in healthy participants and with HCA coming mainly from coffee consumed alone, generalizability and applicability is also limited. From these studies it is unknown what the effects may be in higher risk populations, including those with impaired glucose tolerance or diabetes, and when HCAs are consumed as part of meals, which may be more common in real-world consumption.

Chronic Studies
The main findings associated with the 23 chronic studies investigating the effects of HCA-rich foods on cardiometabolic markers are reported in Table 2. Similar to the acute studies, coffee was the most commonly investigated HCA-rich food (n = 15), followed by artichokes (n = 3), wheat (n = 1) and potatoes (n = 1), while extracts or pure compounds were the subjects of three studies. The HCA dose was sometimes difficult to identify (e.g., in some cases expressed as mg/kg body weight) but it generally ranged from 25 mg CGAs [50] up to 1200 mg CGAs/day [51]. The number of participants ranged from 10 [52] to 183 [53] while mean age ranged from 23 [52] to 54 years [54]. Most study participants were generally healthy (n = 11), and sometimes with pathophysiological conditions such as mild hypertension or hypertension (n = 4), non-alcoholic fatty liver disease (NAFLD, n = 1), MetS (n = 2), impaired glucose tolerance (IGT, n = 1), impaired fasting glucose (IFG, n = 1), hyperlipidemia or hypercholesterolemia (n = 3). Three studies recruited solely individuals with overweight or obesity. The duration of the studies ranged from two weeks [52] to four months [55], although most studies (n = 17) were 4-8 weeks long.
Lipid profile components (n = 22), blood glucose (n = 14) and BP (n = 13) were the CM outcomes most often considered. There were no studies assessing platelet aggregation or exercise capacity. As already observed in acute studies, results from the different investigations were often contrasting. Regarding blood lipids, 55% of the studies (12/22 studies) demonstrated a significant effect on at least one measure of the lipid profile (i.e., decrease in TG, LDL-C and total-C and/or increase in HDL-C) following HCA consumption [28,50,51,53,54,[56][57][58][59][60][61][62]. Fifty percent of studies (9/18 studies) that measured both LDL and TC demonstrated a significant reduction in both [28,50,51,54,[58][59][60][61][62], aside from one which found significance only for TC [60]. In comparing those nine studies reporting a significant reduction to those that did not, those that did had a greater baseline LDL level (~3.34 mmol/L vs. 2.91 mmol/L). Furthermore, those studies that did find a significant reduction were also slightly more overweight (BMI~28.9 kg/m 2 vs. 26.4 kg/m 2 ). The average of the mean ages in each study was similar between these groups (~43 y) and the ratio of men to women in each study was~37% in those studies that found an effect compared to 49% in those that did not. However, only one study [54] assessed the effect separately for men and women and found consistent effects between sexes, except for a reduction in TG, which was only found for women.
Considering the source of the HCAs consumed, in the nine studies that found a significant reduction in TC or LDL, seven used capsules of extracts rich in HCAs [51,54,[58][59][60][61][62], one used instant soup enriched in GCE [50] and one used 6 g/d of a soluble green/roasted coffee blend [28]. This contrasts the nine studies that did not find a significant improvement on TC or LDL, in which six studies used brewed or canned coffee [53,56,57,[63][64][65], one used a beverage enriched in GCE [30], one fruit juice enriched in GCE [66] and one GCE in capsules [67]. Interestingly, one of the studies that used brewed coffee found an increase in TC, as well as an increase in HDL, after the consumption of 750 mL/d of moderate roasted coffee in 84 healthy subjects for 12 weeks [56]. The authors suggested that this increase in TC may be the result of the diterpenes (i.e., cafestol and kahweol) in coffee [68], which are present in unfiltered coffee. The food source of HCAs (interfering effects of other compounds in unfiltered coffee) may have influenced the difference in results between these two groups of studies.
