IntroductIon
The major burden of disease has shifted from communicable to noncommunicable diseases in high-income countries dur-ing the past century (1). Overweight is a major risk factor of major noncommunicable diseases such as cardio vascular dis-ease (2,3). The incrdis-ease of overweight and obesity in children and adolescents is believed to herald a further obesity epidemic in adulthood in coming years (4). The essential assumption underlying these projections is a high tracking of body compo-sition. Tracking is important as allowing an early identification of individuals at high risk for later disease.
Despite being an indirect estimate of body composition, BMI offers a reliable and frequently used measurement in large cohort studies. However, the distribution of BMI is not comparable between children and adults, not even between the various childhood age groups (5), hampering measurement of BMI tracking from childhood to adulthood. Therefore, track-ing estimates such as the correlation coefficient are preferable compared to difference estimates when assessing BMI track-ing. Correlation coefficients are often used (6–8) and offer
a quantitative surrogate of individual courses. In principle, tracking correlation coefficients can be calculated whenever two or more measurements of the same variable in the same individuals are available.
A number of studies on BMI tracking have been published, however, with tracking estimates ranging from no correlation (r = −0.18; ref. 9) to a strong correlation (r = 0.98; ref. 10). The con-sideration of heterogeneous study samples with respect to age at first measurement or observation time until follow-up measure-ments might explain these differences. An overall tracking esti-mate adjusted for the different baseline ages or follow-up times is lacking but is required for the estimation of potential long-term successes of prevention programs. Unlike conventional meta-analysis, a meta-regression takes care of heterogeneity in study characteristics (such as different follow-up times, ages, etc.) and additionally provides information on the impact of such covari-ates on the outcome of interest, i.e. BMI tracking.
The aim of this study was to summarize the findings of pub-lished studies on BMI tracking in a pooled and adjusted tracking estimate and to explore further determinants of BMI tracking.
Factors associated With Tracking of BMI:
a Meta-Regression analysis on BMI Tracking*
Otmar Bayer
1, Helia Krüger
1, Rüdiger von Kries
1and André M. Toschke
2Stable tracking of body composition is a prerequisite for the long-term effect of preventive measures against obesity and its harmful effects. As BMI tracking estimates reported by individual studies vary considerably, we performed a meta-regression analysis to provide a summary estimate and to assess determinants of BMI tracking. Using the Medline and EMBASE databases, a systematic review was conducted to identify publications reporting correlation coefficients as tracking estimates between BMI at baseline and follow-up measurements and the time interval between these measurements. Additional information recorded included age at baseline measurement, gender, and origin of the studied population. Based on the extracted data, a meta-regression analysis was performed using mixed effects models to account for multiple measurements of the same cohorts. Data on 55,072 individuals (797,094 person-years) extracted from 48 publications with follow-up times between 0.5 and 44 years entered the analysis. The overall estimates for the 1-year tracking correlation coefficient were strong (r = 0.78–0.86 depending on age at baseline measurement) and gradually decreasing over time (0.67–0.78 after 10 years, and 0.27–0.47 after 30 years). Study origin classified by continent was another significant predictor of BMI tracking whereas gender was not. In conclusion, this meta-regression analysis showed a high degree of BMI tracking across all age groups investigated and independent of BMI. Successful prevention in weight control is likely to have long-term effects at any age, thereby being beneficial with respect to the associated risks of over- and underweight.
Obesity (2011) 19, 1069–1076. doi:10.1038/oby.2010.250
1Ludwig-Maximilians University of Munich, Institute for Social Paediatrics and Adolescent Medicine, Department of Epidemiology, Munich, Germany; 2Ludwig-Maximilians University of Munich, Department of Medical Informatics, Biometry and Epidemiology (IBE), Munich, Germany.
Correspondence: Otmar Bayer ([email protected]) *This article was originally published as an Intervention and Prevention.
Methods and Procedures
data sources and search strategy
We searched Medline and EMBASE using two queries in each data-base. The first query (S1) considered the keywords “BMI OR “Body Mass Index”[Mesh] AND correlation AND coefficient”, whereas the second (S2) considered “BMI OR “Body Mass Index”[Mesh] AND (tracking OR trajector*)”. Both queries were restricted to articles on humans in English or German language that entered the databases until 31 December 2008. Two review authors assessed the results of the search strategies for eligibility and methodological quality without con-sideration of the study results. Eligibility scanning of the full texts was assisted by using a bash script (see Supplementary Discussion online). Reference lists of articles remaining after full-text review were manually searched for additional publications of interest. Authors of the original publications were not contacted to gain additional information.
study inclusion criteria and variables
Only articles with at least two measurements of BMI in the same indi-viduals at different time points and information on their correlation were considered. Duplicate articles of the same study were removed and only the article with most complete information, largest sample size, or earlier publication date was included (criteria given in order of priority). Additional information on the population investigated such as diagnosis or applied intervention were also extracted. Other popu-lation characteristics were considered if reported frequently enough for meaningful analysis (in at least five publications). Basic informa-tion consisted of cohort size n, (mean) age at baseline, and at follow-up measurement.
