Metabolic status in children and its transitions during childhood and adolescence—the IDEFICS/I.Family study

Abstract

Background

This study aimed to investigate metabolic status in children and its transitions into adolescence.

Methods

The analysis was based on 6768 children who participated in the European IDEFICS/I.Family cohort (T0 2007/2008, T1 2009/2010 and/or T3 2013/2014; mean ages: 6.6, 8.4 and 12.0 years, respectively) and provided at least two measurements of waist circumference, blood pressure, blood glucose and lipids over time. Latent transition analysis was used to identify groups with similar metabolic status and to estimate transition probabilities.

Results

The best-fitting model identified five latent groups: (i) metabolically healthy (61.5%; probability for group membership at T0); (ii) abdominal obesity (15.9%); (iii) hypertension (7.0%); (iv) dyslipidaemia (9.0%); and (v) several metabolic syndrome (MetS) components (6.6%). The probability of metabolically healthy children at T0 remaining healthy at T1 was 86.6%; when transitioning from T1 to T3, it was 90.1%. Metabolically healthy children further had a 6.7% probability of developing abdominal obesity at T1. Children with abdominal obesity at T0 had an 18.5% probability of developing several metabolic syndrome (MetS) components at T1. The subgroup with dyslipidaemia at T0 had the highest chances of becoming metabolically healthy at T1 (32.4%) or at T3 (35.1%). Only a minor proportion of children showing several MetS components at T0 were classified as healthy at follow-up; 99.8% and 88.3% remained in the group with several disorders at T1 and T3, respectively.

Conclusions

Our study identified five distinct metabolic statuses in children and adolescents. Although lipid disturbances seem to be quite reversible, abdominal obesity is likely to be followed by further metabolic disturbances.

Metabolic syndrome, latent transition analysis, IDEFICS, I.Family, waist circumference, dyslipidaemia, hypertension, glucose disturbances

Key Messages

• Latent transition analysis is a powerful tool to identify groups of children with distinct metabolic status and to estimate changes in metabolic status over several years.

• Five distinct metabolic statuses were identified in children and adolescents.

• Lipid disturbances can be quickly reversed during childhood or adolescence, whereas abdominal obesity is likely to be the trigger for further metabolic disturbances.

• Puberty is a time window during which the risk of developing metabolic abnormalities increases.

Background

Chronic diseases such as cardiovascular disease (CVD), cancer and type 2 diabetes are among the top causes of death and represent a major burden for quality of life.¹ Before chronic diseases become manifest, several risk factors, particularly abdominal obesity, hypertension, dyslipidemia and impaired glucose tolerance, typically occur and accumulate; a triad of these is summarized as metabolic syndrome (MetS).^2,3 Accumulation of these risk factors is not only seen in adults but also in young children and adolescents.^4,5 Recently, based on the large European IDEFICS (Identification and Prevention of Dietary- and Lifestyle-Induced Health Effects in Children and Infants) cohort, a new definition for MetS has been suggested based on reference values derived for children.⁶ In this previous investigation of mainly prepubertal children, a prevalence of MetS of 0.4% to 5.5% was observed in the total population, increasing up to 31.5% in obese children, depending on the definition used.⁶ Similar prevalences of MetS and incidence of cardiovascular risk factors have been reported in other paediatric cohorts.^5,7 Previous studies have shown that temporal changes in metabolic risk factors are already occurring in early adulthood, years before the onset of clinical CVD events.^8,9 However in children, little is known about the temporal occurrence of the components of MetS and the chances of their remission over time. This would require large cohorts of children with multiple examinations and blood sample collections, which are scarce due to ethical and cost constraints. Children represent an important target group, as metabolic risk factors are not yet as manifest as they are in adults and can potentially be reversed more easily.

From a methodological perspective, it is challenging to assess the clustering of metabolic risk factors and its progression over time, due to the variety of possible combinations of risk factors (16 combinations of showing/not showing the four MetS components plus respective changes over time). Some risk factor combinations will occur only rarely, leading to estimation problems with respect to these sparse groups. Latent class analysis (LCA) helps to reduce the dimensionality of data in such situations by identifying groups of subjects with a distinct status with respect to the variables considered.¹⁰ Latent transition analysis (LTA) is a longitudinal extension of LCA which enables the estimation of transition probabilities among latent statuses over time.¹⁰ To our knowledge, no study to date has assessed metabolic status in children and its transitions during childhood and adolescence, using this sophisticated statistical method.

