Nowadays, the world is experiencing an obesity epidemic. In 2016, 39% of adults were overweight and 13% were obese, worldwide. Thus, obesity’s prevalence is three-fold higher than in 1975 [1
] and it is still rising. For example, the prognosis for the United Kingdom (UK) estimates that approximately 60% of men and 50% of women will be obese by 2050 [2
]. Today, the majority of countries around the world are affected by obesity prevalence rates above 10% and estimates suggest a rise to 20% of world’s population being affected by obesity by 2025 [3
]. Globally, this rapid increase in the prevalence of overweight and obesity is one of the most important health issues [5
Obesity is a major contributor to the global burden of disease through its deuteropathies of serious non-communicable diseases (NCDs) [7
]. Psychological, pulmonary, orthopaedical, cardiovascular, metabolic, reproductive, and oncological diseases are attributable to obesity. For example, obesity is associated with depression, obstructive sleep apnoea, osteoarthritis, myocardial infarction, diabetes mellitus type 2, infertility, and colon cancer [8
]. Therefore, obesity may cause premature death. In 2015, obesity contributed to 4 million deaths, equivalent to 7.1% of all-cause mortality, worldwide [7
]. Furthermore, there is a huge economic and social burden of obesity. Total health costs and drug costs increase with increasing body mass, which is proportionally beyond their standard values. Obesity correlates to a low socioeconomic status, as well [14
]. Therefore, one target of the World Health Organization’s (WHO) Global NCD Action Plan 2013–2020 is to halt the rise in obesity by 2025 [21
]. Moreover, actions against obesity are necessary to achieve the third Sustainable Development Goal which comprises the target to decrease the number of premature deaths caused by NCDs by 33.3% by 2030 [22
The underlying pathological process of obesity is represented by the increase in both the total number and size of fat cells, which leads to a heightened accumulation of fat cells in relation to one’s body size [8
]. Being overweight is defined as elevated body fat accumulation, while obesity defines a situation characterized by an excess body fat mass [1
]. The most common used measurements to assess human body size are anthropometric measures (e.g., body mass index (BMI) or waist–hip ratio (WHR)) and measurements of body composition (e.g., bioelectrical impedance analysis (BIA) or dual-energy x-ray absorptiometry (DEXA)). However, there are many more methods available for assessing human body weight status [26
The etiology of obesity is multifactorial, but the fundamental determinant is the positive energy balance. This is mostly determined by a high energy intake through inappropriate nutrition and a low energy expenditure through physical inactivity [24
]. In view of nutritional physiology, it is notable that breakfast is the meal eaten after the longest period with an empty stomach (i.e., postprandial fasting), and therefore, it has the potential to decrease the risk of weight gain due to several metabolic mechanisms [31
]. For example, reduced levels of ghrelin (growth hormone release inducing, an appetite suppressant peptide) and increased postprandial energy expenditure have been observed when breakfast is eaten. Moreover, a hypothesis exists stating that nutrient timing is part of the circadian rhythm. In the scope of breakfast skipping, negative effects on the circadian rhythm—such as the irregulating of metabolism—with consequences related to weight management, are conceivable [32
]. Additionally, international recommendations on breakfast agree in their statements that daily consumption of breakfast is advisable for providing a sufficient intake of macro- and micronutrients, maintaining body weight, and improving cognitive functions [34
]. In contrast, breakfast skipping is associated with elevated plasma lipoproteins and fasting glucose [19
], and insufficient intake of micronutrients [39
Considering the huge medical, financial, and social burden of obesity, this study aims to examine whether breakfast skipping is associated with adult body weight. Existing systematic reviews and meta-analyses on this topic have examined target groups including children and adolescents in all study designs [40
]. However, adults have only been studied in cross-sectional [53
] and interventional study designs [56
]. Interventional studies comprise the highest level of evidence but are limited to comparably young and healthy participants, analyzed in small sample sizes under laboratory conditions. For this reason, the systematic review and meta-analysis presented here is based on primary studies using observational longitudinal study designs, to gain further evidence with a high level of external validity.
