Cheese consumption and multiple health outcomes: an umbrella review and updated meta-analysis of prospective studies

This umbrella review aims to provide a systematic and comprehensive overview of current evidence from prospective studies on the diverse health effects of cheese consumption. We searched PubMed, Embase, and Cochrane Library to identify meta-analyses/pooled analyses of prospective studies examining the association between cheese consumption and major health outcomes from inception to August 31, 2022. We reanalyzed and updated previous meta-analyses and performed de novo meta-analyses with recently published prospective studies, where appropriate. We calculated the summary effect size, 95% prediction confidence intervals, between-study heterogeneity, small-study effects, and excess significance bias for each health outcome. We identified 54 eligible articles of meta-analyses/pooled analyses. After adding newly published original articles, we performed 35 updated meta-analyses and 4 de novo meta-analyses. Together with 8 previous meta-analyses, we finally included 47 unique health outcomes. Cheese consumption was inversely associated with all-cause mortality (highest compared with lowest category: RR = 0.95; 95% CI: 0.92, 0.99), cardiovascular mortality (RR = 0.93; 95% CI: 0.88, 0.99), incident cardiovascular disease (CVD) (RR = 0.92; 95% CI: 0.89, 0.96), coronary heart disease (CHD) (RR = 0.92; 95% CI: 0.86, 0.98), stroke (RR = 0.93; 95% CI: 0.89, 0.98), estrogen receptor-negative (ER−) breast cancer (RR = 0.89; 95% CI: 0.82, 0.97), type 2 diabetes (RR = 0.93; 95% CI: 0.88, 0.98), total fracture (RR = 0.90; 95% CI: 0.86, 0.95), and dementia (RR = 0.81; 95% CI: 0.66, 0.99). Null associations were found for other outcomes. According to the NutriGrade scoring system, moderate quality of evidence was observed for inverse associations of cheese consumption with all-cause and cardiovascular mortality, incident CVD, CHD, and stroke, and for null associations with cancer mortality, incident hypertension, and prostate cancer. Our findings suggest that cheese consumption has neutral to moderate benefits for human health.


Introduction
Cheese is generally a nutrient-dense and well-tolerated fermented dairy product consumed worldwide.However, the health effects of cheese consumption remain a matter of controversy.On one hand, cheese is a rich source of high-quality protein (mainly casein), lipids, minerals (e.g., calcium, phosphorus, and magnesium), and vitamins (e.g., vitamin A, K 2 , B 2 , B 12 , and folate), and probiotics and bioactive molecules (e.g., bioactive peptides, lactoferrin, short-chain fatty acids, and milk fat globule membrane), which may provide various health benefits.On the other hand, cheese contains relatively high contents of saturated fat and salt, which are perceived as unfavorable dietary components for cardiovascular health [1,2].Currently, most dietary guidelines recommend consuming dairy products as part of a healthy diet while avoiding intake of full-fat and high-sodium versions [3][4][5][6].Of note, the recommendation is primarily based on extrapolated benefits and harms of single nutrient contained in dairy.However, whole dairy foods are not a simple collection of isolated nutrients but have complex physical and nutritional structures (i.e., dairy matrix), which affect digestibility and nutrient bioavailability, thereby modifying the overall effects of dairy consumption on health and disease [7][8][9].In addition, dairy products are a heterogeneous group of foods regarding the dairy matrix due to different processing methods [8].Because various types of dairy products appear to have distinct influences on specific health outcomes [10], merging them into 1 group (i.e., total dairy consumption) may blur the true association.Thus, a separate assessment of the health effects of cheese consumption is required.
Umbrella reviews can provide a comprehensive overview of evidence from existing meta-analyses on a given topic, with unique strengths of identifying the uncertainties, biases, and knowledge gaps of the evidence [11].Many meta-analyses on the association between cheese consumption and a range of health end points, such as all-cause and cause-specific mortality, cardiovascular diseases (CVD), cancer, metabolic diseases, bone fracture, and other diseases, have been published [12][13][14][15][16][17].An extensive summary of the breadth and validity of these associations with diverse health outcomes will help elucidate the role of cheese consumption in human health.Therefore, we conducted an umbrella review to synthesize the available evidence from meta-analyses of prospective studies to examine the various health impacts of cheese consumption.Furthermore, we contextualized the magnitude, direction, and significance of the identified associations, evaluated risk of potential biases, and assessed the credibility of the evidence.