For markers related to glucose metabolism,~30% of the studies demonstrated a significant effect on fasting glucose (6/19 studies) [26,51,60,62,67,69] or HbA1c (2/7 studies) [54,62] following HCA consumption. In comparing those six and two studies that found a significant reduction to those that did not, those that did had a greater baseline glucose level (~6.00 mmol/L vs. 4.91 mmol/L and 6.55% vs. 5.9%, respectively). Furthermore, those studies that did find a significant reduction were also performed on slightly more overweight individuals (BMI~27.9 kg/m 2 vs. 26.2 kg/m 2 ). The average of the mean ages in each study was similar between these groups (~41 y) and the ratio of men to women at~45%. Of the nine studies that reported results for insulin, none demonstrated a significant reduction, with one that found a significant increase [56]. For homeostatic metabolic assessment-insulin resistance (HOMA-IR), 57% (4/7 studies) found a significant reduction [54,60,62,67] and one study found a significant increase [56]. In comparing the four studies that found a significant reduction in HOMA-IR to the two that found no effect, the mean baseline HOMA-IR level was higher (4.03 vs. 2.19), with similar BMI (31.3 vs. 29.5 kg/m 2 ), age (45 vs. 44 y) and ratio of men to women (35% vs. 41%). Furthermore, those four studies all used HCAs consumed in the form of capsules (ranging from 372 mg to 600 mg/d CGAs from GCE or artichoke) [54,60,62,67] whereas the other three studies all assessed the effect of HCAs consumed in brewed coffee (ranging from 9 mg/d CGAs to 216 mg/d CGAs). Interestingly, the one study of 84 healthy adults that found a significant increase in insulin and HOMA-IR [56], had a much lower difference in CGA between the two coffees (difference of 9 mg/d of CGAs) compared to the rest of the studies, and the authors noted that the concentration of CGAs upon analyses demonstrate that they did not differ between the two coffees [70], thus the differences in outcomes may be the result of other differences between the coffees (e.g., N-methylpyridinium (NMP)). Additionally, one of the two studies that did not find an effect of HOMA-IR [57], conducted in 116 overweight adults, also described a similar intervention of brewed coffees to compare which differed in CGAs by 9 mg/d, thus may also not have had an adequate difference in CGAs in order to assess effectiveness.
The beneficial effects predominating in higher risk groups are further supported if we look at those studies that were done in higher risk groups. For instance, the consumption of HCA-rich foods provided beneficial effects to study participants with impaired glucose metabolism, who are at a higher risk of type 2 diabetes mellitus. In fact, the oral consumption of 1200 mg CGA by 30 study participants with IGT for 12 weeks [51], or 600 mg artichoke extract by 55 participants with IFG for eight weeks [62], significantly reduced fasting glucose (from 5.7 ± 0.4 to 5.5 ± 0.4 mmol/L) as well as other parameters such as the insulinogenic index and the homeostatic metabolic assessment (HOMA) index (−11.7%), compared to controls. Similarly, a positive effect on fasting blood glucose was observed in two other investigations on individuals with NAFLD supplemented with 1 g GCE/day (500 mg CGA) for eight weeks [60], as well as in individuals with MetS who consumed two decaffeinated GCE capsules (372 mg CGA/day) for eight weeks [67]. Interestingly, contrasting results were observed in four trials with healthy study participants [26,69,71,72]. A statistically significant amelioration of fasting blood glucose (from 107.6 ± 3.0 to 99.0 ± 2.5 mg/dL) was evidenced when healthy individuals consumed 185 mL of a test beverage with 329 mg CGA for four weeks [69] or green-roasted coffee blends providing 510 mg HCA/day for eight weeks [26]. On the contrary, Vitaglione et al. [71] failed to find any effect on blood glucose following an eight-week consumption of wholegrain biscuits (70 g,~130 mg HCAs as sum of ferulic, sinapic and coumaric acids), similarly to Robertson and colleagues [72], who found no differences after the consumption of four cups/day of instant coffee (44 mg CGAs/serving) for 12 weeks, compared to non-coffee consumers.