Therefore, the present study aims (i) to identify groups of children with distinct metabolic status, and (ii) to estimate the probabilities of changes in metabolic status when transitioning into adolescence. For this purpose, LTA will be applied to the large and well-phenotyped IDEFICS/I.Family cohort, which provides unique longitudinal data in European children and adolescents from multiple examinations and blood sample collections over time.

Methods

Study population and data

The IDEFICS/I.Family cohort is a multicentre population-based study aiming to investigate and prevent the causes of diet- and lifestyle-related diseases in children and adolescents.¹¹ Participants were aged 2 to <10 years at the baseline survey (T0), which was conducted from September 2007 to May 2008 in eight European countries (Belgium, Cyprus, Estonia, Germany, Hungary, Italy, Spain, Sweden). In total, 16 229 children participated and fulfilled the inclusion criteria. The baseline examination included interviews with parents about lifestyle habits and dietary intakes, as well as physical examinations of the children. Details can be obtained from Ahrens et al.^12,13 A follow-up examination (T1) applying the same standardized assessments was conducted in 2009–10, where 13 596 children were enrolled (2555 newcomers; 11 041 (68%) children who had participated in T0). A second follow-up examination (T3) took place in 2013–14, where 7105 of the children (44%) already participating in T0 or T1 were included.¹¹ A detailed description of all study measures used in the present analysis is given in Supplementary material S1, available as Supplementary data at IJE online.

Before children entered the study, parents provided written informed consent. Additionally, children aged 12 years and older gave simplified written consent. Younger children gave verbal assent for examinations and sample collection. Ethics approval was obtained from the institutional review boards of all eight study centres.

Metabolic syndrome components

As levels of many health parameters change during childhood, a new definition of MetS and of disturbances in its single components has been proposed for children by Ahrens et al.,⁶ which was applied in the present analysis. According to previously described methods,^14–17 sex- and age-specific reference values were derived for waist circumference, diastolic and systolic blood pressure (also height-specific), high-density lipoprotein cholesterol (HDL-C), triglycerides and blood glucose in children and adolescents, using the data collected in the IDEFICS/I.Family cohort. In case the measurement method changed over time, separate reference curves were estimated, depending on the assessment method used (applies to blood glucose, HDL-C and triglycerides; see Supplementary material S1, available as Supplementary data at IJE online). Subsequently, children were defined as being above the so-called monitoring or action levels of the different metabolic parameters if the parameters exceeded the 90th or 95th age- and sex-specific reference percentiles (age-, sex- and height-specific in the case of blood pressure), respectively. In the present investigation, waist circumference was considered a marker for abdominal obesity, systolic (SBP) and diastolic blood pressure (DBP) for hypertension (criterion: either SBP or DBP above 90th/95th percentile, respectively), triglycerides and HDL-C for dyslipidaemia (criterion: either triglycerides above 90th/95th percentile or HDL-C below 10th/5th percentile) and fasting blood glucose for impaired glucose tolerance.

Pubertal status

At T3, pubertal status (yes vs no; yes if menarche had already occurred in girls or if voice alterations had already started/were completed in boys) was self-reported by children aged 8 years and older, based on questions adapted from Carskadon and Acebo.¹⁸

Analysis dataset

Our analysis dataset included all children in the age range from ≥4 to ≤15 years across all waves, who provided at least two repeated measurements of all MetS components (waist circumference, blood pressure, blood lipids and blood glucose). Laboratory measurements obtained based on non-fasting blood samples were not considered (N = 1897 measurements), nor were those from children taking medications that may influence our parameters of interest. For the latter purpose, children being treated for type 1 or type 2 diabetes (ATC codes: A10A, A10B, A10X), elevated blood lipids (C10), hypertension (C02, C03, C07, C08, C09) or obesity (A08) were identified based on ATC codes of reported medications, and excluded (N = 54). This led to a final study sample of 6768 children.

Statistical methods

Applying the above definition, variables were derived indicating children ‘with normal levels’, ‘above monitoring levels (P90)’ and ‘above action levels (P95)’ with respect to the four metabolic markers (waist circumference, blood pressure, lipid levels blood glucose). Based on these variables, LTA¹⁰ was conducted to identify groups of children with distinct metabolic statuses (latent groups) and to estimate: (i) probabilities (prevalence) for latent statuses at T0, T1 and T3; (ii) probabilities for transitions between latent statuses from T0 to T1 and T1 to T3; and (iii) item-response probabilities conditional on latent status membership (i.e. probabilities of showing normal levels or levels above the monitoring or action levels for the metabolic markers in the different latent statuses). Further details on the statistical analyses are given in Supplementary material S2, available as Supplementary data at IJE online. Models with from three to seven latent statuses were estimated, with the five-status model showing the best fit [evaluated based on the Bayesian Information Criterion (BIC)].

