Introduction

Emerging evidence suggests that cardiometabolic risk is initiated in early life, and it is tracked from childhood into adulthood and may predict future diabetes and cardiovascular disease [1, 2]. Measures of cardiometabolic risk among youth include the obesity, dyslipidaemia, elevated glucose, and blood pressure and cluster together in young children [3].

Insulin resistance (IR) has been proposed as the primary mediator of metabolic syndrome (MetS) [4], so that identification of youth with IR has been proposed as a strategy for identifying high-risk subjects for targeting MetS interventions.

Different methods have been proposed to measure insulin sensitivity. Among other methods, the euglycemic-hyperinsulinemic clamp is considered to be the gold standard technique to estimate insulin sensitivity [2, 5]. Although clamp technology has been applied to the study of insulin sensitivity and insulin secretion during childhood, it is too invasive for general epidemiologic studies. In the quest for a non-invasive measurement technique for insulin sensitivity, several fasting or “homeostatic” models have been proposed. The homeostasis model for the assessment of IR (HOMA-IR) is frequently used in clinical research [6], because the HOMA-IR model presents a good correlation with the euglycemic-hyperinsulinemic clamp [7]. HOMA-IR is a set of simple, mathematically derived nonlinear equations. It is an equation which related glucose and insulin dynamics that predicts fasting steady-state glucose and insulin concentrations for a wide range of possible combinations of IR and β cell function [8]. The HOMA-IR has high sensitivity and specificity for measuring IR [9]. There are other methods as the quantitative insulin-sensitivity check index (QUICKI), which is similar to the simple equation form of the HOMA-IR model in many aspects, except that a log transform of insulin and glucose [10], and fasted glucose/insulin ratio (FGIR) [11] that have been described frequently as well.

Despite HOMA-IR was found to be much more reliable than FGIR and QUICKI to determinate IR in obese children [11], there is no consensus to define the optimal cut-off point of HOMA-IR in children and adolescents associated with MetS. Therefore, the aim of this study was to review and meta-analyzed the different HOMA-IR cut-off points used to determine IR in children and adolescents associated with MetS.

Materials and methods

Protocol and registration

We followed the checklist of the Meta-analysis of Observational Studies in Epidemiology (MOOSE) [12]. The review was registered in PROSPERO (registration number: CRD42018090076).

Literature search and inclusion criteria

A literature search was conducted in MEDLINE (via PubMed), EMBASE, Web of Science, Proquest, and Scopus databases from their inception to June 2018. The search terms used were “children or childhood or pediatrics” and “HOMA or QUICKI or FGIR” and “AUC or ROC or prevalence or sensitivity or specificity or cut-point or cut-off” and “insulin resistance or insulin sensitivity or metabolic syndrome.” Reference lists of related studies and previous systematic reviews were also screened for eligible studies. Two authors (PAR and AGH) independently screened the search results, determined the eligibility of the studies, and extracted the data. Any discrepancies were resolved via consensus-based discussions with a third reviewer.

We aimed to identify original articles analyzing the optimal HOMA-IR cut-off point for defining IR associated with MetS in children and adolescents. The eligibility of the studies was formulated according to the PICOS criteria (i.e., participants, interventions, comparisons, outcomes, and study design). Eligible studies were cross-sectional studies with participants younger than 18 years and reported the prevalence of MetS. For a study to be included, the cut-off point of IR should have been calculated by using Youden’s index (sensitivity-(1-specificity)) associated with MetS. We included only studies published in English.

Data extraction and quality assessment

The following data were collected from each study: (1) author identification, (2) year of publication, (3) country of the study, (4) year of data collection, (5) age of the participants, (6) number of participants, (7) prevalence of MetS, and (8) parameters summarizing the accuracy of the test (cut-off, sensitivity, specificity, and area under the curve (AUC)).

We also used the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS) tool to evaluate four domains of each study: patient selection, index test, reference standard, flow, and timing. Each domain was evaluated in terms of the risk of bias and in terms of concerns regarding the applicability of the results. The QUADAS-2 was designed to assess the validity of the studies. It assesses the quality of included studies in terms of risk of bias and concerns regarding applicability over four domains, as discussed below [12].

The patient selection could be biased by the method of selection of patients or included patient. The concerns regarding applicability evaluate if the review question matches the included patients. The index test may be biased by the interpretation of the test, and the concerns regarding applicability are based on the interpretation that differs from the review question analyzed. The reference standard could be biased by the election of the reference standard test and its application on the chosen sample. The concerns regarding applicability evaluate if the reference standard matches the review question. The flow and timing describe if the reference standard and the index test were applied at the same time and to all patients [12].

Data extraction and quality assessment were independently performed by PAR and AGH, and inconsistencies were managed by consensus.