Methods
The present umbrella review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [18].The protocol of this study was registered in PROSPERO (CRD42022331328).

Literature search
We systematically searched PubMed, Embase, and Cochrane Library databases to identify existing meta-analyses (including pooled analyses) of prospective studies investigating the association between cheese consumption and any health outcome from inception to August 31, 2022.The search terms were as follows: (cheese) AND ("meta analysis" OR metaanalysis OR "meta analyzed" OR meta-analyzed OR "pooled analysis" OR "systematic review").We also extensively searched the 3 databases for recently published original prospective studies to update previous meta-analyses or derive de novo metaanalyses.Predefined search strategies for meta-analyses and primary studies are presented in Supplemental Tables 1 and 2. Two investigators (XD, MZ) independently performed a 3-step parallel screening of titles, abstracts, and full texts for all identified studies according to the inclusion and exclusion criteria.Any discrepancies were discussed and resolved by a third investigator (ZH).

Eligibility criteria
Meta-analyses of population-based prospective studies (i.e., prospective cohort studies, case-cohort studies, nested case--control studies, and randomized controlled trials) exploring the association between cheese consumption (primary or secondary exposure of interest) and major health outcomes were included in the umbrella review.Original prospective studies eligible for updated or de novo meta-analyses were also included.Conference abstracts, interviews, letters, and narrative reviews were excluded.Meta-analyses or original studies without full text, effect size [e.g., risk ratio (RR), odds ratio (OR), or hazard ratio (HR)], or not written in English were also excluded.Studies with changes in cheese consumption rather than absolute intake as exposure, using substitution analysis, or using surrogate end points (e.g., blood lipids, blood pressure, and body weight) as outcomes were removed.If more than 1 article reported the results for an identical outcome from the same study population (or cohort), only the one with the largest sample size, the longest follow-up, or the most complete information was included.

Data extraction
From each included meta-analysis, the following information was extracted and verified by three investigators (XD, MZ, ZH): the first author's name, publication year, outcome of interest, study population (general or disease status), study design of the primary studies, type of comparison (highest compared with lowest category of cheese consumption or each increment in cheese consumption), number of included studies, number of participants and cases, and the reported summary risk estimates (RR, OR, or HR) with corresponding 95% CIs.For meta-analyses on over 1 health outcome, each outcome was recorded separately.For original studies, the extracted data covered information on the first author's name, publication year, study design, study population characteristics, geographic location, number of participants and cases, length of follow-up (cohort study), dietary assessment method (e.g., food frequency questionnaire and 3-d 24-h dietary records), cheese type, categorization and amount of cheese consumption, adjustment factors, and effect size with 95% CIs.

Evaluation of methodological quality
The AMSTAR-2 (A Measurement Tool to Assess Systematic Reviews) tool [19] was used to evaluate the methodological quality of the included published meta-analyses and systematic reviews.It includes 16 individual items and 7 of them are identified as critical.Systematic reviews with no or 1 noncritical weakness are rated as high confidence; those with more than 1 noncritical weakness are rated as moderate confidence; those with 1 critical flaw with or without noncritical weaknesses are rated as low confidence; and those with more than 1 critical flaw with or without noncritical weaknesses are rated as critically low confidence.Two investigators (XD, MZ) implemented the evaluation independently, with disagreements reconciled by discussion and consensus.