LTA was conducted for the total study sample, as well as being stratified by sex and age (2 to <6 vs 6 to <10 years at T0) and separately for children who had entered puberty at the time of the T3 examination. All analyses were performed using SAS^® statistical software version 9.3 (SAS Institute, Inc., Cary, NC, USA). Proc LTA was used to conduct the LTA.

Results

A description of the study population and study measures is provided in Tables 1 and 2. Mean ages of children at T0, T1 and T3 were 6.6 years, 8.4 years and 12.0 years, respectively. Mean values of all MetS components increased as children got older (i.e. were highest at T3).

Table 1.

Means and standard deviations (SD) of age and cardio-metabolic parameters for boys and girls and for the three examination waves T0, T1 and T3

		T0			T1			T3
		N	Mean	SD	N	Mean	SD	N	Mean	SD
Age (years)	Boys	3072	6.6	1.4	3227	8.3	1.6	2003	12.0	1.7
	Girls	3029	6.7	1.4	3222	8.4	1.6	1938	12.0	1.7
Waist circumference (cm)	Boys	3047	56.3	7.3	3216	60.5	9.0	1967	68.8	11.1
	Girls	3004	55.7	7.2	3212	59.6	8.7	1906	67.1	9.8
Systolic blood pressure (mmHg)	Boys	2990	101.9	9.3	3194	104.7	9.2	1954	107.9	9.9
	Girls	2949	101.4	9.3	3190	103.9	9.2	1886	106.7	9.4
Diastolic blood pressure (mmHg)	Boys	2990	63.0	6.6	3194	63.9	6.6	1954	63.7	6.5
	Girls	2949	63.9	6.6	3190	64.5	6.3	1886	64.8	6.5
Triglycerides (mg/dl)	Boys	2899	58.0	28.4	3102	58.5	27.8	1730	63.5	34.0
	Girls	2875	60.9	31.5	3111	62.2	27.5	1669	68.2	31.5
HDL-C (mg/dl)	Boys	2900	53.9	13.9	3127	54.6	13.6	1730	59.2	14.6
	Girls	2875	52.8	13.8	3135	52.3	13.5	1669	58.8	13.1
Glucose (mg/dl)	Boys	2900	86.4	9.5	3134	88.7	9.2	1711	94.8	7.1
	Girls	2875	83.9	9.1	3137	86.8	9.1	1658	93.1	6.7

This table is based on a total of 6768 children providing each at least two repeated measurements of the different risk markers. The statistical model did not require children to have complete data in all three survey waves, which is the reason for the varying numbers of observations with regard to the different markers.

SD, standard deviation.

Table 2.

Children with normal levels, levels above monitoring but below action level (>P90 and ≤P95) or above action level (>P95) at T0, T1 and T3 with respect to the four cardio-metabolic markers (total study group and stratified by sex; number of children and percentages)

	All						Males						Females
	Normal level		>P90 and ≤P95		>P95		Normal level		>P90 and ≤P95		>P95		Normal level		>P90 and ≤P95		>P95
	N	%	N	%	N	%	N	%	N	%	N	%	N	%	N	%	N	%
Waist T0	4475	74.0	339	5.6	1237	20.4	2248	73.8	166	5.5	634	20.8	2227	74.2	173	5.8	603	20.1
Waist T1	4443	69.1	424	6.6	1561	24.3	2224	69.1	180	5.6	813	25.3	2219	69.1	244	7.6	748	23.3
Waist T3	2631	67.9	247	6.4	995	25.7	1309	66.6	115	5.9	543	27.6	1322	69.4	132	6.9	452	23.7
BP T0	4795	80.7	530	8.9	614	10.3	2397	80.1	276	9.2	318	10.6	2398	81.3	254	8.6	296	10.0
BP T1	5141	80.5	573	9.0	670	10.5	2527	79.1	306	9.6	362	11.3	2614	82.0	267	8.4	308	9.7
BP T3	3219	83.8	320	8.3	301	7.8	1631	83.5	161	8.2	162	8.3	1588	84.2	159	8.4	139	7.4
Lipids T0	4792	83.0	473	8.2	508	8.8	2389	82.4	249	8.6	262	9.0	2403	83.6	224	7.8	246	8.6
Lipids T1	4990	80.6	622	10.0	583	9.4	2501	80.8	315	10.2	278	9.0	2489	80.3	307	9.9	305	9.8
Lipids T3	2791	82.1	297	8.7	311	9.2	1399	80.9	161	9.3	170	9.8	1392	83.4	136	8.2	141	8.5
Glucose T0	4844	83.9	449	7.8	482	8.4	2436	84.0	227	7.8	238	8.2	2408	83.8	222	7.7	244	8.5
Glucose T1	4876	77.9	592	9.5	790	12.6	2490	79.6	274	8.8	366	11.7	2386	76.3	318	10.2	424	13.6
Glucose T3	2612	77.5	341	10.1	416	12.4	1313	76.7	186	10.9	212	12.4	1299	78.4	155	9.4	204	12.3