Follow-up time was calculated using mean age at baseline and at follow-up. Age at baseline was coded in years and then categorized into four groups to allow for nonlinear effects of age on BMI tracking: <10 years (reference), 10–14 years, >14 to <18 years, and ≥18 years.
statistical analysis
BMI correlation coefficients were transformed using Fisher’s z- transformation (11) before regression analysis to normalize the response variable distribution. Different estimates of correlation coef-ficients of BMI values including Pearson’s product-moment correlation coefficient, Spearman’s rank correlation coefficient, and standardized regression coefficients were considered. In addition, articles reporting the proportion of adjusted and unadjusted explained BMI variance at a later time point explained by an earlier BMI measurement (r2) were
also considered as were correlation coefficients based on BMI z-scores rather than raw BMI values. Type of tracking estimation was recorded for sensitivity analyses.
Most publications reported tracking estimates of different subgroups, e.g., differing by gender or baseline ages. Additionally, some publica-tions reported multiple tracking estimates for the same subgroup due to multiple follow-up measurements and consideration of different time intervals.
Therefore, random effects models were used that are further detailed in the Supplementary Discussion online. Observations were weighted by cohort size. Calculations were performed in R, version 2.8.1 (12,13) on Linux.
results
Both electronic search strategies yielded triple digit number of results, S1 yielded 815 (380 and 527 in Medline and EMBASE, respectively), S2 yielded 535 (271 and 437) references. Figure 1
illustrates the selection process for both queries S1 and S2 with the results pooled from both databases. Manual searches in the reference lists of the articles revealed five further references. A total of 10 articles were excluded for reasons given in Table 1, mostly because of overlap in study population. Overall 48
articles (Table 2) were left for further analysis providing data on 55,072 individuals, and 797,094 person-years studied.
Follow-up time ranged from 0.5 to 44 years with a mean and median of 14.6 and 15.0 years, respectively. The identi-fied studies covered populations of all continents (Africa n = 1, Asia n = 4, Australia n = 3, Europe n = 24, North America n = 13, Central and South America n = 2, unclassified n = 1), and two reported correlation coefficients stratified by ethnicity. As
S1 S2 815 References 535 References 682 Title 91 Abstract 37 Full text 55 Abstract 115 Full text 5 Articles 53 Articles 48 Articles 5 Contained in both S1 and S2 10 Excluded 5 Reference list 53 Articles 312 Title
Figure 1 Flowchart showing the number of articles excluded after title,
abstract, or full-text review.
table 1 articles excluded albeit meeting the inclusion criteria
Angelico et al. (18) Same study population as Angelico et al. (19), intervention trial with less individuals Juonala et al. (20) Same study population as Yang et al. (21),
assignment of r and cohorts unclear Kivimäki et al. (22) Same study population as Yang et al. (21) Lauer and Clarke (23) Same study population as Clarke and
Lauer (24)
Monyeki et al. (25) Same study population as Monyeki et al. (26)
Raitakari et al. (27) Same study population as Yang et al. (21), r not stratified by gender and age
Shephard (28) No primary data
Steinhausen et al. (29) Anorexia nervosa patients with individual treatments
Deshmukh-Taskar et al. (30)
Same study population as Wattigney et al. (31)
table 2 articles included in the meta-analysis
Publication
(first author and year) Individuals Correlation coefficient Follow-up time in years Population origin (continent) First and last baseline
Andersen and Haraldsdóttir (32)P,G 200 0.77–0.80 8.00 Europe 1983
Angelico et al. (19)G 214 0.52–0.64 11.00 Europe 1983
Burke et al. (33)P,G 4,207 0.86–0.90 3.60–4.80 North America 1985
Burke et al. (34)P 600 0.58–0.82 7.00–16.00 Australia 1985–1994
Casey et al. (35)P,G 114 −0.03–0.87 10.00–44.00 North America 1940–1974
Clarke and Lauer (24)P,G 5,778 0.58–0.91 5.50–19.50 North America 1971
Clausen et al. (36)G 227 0.52–0.64 14.20 Europe 1979
Eisenmann et al. (37)P 48 0.64 10.80 North America 1970
Ekblom et al. (38) 296 0.80 6.00 Europe 2001
Fabsitz et al. (39)P,G 486 0.32–0.94 6.00–43.00 North America 1941–1982
Fuentes et al. (40)P 100 0.11–0.69 6.50–14.50 Europe 1981–1989
Gaskin and Walker (41)P,G 304 0.82–0.83 3.91–3.92 Central America
Guo et al. (42)I,P,G 555 0.08–0.76 17.00–34.00 1930–1947
Hulens et al. (43)P,G 161 0.49–0.91 5.00–28.00 Europe 1968–1986
Hulman et al. (44)I,P 136 −0.02–0.62 7.00–27.70 North America 1959–1980