Regarding BP, most studies (62%, 8/13 studies) demonstrated a significant effect on either SBP or DBP resulting from the consumption of HCA-rich foods [28,50,53,[64][65][66][67]73]. In comparing those eight studies that found a significant reduction to those that did not, those that did had a greater baseline SBP level (~138 mmHg vs. 118 mmHg). Those studies that did find a significant reduction were performed on slightly less overweight subjects (BMI~26.3 kg/m 2 vs. 27.8 kg/m 2 ) with similar mean ages (~45 y) and ratio of men to women (~46 vs. 43%). No study completed subgroup analyses by BP status or by sex.
With regard to measures of adiposity, 29% (5/17 studies) demonstrated a significant reduction in BMI, body weight or waist circumference [51,53,61,62,67]. In comparing those studies that found a significant reduction to those that did not, those that did had a greater baseline BMI (29.8 kg/m 2 vs. 27.0 kg/m 2 ) and greater baseline waist circumference (106.2 cm vs. 97.7 cm). Those studies that did find a significant reduction had similar mean ages (~44 vs. 43 y, respectively) and lower ratio of men to women (~30% vs. 54%). Interestingly the populations of those studies that found a reduction included (n = 1) mild hypertension, (n = 1) obese, (n = 2) impaired glucose tolerance and (n = 1) metabolic syndrome.           Overall, the results from each of the outcomes assessed in the chronic studies consistently demonstrate there is a significant effect in those studies with participants with higher baseline levels of each risk factor for CVD. Therefore, HCAs may be more effective in those at higher CVD risk.

Inter-Individual Variability
Only seven out of the 45 studies took into account one or more determinants that might explain inter-individual variability, such as pathophysiological status (n = 4), sex (n = 2), dietary patterns (n = 1) or specific polymorphisms (n = 1). The main findings of these studies are reported in Table 3. It is worth noting that the role of polymorphisms was also considered by the study of Robertson and colleagues, but the considered polymorphism (on the CYP1A2 gene) was related to the caffeine metabolism and no insights in the metabolism or effect of HCAs were provided [72].
Among the studies investigating the putative role of pathophysiological status in the effects of HCA-rich sources, two acute studies stratified results based on the glycemic response [35] or the insulinogenic index (an index of pancreatic β-cell function) [46]. In the former, the consumption of 300 mg of decaffeinated green coffee beans (EDGCB) extract significantly lowered peak glycemic levels, but not plasma glucose IAUC, after ingestion of a 200 g carbohydrate rich-meal, compared to water [35]. Interestingly, when results were analyzed on a sub-set of 18 study participants categorized as having a high-glycemic response (i.e., those with the highest mean postprandial glucose level 30 min after consumption of the loading diet/placebo food), plasma glucose 2 h IAUC was reduced following the 100 mg EDGCB beverage compared to control. Similarly, plasma glucose after 30 min was significantly lower after the ingestion of both low (100 mg) and high (300 mg) intake of EDGCB beverages, compared to controls. Jokura and colleagues [46] focused on the insulinogenic index and its potential role as a determinant of the effect of HCA on postprandial hyperglycemia and vascular endothelial function. In all study participants, the ingestion of a meal with a coffee phenolic extract (355 mg CQAs) beverage significantly lowered the plasma glucose levels (p < 0.05) after 30 min and FMD response after 60 min compared to the same meal without the coffee polyphenol extract. However, when study participants were stratified by insulinogenic index, those with an index <0.88 had significantly lower postprandial blood glucose concentration at 30 min (p < 0.01) compared to control, whereas there was no effect in those with an index ≥0.88. Thus in those with lower β-cell function, and thus at greater risk of chronic disease such as diabetes, there was an effect of HCA. With regards to the assessment of FMD, the response seemed not to be influenced by the insulinogenic index, being significantly higher at 120 min in both strata compared to control.