Statistical analysis
We reanalyzed previous meta-analyses to obtain necessary information for subsequent assessment of the credibility of evidence.If the existing meta-analysis included crosssectional, retrospective, and prospective studies, we only kept the results from prospective studies in our meta-analysis.Furthermore, we incorporated newly identified original studies into previous meta-analyses to update or derive de novo meta-analyses, where appropriate.For each outcome, we recalculated the summary risk estimates and their corresponding 95% CIs for the highest compared with the lowest category of cheese consumption and/or per 30-g/d increment in cheese consumption by using the random-effects model by DerSimonian and Laird [21].When results from the same cohort were reported separately for different cheese types (e.g., hard cheese and cottage cheese or low-fat cheese and high-fat cheese instead of total cheese) and disease subtypes [e.g., coronary heart disease (CHD) and stroke rather than total CVD], we used a fixed-effects model to generate an overall estimate before pooling with other studies.
Heterogeneity across studies was investigated using I 2 statistic.We also performed subgroup analyses according to adjustment for total energy intake in the models (adjusted and unadjusted) and geographical location (Europe, North America, and Oceania; Asia and other regions; multiregion) to explore potential sources of heterogeneity.We computed 95% prediction intervals (95% PIs) to predict the effect size range of a future original study will lie after considering both the uncertainty in the mean effect and the heterogeneity from the random-effects model [22,23].We assessed potential small-study effects by Egger's test [24] if !3 studies were available.A P value of <0.10 was interpreted as the presence of small-study effects.The excess statistical significance test was performed to evaluate whether the observed number of nominally statistically significant studies was larger than their expected number using the χ 2 test [25].A P value of <0.10 was considered statistically significant.
We further examined the nonlinear dose-response association among studies that provided risk estimates with !3 exposure categories using a 2-stage restricted cubic splines (3 knots at 25, 50, and 75 percentiles) analysis [26,27].The P value for nonlinearity was calculated by testing whether the coefficient of the second spline was equal to 0 [27].We used the median/mean of each consumption category if available or the midpoint between the lower and upper bounds of each intake category to represent the intake levels.We assumed zero as the lower bound for the open-ended lowest category and multiplied the lower bound value by 1.2 as the upper bound for the open-ended highest category.All statistical analyses were conducted using the "metafor," "meta," "dosresmeta," and "forestplot" packages in R software version 4.1.0(The R Foundation).

Subgroup analyses
Of the 184 original studies included in the meta-analyses, 149 (81.0%) studies adjusted for total energy intake in the models, and 152 (82.6%) studies were conducted in North America, Europe, and Oceania.Among the 47 major outcomes in our study, 4 outcomes were solely based on studies without energy adjustment, 18 outcomes were only based on studies with energy adjustment, and 30 outcomes were entirely based on studies conducted in Europe, North America, and Oceania (Supplemental Tables 13-16).
Subgroup analyses indicated that the association between cheese consumption and most health outcomes remained consistent regardless of adjustment for total energy intake or geographic location (Supplemental Tables 13-16).However, heterogeneity existed in subgroups by total energy adjustment for overall and breast cancer incidence, where higher cheese consumption was linked with an increased risk of overall cancer (RR ¼ 1.14; 95% CI: 1.01, 1.29; P ¼ 0.0305; I 2 ¼ 0%; P-subgroup ¼ 0.02) (Supplemental Table 14) and was marginally associated with elevated risk of breast cancer (RR ¼ 1.43; 95% CI:0.99, 2.06; P ¼ 0.0557; P-subgroup ¼ 0.04) (Supplemental Table 14) in studies without adjustment for total energy intake, whereas null associations were found in studies with adjustment for total energy intake.Although heterogeneity was observed in subgroup analyses by energy adjustment for metabolic syndrome (P-subgroup ¼ 0.03), the null association was consistent between subgroups.Meanwhile, an inverse association was detected for CHD mortality (RR: 0.67; 95% CI: 0.51, 0.89; P-subgroup ¼ 0.03) (Supplemental Table 15) in studies conducted in Asia and other regions but not in studies conducted in North America, Europe, and Oceania.