Julia et al. (45)P,G 308 0.65–0.78 5.00 Asia 1999
Kelly et al. (46)i,P 712 0.35–0.87 2.00–10.00 Australia 1972–1981
Kvaavik et al. (47)P 485 0.54 17.70 Europe 1981
Lake et al. (16)P,G 7,339 0.19–0.54 26.00 Europe 1965
Lambrechtsen et al. (48)P,G 900 0.59–0.66 11.00 Europe 1985
Leppik et al. (49)G 167 0.81–0.94 0.90–2.90 Europe
Maffeis et al. (50)P,G 103 0.48–0.52 14.10–16.30 Europe 1980
Magarey et al. (51)P,G 397 0.30–0.92 2.00–18.00 Australia 1977–1991
Marshall et al. (52)G 324 0.83–0.90 0.65–2.70 North America 1990–1991
Matton et al. (53)I,G 118 0.53 23.90 Europe 1979–1980
Mohler et al. (54) 399 0.62–0.78 4.00–9.00 Europe 1976–1981
Monyeki et al. (26)P,G 702 0.82–0.91 1.00 Africa 2001
Mueller et al. (10)i,P,G 678 0.87–0.98 1.00–3.00 North America 1991–1993
Oja and Jürimäe (55)G 252 0.41–0.75 0.50–1.50 Europe
Olvera et al. (56)P 69 0.90–0.95 1.00–3.00 South America
Oren et al. (57)G 750 0.62–0.65 14.90–15.00 Europe 1999–2000
Palti et al. (58)P,G 558 0.50–0.57 8.00 Asia 1975
Pate et al. (59)P,G 145 0.81–0.90 2.00 North America
Psarra et al. (60)G 918 0.83–0.85 1.90 Europe 2000–2001
Sinaiko et al. (61)P 208 0.85 6.00 North America
Steinberger et al. (62)P 31 0.67 8.50 North America 1985
Taeymans et al. (9)P,G 115 −0.18–0.79 17.00–29.00 Europe 1969–1985
Tan et al. (63)G 507 0.77–0.84 3.80–3.86 Asia 1992–1993
Trudeau et al. (64)P,G 191 0.43–0.70 22.00–24.00 North America 1970–1979
Valerio et al. (65) 341 0.86 2.60 Europe 1999
Vogels et al. (66)P 105 0.05–0.76 5.40–12.40 Europe 1990–1997
Wang et al. (67)P,G 975 0.38–0.42 6.00 Asia 1991
Wardle et al. (68)i,P 2,654 0.82–0.95 1.00–4.00 Europe 1999–2002
only two studies were intervention studies, we finally decided to exclude the correlation coefficients reported for the inter-vention groups and to use only those reported for the control groups.
Meta-regression
The multivariable regression model with random intercept and slope by cohort yielded a significant inverse association of the correlation coefficients with follow-up time (P < 0.0001).
Figure 2 illustrates the high tracking of BMI decreasing with follow-up time. Age group at baseline measurement was a further predictor (P < 0.0001). Gender was not significantly associated with correlation coefficients (P = 0.48).
Pooled correlation coefficients for different follow-up times and baseline age groups are shown in Table 3. For example, the 5-year tracking correlation coefficient for a child is 0.74, whereas it is 0.83 for an adult.
additional analyses
In several studies, correlation of BMI was rather a sidetrack than main focus and, therefore, not clearly obvious from the title, resulting in quite a high proportion of abstracts and full texts to be examined. As another consequence, the sample of which the correlation coefficient was reported from was not always described in detail. In order to minimize exclusion of studies for the analysis, we imputed appropriate information whenever possible: if for example, minimum and maximum age at measurement instead of an age mean was reported, the center of this interval was used as estimate for the mean age. For some studies with multiple measurements of the same cohort, the number of individuals was not given for each measurement presented. In these cases, we considered the smallest number of individuals available only for the respec-tive cohort, typically the number of individuals completing the first and last measurement, in order to avoid overweight-ing. In one publication, the r-values had to be estimated from a plot. To estimate, how such imprecision biased our results, or if our counteractions were appropriate, we performed a sen-sitivity analysis: excluding all cohorts with any kind of impre-cise information and refitting the model yielded similar effect sizes (data not shown) and the same covariates being signifi-cant as in the main model (Table 3). Similarly, restriction to the 22 publications that used Pearson’s r (not Spearman’s r or
standardized regression coefficients) did not reveal different results (data not shown).
subgroup analyses
Overall 22 publications additionally reported enough information to calculate BMI z-scores at baseline (min. −1.95, max. 2.65 based on http://www.who.int/childgrowth/ standards/bmi_for_age/en/index.html and http://www.who.int/ growthref/en) allowing to assess the relation between initial BMI and BMI tracking. Cohorts with higher BMI at baseline showed slightly—though not significantly—higher tracking correlation coefficients (β_BMIz = 0.063, corresponding roughly to a 0.018 increase in correlation coefficient per additional kg/m2 for adults, P = 0.20), and there was no indication of a U-shaped relationship (P = 0.50 for second-degree polynomial).