2. Materials and Methods
A systematic review and subsequent meta-analyses were conducted. The study was planned and conducted in accordance with the “Meta-analysis Of Observational Studies in Epidemiology” (MOOSE) standards [64
]. The systematic literature search, screening of the identified literature, data extraction and quality assessments were carried out independently by two reviewers (Julia Wicherski, Florian Fischer). There were no discrepancies in judgement between the two independent reviewers.
2.1. Search Strategy
According to the “Population-Item-Comparison-Outcome” (PICO) framework, the population of interest was exclusively adults—from all around the world—aged 18 years or older. The exposure of interest was breakfast skipping. This was compared to breakfast eating as regards the occurrence of overweight, obesity or weight gain, respectively. The literature review was conducted in PubMed and Web of Science and included all literature related to the topic that was published up until September 2020.
The included studies were from observational longitudinal study designs and had specified effect estimates expressed as risk ratios (RRs), such as odds ratios, hazard ratios, or relative risk. These RRs were reported with corresponding 95% confidence intervals (95% CI). Alternatively—instead of RRs—some of the included studies reported a coefficient with corresponding 95% CI, in the case that the regression models contained a linear term, such as continuous variables for breakfast frequency and/or BMI. Participants of the included studies were aged ≥ 18 years. The outcome was measured through BMI, waist circumference (WC), WHR, waist-height ratio (WHtR), body fat percentage (BF%), or weight change (gain or loss). The included studies were published/peer-reviewed articles and written in English or German. Studies were excluded if the study was an interventional study, a cross-sectional study, a case study or case report. Likewise, studies were excluded if they did not have specified effect estimates expressed as RRs or coefficients. Individuals who were aged < 18 years (e.g., children, adolescents) or who were pregnant were also excluded.
To build our search queries, we used the Boolean operators “AND” and “OR”. Furthermore, we used truncations to search for all terms that begin with a word of interest (e.g., using obes* to find obesity as well). We searched in specific fields like “Mesh” or “Publication Type” or “Title (TI)” or “Topic (TS)”, and we used abbreviations of effect estimates in our search (i.e., odds ratio (OR), hazard ratio (HR) and relative risk (RR)). The following query was used for PubMed: “(((((((breakfast* OR “breakfast skipping” OR “breakfast frequency” OR “breakfast omission” OR “breakfast” (Mesh) OR ((breakfast* AND (skipp* OR frequen* OR omit* OR omis* OR consum*))) AND (“body weight” (Mesh) OR “overweight” (Mesh) OR “obesity” (Mesh) OR “body weights and measures” (Mesh) OR “body weight” OR “body fat” OR “body mass” OR *weight OR obes* OR adipos* OR “BMI” OR “WHR” OR “WC” OR “WHtR” OR waist OR circumference OR “body size” OR “body fat distribution” (Mesh)) AND (cohort* OR “cohort studies” (Mesh) OR “case control studies” (Mesh) OR “OR” OR “RR” OR “HR” OR retrospective OR prospective OR observational OR “longitudinal studies” (Mesh) OR “follow-up studies” (Mesh) OR “Observational Studies as Topic” (Mesh) OR “Observational Study” (Publication Type))))))))”.
For Web of Science, the following query was utilized: “((TI = (breakfast*) AND TI = (skipp* OR omi* OR freq* OR eat*) OR (TI = (“breakfast skipping” OR “breakfast omission” OR “breakfast frequency”) OR TS = (“breakfast skipping” OR “breakfast omission” OR “breakfast frequency”))) AND (TI = (*weight OR obes* OR adipos* OR fat OR mass OR “body weight” OR “body fat” OR “body mass”) OR TS = (*weight OR obes* OR adipos* OR fat OR mass OR “body weight” OR “body fat” OR “body mass”)))”.