Statistical analysis and data synthesis

The diagnostic test requires a cut-off point above the reference from which the test is regarded as positive, in other words, allows it to classify individuals as healthy or sick. The value of threshold depends on two statistical parameters, sensitivity indicates the proportion of sick patients that have a positive result and specificity indicates the proportion of healthy patients that have a negative result, so that the sensitivity and specificity vary depending on the situation which the test is used [13]. This variability is the main source of heterogeneity in a meta-analysis. The heterogeneity of results across studies was evaluated using the I2 statistical parameter, I2 was calculated as 100%*(Q − df)/Q, where Q is the Cochrane heterogeneity statistic [14]. I2 values of < 25%, 25–50%, and > 50% usually corresponds to small, medium, and large heterogeneity, respectively [15].

The dOR is a measure of the accuracy of the test data that combines sensitivity and specificity into a single value. The dOR values range from 0 to infinity, with higher values indicating a better discriminatory test performance. A dOR of 1.0 indicates that a test does not discriminate between patients with the disorder and those without it [16].

Forest plots were used to display the sensitivity, specificity, and dOR for HOMA-IR in the reviewed studies. The sensitivity, specificity, and dOR as well as their corresponding 95% confidence intervals (CI) were calculated for HOMA-IR in each study.

Hierarchical summary receiver operating characteristic curves (HSROC) is a quantitative model that allows to relate sensitivity, specificity, heterogeneity, and threshold effects [13]. The HSROC have been proposed to estimate the performance of diagnostic tests on data from a meta-analysis. HSROC were used to summarize the overall test performance.

Each study was weighted in the analyses according to the inverse variance method. Statistical analyses were performed using StataSE software, version 13 (StataCorp).

Results

Literature search and baseline characteristics

The details of our study selection method are shown in Fig. 1. A total of 5915 articles were retrieved from literature search. After removing 1733 duplicated articles, the titles and abstracts of 4182 studies were screened. We excluded 4148 studies due to at least one exclusion criteria was fulfilled, leaving 34 studies that were reviewed in full. Finally, six studies were included in this systematic review and meta-analysis. All studies had cross-sectional designs.

Fig. 1
figure 1

Flow chart for the identification of the meta-analyses included in the study

Full size image

The six studies comprising this systematic review included 8732 children and adolescents. The age of the participants ranged from 6 to 18 years. In each selected study, the prevalence of MetS was reported according to different criteria Adult Treatment Panel III (ATPIII) [17], International Diabetes Federation (IDF) [18], Cook et al. [19], Cruz et al. [20], Ford [21], de Ferranti et al. [22], and Salas-Fernandez et al. [23]). Among six studies, we found one report, which used three different definitions [24] and another which used six different definitions [18]. All the studies provided information on the MetS prevalence, except for one [5], which was calculated. The prevalence of MetS varied from 1.6 to 29.7% and the HOMA cut-off points ranged from 2.30 to 3.54 (Table 1). Each definition was analyzed as an independent population. The studies were conducted in China [5], Korea [17], Mexico [25], Chile [26], and India [27, 28].

Table 1 Characteristics of studies included
Full size table

Study quality

All included studies were assessed as minimal risk of bias in terms of the four domains of the QUADAS-2 tool: patient selection, index test, reference standard, flow, and timing (Table 2).

Table 2 QUADAS-2 risk of bias assessment
Full size table

Meta-analysis

Figure 2 shows the dOR forest plot of HOMA-IR. There was heterogeneity across the studies in the dOR of HOMA-IR (I2 = 77%) based on Adult Treatment Panel III (ATP III) definition of MetS and the pooled dOR was 5.45 (95% CIs, 3.59 to 8.28; p = 0.013), whereas the heterogeneity across the studies based on IDF definitions was lower (I2 = 0.0%) and the pooled dOR was 9.58 (95% CIs, 7.10 to 12.91; p = 0.487). The pooled accuracy parameters from the studies that evaluated the diagnostic performance of HOMA-IR ranged from 4.39 to 37.67.

Fig. 2
figure 2

Forest plot of diagnostic odds ratio (dOR) of each index test reviewed studies. CI, confidence interval; a, b, and c indicate different subgroup of participants in that study, as defined by setting

Full size image

Figure 3 shows the sensitivity forest plot of HOMA-IR. The sensitivity of the optimal cut-off proposed ranged to 0.59 to 0.89. The heterogeneity across the studies based on ATP III and IDF definition was small, and the pooled sensitivity was 0.77 (95% CIs, 0.71 to 0.84; I2 = 0; p = 0.481) and 0.72 (95% CIs, 0.62 to 0.83; I2 = 0; p = 0.533), respectively.

