Acne is a common chronic inflammatory skin disease of sebaceous follicles [1
]. Clinically, acne is characterized by the presence of open and closed comedones, papules, pustules, and dermal tissue damage with eventually heavy scar formation. Follicular hyperkeratosis, modifications of the sebofollicular microbiome, increase production of sebum with increased amounts of pro-inflammatory monounsaturated fatty acids, and Th17-cell-mediated inflammatory responses are all involved in acne pathogenesis. Sebum production can be induced by insulin-like growth factor-1 (IGF-1) and androgens, whose adrenal and gonadal synthesis is stimulated by IGF-1 [3
]. Although prevalence varies across studies, acne is common in children and adolescents aged 12–24 years and is moderate to severe in 15–20% of cases [1
Heritability of acne alone does not explain high acne prevalence rates of over 80% in western countries [5
]. It has long been debated if a Western diet per se
or specific dietary components contribute to the prevalence and severity of acne [4
]. This has predominantly been investigated in observational studies and only a few trials exist [9
]. In particular, dairy products have been incriminated. Milk-derived amino acids promote insulin secretion and induce hepatic insulin-like growth factor-1 (IGF-1) synthesis [10
]. IGF-1 has been suggested as the pivotal driver of acne and stimulates follicular epithelial growth and keratinization [11
]. IGF-1 gene polymorphism has been shown to increase susceptibility to acne [14
] and IGF-1 plasma levels correlate with acne severity [12
Several worldwide observational studies have been published on dairy intake and acne in children, adolescents, and young adults (7–30 years) in various countries [15
]. Some narrative and systematic reviews about dairy intake and acne have been published [4
]. Recently, a meta-analysis of dairy and acne was published [29
] but with several methodological flaws, including lack of bias assessment and inadvertent double-counting of studies due to duplicate publications [19
] that caused inappropriate weighting of results and skewed pooled estimates. So far, no previous meta-analysis has statistically combined the observational studies in an attempt to estimate the effect of the association of dairy intake and acne with the heterogeneity across studies, a bias assessment, a stratified analysis by study characteristics, and publication bias.
The primary objective of this study was therefore to perform a meta-analysis to estimate the association of acne in children, adolescents, and young adults consuming any dairy products. Furthermore, our aim was to explore the association between acne and intake of varies types of dairy (milk, yogurt, cheese), dairy subgroups (full fat, low fat, skim), and various amounts and frequencies of dairy intake (times per week or day).
This systematic review and meta-analysis was undertaken according to Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines and according to a specified protocol (Supplementary Materials
). The search, selection of studies, full-text reading, and data extraction were performed by CRJ and verified by CE.
2.1. Search Strategy
The search was performed on 11 December 2017 and included all studies up until that date. Studies were identified in the PubMed database using the search terms: (“Dairy products”[Mesh] OR dairy[All Fields] OR milk[Mesh] OR milk[All Fields] OR yogurt[All Fields] OR cheese[All Fields] OR lifestyle[All Fields]) AND (“Acne Vulgaris”[Mesh] OR Acne[All Fields]). We identified 241 records.
2.2. Eligibility Criteria
All observational studies (case-control, cross-sectional, population-based, retrospective) on childhood, adolescent, or young adult acne (max age of 30 years) were eligible if they reported a risk estimate and a 95% confidence interval for acne in a dairy group vs. a non-dairy group, or the raw numbers from 2 by 2 tables of dairy intake and acne.
2.3. Procedure for Selection of Studies
We screened the title and abstracts of 241 articles (Figure 1
). If relevant, we retrieved the full-text articles. We identified 25 full-text articles, but excluded the following 11 studies: duplicate [19
] (there was a statement in the article by Grossi that it was the same cohort and results as [23
]), beliefs/opinions about acne aggravating food items [32
], semi-fat/whole milk vs. skim milk/no milk drinkers [34
], Chinese ying-yang medicine [35
], no control group [36
], adult acne (mean age ≥ 30 years) [37
], milk as part of a Mediterranean diet [39
], milk only as a continuous variable in acne and non-acne groups [40
], and poorly defined intake [41
]. In total, we included 14 studies. Two other studies were identified outside the search, but these studies were duplicates and published simultaneously without a clear statement of which one was the original; therefore, we did not include these papers [30
]. The study selection process is shown in a flow diagram (Figure 1
2.4. Data Extraction and Management
We extracted the following data for each study and entered the information in an excel spreadsheet: author, year, population, country, age, gender, study design, how outcome was estimated, dairy type (dairy, milk, yogurt, cheese), dairy subtype (whole (full-fat), low-fat, skim), dairy amount, frequency of intake (times per day or week), numbers of acne patients and controls subjects in each category of dairy intake, crude and/or adjusted odds ratio (OR) or prevalence ratio with 95% confidence interval (CI), raw numbers to calculate crude OR (95% CI).