2.2. Data Extraction and Quality Assessment of Included Studies
Information on the first author’s surname, year of publication, country, study name, design and follow-up period, total number of participants, number of cases, distribution of sex and mean age, exposure and outcome definitions and measurements, specified effect estimates—expressed as RRs or coefficients—with the corresponding 95% CI, and adjusted covariables were extracted from each study and included in the qualitative and quantitative syntheses.
The risk of bias in included studies was assessed by applying the “Risk Of Bias In Non-randomized Studies of Interventions” (ROBINS-I) tool [65
] for each study (Supplementary Table S2
). Along with the risk of bias due to confounding, age, sex, education or socioeconomic status, smoking, alcohol, physical activity and total energy intake (TEI) were established in accordance with the literature [31
]; as these are important covariables because of their confounding nature of being associated with breakfast and body weight. Moreover, the quality of evidence of the conducted meta-analyses was assessed by using the “Nutrition Grading of Recommendations Assessment, Development and Evaluation” (NutriGRADE) approach [82
] (Supplementary Table S3
2.3. Statistical Analysis
Pairwise meta-analyses were conducted by comparing skipping breakfast on ≥3 days per week to ≤2 days per week, and skipping breakfast to eating breakfast without detailed category definition, respectively. The first meta-analysis for breakfast skipping focused on the occurrence of overweight (defined as BMI ≥ 25 kg/m2) and/or obesity (defined as BMI ≥ 30 kg/m2). The second meta-analysis for breakfast eating versus skipping focused on changes in BMI (change in kg/m2).
Due to the fact that there are no standard definition criteria for breakfast skipping and eating in regard to its frequency, exposure definitions of this meta-analyses were adopted from the definitions used by the primary studies.
A random-effects model was applied due to the assumption that a normal distribution of the true effect and the heterogeneity within and between the studies was caused by unexpected effects rather than residual effects [83
]. At first, the natural logarithm of each risk estimate expressed as a risk ratio (logRR) for overweight/obesity, respectively, was calculated. Subsequently, the logRR for each included study was weighted and pooled accordingly to the variance-based method of DerSimonian and Laird [85
], which considers the variability within and between the studies. In the scope of the second meta-analysis, β regression coefficients were weighted and pooled directly.
Moreover, for each of the two meta-analyses conducted, we performed a sensitivity analysis for the standard error adjustment by applying the Knapp–Hartung method. This ensured consideration of the less favorable statistical properties of the DerSimonian and Liard method in meta-analyses with a small number of included studies [86
] and the heterogeneity of included studies, as was the case in our analyses.
We used the I2
test to evaluate the heterogeneity. I2
is a measure of inconsistency which describes the variability between the studies included in the meta-analysis due to heterogeneity rather than chance [87
]. According to the Cochrane recommendation [88
], analyses should include ≥ 10 studies to check for publication bias. Since only two studies were included in the meta-analyses, funnel plots and Egger’s test were not applied. Checking for missing studies by applying the trim and fill analysis was also deemed to not be practicable. The meta-analyses were conducted by using the metan and metareg package in Stata v16.1 (College Station, TX, USA).
All nine studies included in the review reported a statistically significant association between breakfast skipping and overweight/obesity or weight gain. Moreover, eight out of nine studies displayed that breakfast skipping increased the relative risk for overweight/obesity or weight gain, respectively, compared to eating breakfast. Both meta-analyses provided a very low meta-evidence; one showed an increased relative risk for overweight/obesity, while the other might imply a small tendency for weight gain displayed as increasing BMI values, when breakfast is skipped. Skipping breakfast on ≥3 days per week increased the risk to become overweight/obese about 11% (95% CI: 4%, 19%) compared to skipping breakfast on ≤2 days per week. The second meta-analysis showed no association between breakfast skipping and changes in BMI. With regard to our sensitivity analysis, no association between breakfast skipping and overweight/obesity or BMI change, respectively, was found.