This table is based on a total of 6768 children providing each at least two repeated measurements of the different risk markers. The statistical model did not require children to have complete data in all three survey waves, which is the reason for the varying numbers of observations with regard to the different markers and time points.

BP: Blood pressure; P90, P95, age- and sex-specific percentiles; for blood pressure also height-specific.

At T0, 26.0% of the children fell above the monitoring or action level for abdominal obesity, with the percentage rising to 30.9% and 32.1% at T1 and T3, respectively (see Table 2). Prevalence of the other components of MetS falling above the monitoring or action levels at T0 were 19.2% for blood pressure, 17.0% for blood lipids and 16.2% for blood glucose. Waist circumference was not only the most common risk factor at all measurement points but also occurred most often in combination with the other risk factors (see Supplementary material S3, available as Supplementary data at IJE online, showing prevalence of all risk factor combinations over time).

Results of the LTA

The identified latent groups are characterized as follows (see Table 3). Children in group 1 showed a high probability of being within the normal range of all metabolic markers (all above 87.2%; labelled as ‘metabolically healthy’). In group 2, labelled as ‘abdominal obesity’, the probability of having normal levels for waist circumference was only 5.8% but was high for the other metabolic markers. Group 3 was characterized by a low probability of having normal blood pressure (17.8%; labelled as ‘hypertension’), whereas in group 4, the probability of having normal lipid levels was low (24.6%; labelled as ‘dyslipidaemia’). Finally, in group 5, the probability of having normal waist circumference was almost zero (0.7%) and probabilities of having normal levels for the other metabolic markers was also rather low (max. 55.2%; labelled as ‘several MetS components’). No group of children was identified as suffering mainly from glucose disturbances.

Table 3.

Item-response probabilities in the identified latent groups, i.e. the numbers provide the probabilities of children having normal levels, being above the monitoring (P90) or being above the action level (P95) of the four metabolic markers, respectively, in the five latent groups reflecting children with distinct metabolic status. Item-response probabilities were constrained to be equal at all three time points