Evidence credibility
The credibility of the identified associations with cheese consumption is summarized in Supplemental Tables 10 and 11, and the detailed NutriGrade scores for each meta-analysis are presented in Supplemental Table 17.No health outcome met the standards for high meta-evidence.Eight health outcomes (17%)-death from any cause, cancer, and CVD and incidence of overall CVD, CHD, stroke, hypertension, and prostate cancer--presented moderate meta-evidence.Twenty-two health outcomes (47%)-site-specific cancer mortality (the colorectum, colon, rectum, lung, and stomach), CHD mortality, overall and site-specific cancer (colorectum, total and distal colon, rectum, total, ERþ and ERÀ breast, bladder, and pancreas), T2D, overweight/obesity, total and hip fractures, fall, and dementia-presented low meta-evidence.The rest health outcomes indicated very low meta-evidence.

Discussion
This umbrella review provides a broad overview of the existing evidence on the association between cheese consumption and 47 unique outcomes through 35 updated, 4 de novo, and 8 previous meta-analyses based on 184 prospective observational studies from 145 primary articles.Moderate quality of evidence showed that cheese consumption was associated with reduced risk of all-cause mortality, CVD mortality, and incident CVD, CHD, and stroke but not related to the risk of cancer mortality, hypertension, and prostate cancer.Low quality of evidence was observed for inverse associations of cheese intake with incidence of ERÀ breast cancer, T2D, total fracture, and dementia and null association with site-specific cancer mortality (i.e., colorectum, colon, rectum, lung, and stomach), CHD mortality, and incidence of overall, site-specific cancer and its subtypes (i.e., colorectum, total and distal colon, rectum, total and ERþ breast, bladder, and pancreas), overweight/obesity, hip fracture, and fall.The nonlinear dose-response analyses additionally suggested a U-shaped association between cheese consumption and the risk of all-cause mortality and cardiovascular mortality and an L-shaped association with the risk of overall CVD, CHD, stroke, and total and hip fractures with the optimal intake at ~40 g/d.
Although cheese is theorized to have detrimental effects on blood pressure and blood lipid profile based on its high sodium and saturated fat contents, a moderate quality of evidence suggest that cheese consumption does not increase the risk of cardiovascular diseases and may even have protective associations with overall CVD, CHD, and stroke incidence and cardiovascular and all-cause mortality in our updated meta-analyses.The inverse associations are in line with findings from most previous meta-analyses [13,36,37,39,40,42].In terms of the nonlinear analysis, 1 meta-analysis in 2016 reported an L-shaped association with the risk of stroke leveling off at 25 g/d of cheese intake [41].Still, another meta-analysis in 2017 derived a U-shaped association with the lowest CVD risk at 40 g/d, which was consistent with the findings from our study [36].Regarding cheese intake and hypertension, a null association with moderate quality of evidence was observed in our study, which was in line with previous studies [43][44][45].In subgroup analyses, an inverse association between cheese consumption and CHD mortality was only notable in Asian populations but not in European and American populations, which may be attributable to the differences in the amount and patterns of cheese consumption among different regions [43].
Results from pervious meta-analyses [31,57,59,60,216] and large prospective cohort studies [217,218] have raised the concern regarding high consumption of dairy products (particularly whole milk) increasing the incidence and mortality of several cancers, for example, prostate, breast, ovarian, and liver cancers and lymphoma.By contrast, cheese consumption has been reported to be inversely related to the risk of colorectal cancer, breast cancer, and prostate cancer in earlier meta-analyses including both prospective studies and case--control studies [46,51,58].However, our latest meta-analyses of prospective observational studies found null associations between cheese consumption and overall and site-specific cancer incidence and mortality, consistent with previous meta-analyses for overall cancer incidence and mortality [14,30,31] and colorectal cancer [32,53].Of note, total energy intake is a crucial confounder in the association between cheese consumption and cancer risk.Failure to adjust for total energy intake in the analyses could lead to spurious conclusions, such as in the case that higher cheese consumption was associated with an elevated risk of overall and breast cancer when not controlling total energy intake.The quality of evidence was moderate for null associations with total cancer mortality and incident prostate cancer and low for null associations with mortality from colorectal, colon, rectal, lung, and gastric cancers and incidence of overall, colorectal, colon (total and distal), rectal, breast (total and ERþ), bladder, and pancreatic cancers.In addition, low-quality evidence revealed an inverse association between cheese intake and ERÀ breast cancer incidence, which was driven by a protective association of cottage/ricotta cheese consumption rather than hard cheese consumption with ERÀ breast cancer risk in a pooled analysis of 21 cohort studies [62].The findings for prostate cancer incidence are also warranted to be confirmed in large-scale, long-term, prospective cohort studies because our linear dose-response meta-analysis of 7 cohort studies suggested a borderline positive association between cheese consumption and prostate cancer risk.