In the eight publications reporting a total of 83 correlation coefficients with respect to parental BMI or weight status, BMI tracking showed no major differences between cohorts with normal-, under-, or overweight parents (Figure 3). If the origi-nal publication did not provide correlation coefficients strati-fied by parental weight status, but reported mean parental BMI within the normal range, parents were classified as normal
1.0 0.8 0.6 0.4 0.2 0.0 0 10 20 30 40 50 −0.2 r Years follow-up
Figure 2 Correlation coefficients (r) vs. follow-up time. Numbers
correspond to publications as listed in references. table 2 continued
Publication
(first author and year) Individuals Correlation coefficient Follow-up time in years Population origin (continent) First and last baseline
Wattigney et al. (31)G 1,583 0.43–0.74 15.00 North America 1973–1984
Weststrate et al. (69)P,G 167 0.10–0.83 5.00 Europe 1978
Wilsgaard et al. (70)G 17,710 0.73–0.84 16.00 Europe 1979
Wright et al. (71)i,P,* 529 0.24–0.39 37.00–41.00 Europe 1956
Yang et al. (21)I,P,G 1,319 0.34–0.41 19.00 Europe 1980
The small letters identify publications that were excluded for sensitivity analysis completely (I) or in part i.e. only some cohorts (i), reported Pearson’s correlation coefficient (P), reported correlation coefficients stratified by gender (G).
weight for the corresponding study. Therefore, it should be noted that some of the data points labeled as normal weight might also include over- or underweight parents.
As most publications reported origin of the studied popu-lation, we performed a further analysis to address potential differences in tracking by region. Continent was an almost significant predictor of tracking (P(likelihood-ratio χ2 test) = 0.055, see Supplementary Discussion online for model details), revealing lowest correlation coefficients for Asia, Australia, and Europe, intermediate for Africa and North America, and highest for South America. For Europe (as an example of one of the continents with the lowest estimates), we computed the 1-, 10-, and 20-year correlation coefficients for an individual in puberty as 0.80, 0.69, and 0.52, respectively, to enable compari-son with the results from the main model in Table 3.
Although there was quite a strong correlation (r = –0.70) between follow-up time and year of baseline measurement
(the older study the study the longer the follow-up possible) of the cohort, we refit the model using the explanatory variables follow-up time, age group, and additionally baseline year to assess secular trends in BMI tracking. The result indicated no substantial strengthening or weakening over the years covered by the included studies—the parameter estimate was 0.0028 per year, P = 0.20 based on 370 correlation coefficients from 40 publications. The fixed effect of follow-up time was attenuated (parameter estimate 0.022 vs. 0.027 in Table 3) but remained significant with P < 0.0001.
dIscussIon
Based on 48 cohort studies, a high degree of body composition tracking as measured by BMI was observed. Although gradu-ally decreasing over time, BMI tracking after 10 years was still high with r-values around 0.7. Our result for 1-year tracking in adults (Table 3) is in good agreement with the standardized regression coefficients (0.85–0.89) reported in a large Austrian study, which could not be included due to missing information on follow-up time (14). In comparison to that study, our meta-analysis additionally considers the entire range of age groups and allows to predict tracking for arbitrary follow-up times up to three decades.
Independent of follow-up time, tracking was significantly associated with age group at first measurement and highest for adults. This seems plausible when thinking e. g. of a 5- and a 50-year-old individual both followed for 5 years. While the 50-year-old individual’s lifestyle might be more settled, for the 5-year-old individual a 5-year follow-up period makes up 50% of the lifetime so far experienced. Before adulthood, the highest correlation coefficients were found for puber-tal children. This is in agreement with results from the 1958 British National Child Development Cohort (15), showing that pubertal body composition changes were a superior predictor for adult obesity. Although statistically significant differences between age groups were detected, the degree of tracking was strong in all age groups.
Interestingly, based on BMI z-scores in subgroup analysis, BMI at first measurement had no influence on BMI track-ing. Therefore, the observed tracking estimates appeared to apply equally to obese, under-, or normal-weight individu-als. Furthermore, parental BMI was not found to be associ-ated with BMI tracking. However, the subgroup analysis by parental weight status might lack of sample power, due to a limited number of studies reporting appropriate information. Additionally, parental overweight was defined as one or both parents being overweight. In contrast, Lake et al. (16) com-pared different subgroups. A significantly higher tracking esti-mate between measurements at the age of 7 and 33 years was observed in offspring of two obese parents compared to those of normal-weight parents. These differences in definition as well as waning parental influence in later life may explain the conflicting results.