	Status 1: Metabolically healthy		Status 2: Abdominal obesity		Status 3: Hypertension		Status 4: Dyslipidemia		Status 5: Several MetS components
	Prob	95% CI	Prob	95% CI	Prob	95% CI	Prob	95% CI	Prob	95% CI
Normal level: BP	91.5	(89.2; 92.0)	82.3	(75.4; 89.4)	17.8	(1.5; 34.2)	85.4	(80.3; 89.2)	55.2	(46.4; 61.2)
Normal level: waist	96.4	(94.9; 97.9)	5.8	(2.1; 12.3)	87.4	(77.2; 92.4)	87.0	(77.9; 91.0)	0.7	(0.0; 2.9)
Normal level: lipids	93.6	(91.5; 96.7)	88.0	(84.1; 92.3)	90.6	(86.2; 94.1)	24.6	(0.6; 48.5)	39.6	(19.8; 51.7)
Normal level: glucose	87.2	(86.2; 88.2)	74.7	(70.6; 78.8)	77.8	(72.8; 82.1)	77.7	(72.6; 81.7)	48.3	(41.6; 54.6)
BP >P90 and ≤P95	5.9	(5.1; 7.0)	8.5	(5.9; 10.8)	29.1	(23.3; 35.1)	6.2	(4.1; 8.7)	16.3	(13.5; 19.4)
Waist >P90 and ≤P95	2.9	(2.1; 3.7)	19.3	(15.6; 23.9)	6.5	(3.7; 9.8)	7.1	(4.4; 11.1)	2.0	(0.2; 3.9)
Lipids >P90 and ≤P95	4.1	(2.7; 5.2)	7.8	(5.3; 10.2)	5.8	(3.3; 8.7)	33.3	(22.5; 43.6)	23.7	(18.8; 30.5)
Glucose >P90 and ≤P95	6.7	(6.1; 7.4)	13.0	(11.0; 14.9)	11.2	(8.4; 14.2)	9.5	(7.1; 12.5)	13.9	(11.4; 16.5)
BP >P95	2.6	(1.7; 3.9)	9.3	(4.1; 14.3)	53.1	(40.5; 66.4)	8.4	(5.6; 12.5)	28.5	(23.6; 35.4)
Waist >P95	0.7	(0.0; 1.5)	74.9	(65.8; 81.4)	6.1	(2.9; 17.3)	6.0	(3.0; 12.8)	97.3	(94.5; 99.4)
Lipids >P95	2.2	(0.4; 3.5)	4.2	(2.0; 6.4)	3.6	(1.6; 6.2)	42.1	(30.0; 57.0)	36.7	(28.3; 51.3)
Glucose >P95	6.1	(5.4; 6.8)	12.3	(9.1; 15.5)	11.0	(8.2; 14.7)	12.8	(9.9; 16.5)	37.8	(31.4; 44.4)

BP, blood pressure; P90, P95, age- and sex-specific percentiles (for blood pressure also height-specific); Prob, probability; 95% CI, 95% bias-corrected bootstrap confidence interval estimated using 5000 replicates (sample size: N = 6768).

Probabilities of being assigned to the different latent groups at T0, T1 and T3 are shown in Table 4. Children had the largest probability of being classified as metabolically healthy (61.5% at T0, 56.5% at T1 and 59.8% at T3), whereas the probabilities were lowest for having dyslipidaemia or hypertension (below 10%). At T0, T1 and T3, the probabilities for children having abdominal obesity were 15.9%, 17.2% and 18.0%, respectively (i.e. increasing over time). The probabilities of showing several MetS components were also markedly higher at T1 (10.5%) and T3 (12.1%) compared with T0 (6.6%). Supplementary material S4, available as Supplementary data at IJE online, shows age, sex and body mass index (BMI) distributions with regard to the different metabolic statuses. Both mean age and BMI (z-score) were highest in the group showing several MetS components, followed by the abdominal obesity group.

Table 4.

Prevalence of latent statuses at T0 (mean age: 6.6 years), T1 (mean age: 8.4 years) and T3 (mean age: 12.0 years) estimated based on latent transition analysis (probabilities for group memberships at T0, T1 and T3, and 95% confidence intervals)

	Status 1: Metabolically healthy			Status 2: Abdominal obesity		Status 3: Hypertension		Status 4: Dyslipidaemia		Status 5: Several MetS components
	Prob	95% CI		Prob	95% CI	Prob	95% CI	Prob	95% CI	Prob	95% CI
T0	61.5	(60.5; 62.4)		15.9	(15.2; 16.7)	7.0	(6.5; 7.5)	9.0	(8.6; 9.5)	6.6	(6.1; 7.0)
T1	56.5	(55.5; 57.5)		17.2	(16.5; 17.9)	7.3	(6.9; 7.8)	8.4	(8.0; 8.9)	10.5	(9.9; 11.1)
T3	59.8	(58.8; 60.7)		18.0	(17.4; 18.7)	3.5	(3.2; 3.8)	6.6	(6.2; 6.9)	12.1	(11.5; 12.7)

Prob, probability for group membership; 95% CI, 95% confidence interval calculated based on sample post probabilities.

Transition probabilities for changes in metabolic status from T0 to T1 and from T1 to T3 are displayed in Table 5, Figures 1 and 2 and Supplementary material S5 to S7, available as Supplementary data at IJE online. The probability of metabolically healthy children at T0 remaining healthy at T1 was 86.6%; when transitioning from T1 to T3, it was 90.1% (see Figure 1). Metabolically healthy children at T0 further had a 6.7% probability of switching to the abdominal obesity group at T1, and 7.2% at T3. Children with abdominal obesity at T0 showed a probability of 18.5% of developing several MetS components at T1 (see Figure 2), as opposed to only 0.7% for those who were metabolically healthy at T0. Of all children with any metabolic disturbances, the subgroup of children with dyslipidaemia at T0 had the highest probability of becoming metabolically healthy (32.4% at T1 and 35.1% at T3; see Table 5 and Supplementary material S5, available as Supplementary data at IJE online). Children with hypertension at T0 showed a 92.2% probability of remaining hypertensive at T1; when transitioning from T1 to T3, the probability of hypertensive children becoming metabolically healthy was 40.3%. Finally, children with several MetS components at T0 had a high probability of being in the same group at T1 (99.8%). With regard to the transition from T1 to T3, the probability was still 88.3%, being the most stable pattern over time, followed by the metabolically healthy group.