2.5. Overall and Subgroup Analyses
The primary objective was to perform a meta-analysis to estimate the odds ratio of acne in children, adolescents, and young adults consuming any dairy compared to those who do not. The secondary objective was to estimate the odds ratio of acne associated with intake of varies types of dairy (milk, yogurt, cheese), dairy subgroups (full-fat, low-fat, skim), and various amounts and frequencies of dairy intake (times per week or day) compared to those who did not consume any dairy/milk.
2.6. Risk of Bias and Study Quality Assessment
The quality of each study was evaluated and scored using the nine-star Newcastle-Ottawa Scale (NOS), a tool used for quality assessment of nonrandomized studies [42
]. Studies were evaluated based on selection, comparability, exposure, and outcome, and scored by a maximum of nine points. Scores above five indicate moderate to high study quality. The NOS for cohort and case-control studies was retrieved from [43
2.7. Statistical Analyses
The meta-analyses were performed with STATA SE 14.0 (Stata Corp., College Station, TX, USA). Using raw numbers, we calculated the crude odds ratios OR (95% CI). Analyses were performed for any dairy intake, any milk intake, full-fat dairy, whole milk, and low-fat/skim milk compared to those who did not consume any dairy/milk (study specific definitions). For any milk intake, whole milk and low-fat/skim milk, analyses of frequencies (times per week or day) were performed using studies by Adebamowo et al. [15
], as these studies had identical ascertainment of the frequency of milk intake. DerSimonian and Laird (D + L) pooled random effects estimates were used. We also present inverse variance (I-V) fixed effects in supplementary Figures. Heterogeneity was assessed by Cochrane Q statistic test and I2
-statistical analysis. The I2
-statistical analysis assess what proportion of the observed variance reflects variance in true effect sizes rather than sampling error [44
]. Publication bias was examined visually by funnel plots and statistically using Egger’s test (one-sided) [45
] and by using the Duval and Tweedie's Trim and Fill to simulate where potential unpublished studies would belong in the funnel plot and to calculate a hypothetical new pooled odds ratio based on the added simulated studies. Robustness of the meta-analysis was examined by “leaving-one-out” analysis. Publication bias and robustness were carried out by use of the statistical program Comprehensive Meta-Analysis (CMA) version 3 (Biostat, Englewood, NJ, USA) for any dairy intake vs. no dairy intake and any milk intake vs. no milk intake. Four studies provided adjusted estimates for milk intake, with one study providing them as odds ratios [46
], and three studies as prevalence ratios [15
]. In a sensitivity analysis, we used only adjusted prevalence ratios from the studies by Adebamowo et al. [15
]. Stratification on acne severity was not possible because of too few studies.
Intake of any dairy, any milk, full-fat dairy, whole milk, low-fat/skim milk, and yogurt regardless of amount or frequency were associated with a higher odds ratio for acne compared to no intake in individuals aged 7–30 years. Intake of cheese was associated with a borderline higher odds ratio for acne compared to no intake. Stratifying the association of any milk by frequency of intake revealed that intake of 1 glass of milk or more per day was associated with a higher odds ratio for acne, whereas 2–6 glasses per week was not, compared to intake less than weekly. Stratified analyses for any dairy intake and acne fat content demonstrated that full-fat dairy and whole milk had lower odds ratios, whereas low-fat/skim milk had higher odds ratios than the overall summary estimates; a likely explanation for this observation could be that the amount of milk consumed for low-fat/skim milk is higher than that for whole milk. However, results should be interpreted with caution due to heterogeneity and bias across studies.