We found a low quality of evidence for an inverse association between cheese consumption and T2D risk in the highest compared with that of the lowest intake, which is in accordance with previous meta-analyses [67,68].Substitution analysis demonstrated that replacing red and processed meat (per 50 g/d) with cheese (per 30 g/d) was associated with 10% decreased risk of T2D [219].However, our linear and nonlinear dose-response analyses did not find significant associations, which was consistent with the most recent dose-response meta-analysis [220].Inconsistently, a meta-analysis published in 2013 showed that per 50-g cheese/d was associated with 8% reduced risk of T2D, and there was a marginal nonlinear association between cheese consumption and T2D risk, with a reduction 50 g/d [15].More studies are needed to clarify the discrepancy between categorized and dose-response analyses.
Dairy products are rich in calcium, magnesium, phosphorus, and protein, which are essential for good bone health [221,222].Nevertheless, the role of dairy intake in preventing bone fractures remains debated [223].Previous meta-analyses reported both inverse and null associations between cheese intake and the risk of hip fracture [73][74][75] and fracture at any site [73].Our updated meta-analyses including only prospective studies supported a favorable association of cheese intake with total fracture risk in the highest compared with that of the lowest intake and with hip fracture risk per 30-g/d increase in cheese consumption.Nonlinear dose-response analyses showed an L-shaped association between cheese consumption and total and hip fracture risk, leveling off at ~40 g/d.Given that the quality of evidence was low, further research is warranted to confirm these findings.
Additionally, low quality of evidence also showed that higher cheese intake was associated with a lower dementia risk in our de novo meta-analysis of 2 prospective cohort studies [197,198].The beneficial association was supported by previous randomized controlled crossover trial [224] and observational studies [225,226], suggesting that cheese consumption may improve cognitive function.
The protective association of cheese consumption with mortality, CVD, bone fracture, and dementia may be attributed to the abundance of nutrients, bioactive compounds, and probiotics in cheese.Dairy products, especially cheese, are a predominant dietary source of vitamin K 2 in many regions [227,228].Vitamin K 2 can improve cardiovascular health by inhibiting and reversing vascular calcification [229,230], reduce age-related bone loss through promoting the γ-carboxylation of osteocalcin and increasing osteoprotegerin [230], and maintain neurocognitive functions through contributing to the biological activation of proteins Gas6 and protein S and the synthesis of sphingolipids [231].Probiotic bacteria in cheese may also interact with the gut microbiome [232], exerting various health enhancing functions [233].Additionally, the cheese matrix can mitigate the harmful effects of saturated fat and sodium [234][235][236][237].Besides the components of cheese itself (i.e., protein or specific micronutrients), the observed inverse associations could also be owing to the fact that increased cheese intake may replace consumption of other foods (e.g., processed/red meat and refined carbohydrates) that have been consistently associated with higher risk of incidence or mortality from chronic diseases [238][239][240] because the studies adjusting for total energy intake hold calories constant, as in isocaloric intervention trials.
It is noteworthy that a borderline positive association between cheese intake and Parkinson disease risk was observed in a previous meta-analysis of 5 cohort studies, which accords with findings from the latest meta-analysis on total dairy and milk [241] and 1 recent prospective study on low-fat dairy foods (including cottage cheese and low-fat cheese) [242].If causal, suggested mechanisms include reduction on uric acid by dairy proteins and the inhibition of calcium and phosphate in dairy products on the formation of 1, 25(OH) 2 D 3 (1,25-dihydroxy-vitamin D 3 ¼ calcitriol) because urate and 1,25(OH) 2 D 3 may protect against Parkinson's disease [242,243].However, the quality of the meta-evidence was very low, and inconsistent results were observed for cheese and other fermented dairy products in some prospective studies [244,245].Accordingly, the findings should be interpreted with caution and validated by further studies.