Consideration of studies from all over the world in the meta-regression analysis allowed to compare BMI track-ing by continent. The observed differences suggest that BMI table 3 correlation between BMI measured in childhood,
puberty, adolescence, or adulthood and BMI 1, 3, 5, 10, 20, and 30 years later
Follow-up time in years
1 3 5 10 20 30 Age at baseline measurement child (<10 years) 0.78 0.76 0.74 0.67 0.50 0.27 Pubertal (10–14 years) 0.84 0.82 0.80 0.75 0.60 0.40 Adolescent (>14 years to <18 years) 0.83 0.81 0.79 0.73 0.58 0.38 Adult (≥18 years) 0.86 0.84 0.83 0.78 0.65 0.47
The values in this table can be calculated as follows: z = 1.080 − (follow-up time in years) × 0.027 + b_agegrp. The constant b_agegrp is 0 for prepubertal children, 0.152 for pubertal children, 0.122 for adolescents, and 0.231 for adults. Finally, the result is transformed back into a correlation coefficient: r = (e2z − 1)/ (e2z + 1). 1.0 0.8 0.6 0.4 0.2 0.0 0 5 10 15 20 25 30 r Years follow-up Overweight Normal weight Underweight parents
Figure 3 Correlation coefficients vs. follow-up time for cohorts with
tracking is influenced by factors such as ethnicity and lifestyle. Unfortunately, tracking estimates for specific ethnic groups were provided by two studies only, disabling a more detailed analysis. Because only few studies outside North America and Europe were available for analysis, the validity of our findings outside these regions may be limited.
The observed degree of BMI tracking was considerably higher than the tracking of blood pressure, which is another important vascular risk factor (7,8). This might be due to the higher variation of blood pressure values that are associated to work charge or even show remarkable circadian variations.
The International Obesity TaskForce (www.iotf.org/ childhoodobesity.asp) cut points for obesity in children by Cole et al. (5) are based on an attempt to translate adult BMI cut points for overweight and obesity into childhood. The track-ing correlation coefficients allow to quantify the imprecision of predicting later BMI values based on earlier BMI values.
The high degree of BMI tracking observed has implications on trial design: to assess the effect of an intervention pro-gram a trial can compare pre- and postchanges between two study groups. Alternatively, its effect can simply be assessed by measuring the difference between both study groups at the end of the study, however, not accounting for potential base-line values. The first design has a variance of 2σ2·(1 − r) for the pre- and postdifference. In the second design, the vari-ance of the study end point is simply σ². As the varivari-ance in the first design is proportional to 1 − the tracking correlation coefficient (r), trials following the first design need to include less patients compared to trials following the second design, if the correlation (r) between pre- and postmeasurement is >0.5. Hence, a tracking estimate of r = 0.8 would result in a sample size decrease of 60 % when using the first design.
Thinking beyond the time span of intervention or preven-tion trials, it would be desirable to take the high, but not per-fect tracking into account when planning such trials: If a trial successfully demonstrates the effectiveness of a BMI-targeted intervention taking place between 3 and 5 years of age, it is also of interest to what extent the effect can be expected to be main-tained 10 years later in adolescence, given the tracking estimates shown in Table 3 (thereby assuming natural courses of BMI tracking, as estimates after intervention trials are lacking).
Methodological considerations
The systematic review underlying this meta-regression analy-sis was carried out in accordance with the MOOSE consensus statement (17). In order to minimize loss of relevant studies, a high proportion of articles passed the title or abstract review. Reviewing these articles was supplemented by an automatic full-text screening, using a bash script that allowed identify-ing articles reportidentify-ing correlation coefficients or their squared analogues.
The inclusion of studies from all continents as well as the predominance of population-based studies on the regression analysis provides a basis for generalizable results.
Multiple correlation coefficients with different follow-up periods were reported in some studies and considered in
our analysis. These data particularly allow examining the influences of follow-up time and baseline age on BMI track-ing. The statistical approach in this article controls for the possible dependence between multiple measurements from the same cohort.
Sensitivity analysis revealed a high stability of tracking esti-mates. The potential impact of methodological issues such as type of tracking estimate or completeness of information was low.
Imprecision in measurement can bias the results of tracking studies toward lower values. As height and weight measure-ment can be regarded reliable, this is not an issue in this analy-sis of BMI tracking.
conclusions
The observed tracking estimates imply a low probability of spontaneous weight changes among individuals not under weight-loss treatment. Obese individuals should therefore be identified early and a weight-loss treatment should be con-sidered regardless of their baseline age. As correlation data on intervention studies were too sparse to be considered, the tempting conclusion of a successful long-term effect after obesity intervention has to be further examined. However, population-based primary prevention programs might have beneficial long-term effects due to the observed high tracking and possibly not changing natural courses of BMI tracking.
