Associations between dietary factors and obesity-related biomarkers in healthy children and adolescents - a systematic review

We conducted our systematic literature review in accordance to the PRISMA Statement [14]. We searched the scientific databases PubMed/Medline and Web of Science Core Collection (WoS CC) for potentially eligible articles. We developed a systematic search strategy, which included the following Medical Subject Heading terms from PubMed: “child”, “adolescent”, “food and beverages”, “diet” “food quality”, “triglycerides”, “blood glucose”, “blood pressure”, and “C-reactive protein”. In addition, we took into account free-text terms like “dietary intake” and “biomarker”. For the WoS CC database we adapted our final PubMed search strategy. In Additional file 1 we provide the exact search strategies of both literature databases. The final screen on 2016/02/29 resulted in 2727 hits (Fig. 1) from both databases after excluding 335 duplicates identified via Endnote software (Thomson Reuters).

Table 1 provides an overview of the a priori defined inclusion and exclusion criteria that we applied to select eligible articles. The exposure variable was a dietary factor (macronutrient, single food, dietary pattern). The outcomes of interest were obesity-related biomarkers (Table 1), which we had previously derived from two existing literature reviews that investigated associations between biomarkers and obesity in children and adolescents [8, 15]. Barkin, et al. [15] revealed that biomarkers related to obesity in adults (e.g., cortisol) did not show such associations in children. We therefore took only those biomarkers into account that were associated with obesity in children.

Three of the authors (JHK, IM, CB) independently screened all 2727 abstracts. By applying inclusion/exclusion criteria to the information contained in the abstract, we reduced the pool of potentially eligible articles to 316 (Fig. 1). The same three authors also evaluated the retrieved full-text articles applying the same inclusion and exclusion criteria that were used for the abstract selection. Any disagreements during the selection process were discussed among all three reviewers since consensus was reached. Finally we included 81 full-text articles into our review.

Three of the authors (JHK, IM, CB) extracted relevant information from all 81 articles using a standardized data extraction template. We designed our template with the intent of shedding light on possible reasons for inconclusive findings. Thus we included detailed information on food/nutrient intake (e.g., assessment method, nutrient databases) and biomarkers (e.g., assessment method, blood collection procedures), in addition to general information on study design and characteristics of the study population (e.g., recruitment methods, sample size). We also extracted from each article details on data analyses (e.g., statistical methods, adjustment for confounders) and relevant information to evaluate a potential risk of bias.

Study quality was assessed by two of the authors (JHK, CB) independently. We slightly adapted a quality assessment tool that was originally developed by Voortman, et al. [12] to evaluate study quality of all articles included in our review. The quality assessment tool used a rating scheme ranging from low to high (low quality: 0-4, moderate quality: 5-8, high quality: 9-11). We took into account the following characteristics to rate the study quality: study design, sample size, intake validity, adjustment for potential confounders, and the occurrence of a selection bias (Additional file 2). We decided to use this assessment tool because it takes criteria into account that are unique for nutrition studies (e.g., the validity of the dietary assessment) and thus are not reflected by the standard tools. As measurement error in dietary intake assessment might also be on potential explanation for inconclusive findings we preferred the tool introduced by Voortman et al. 2015 to quality assessment tools conventionally used in systematic reviews.

Two of the authors (CB, JHK) rechecked a random sample of all data extracted to ensure high data quality. Afterwards, we prepared data for descriptive data analyses. In some cases, we extracted multiple subentries (from now on labeled as observations) from one article because they reported one value for the total study sample and, for example, additional values for either males or females. In such cases, we decided to exclude the overall value from further analyses to avoid double counting of individuals. In addition, several articles reported on associations between a dietary factor and more than one biomarker (e.g., blood pressure, total cholesterol) or on associations between different dietary factors (e.g., vegetables, dairy products, sweets) and one biomarker. In these cases, we kept multiple observations from one article in the descriptive analyses as these observations provided separate information of interest.

We conducted descriptive analyses to give an overview of the main characteristics of all articles included (see Additional file 3 for a detailed overview of the characteristics of each single study). Due to heterogeneity of the articles with regard to, for example, study designs, dietary assessment methods, and statistical analyses, we decided that conducting a formal meta-analysis was not appropriate. Instead we provide a descriptive summary of our results.

Our systematic review revealed that only a minority (19.2%) of the total observations showed significant associations between a dietary factor and any of the obesity-related biomarkers included. Furthermore, for many biomarkers we were not able to identify enough observations to draw clear conclusions on possible associations between a dietary factor and the respective biomarker. Our results confirm the findings of existing reviews [12, 13]. Since nonsignificant and inconclusive findings may impede the development of effective strategies against childhood obesity, we decided to take a closer look at possible reasons for these findings.