The role of background lipid status in determining the effect of HCA-rich foods on CM markers was considered in one chronic study (two publications [26,28]) in which half of the recruited study participants had hypercholesterolemia. In this study, daily consumption of green/roasted coffee blend (344 mg CQAs and~510 mg HCAs total, plus~120 mg caffeine) for eight weeks resulted in decreased the levels of several CVD risk factors including SBP, DBP, blood glucose, HOMA-IR, total-C, LDL-C and TG. However, when the results were separated by hyperlipidemia status, SBP, DBP, total-C, LDL-C and TG were significantly reduced after coffee consumption only in those with hyperlipidemia. Thus, demonstrating potential effectiveness of CQAs in those subjects at high CVD risk.
Regarding sex differences, the eight-week supplementation of a CQA-rich artichoke leaf extract significantly increased HDL-C in primary mild hypercholesterolemic study participants (0.207 mmol/L) [54]. However, when sex groups were compared, an increase in HDL was only significant for males and reductions in TG were only seen for women. However, when groups were further stratified by median HDL level, there was a significant increase in HDL in women who had ≤median HDL. The impact of sex was also considered in a study enrolling 80 patients with MetS and screened for polymorphisms of cholesteryl ester transfer protein (CETP), which is directly related to MetS risk [58]. Although there was no interaction in the whole population between CETP gene mutation and response to 12-week ALE supplementation (~22 mg CQAs), the subgroup analysis revealed that only men with CETP gene mutation had significantly lower LDL-C levels after 12-week supplementation, compared to placebo.
The inter-individual variability associated with the dietary pattern was taken into account by Ioakeimidis and colleagues [49] in an acute study focusing on habitual coffee consumption. Results revealed positive effects of both caffeinated and decaffeinated coffee consumption (79 mg CQAs each) on markers of arterial stiffness in non-habitual compared to habitual coffee consumers (maximal differences of changes in responses by 4.5%). As some markers, not subject of this systematic review (i.e., pulse wave velocity and augmentation index), increased only after decaffeinated coffee consumption in non-habitual consumers, authors hypothesized that the effect was related to compounds other than caffeine and supported the potential role of coffee habituation as determinant of its effect on vascular function.

Discussion
There is a clear interest in the exploitation of phenolic-rich foods as potential modulators of markers of CM health, and the present review aimed at summarizing the main findings from RCTs focused on HCA-rich foods. The observed intervention effects had clinical relevance mostly in study participants at high CVD risk. Conversely, many studies performed on self-reported healthy individuals failed to find any effect of HCAs, increasing the heterogeneity of the results and making it difficult to draw any clear conclusion.
In the present review, only seven studies out of the 45 included publications considered one or more determinants of inter-individual variability in response to HCAs, while the remaining studies did not stratify results based on these parameters. This hinders our understanding of the role played by these determinants in the individual response to HCA-rich foods.
Despite limited, results from these seven studies suggest that some individual characteristics may influence the beneficial effect of HCAs. Firstly, determinants of health or pathophysiological status, like baseline cholesterol levels [26,28], insulinogenic index [46] or glycemic response [35] could play an important role in the variation among study participants in the biological response to HCAs regarding CM outcomes, being associated with an increased beneficial effect following HCA intake. This was not only demonstrated in the stratified analyses conducted by Martinez-Lopez et al., Sarriá et al., Iwai et al. and Jokura et al., [26,28,35,46] but also in the assessments conducted within this systematic review for all outcomes in the chronic studies, which demonstrated that effectiveness of HCAs, regardless of specific source, was greater in those at higher risk (i.e., greater baseline cholesterol, glycemic or SBP). This supports previous studies that suggested that the pathophysiological status can lead to inter-individual variation in response to polyphenols [74,75]. Some recent systematic reviews and meta-analyses have also indicated that the health status or BMI may influence the impact of several polyphenols (flavonols, flavan-3-ols, anthocyanins and ellagitannins) on blood lipid levels [18,19,24].