Furthermore, the high tracking has implications on trial design: when assessing e. g. intervention effects on BMI a lon-gitudinal pre- and postdesign is favorable to increase statistical power.
suPPleMentarY MaterIal
Supplementary material is linked to the online version of the paper at http:// www.nature.com/oby
acknowledgMents
This study was partially financially supported by the LMUinnovativ research priority project MCHealth (sub-project II). The funders had no role in design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript. O.B.: study design, literature search and review, data analysis and interpretation, writing. H.K.: literature search and review, data collection. R.v.K.: study design, writing. a.M.T.: study design, data interpretation, writing.
dIsclosure
The authors declared no conflict of interest. © 2010 The Obesity Society
reFerences
1. Mascie-Taylor CG, Karim E. The burden of chronic disease. Science 2003;302:1921–1922.
2. Kannel WB, LeBauer EJ, Dawber TR, McNamara PM. Relation of body weight to development of coronary heart disease. The Framingham study.
Circulation 1967;35:734–744.
3. Calle EE, Thun MJ, Petrelli JM, Rodriguez C, Heath CW Jr. Body-mass index and mortality in a prospective cohort of U.S. adults. N Engl J Med 1999;341:1097–1105.
5. Cole TJ, Bellizzi MC, Flegal KM, Dietz WH. Establishing a standard definition for child overweight and obesity worldwide: international survey. BMJ 2000; 320:1240–1243.
6. Twisk JWR. The problem of evaluating the magnitude of tracking coefficients. Eur J Epidemiol 2003;18:1025–1026.
7. Chen X, Wang Y. Tracking of blood pressure from childhood to adulthood: a systematic review and meta-regression analysis. Circulation 2008;117:3171–3180.
8. Toschke AM, Kohl L, Mansmann U, von Kries R. Meta-analysis of blood pressure tracking from childhood to adulthood and implications for the design of intervention trials. Acta Paediatr 2010;99:24–29.
9. Taeymans J, Hebbelinck M, Borms J et al. Tracking of adult adiposity in early, average and late maturing children: a thirty year longitudinal growth study. J Sports Med Phys Fitness 2008;48:326–334.
10. Mueller WH, Dai S, Labarthe DR. Tracking body fat distribution during growth: using measurements at two occasions vs one. Int J Obes Relat
Metab Disord 2001;25:1850–1855.
11. Fisher RA. Frequency distribution of the values of the correlation coefficient in samples of an indefinitely large population. Biometrika 1915;10:507–521. 12. R Development Core Team. R: A Language and Environment for Statistical
Computing. ISBN 3-900051-07-0. <http://www.R-project.org>, 2008.
13. Pinheiro J, Bates D, DebRoy S, Sarkar D, the R Core team. nlme: Linear
and Nonlinear Mixed Effects Models. R package version 3.1-90, 2008.
14. Ulmer H, Kelleher C, Diem G, Concin H. Long-term tracking of cardiovascular risk factors among men and women in a large population-based health system: the Vorarlberg Health Monitoring & Promotion Programme. Eur Heart J 2003;24:1004–1013.
15. Toschke AM, Rückinger S, Reinehr T, von Kries R. Growth around puberty as predictor of adult obesity. Eur J Clin Nutr 2008;62:1405–1411. 16. Lake JK, Power C, Cole TJ. Child to adult body mass index in the 1958
British birth cohort: associations with parental obesity. Arch Dis Child 1997;77:376–381.
17. Stroup DF, Berlin JA, Morton SC et al. Meta-analysis of observational studies in epidemiology: a proposal for reporting. Meta-analysis Of Observational Studies in Epidemiology (MOOSE) group. JAMA 2000;283:2008–2012. 18. Angelico F, Del Ben M, Fabiani L et al. Management of childhood obesity
through a school-based programme of general health and nutrition education. Public Health 1991;105:393–398.
19. Angelico F, Del Ben M, Barbato A et al. Eleven-year tracking of established cardiovascular risk factors in Italian school-aged children. Ann Ig 1997;9:193–200.
20. Juonala M, Raitakari M, S A Viikari J, Raitakari OT. Obesity in youth is not an independent predictor of carotid IMT in adulthood. The Cardiovascular Risk in Young Finns Study. Atherosclerosis 2006;185:388–393.
21. Yang X, Telama R, Leskinen E et al. Testing a model of physical activity and obesity tracking from youth to adulthood: the cardiovascular risk in young Finns study. Int J Obes (Lond) 2007;31:521–527.
22. Kivimäki M, Lawlor DA, Smith GD et al. Association of age at menarche with cardiovascular risk factors, vascular structure, and function in adulthood: the Cardiovascular Risk in Young Finns study. Am J Clin Nutr 2008;87: 1876–1882.
23. Lauer RM, Clarke WR. Childhood risk factors for high adult blood pressure: the Muscatine Study. Pediatrics 1989;84:633–641.
24. Clarke WR, Lauer RM. Does childhood obesity track into adulthood? Crit
Rev Food Sci Nutr 1993;33:423–430.
25. Monyeki KD, Monyeki MA, Brits SJ, Kemper HC, Makgae PJ. Development and tracking of body mass index from preschool age into adolescence in rural South African children: Ellisras Longitudinal Growth and Health Study.
J Health Popul Nutr 2008;26:405–417.