Figure 1.

Transition probabilities from T0 to T1 and from T1 to T3 of children being in the metabolically healthy group (61.5% at T0).

Figure 2.

Transition probabilities from T0 to T1 and from T1 to T3 of children being in the abdominal obesity group (15.9% at T0).

Table 5.

Transition probabilities (and 95% confidence intervals) from T0 to T1 as well as from T1 to T3, i.e. probabilities to change from a certain status at T0/T1 to another status at T1/T3 or to remain in the same status. Entries in bold font indicate membership in the same latent status at two consecutive time points

Transition probabilities from T0 to T1 (95% CI in brackets)	Status 1: Met. healthy T1	Status 2: Abdominal obesity T1	Status 3: Hypertension T1	Status 4: Dyslipidaemia T1	Status 5: Several MetS comp. T1
Status 1:	86.6	6.7	1.4	4.6	0.7
Met. healthy T0	(82.8; 90.2)	(5.3; 8.2)	(0.0; 3.0)	(2.1; 8.7)	(0.0; 1.6)
Status 2:	2.3	79.3	0.0	0.0	18.5
Abdominal obesity T0	(0.0; 8.0)	(69.8; 86.1)	(0.0; 0.7)	(0.0; 2.8)	(12.0; 26.1)
Status 3:	0.6	0.4	92.2	0.0	6.8
Hypertension T0	(0.0; 17.9)	(0.0; 5.9)	(85.3; 99.4)	(0.0; 9.9)	(2.7; 13.2)
Status 4:	32.4	4.4	0.0	62.1	1.2
Dyslipidaemia T0	(10.4; 50.3)	(0.0; 9.9)	(0.0; 6.7)	(44.6; 87.7)	(0.0; 5.5)
Status 5:	0.0	0.0	0.0	0.2	99.8
Several MetS comp. T0	(0.0; 6.2)	(0.0; 14.7)	(0.0; 5.7)	(0.0; 3.4)	(96.4; 100)
Transition probabilities from T1 to T3	Status 1:	Status 2:	Status 3:	Status 4:	Status 5:
(95% CI)	Met. healthy T3	Abdominal obesity T3	Hypertension T3	Dyslipidaemia T3	Several MetS comp. T3
Status 1:	90.1	7.2	0.0	2.3	0.4
Met. healthy T1	(86.8; 93.7)	(5.4; 9.7)	(0.0; 2.8)	(0.0; 5.9)	(0.0; 1.6)
Status 2:	16.3	73.8	0.0	0.0	9.9
Abdominal obesity T1	(11.6; 22.8)	(62.7; 81.7)	(0.0; 2.0)	(0.0; 0.0)	(3.7; 18.0)
Status 3:	40.3	6.8	47.2	0.0	5.7
Hypertension T1	(27.2; 51.1)	(0.0; 16.5)	(37.3; 61.8)	(0.0; 10.6)	(0.0; 15.1)
Status 4:	35.1	0.0	0.6	58.1	6.2
Dyslipidaemia T1	(18.3; 52.2)	(0.0; 6.2)	(0.0; 8.3)	(42.7; 77.5)	(1.1; 14.0)
Status 5:	0.8	7.3	0.0	3.5	88.3
Several MetS comp. T1	(0.0; 5.0)	(0.0; 17.0)	(0.0; 1.7)	(0.2; 8.9)	(78.5; 97.2)

Comp., components; 95% CI, bias-corrected bootstrap 95% confidence intervals estimated using 5000 replicates (sample size: N = 6768).