Regarding sex differences, the response to HCAs has been shown to be different between men and women following artichoke leaf extract consumption [54]. So far, a sex effect in response to plant-food bioactive compounds such as HCAs has been reported in very few studies [76], mainly focusing on flavanol-rich products, with some but limited differences in the response between men and women. However, results are often contrasting; for instance, a decreased augmentation index was observed only in women after cocoa consumption for four weeks [77], while the antioxidant status was improved mostly in men after a four-week consumption of ready-to-eat meals supplemented with cocoa extract [78]. Again, TC and LDL were significantly reduced only in females after intervention with flavanol-containing products [19]. Similarly to what happens for other individual characteristics, the lack of differences in the response among men and women after consumption of HCAs could be due to the lack of statistical power resulting from smaller sample sizes after the stratification process.
Genetic polymorphisms have also been shown to impact the effect of plant compounds from different sources [79], since they are often present in genes encoding for enzymes involved in the metabolism of these bioactives. Regarding coffee, mostly polymorphisms related to caffeine metabolism have been investigated, above all the cytochrome P450 as CYP1A2 accounts for about 95% of caffeine metabolism, with a high inter-individual variability in activity [80]. In the present review, only one study investigating variation based on polymorphisms (in CETP Taq 1B) was included [58], suggesting that genetic polymorphism may predict the lipid responsiveness to HCA intake.
An additional determinant, not explored in any of the studies included in the present review, may likely be age, which is the strongest independent cardiovascular risk factor. Overall, the impact of age on the response to the consumption of plant food bioactives has received very limited attention [17]. Differences due to age might depend on variability in absorption, distribution, metabolism and excretion (ADME), as recently showed by Alkhaldy and colleagues, who found variation in the urinary phenolic acid profile between younger and older adults after a polyphenol-rich meal [81].
The inter-individual variability in the biological response to the consumption of phenolic compounds, such as HCAs, may be partially related to the inter-individual variation in the bioavailability and metabolism of these compounds [16]. Several human studies have demonstrated that the plasma concentrations and the urinary excretion of phenolics or their derived metabolites can differ markedly between individuals following a similar intake of plant food bioactives, due to variances in their ADME [82,83]. Despite a small amount of some (poly)phenols being absorbed in the upper gastrointestinal tract (and being subjected to glucuronidation, sulfation or methylation by the gut epithelium and/or liver) most compounds reach the lower gastrointestinal tract unmodified, where they undergo extensive metabolism [83]. Thus, a key role is played by the gut microbiota, which may modify the structure of polyphenols, releasing these colonic catabolites into the portal vein towards the liver, where they can: (i) Directly enter the bloodstream, (ii) undergo phase I and II metabolism and then enter the bloodstream and (iii) undergo enterohepatic recirculation until complete detoxification [84].
This long and complex physiological mechanism and the influence of gut microbiota composition form the pathway through which different types and levels of metabolites enter circulation where they may have metabolic effects. This is supported by the work by Stalmach and colleagues who assessed the metabolic fate of CGAs from coffee [85] and demonstrated that~1 h after the consumption of instant coffee, low nanomolar concentrations of CQA lactones and caffeic acid sulfates reached the plasma. Although it has been hypothesized that quinic-HCA linkage is hydrolyzed with consequently few HCA conjugates being absorbed [83],~5 h after coffee consumption, gut-derived phase II metabolites were detected in much higher concentrations in plasma showing high inter-individual differences. Gut microbial esterases can rapidly hydrolyze the quinic-phenolic linkage and convert the HCA into dihydroxy forms and further metabolism of these compounds by gut microbial strains leads to the formation of smaller catabolites [83,86,87]. Thus, future studies assessing the profile of phenolic metabolites in circulation in addition to variations in gut microbiota composition would be highly valuable in understanding the potential effectiveness of HCAs. Among the 44 included RCTs, only the one by Mills and colleagues [25] considered circulating metabolites and found that FMD responses to coffee intake were closely paralleled by the appearance of CGA metabolites in plasma.