Fig. 3
figure 3

Forest plot of sensitivity of each index test reviewed studies. CI, confidence interval; a, b, and c indicate different subgroup of participants in that study, as defined by setting

Full size image

Figure 4 shows the specificity forest plot of HOMA. The specificity of the optimal cut-off points proposed ranged to 0.62 to 0.87. The heterogeneity across the studies based on ATP III and IDF definition was small, and the pooled specificity was 0.65 (95% CIs, 0.63 to 0.68; I2 = 0; p = 0.569) and 0.80 (95% CIs, 0.75 to 0.85; I2 = 53.7; p = 0.090), respectively.

Fig. 4
figure 4

Forest plot of specificity of each index test reviewed studies. CI, confidence interval; a, b, and c indicate different subgroup of participants in that study, as defined by setting

Full size image

The area under the HSROC estimating the discriminating accuracy of HOMA-IR for identifying IR is show in Fig. 5. The 95% confidence region for the point that summarized the overall test performance included studies in which the cut-off points ranged from 2.30 to 3.54.

Fig. 5
figure 5

Hierarchical summary receiver operating characteristic (HSROC) curves summarizing the ability of Homeostasis Modal Assessment-Insulin Resistance (HOMA-IR) to identify insulin resistance

Full size image

Discussion

Our meta-analysis shows that HOMA-IR test as measurement of IR surrogate may be a useful tool for identifying IR associated with MetS among children and adolescents across countries. It presents an enough dOR on diagnostic IR independently of the definition of MetS used.

MetS is a state of chronic inflammation, as a consequence of complex relationship between genetic and environmental factors [29]. It is difficult to establish cut-off points for IR associated with MetS, since the clinical outcomes are observed into adulthood [30]. Some studies demonstrated that HOMA-IR levels increase directly with the numbers of MetS components present in a subject [5]. It has also been proposed that the measure of IR surrogate could be a useful tool to unify the diagnosis criteria of MetS [18]. However, in many studies, cut-off points have been based on the distribution in reference population, instead of using the Youden index.

Our findings show a considerable variation of MetS prevalence, depending on the definition used. We observed that the lowest prevalence was noticed when the IDF definition was used [17,18,19]. The IDF definition requires for diagnosis, the abdominal obesity plus any additional criteria (raised triglycerides levels, reduced high density lipoprotein cholesterol, raised blood pressure or raised fasting plasma glucose). The high prevalence observed for De Ferranti’s definition [22] may be because it uses a lower threshold for waist circumference in comparison with other definitions.

HOMA-IR as a surrogate measure of IR associated with MetS has similar accuracy in terms of the dOR and AUC between different definitions. It also should be noted that when criteria of MetS were adapted to pediatric population, the accuracy diagnostic increased. Ford’s definition [21] reported the best accuracy for HOMA-IR. It could be due to the fact that the cut-off point used to measure fasting glucose is lower in comparison with other definitions. The IDF definition presents better dOR in comparison to ATPIII classification. It could be explained that, for unknown reasons, insulin stimulates hepatic lipogenesis, causing dyslipidemia, which results in ectopic adipose deposition [31] and abdominal obesity is an essential requirement of IDF definition. As is common, all the estimations of the diagnostic accuracy were performed considering the large variability across individual studies, derived from a threshold effect to determine positivity in the tests as the different criteria for defining MetS.

Regarding pooled specificity and sensibility in the meta-analysis, the results show similar values among studies, ranged from 0.59 to 0.89 for sensitivity and from 0.62 to 0.87 for specificity. The number of subjects, the average of age, and the lack of information of maturity sexual among participants could be explained differences among studies that have used the same definition of MetS.

Meta-analysis of diagnostic test synthetizes the performance of a test providing a pooled estimation of diagnostic accuracy parameters and estimates a summary point (a summary sensitivity and specificity estimates) and a HSROC, but does not allow the identification of the optimal cut-off point [32]. However, the cut-off points reported are within the 95% confidence intervals for HOMA-IR, ranged from 2.30 to 3.54; thus, all of them presented a good accuracy diagnosis that could be used in clinical practice.

The present meta-analysis has some limitations: (i) One of the limitations was the insufficient available data prevented us from conducting separate analyses; (ii) we included only population-based studies conducted in apparently healthy children and adolescents aged ≤ 18 years; (iii) the lack of discrimination among children and adolescents is also a limitation, because, during pubertal period, IR increases at the onset of puberty; (iv) the lack of information about nutrition and lifestyle participants was another limitations; (v) due to the lack of available studies, we cannot calculate the AUC and their 95% confidence intervals; (vi) a comparison with gold standard methods was not possible in any study; (vii) the number of studies was too small to allow subgroup analyses by gender and pubertal development; and (viii) the lack of consensus of MetS definition makes difficult the comparison between studies.

Conclusions

Our study determined that the use of HOMA-IR, an indirect marker for IR, seems to be useful for early evaluation among children and adolescents at risk, independently of definition of MetS. Due to the IR plays a main role in the pathogenesis of MetS, the use of these cut-off points early and its proper identification could have benefits of preventive and diagnostic therapeutic intervention.