The results of this report are similar to recent systematic reviews and meta-analyses on studies with cross-sectional designs, but with a lower strength of association observed in our results. For comparison, the relative risk for overweight/obesity was increased by about 75% (95% CI: 57%, 95%) for the breakfast skippers compared to the breakfast eaters analyzed in the meta-analysis on cross-sectional studies from Asian and Pacific countries [95
]. In contrast, the results of meta-analyses on studies with interventional designs reported associations between breakfast skipping and body weight concerning weight loss in breakfast skippers [63
]. In accordance, the recent meta-analysis by Bonnet et al. [63
] reported a weighted mean difference of −0.54 kg (95% CI: −1.05 kg, −0.03 kg) in body weight when breakfast was skipped in trials conducted in the UK and USA, with a follow-up time between 4 and 16 weeks.
In addition, breakfast skipping is part of several intermittent fasting programs [97
]. One intermittent fasting method is time-restricted feeding, whereby the individual fasts for 16 to 20 h per day and eats only in the remaining 4 to 8 h per day, mainly in the evening. This fasting program is called “20:4” or “16:8”, respectively [97
]. Systematic reviews on intermittent fasting programs [97
] suggest that body weight reduction is possible. This association is stronger in interventional and randomized controlled studies than in observational studies [97
]. Those results stand against the findings of the present review and meta-analyses and might be due to the different study designs and outcome measurements.
Randomized controlled trials are considered the gold standard study design, they are conducted under laboratory conditions, provide good internal validity, and are labeled with the highest level of evidence. Typically, their outcome measurements are more trustworthy than outcome measurements in observational study designs. In view of overweight and obesity, trials utilized body composition values (e.g., fat mass and fat free mass) by using DEXA or BIA, respectively [63
], whereas observational longitudinal studies utilized BMI or weight change in kilograms by using a tapeline and scale [36
]. However, trials such as those conducted by Sievert et al. [96
] and Bonnet et al. [63
] are limited to a small sample size (i.e., <500 people included in meta-analysis) and short observation intervals (i.e., between 2 and 16 weeks). In contrast, our meta-analyses on cohort studies contained ≥ 100,000 participants. Furthermore, those observational longitudinal studies followed-up the study populations after between 5 and 18 years [36
]. The main advantage of observational longitudinal studies is that they are able to determine real-world conditions. Their results provide better external validity and are more transferable to the general population than the results of trials. A final difference to take into account is the fact that the analyzed population in five out of seven trials included in the meta-analysis by Bonnet et al. [63
] (or five out of ten trials included in the study by Sievert et al. [96
]) were overweight/obese, and the population was younger, with a mean age of 35 years.
The physiological principles of intermittent fasting are interesting in view of weight changes. The metabolic conversion of receiving energy from the glycogen stores of the body starts 12 h after the last ingestion of food. A few days later, up to 90% of energy supply stems from the adipose tissue. This has the clinical effect of losing the visceral fat. With losing this fat, the body weight and the metabolic health risk decreases, but levels of the hormone leptin increase. The latter causes a ravenous appetite [99
]. People who skipped breakfast are more likely to crave for high caloric food than low caloric food [103
]. In accordance with this, the issues of whether breakfast’s satiating effect takes influence on the TEI and whether breakfast eating or skipping increases the TEI have been discussed. Some studies have reported a lower TEI in breakfast eaters compared to skippers [104
], while other studies reported a higher TEI [41
Breakfast is only one of several determinants on body weight, and even in the scope of breakfast itself, there are different factors influencing the body weight status through nutritional physiological processes. For instance, the time of the day at which breakfast is eaten (e.g., before 10 a.m.), time spent on eating (e.g., at least 20 min), its energy intake (e.g., containing 25% of TEI) and its composition (e.g., wholegrain-based, fiber-rich foods) are associated with the body’s weight status due to the release of gut hormones [62
]. For example, the analysis by Deshmukh-Taskar et al. [40
] suggests that ready-to-eat cereals are the best kind of breakfast for losing and maintaining body weight. This difference between type of breakfast and body weight status has also been seen in other studies [41
]. Additionally, the analysis by Kent et al. [109
] shows that the larger the breakfast proportion size, the lower the BMI of men. The observed relationship was even more pronounced in vegetarian men, compared to their non-vegetarian counterparts [109
]. This opens a new viewpoint which should be considered when looking at the relationship between breakfast skipping and body weight.