The meta-analyses showed considerable heterogeneity reflecting the heterogeneous age and gender of the participants, various study characteristics, ascertainment of information about milk intake and acne, reporting of milk intake, and acne severity across the studies. In general, stratifying on subgroups in sensitivity analyses revealed that heterogeneity diminished for most subgroups, but also revealed that especially meta-analyses conducted on females, whole milk, North America, and questionnaire ascertained acne diagnosis demonstrated high heterogeneity. Prospective studies and studies with physician-verified diagnosis of acne had low heterogeneity.
Despite the stratifications, confidence intervals were overlapping.
Stratifying on age and gender demonstrated similar odds ratios; however, the gender stratified analyses had higher odds ratios than in the gender combined analyses. Smaller studies had higher odds ratios than large studies, African and Asian studies had higher odds ratios than other studies, and prospective designs had higher odds ratios than other designs. A recent multinational European online questionnaire study in adolescents showed that acne prevalence did not differ by gender but differed by country, and acne was more prevalent in younger people and obese people [27
]. Intake of milk varies globally and is largely dependent on genetically determined lactase persistency, which is high in people of Northern European descent, but lower in people of Southern European descent, patchy in Africa, and low in the Middle East and Asia [48
]. The weaning of the lactase enzyme activity usually happens in childhood and early adolescent years. How the age of weaning of the lactase enzyme activity impacts acne development is not known.
We used random effects method in all meta-analyses, which includes between-study variance and has a higher degree of statistical uncertainty built into the model. Thus, 95% confidence intervals are wider compared to fixed effects models. Even in these models, the results of the meta-analyses were significant. There was evidence of publication bias with Egger’s test with an overweight of smaller studies overestimating the odds ratio compared to the pooled summary estimate. If the meta-analyses had captured all the relevant studies, we would expect the funnel plots to be symmetric. The selective reporting may be explained by studies with null-findings or negative results being deliberately not published because of authors not submitting or editors rejecting them or authors not finding enough merit in a potential publishable study [49
]. Furthermore, some studies reported only the pooled exposures for different dairy groups rather than showing the stratified results for each of the dairy groups and/or for each reported frequency of intake [23
], and some studies had only collected an overall dairy or milk variable with no possibility for stratification [20
]. However, the trim and fill method did not change the overall estimates for “any dairy” or “any milk” remarkably.
There are many limitations of the included studies [4
]. Self-reported acne with lack of a physician verified diagnosis of acne [15
] may lead to misclassification bias as validity of self-reported acne is at best only moderate, with sensitivity of 55%, specificity of 72%, positive predictive value of 70%, and negative predictive value of 57% [50
]. Including other dermatology patients as controls [26
] may attenuate associations, as seborrhea may play a role in several diseases. The observational studies were cross-sectional [20
], case-control [18
], retrospective [15
], or longitudinal [16
]; thus, in most studies we cannot rule out reverse causation. Questionnaire ascertainment of dairy intake varied between the articles and only a few studies used validated food frequency questionnaires [15
]. Despite the food questionnaire used, participants may deliberately over- or underestimate (information bias) or not accurately remember (recall bias) when filling out questionnaires about dairy intake and acne. Furthermore, it was not possible to differentiate acne development, acne triggers, and severity of acne in the meta-analyses. Only a few studies provided adjusted results [15
] so we based most of the analyses on raw numbers, which makes it difficult to rule out confounding from other dietary factors (e.g., glycemic index or calorie intake) or other lifestyle factors previously associated with acne [4
Acne prevalence varied remarkably across the included studies, between 7–89%. The retrospective study by Adebamowo in 2005 with 7.3% acne cases focused on recall data provided by subjects in the Nurses’ Health Study II (NHS), which were aged 25–42 years old in 1989 when information on teenage acne was collected [15
]; thus, the acne prevalence is likely underestimated and the results from this study may not be representative. Furthermore, the studies from 2006 and 2008 were offspring studies from the NHS in girls and boys [16
]; however, leave-one-out analyses revealed that only the Adebamowo 2005 study was an outlier [15
The observational studies may suffer from bias from confounding and reverse causation [9
], are unable to indicate causality of the relationship between dairy and acne, and unable to prove preventive effects of abstaining from dairy. Only one study exists on milk intake and acne. The study is uncontrolled and unblinded and is based on medical students who drank milk or consumed other potential acne provoking foods. In addition, the total number of people with and without acne lesions were counted for all foods combined, but with no formal statistical testing [51
]. Thus, there is still a knowledge gap with respect to whether dairy intake is causally associated with acne, acne flare, or acne severity and to what extent. To answer this question, we would ideally need results from large clinical randomized double-blind placebo-controlled trials (RCT); however, the question is whether this is realistically possible ethically, clinically, and/or operationally. Another approach (which no previous studies have yet undertaken) would be to perform a Mendelian Randomization study of lactase persistence, dairy intake, and acne using genetic lactase persistence as a proxy for lifetime dairy intake under the assumption that alleles are randomly distributed at conception [52
]. Such a study design mimics an RCT and allows for the causal estimate of dairy intake and acne.