Strengths and limitations
This umbrella review provides the most recent evidence from prospective observational studies on the association between cheese intake and a wide range of health outcomes.Different from traditional umbrella reviews focusing only on published metaanalyses, we thoroughly and systematically resynthesized the available evidence by incorporating newly identified prospective studies into prospective studies included in previous metaanalyses.On one hand, we updated those outdated meta-analyses to reflect up-to-date conclusions with more statistical power.On the other hand, we performed de novo meta-analyses for specific health outcomes without previous meta-analyses despite enough published original studies to include as many potentially related health outcomes as possible.Furthermore, dose-response analyses were conducted to reveal further the linear and nonlinear association between cheese intake and multiple health outcomes, thereby determining the optimal consumption level of cheese.
There are also some limitations in our research.Given that the original studies included in this review are all observational, some of their inherent limitations could not be excluded, such as residual confounding and reverse causality.Besides, the updated metaanalyses for some health outcomes are highly heterogeneous (I 2 !50%), probably due to the inclusion of original studies involving different populations.Also, caution should be taken when generalizing the conclusions to populations with different genetic backgrounds and dietary habits because most primary studies included were conducted in Europe and North America.Moreover, different types of cheese also vary a lot in dairy matrix and nutrient content like fat and sodium, which may deliver divergent health effects.However, the lack of information on cheese type deterred finer stratified analyses by cheese type.Finally, the limited number of prospective observational studies in meta-analyses for cheese consumption and specific health outcomes-such as cancer at sites other than colorectum, breast, and prostate, overweight/obesity, dementia, fall, and frailty-leads to insufficient statistical power and low credibility of evidence.Thus, further large-scale prospective studies are warranted to ascertain the association of cheese intake with these health outcomes.

Conclusions
Our results indicate that cheese consumption has neutral to moderate benefits for human health, particularly !40 g/d, with a moderate quality of evidence for inverse associations with allcause and CVD mortality and overall CVD, CHD, and stroke incidences.Null associations were observed with cancer mortality, hypertension, and prostate cancer incidence.Although high saturated fat and sodium in some cheeses tend to be emphasized as a health concern in dietary guidelines, cheese also provides some nutrients and bioactive compounds, which potentially may confer some benefits.Environmental effects of cheese production should also be considered.manuscript; AF, ELG: had primary responsibility for final content; and all authors: contributed substantially to the interpretation of the data and read and approved the final manuscript.

FIGURE 1 .
FIGURE 1. Flow diagram of the study search and selection process.

FIGURE 2 .
FIGURE 2. Association between cheese consumption (highest compared with lowest intake level) and all-cause and cause-specific mortality.

FIGURE 3 .
FIGURE 3. Association between cheese consumption (highest compared with lowest intake level) and disease incidence.

FIGURE 4 .
FIGURE 4. Association between cheese consumption (per 30-g/d intake level) and mortality and multiple disease incidence.