Since the nourishment of European, American, and Asian breakfasts are not comparable [36
], one needs to consider that the current meta-analyses pooled breakfasts from countries with different breakfast types. Therefore, this analysis is limited to the extent that the examined results are most likely not transferable to the context of an individual country.
Furthermore, breakfast is an indicator of general health-promoting lifestyle and behavior: breakfast skippers are more likely to be smokers compared to breakfast eaters [34
]. With a decreasing number of days on which breakfast is consumed per week the likelihood to be a smoker increases [76
]. Likewise, breakfast skipping is associated with a higher level of alcohol consumption [34
]. Breakfast skippers drank on average 20.5 g alcohol per day, whereas breakfast eaters drank averagely 11.9 or 8.6 g alcohol per day, respectively. Ready-to-eat cereal breakfast consumers drank less alcohol than consumers of other types of breakfast [110
]. As well, breakfast skippers are more likely to be physically inactive than breakfast eaters [34
]. Moreover, breakfast skipping is correlated to a worse quality of sleep [76
]. Furthermore, breakfast skippers are more likely to have deficits in macro- and micronutrient intake [41
]. In addition, breakfast skippers consumed the highest content of added and free sugar [71
]. Lastly, breakfast skippers are more likely to have lower scores of general health perceptions, vitality, social functioning, emotional roles, mental health, and total health status scores compared to breakfast eaters [74
]. Therefore, these factors should be adjusted, as was done in some of the included studies.
Moreover, there are socioeconomic and demographic differences in characteristics of breakfast skippers compared to breakfast eaters [34
]. People affected by poverty are most likely to skip breakfast [40
], whereas people affiliated by the highest socioeconomic status are most likely to consume cereals for breakfast [71
]. As well, being married seems to increase the likelihood of breakfast consumption [76
]. Breakfast skippers tend to be of younger age, so the likelihood of breakfast consumption increased with increasing age [34
]. Likewise, the literature shows a sex gradient: men are more likely to skip breakfast than women [34
]. Lastly, there are differences between ethnicities displayed in the literature. Breakfast skippers are more likely to be of non-Hispanic black ethnicity and are less likely to be of non-Hispanic white ethnicity than breakfast eaters [75
With regard to the behavioral, demographic and socioeconomic factors which are associated with breakfast skipping, most of them are also associated with overweight/obesity and may, therefore, lead to confounding factors: people with overweight/obesity are more often physically inactive compared to individuals of healthy weight [73
]. An unhealthy diet is also seen more often in people with overweight/obesity than in people with normal weight [68
]. Additionally, BMI decreased in people who smoked compared to non-smokers [68
]. Overweight/obese people are more likely to have a lower socioeconomic status and/or have a lower educational level [72
]. Another study [116
] reported that BMI increased with increasing economic status in both women and men. With increasing age, the likelihood of overweight/obesity also increased [67
]. The sex gradient indicates that women are more likely to be overweight/obese, globally [118
]. Moreover, the prevalence of overweight/obesity is positively linearly related to the income level of a country. The higher a country’s income level, the higher the prevalence [118
]. Besides income-dependent differences in the prevalence of overweight/obesity, a variety in fat distribution regarding ethnicity could have an influence [9
The described issue of an insufficient number of studies included in the meta-analyses, is one of the main limitations of this study. This is likely leading to a publication bias and a small study effect but it was not practicable to check for this with only two included studies for each meta-analysis. With regard to this, an overestimation of the true effect size is likely [83
]. Due to this, this study is not able to estimate the range in which the true effect is located with an appropriate level of precision. This is also visible when looking at the reported 95% CIs, which are relatively broad. Additionally of note, a second limitation is based on the fact that studies included in the analysis show some differences in their methodologies. For example, we pooled data of a primary and secondary nature and with different outcome measurements (i.e., BMI change/year and BMI change without a clear period). Moreover, the sensitivity analyses (standard error adjustment to face the heterogeneity) displayed no association between breakfast skipping and overweight/obesity or BMI change, respectively. Accordingly, the reported results are not robust for variations [83
]. Additionally, information on differences in this association dependent on the composition, quantity, and quality of breakfast cannot be displayed. Brikou et al. [119
] report that the definitions of breakfast and the classifications of being a breakfast skipper/eater vary highly. As most studies did not report a definition of breakfast or used different definitions of breakfast, a misclassification bias must be considered. Stemming from the fact that this meta-analysis did not comprise a sufficient number of studies to stratify for the quality or measurement methods of the studies, we were not able to provide suggestions on the influence on the analyzed association.