We observed only negligible differences between males and females (data not shown). However, results differed markedly by age group (<6 years at baseline vs ≥6 years at baseline; see Supplementary material S8, available as Supplementary data at IJE online). In particular, the percentage of children with several MetS components was much higher in older children compared with younger children at all three time points, but increased from T0 to T3 in both younger and older children. The proportion of children with several MetS components was highest in children who had entered puberty at T3 (see Supplementary material S8, available as Supplementary data at IJE online). The results of several sensitivity analyses are presented in Supplementary material S9, available as Supplementary data at IJE online.

Discussion

In the present study, we identified five distinct metabolic statuses among children of the IDEFICS/I.Family cohort, namely ‘metabolically healthy’, ‘abdominal obesity’, ‘hypertension’, ‘dyslipidaemia’ and ‘several MetS components’. Over time, the prevalences of abdominal obesity and of showing several MetS components particularly increased, and these two statuses were less likely to be reversed to the metabolically healthy status.

For metabolically healthy children at baseline, the highest risk was for developing abdominal obesity, followed by dyslipidaemia, whereas risks for developing hypertension or several MetS components were very small. This may indicate that (abdominal) obesity is indeed the starting point for subsequent metabolic disturbances. Our observation is in line with the Framingham Heart Study, which reported the presence of abdominal obesity to be the main risk factor for development of MetS in adults.¹⁹ In another adult cohort, it was observed that an increase in BMI and decrease in HDL-C preceded the onset of type 2 diabetes.⁸ Recent studies in children and teens also suggest that unfavourable weight development is associated with subsequent adverse cardiovascular profiles.^20–22

Among children with abdominal obesity who changed their metabolic status over time, the largest proportion either developed several MetS components or became metabolically healthy. Lipid disturbances, on the other hand, seem to be quite reversible, as about one-third of the children in that group became metabolically healthy at T1 and T3. Indeed, previous studies have reported that increased triglyceride levels were more prevalent in 0–9- than in 10–16-year-old children,²³ and the median triglyceride concentration decreased gradually in Korean girls from the 11-year-old group to the 19-year-old group.²⁴ Additionally, studies have shown that lifestyle modifications substantially improved blood lipid levels in the short term.^25,26 Thus, blood lipid levels may change more easily compared with other metabolic markers when entering youth.

In general, few children suffered solely from dyslipidaemia or solely from hypertension. This observation underscores the hypothesis that lipid disturbances or hypertension rarely occur in isolation, but are more likely to manifest as a comorbid condition of abdominal obesity. Indeed, exclusive abdominal obesity appears in a substantial proportion of children and may therefore present the starting point for the other metabolic disorders. A distinction between metabolically healthy and metabolically unhealthy obesity has been suggested but is poorly understood in children.²⁷ This concept is controversial, as previous studies have shown that the majority of metabolically healthy obese people progress to an unhealthy status.^28,29 Thus, metabolically healthy obesity may just be considered an intermediate state in the development of MetS.^28,30 Nevertheless, increased abdominal fat tissue was shown to be associated with the metabolically unhealthy obese phenotype.^27,31 Several mechanisms have been discussed, indicating that the characteristics of abdominal adipose tissue can vary and may differently influence metabolic health.³¹ Obesity per se may directly raise blood pressure through different pathways, such as adversely affecting intravascular volume, cardiac systolic and diastolic function and output, and renal-pressure natriuresis and renal medullary compression.³² In addition, obesity may induce dyslipidaemia, particularly through elevated fasting and postprandial triglycerides, partly caused by increased flux of free fatty acids to the liver.³³ Hypertriglyceridaemia then further causes delayed clearance of the triglyceride-rich lipoproteins, which eventually leads to low levels of HDL-C and high levels of pro-atherogenic small dense LDL-particles.³³

No latent status was found that was mainly characterized by glucose disturbances. When estimating the LTA with six groups, the additional group was characterized by children showing a high probability for glucose level being above the monitoring or action level (data not shown). However, the probability for group membership was very small (1.9% at T0) and the model fit was worse compared with the selected five-group model. Our results suggest that glucose disturbances mainly go along with obesity and are rarely present in children not suffering from additional metabolic disturbances. This is in line with previous studies suggesting that insulin resistance and obesity typically co-exist, being integral in the development of multiple metabolic disturbances.^34,35 There is some evidence that insulin resistance may even be causally involved in the development of obesity.³⁶ However, the majority of literature suggests the reverse direction (i.e. obesity causes insulin resistance).^37–42 The excessive adipose tissue may lead to an increased flux of free fatty acids and dysregulated adipokine secretion, including lower secretion of insulin-sensitizing adipokines such as adiponectin and upregulated secretion of proinflammatory adipokines.⁴³ These mechanisms may trigger insulin resistance.