26. Monyeki MA, Koppes LL, Monyeki KD, Kemper HC, Twisk JW. Longitudinal relationships between nutritional status, body composition, and physical fitness in rural children of South Africa: The Ellisras longitudinal study. Am J
Hum Biol 2007;19:551–558.
27. Raitakari OT, Juonala M, Viikari JS. Obesity in childhood and vascular changes in adulthood: insights into the Cardiovascular Risk in Young Finns Study. Int J Obes (Lond) 2005;29 Suppl 2:S101–S104.
28. Shephard RJ. Role of the physician in childhood obesity. Clin J Sport Med 2004;14:161–168.
29. Steinhausen HC, Grigoroiu-Serbanescu M, Boyadjieva S, Neumärker KJ, Metzke CW. The relevance of body weight in the medium-term to long-term course of adolescent anorexia nervosa. Findings from a multisite study. Int J
Eat Disord 2009;42:19–25.
30. Deshmukh-Taskar P, Nicklas TA, Morales M et al. Tracking of overweight status from childhood to young adulthood: the Bogalusa Heart Study.
Eur J Clin Nutr 2006;60:48–57.
31. Wattigney WA, Webber LS, Srinivasan SR, Berenson GS. The emergence of clinically abnormal levels of cardiovascular disease risk factor variables among young adults: the Bogalusa Heart Study. Prev Med 1995;24: 617–626.
32. Andersen LB, Haraldsdóttir J. Tracking of cardiovascular disease risk factors including maximal oxygen uptake and physical activity from late teenage to adulthood. An 8-year follow-up study. J Intern Med 1993;234:309–315. 33. Burke GL, Bild DE, Hilner JE et al. Differences in weight gain in relation
to race, gender, age and education in young adults: the CARDIA Study. Coronary Artery Risk Development in Young Adults. Ethn Health 1996;1:327–335.
34. Burke V, Beilin LJ, Dunbar D, Kevan M. Associations between blood pressure and overweight defined by new standards for body mass index in childhood. Prev Med 2004;38:558–564.
35. Casey VA, Dwyer JT, Coleman KA, Valadian I. Body mass index from childhood to middle age: a 50-y follow-up. Am J Clin Nutr 1992;56:14–18. 36. Clausen JO, Ibsen H, Ibsen KK, Borch-Johnsen K. Association of body
mass index, blood pressure and serum levels of triglycerides and high-density lipoprotein cholesterol in childhood with the insulin sensitivity index in young adulthood: a 13-year follow-up. J Cardiovasc Risk 1996;3:427–433. 37. Eisenmann JC, Welk GJ, Wickel EE, Blair SN; Aerobics Center Longitudinal
Study. Stability of variables associated with the metabolic syndrome from adolescence to adulthood: the Aerobics Center Longitudinal Study. Am J
Hum Biol 2004;16:690–696.
38. Ekblom OB, Bak EA, Ekblom BT. Trends in body mass in Swedish adolescents between 2001 and 2007. Acta Paediatr 2009;98:519–522. 39. Fabsitz RR, Carmelli D, Hewitt JK. Evidence for independent genetic
influences on obesity in middle age. Int J Obes Relat Metab Disord 1992;16:657–666.
40. Fuentes RM, Notkola IL, Shemeikka S, Tuomilehto J, Nissinen A. Tracking of body mass index during childhood: a 15-year prospective population-based family study in eastern Finland. Int J Obes Relat Metab Disord 2003;27: 716–721.
41. Gaskin PS, Walker SP. Obesity in a cohort of black Jamaican children as estimated by BMI and other indices of adiposity. Eur J Clin Nutr 2003;57:420–426.
42. Guo SS, Roche AF, Chumlea WC, Gardner JD, Siervogel RM. The predictive value of childhood body mass index values for overweight at age 35 y. Am J
Clin Nutr 1994;59:810–819.
43. Hulens M, Beunen G, Claessens AL et al. Trends in BMI among Belgian children, adolescents and adults from 1969 to 1996. Int J Obes Relat Metab
Disord 2001;25:395–399.
44. Hulman S, Kushner H, Katz S, Falkner B. Can cardiovascular risk be predicted by newborn, childhood, and adolescent body size? An examination of longitudinal data in urban African Americans. J Pediatr 1998;132:90–97.
45. Julia M, van Weissenbruch MM, Prawirohartono EP, Surjono A, Delemarre-van de Waal HA. Tracking for underweight, overweight and obesity from childhood to adolescence: a 5-year follow-up study in urban Indonesian children. Horm Res 2008;69:301–306.
46. Kelly JL, Stanton WR, McGee R, Silva PA. Tracking relative weight in subjects studied longitudinally from ages 3 to 13 years. J Paediatr Child
Health 1992;28:158–161.