Nutrition epidemiology is also affected by bias resulting from imprecision in the measurement of dietary intake [24]. Measurement error may cause over- or underestimation of the impact of exposure [24]. Thus the accurate assessment of dietary intake in children and adolescents is essential not only to monitor nutritional status but also to draw reliable conclusions on diet-disease relationships within this age group [25]. However, valid assessment of dietary intake can be particularly difficult for this age group [26]. Children younger than 8 years do not have the cognitive abilities to report their dietary intake [27]. Therefore, involving proxy respondents like parents or other caregivers is necessary to obtain information on children’s dietary intake [27]. Overall, 43.2% of all articles included in our review reported the involvement of parents in dietary intake assessment. In the youngest age group (three- to seven-year-old children), the percentage was even higher, with more than 75% of articles stating that parents assisted in dietary intake assessment. However, parents often do not know what their children have consumed when the children are supervised by other persons, for example, by their teachers at school [28]. Another issue in the assessment of dietary intake is the estimation of portion sizes: most children, in addition to lacking the cognitive abilities necessary to accurately report portion sizes, simply do not pay attention to frequencies and portion sizes while they are eating [27]. Among adolescents valid reporting of dietary intakes is affected by unstructured eating patterns, increased out-of-home eating, and lack of motivation [27]. Another possibility of measurement error is the dietary assessment method itself. Although 24-h recalls, FFQs, and dietary records are the common methods to assess dietary intake, their accuracy and appropriateness in the age group of children has been questioned [29]. A current systematic review indicates that the FFQ might be the most appropriate method to assess dietary intake in children aged 11 years and younger [29]. However, the authors concluded that further research on the validity and reliability of dietary assessment methods in children is needed due to a limited generalizability of the results.

Moreover, current dietary assessment methods are prone to social desirability bias [13]. Underreporting of foods considered as unhealthy (e.g., sugar-sweetened beverages) and overreporting of foods perceived as healthy (e.g., fruits and vegetables) might be another reason why we do not find consistent associations between a dietary factor and the obesity-related biomarkers included. Archer et al. [30] examined the issue of dietary reporting error in different waves of the US National Health and Nutrition Examination Survey (NHANES). They found that the self-reported energy intakes were implausible for more than half of all participants [30]. Moreover, a large European multicenter study in children and infants revealed that parental underreporting was strongly affected by parental concerns/perceptions of their child’s weight status [31].

In contrast to the measurement of dietary intake, subjective reporting errors like social desirability bias are not affecting biomarkers [35]. However, biomarker measurements are not free of error. For example, blood sampling techniques, storage procedures, and laboratory assay errors can influence the results [36]. In addition, biomarker levels may vary over time within an individual; a single biomarker measurement, as common in many epidemiological studies [36], may thus not be able to reflect long-term consequences of diet. Another shortcoming of biomarker measurement is determining adequate reference values for the age group of children and adolescents [37]. For example, there are currently four methods used to determine insulin levels: bioassays, high-performance liquid chromatography, stable isotope dilution mass spectrometry assay, and immunoassays [37]. However, separate reference values for the pediatric population have not been defined for any of these methods [37]. As discussed above, growth and pubertal stage may affect biomarkers in this age group, making the establishment of reference intervals for children and adolescents a major challenge [37].

Longitudinal studies in particular should be conducted to obtain information on long-term consequences of dietary intake on obesity-related biomarkers in children and adolescents. In addition, large sample sizes are necessary to ensure adequate adjustment for important confounding factors such as: sex, age, energy intake and anthropometric measurements. In addition, other confounding factors include: pubertal stage, sex hormones, parental overweight, physical activity, and socioeconomic status.

The development of novel dietary assessment methods for children and adolescents or at least a refinement of existing methods is necessary and should take into account age, cognitive abilities, and adequate tools for portion size estimation [27]. In addition, tools to assess dietary intake in adolescents should be less burdensome and more able to motivate young people, as a lack of motivation in reporting dietary intake seems to exist in this age group [27]. As the age group of adolescents is very computer literate, smartphone applications for assessing dietary intake may greatly improve dietary intake reporting within that age group [27]. Furthermore, we also need better reporting on the details of dietary assessment. For example, a statement on the validity of the dietary assessment method used should be given. Moreover, it is important that novel dietary assessment methods reflect the usual diet of an individual and not merely provide a snapshot of what an individual has eaten during a single day or week. Smartphone technologies may be helpful in collecting long-term data on dietary intake without too much effort for the individual.

In addition, novel biomarkers that better reflect dietary intake in the age group of children and adolescents are necessary [35]. These biomarkers should be valid, noninvasive, cost-effective, and able to reflect changes in dietary intake over time [35]. Moreover, they should enable researchers and health professionals to detect early on children that have an increased risk of becoming overweight or obese. Kuhnle suggested to analyze hair specimens as an alternative to the current blood sample analyses, due to its ability to reflect long-term diet [39]. The emerging field of metabolomics may also be helpful in discovering novel biomarkers [35]. Current metabolomic screenings identified urinary markers that are associated with the consumption of a specific food or food groups [40]. For example, researchers found urinary markers reflecting the intake of oily fish, a meat-rich diet, and a vegetable-rich diet [40]. Furthermore, metabolomics may be helpful in validating the findings from observational or epidemiological studies in the future [41].

The major strengths of our literature review include the application of a systematic search strategy and adherence to PRISMA guidelines [14]. Furthermore, we searched for eligible studies in two of the most renowned biomedical literature databases, which also decreased the probability of missing relevant articles. However, as we only included peer-reviewed articles we cannot fully exclude the occurrence of a publication bias. In addition, a language bias may have affected our results as we only took English-language articles into account. Nevertheless, our literature review provides valuable insights into possible reasons for the nonsignificant and inconclusive findings with regard to associations between dietary factors and obesity-related biomarkers in the age group of children and adolescents.