Another observation in the present study was that probabilities for changing group were higher when transitioning from T1 to T3 compared with the transition from T0 to T1 (except for the metabolically healthy group). This may be explained by the longer follow-up time (about 2 years from T0 to T1 and about 4 years from T1 to T3), but also by the fact that the period from T1 to T3 coincided with entering puberty for about one-third of the children (N = 1830). Puberty incorporates various hormonal and body changes, including puberty-related accumulation of fat mass and reduced insulin sensitivity, that will also affect the parameters considered here.^44,45 As indicated by our subgroup analyses in only pubertal children, metabolic disturbances increased when entering puberty, which suggests that puberty is a sensitive time window for the development of MetS. Accordingly, Reinehr et al.⁴⁶ showed that entering puberty doubled the risk of changing from metabolically healthy obesity to metabolically unhealthy obesity in a cohort of obese children. However, they further observed that this risk was reduced again in late puberty.

Finally, we found that hardly any children showing several MetS components became metabolically healthy over time. This observation underlines the need for early interventions, which could be accomplished, for example, by use of the recently suggested monitoring cut-offs for young children to detect MetS at a preclinical stage.⁶ A main prevention priority in children should be to reduce obesity, as the starting point for further metabolic disturbances.

The present study is not without limitations. Gustafson et al.⁴⁷ showed that both the short-term and the long-term diagnosis of MetS based on cut-offs is quite unstable in children due to influences of factors like time of day, concurrent (unknown) illness or previous energy/macronutrient intake. We tried to mitigate this common problem by application of very strict and highly standardized procedures, as well as by collection of fasting blood samples. For instance, for blood pressure, adequate cuff sizes for children were used and up to three repeated measurements were taken after a 5-min rest to ensure high data quality. However, for identification of impaired glucose tolerance, an oral glucose tolerance test would have been the preferred method, which was not feasible in this large cohort. This means that there may be more uncertainty in the classification of markers showing circadian variability. This may be reflected by the wider confidence intervals that we obtained for these markers. As we covered a broad age range of children, the metabolic markers may be further influenced by developmental stage and puberty. We dealt with this problem in our subgroup analyses by age and puberty status, but also by application of recently published age-, sex- (and height-) specific cut-off values. The IDEFICS/I.Family definition of the ‘monitoring’ and ‘action’ levels of the different metabolic markers is the first to be also based on age- and sex-specific reference values for blood markers.⁶ However, as these reference values were derived based on healthy IDEFICS/I.Family children, it is expected that there is proportion of at least 10% and 5% of children falling above the monitoring and action levels, respectively, when applying the definition to a general subset of the IDEFICS/I.Family cohort, as in the present study.

Our study also has several strengths, as it included a large, well-phenotyped sample of European children which was examined based on a highly standardized protocol and quality control procedures. The main strength is the longitudinal nature, including repeated blood collections over a 6-year time span. Such data are particularly rare in children. This enabled us to investigate changes of the metabolic status over time. In addition, these complex survey data were analysed with sophisticated statistical methods that helped to reduce the dimensionality of the data and to assess clustering and progression of metabolic risk factors over time. Assessment of determinants of metabolic health trajectories (such as lifestyle, socioeconomic or genetic factors) was beyond the scope of the present study, but will be a promising field for future research.

Conclusions

Abdominal obesity was found to be persistent and may precede future metabolic disorders. In contrast, disturbances of lipid levels or hypertension as single metabolic risk factors seem to return to normality more easily. Thus, weight management and reducing obesity to prevent further metabolic disturbances should be a prevention priority in children.

Funding

This work was supported by the European Commission within the Sixth RTD Framework Programme [Contract No. 016181 (FOOD)] for the IDEFICS and within the Seventh RTD Framework Programme [Contract No. 266044] for the I.Family Study.

Acknowledgement

This work was done as part of the IDEFICS [http://www.idefics.eu] and the I.Family Study [http://www.ifamilystudy.eu/]. We are grateful for the support of school boards, head teachers and communities. The authors wish to thank the IDEFICS children and their parents for participating in this extensive examination.

Author Contributions

Each author has seen and approved the contents of the submitted manuscript. All authors contributed to concept and design, acquisition of data and analysis or interpretation of data. Final approval of the version published was given by all the authors. All the authors revised the article critically for important intellectual content.

Conflict of interest: None declared.