In the scope of significance testing, it is also worthy of note that the larger the analyzed sample size, the higher the likelihood for a statistically significant result (and rejecting H0). The present meta-analyses contained a large sample (approx. 100,000) but very small number of included studies (i.e., two) and showed small effect sizes (i.e., RR = 1.11 and β = −0.02). Therefore, a false negative effect (i.e., beta error) should be taken into account, as well [120
Furthermore, the study with the comparably highest risk of bias (i.e., critical risk) [37
] received the largest weighting (i.e., 75%) within the first meta-analysis. The risk of bias in the other two studies included in the meta-analyses also rated high (i.e., serious risk). Additionally, the provided meta-evidence is of a very low level for both meta-analyses.
The measurement methods of breakfast skipping behavior were different between the included studies. Only one out of nine studies used an interview-based assessment, while the residual eight studies used a questionnaire-based assessment of breakfast skipping. This should be taken into account, because the meta-analysis by Horikawa et al. [95
] implied that interview-based assessment of breakfast skipping is more strongly associated with overweight/obesity than the questionnaire-based assessment of breakfast skipping. Therefore, it could be suggested that the results of this thesis are limited in their precision. Another limiting factor is confounding in the included studies. Only two out of nine studies adjusted for all seven important variables [90
]. Because most studies did not adjust their analyses for these important covariables, confounding due to missing adjustment, and a residual confounding must be considered. Finally, this systematic review has not been registered prior to execution.
4.2. Strengths and Further Research Needs
To the authors’ best knowledge, this is the first meta-analysis on the association between breakfast skipping and body weight, examined in the form of an observational longitudinal study of adults, worldwide. Furthermore, the results of these meta-analyses are strengthened due to the observational longitudinal nature of the included cohort studies. They represent real-world conditions of the association between breakfast skipping and body weight. Cohort studies avoid recall bias and decrease the potential for selection bias [120
Considering the reported results of previous and current research, there are still open questions left and new questions accrued which collectively implies a further need for research. In general, there is a need for a clear and consistent definition of breakfast eating and breakfast skipping. Breakfast behavior is mostly not part of a validated measurement tool when looking at the included studies of this analysis. It could be an aim in future research to develop a validated measurement tool for breakfast eating and breakfast skipping.
Moreover, this review displayed a lack of studies conducted in European and African countries compared to the number of studies from the USA and Japan. In addition, future observational studies should consider other measurement methods for body weight/body composition than only BMI. Randomized controlled trials already used these specific outcome measurement methods but analyzed only short-term changes in body weight. Large observational longitudinal studies equipped with those outcome measurement methods would hold an advantage to answer the question of whether breakfast is a relevant setscrew aspect of lifestyle to consider in body weight management to avoid overweight-related deuteropathies. Comparatively, long-term randomized trials would be helpful to examine the association itself with a great deal of evidence but with less transferability into lifestyle. Lastly, both designs—trials and cohorts—are important in examining the association between exposure and outcome at first and working out this association under real and heterogenic life conditions subsequently. For another comparison, there is a need for studies with observational longitudinal design on the association between intermittent fasting and body weight, to compare those results with the results of breakfast skipping/eating and body weight.