The observational studies all assessed dairy intake as an isolated factor. However, dairy is part of various individual and cultural specific diets and not a single factor with a single factor prediction (“reductionist approach” [54
]). Instead, other factors which can affect the bioactive properties of nutrients in dairy and milk intake should be taken into consideration, such as macro- and micronutrients (fat, protein, carbohydrates, vitamins, sodium, and minerals), the dairy structure (liquid or solid), fermentation, and processing (holistic approach [55
]). Only two studies in the meta-analysis also reported the glycemic load and glycemic indices of food consumed in conjunction with milk/dairy products [18
], but did not report the glycemic load from the dairy consumption specifically. Hyperglycemic carbohydrates enhance insulin signaling, which promotes insulin and IGF-1 signaling, which in a synergistic fashion with milk stimulate mTORC1(mammalian target of rapamycin complex 1) signal transduction [56
]. There is accumulating evidence that acne belongs to the spectrum of mTORC1-driven diseases of civilization including metabolic syndrome, obesity, insulin resistance, and cancer [57
]. A randomized trial has shown that a low-glycemic-load diet improves symptoms in acne vulgaris patients [58
]. Interestingly, no acne was observed in the Kitavan Islanders (Pacific Ocean) and in the Ache Hunter-Gatherers from Paraguay, who live under Paleolithic conditions without milk/dairy and hyperglycemic food, although it should be acknowledged that many other differences exist to Western societies [59
]. To present the pathological effects of milk in the Western diet it is therefore important to provide controlled studies that consider milk consumption in association with glycemic load and index as part of a mixed diet [60
Recently, a meta-analysis of dairy and acne was published [29
] but with several methodological flaws, including the inadvertent double-counting of studies (Landro [19
], and Tsoy [30
]) due to duplicate publications, which caused inappropriate weighting of results and skewed pooled estimates. Using the double-counted studies by Tsoy, the authors also only used the most severe category of acne, which caused extremely high odds ratios of 10 and 12 to be included in the meta-analysis, further skewing the pooled estimates. Furthermore, the meta-analysis included a study by Agamia [41
], which we decided to exclude as the intake of “milk and dairy produce” was poorly defined as “low” and “high” intake but not defined with any frequency, type, or amount of milk. The previous meta-analysis also did not provide evidence for the exact search strategy to be replicated, for the bias assessment using the Newcastle Ottawa scale, for leave-one-out analyses, or funnel plots of publication bias. As a comparison, in our meta-analysis, we included the exact search string so it can be replicated, the heterogeneity across studies, a bias assessment using the Newcastle-Ottawa scale presented with a table, a stratified analysis by study characteristics presented in figures, the details of the “leave-one-out” analysis presented in figures, and the publication bias presented in figures. Furthermore, we excluded duplicate studies, and we included four more papers [24
] that were not included in the previous meta-analysis but should have been as the studies were published before the search for the previous meta-analysis was done in August 2017 [29
]. It is of crucial importance that authors of meta-analyses have a critical judgement of the reliability and validity of the papers they consider including in a meta-analysis, otherwise the conduct and assessment of systematic reviews may be hampered.