In the framework of an intervention study, these aspects are further complicated by considering that the same colonic metabolites can be produced after the consumption of different classes of phenolic-rich foods. For example, in vitro and in vivo studies have shown that after berry, coffee, cocoa or orange consumption, the phenolic metabolic degradation pathways reach some common intermediates, most of them HCA or hydroxybenzoic acid derivatives, which can be further metabolized into smaller compounds [83,88,89]. To avoid this further confounding factor, some researchers have applied different strategies, such as: (i) Asking participants to maintain their usual diet; (ii) telling participant to refrain from consuming selected foods, rich in (poly) phenol compounds similar to those of the intervention and (iii) advising a low-/free-polyphenol diet in the day(s) prior to sample collection.
The limitations of the present systematic review include that there were relatively few studies (n = 7), which explored potential determinants of inter-individual variability, thus there may be other variables that may influence the effectiveness of HCAs on CM biomarkers. Additionally, of those included studies, there was a wide range of methods by which HCAs were delivered, as well as ranges of doses, including those coming from coffee (plain coffee and green coffee extracts consumed in a variety of forms (soups, capsules, in a test beverage and in fruit juice)) and foods (purple potatoes, whole grain biscuits and artichoke leaf extract), which may also influence the effect of HCAs. Furthermore, some studies controlled for caffeine whereas others did not, and considering the majority of studies focused on coffee, this also potentially confounds the ability to determine the effect of HCAs. In addition, further efforts should be paid on the contribution of other phytochemicals to the observed effects when dealing with food products rich in different bioactive compounds. A good example may be coffee, where the presence and amount of some compounds like caffeine, trigonelline and HHQ may affect the response to the intervention [8,38,64]. Lastly, it may be a limitation that only two databases were used for conducting the search and thus may not have captured all existing studies on HCAs, however manual searches of the reference lists of all included studies, supplemented the search to minimize this potential issue.
Overall, the studies were characterized by a high variability in study design and methods of outcome assessment and, consequently, comparisons among findings of the different studies challenging. In addition to the health status of the volunteers, major sources of variability include the type of HCA-based product (e.g., pure compounds, foods and beverages) and the dose of HCAs, which broadly ranged in both acute and chronic studies. It is worth noting that, even when the same HCA-food source was used as test food (e.g., coffee) in different investigations, a high variability in terms of phenolic content was observed. This further supports the importance of an in-depth characterization of the phenolic profile of the test items and, when possible, taking into account the actual phenolic amount provided by a single serving [90,91]. In addition to the above-mentioned sources of variability, studies often lacked in information on determinants that are known to influence CM biomarkers, such as BMI, age, sex, smoking habits, dietary patterns, physical activity levels or specific polymorphisms. Improvements in reporting possible determinants of inter-individual variability are needed to make secondary data analysis feasible [92].

Conclusions
The present review summarized the main findings of RCTs investigating the effect of HCA-rich foods on CM health, with a particular focus on the determinants of inter-individual variability such as pathophysiological status, sex or specific polymorphisms. Generally, there are limited studies that conduct stratified analyses to explore factors leading to inter-individual variation in the effect of HCAs to modulate CM outcomes, thus limiting the ability to conclude which factors may modulate the effectiveness of HCAs. However, the present systematic review demonstrates that overall, HCAs have greater effectiveness, regardless of the variation in the source of HCAs, in those subjects with greater baseline levels of cardiovascular risk factors such as cholesterol, BP and glycaemia. Further high quality, well powered RCTs with a low risk of bias should be performed, reporting all characteristics that may influence the individual response to these compounds, including but not limited to sex, age, BMI, pathophysiological status, presence of genetic polymorphisms, smoking habits, physical activity level and dietary patterns.