47. Kvaavik E, Tell GS, Klepp KI. Predictors and tracking of body mass index from adolescence into adulthood: follow-up of 18 to 20 years in the Oslo Youth Study. Arch Pediatr Adolesc Med 2003;157:1212–1218. 48. Lambrechtsen J, Rasmussen F, Hansen HS, Jacobsen IA. Tracking
and factors predicting rising in ‘tracking quartile’ in blood pressure from childhood to adulthood: Odense Schoolchild Study. J Hum Hypertens 1999;13:385–391.
49. Leppik A, Jürimäe T, Jürimäe J. Tracking of anthropometric parameters and bioelectrical impedance in pubertal boys and girls. Coll Antropol 2006;30:753–760.
50. Maffeis C, Moghetti P, Grezzani A et al. Insulin resistance and the persistence of obesity from childhood into adulthood. J Clin Endocrinol
Metab 2002;87:71–76.
51. Magarey AM, Daniels LA, Boulton TJ, Cockington RA. Predicting obesity in early adulthood from childhood and parental obesity. Int J Obes Relat Metab
52. Marshall SJ, Sarkin JA, Sallis JF, McKenzie TL. Tracking of health-related fitness components in youth ages 9 to 12. Med Sci Sports Exerc 1998;30:910–916.
53. Matton L, Thomis M, Wijndaele K et al. Tracking of physical fitness and physical activity from youth to adulthood in females. Med Sci Sports Exerc 2006;38:1114–1120.
54. Mohler B, Ackermann-Liebrich U, Steffen T, Staehelin HB. Cholesterol screening in childhood: results of a 9-year follow-up study in Swiss and Italian children in Switzerland. Soz Praventivmed 1996;41:333–340. 55. Oja L, Jürimäe T. Changes in anthropometrical characteristics during two
years in 6 year old children. Anthropol Anz 2002;60:299–308. 56. Olvera N, Sharma S, Suminski R, Rodríguez A, Power T. BMI tracking
in Mexican American children in relation to maternal BMI. Ethn Dis 2007;17:707–713.
57. Oren A, Vos LE, Uiterwaal CS et al. Change in body mass index from adolescence to young adulthood and increased carotid intima-media thickness at 28 years of age: the Atherosclerosis Risk in Young Adults study.
Int J Obes Relat Metab Disord 2003;27:1383–1390.
58. Palti H, Gofin R, Adler B, Grafstein O, Belmaker E. Tracking of blood pressure over an eight year period in Jerusalem school children. J Clin
Epidemiol 1988;41:731–735.
59. Pate RR, Trost SG, Dowda M et al. Tracking of physical activity, physical inactivity, and health-related physical fitness in rural youth. Pediatr Exerc Sci 1999;11:364–376.
60. Psarra G, Nassis GP, Sidossis LS. Short-term predictors of abdominal obesity in children. Eur J Public Health 2006;16:520–525.
61. Sinaiko AR, Steinberger J, Moran A et al. Influence of insulin resistance and body mass index at age 13 on systolic blood pressure, triglycerides, and high-density lipoprotein cholesterol at age 19. Hypertension 2006;48: 730–736.
62. Steinberger J, Moran A, Hong CP, Jacobs DR Jr, Sinaiko AR. Adiposity in childhood predicts obesity and insulin resistance in young adulthood.
J Pediatr 2001;138:469–473.
63. Tan F, Okamoto M, Suyama A, Miyamoto T. Tracking of cardiovascular risk factors and a cohort study on hyperlipidemia in rural schoolchildren in Japan. J Epidemiol 2000;10:255–261.
64. Trudeau F, Shephard RJ, Arsenault F, Laurencelle L. Changes in adiposity and body mass index from late childhood to adult life in the Trois-Rivières study. Am J Hum Biol 2001;13:349–355.
65. Valerio G, Licenziati MR, Iannuzzi A et al. Insulin resistance and impaired glucose tolerance in obese children and adolescents from Southern Italy.
Nutr Metab Cardiovasc Dis 2006;16:279–284.
66. Vogels N, Posthumus DL, Mariman EC et al. Determinants of overweight in a cohort of Dutch children. Am J Clin Nutr 2006;84:717–724.
67. Wang Y, Ge K, Popkin BM. Tracking of body mass index from childhood to adolescence: a 6-y follow-up study in China. Am J Clin Nutr 2000;72: 1018–1024.
68. Wardle J, Brodersen NH, Cole TJ, Jarvis MJ, Boniface DR. Development of adiposity in adolescence: five year longitudinal study of an ethnically and socioeconomically diverse sample of young people in Britain. BMJ 2006;332:1130–1135.
69. Weststrate JA, Van Klaveren H, Deurenberg P. Changes in skinfold thicknesses and body mass index in 171 children, initially 1 to 5 years of age: a 5 ½-year follow-up study. Int J Obes 1986;10:313–321. 70. Wilsgaard T, Jacobsen BK, Schirmer H et al. Tracking of cardiovascular
risk factors: the Tromsø study, 1979–995. Am J Epidemiol 2001;154: 418–426.
71. Wright CM, Parker L, Lamont D, Craft AW. Implications of childhood obesity for adult health: findings from thousand families cohort study.
