This study compared the importance of age at adiposity rebound versus childhood BMI to subsequent BMI levels in a longitudinal analysis.

From the electronic health records of 4.35 million children, a total of 12,228 children were selected who were examined at least once each year between ages 2 and 7 years and reexamined after age 14 years. The minimum number of examinations per child was six. Each child’s rebound age was estimated using locally weighted regression (lowess), a smoothing technique.

Children who had a rebound age < 3 years were, on average, 7 kg/m^{2} heavier after age 14 years than were children with a rebound age ≥ 7 years. However, BMI after age 14 years was more strongly associated with BMI at the rebound than with rebound age (

Although rebound age is related to BMI after age 14 years, a child’s BMI at age 3 years provides more information and is easier to obtain.

Levels of BMI increase from birth to about 1 year of age, decrease to a minimum between 2 and 8 years of age, and increase again during later childhood and adolescence. Based on serial data from about 155 individuals, Rolland-Cachera et al. reported that the earlier the nadir of BMI occurs, the higher the BMI at age 16 years and early adulthood (

Numerous investigators have since confirmed the inverse relationship between age at rebound and subsequent BMI levels, body fatness, and obesity in adolescence and adulthood (

However, it is uncertain whether age at rebound provides additional information on adolescent or adult obesity beyond that conveyed by BMI levels at various ages in childhood. For example, Williams et al. (

Our objectives are to examine the longitudinal relationship of rebound age to subsequent BMI among 12,228 children from an electronic health record (EHR) database and quantify the amount of information provided by BMI at various ages between 2 and <8 years. All children had at least one BMI value each year between the ages of 2 and <8 years, along with a subsequent BMI between ages 14 and <20 years.

PEDSnet (

The longitudinal data in the current study are from four of the hospitals in PEDSnet: the Children’s Hospital of Philadelphia; the Children’s Hospital of Colorado; the Nationwide Children’s Hospital in Ohio; and the Nemours Children’s Health System (a Delaware and Florida health system). Because this is a secondary analysis of children in this database, no sample size calculations were performed.

The Children’s Hospital of Philadelphia Institutional Review Board (IRB) determined this project does not meet human subjects research criteria. The Centers for Disease Control and Prevention (CDC) determined that this secondary data analysis is not human subjects research, and, therefore, CDC IRB review/approval is not required. The analyses used a deidentified data set; participant identifiers were replaced, and temporal information was limited to the year of birth and age in days.

The use of EHR data from clinical care for research purposes requires assessing the data’s quality (

There were 33.0 million patient weights and 20.3 million heights from 4.35 million 2- to 19-year-old individuals at an in-person clinical encounter at PEDSnet member institutions between 1999 and 2019. These in-person encounters included inpatient hospital stays, outpatient specialty and primary care visits with a physician or non-physician, emergency department visits, and observation stays. We refer to these encounters as “visits” throughout the text.

We identified weights and heights that were likely to be errors using Daymont’s algorithm for the longitudinal detection of outliers (

We excluded all weights and heights identified as carried forwards or as other types of errors, resulting in a data set of 28 million weights and 18 million heights. Because of the large number of weights without a same-day height, we allowed heights to match with weights obtained within 30 days. This matching resulted in 19.3 million records, with both weight and height from 3.5 million children. We calculated BMI as kilograms per meters squared, and extreme values (52,348; 0.24%) of weight, height, and BMI were further identified and excluded based on their modified

BMI

We performed data management and analyses using R (version 4.05) (R Core Team). Inspection of BMI plots for individual children indicated that most children had several BMI values between ages 2 and <8 years that were very close to the BMI nadir. Furthermore, these low BMI values frequently occurred at nonconsecutive visits. Therefore, most analyses are based on using locally weighted regression (lowess) (

We performed sensitivity analyses in which age at rebound was based on the following: 1) minimum observed BMI; 2) the minimum BMI after smoothing with Friedman’s super-smoother (

We examined mean levels of various characteristics by categories of age at the rebound. We then assessed whether the relationship of rebound age to adolescent BMI and obesity was independent of BMI levels between ages 2 and <8 years by comparing the multiple ^{2}s of several regression models that predicted BMI after age 14 years. Predictor variables in these models included rebound age, sex, height-for-age

We constructed a nomogram using the

Logistic regression models were also used to examine the relationship of BMI at rebound ages of 2, 4, 5, and 7 years to the probability of obesity after age 14 years.

All children were first examined at age 2 years, the mean age at the last visit was 15.5 (1.2) years, and the prevalence of obesity after age 14 years was 19%. Girls comprised 48% of the sample; 54% were White, 31% were Black, 8% were Hispanic, and 2% were Asian.

^{2} heavier at their last visit than were those with a rebound at age 7 years (27.4 vs. 20.2 kg/m^{2}). Furthermore, the prevalence of obesity at the last visit was 36% (rebound < 3 years) and 2.5% (rebound ≥ 7 years). However, as shown by the longitudinal correlations in the final columns, BMI and BMIz at the last visit were more strongly associated with the BMI at rebound than with age at rebound (

As shown in

We then constructed a series of regression models to determine whether the relationship of rebound age to BMI in adolescence was independent of a child’s BMI at various ages. A regression model that included the covariates (sex, age at last visit, and height ^{2}s for these additional models predicting BMI after age 14 years by either the BMI at ages 2 to 7 years or with the additional information conveyed by rebound age. A model based on the covariates and BMI at age 2 years, but not rebound age, accounted for almost as much (0.23) of the BMI variability after age 14 years as did rebound age. Furthermore, a model containing BMI at age 3 years accounted for more variability (0.32) in BMI after age 14 years than the model containing rebound age. The amount of variability accounted for by BMI (alone) increased with each subsequent age, whereas the additional information supplied by rebound age decreased from 19% (0.42 vs. 0.23) at age 2 years to 7% at age 4 years and to virtually 0 at ages 6 and 7 years.

Our results indicate that age at BMI rebound is inversely associated with subsequent levels of BMI and obesity after age 14 years. However, the BMI at age 2 years accounts for a similar amount of the variability in BMI levels among adolescents as does rebound age (^{2} of 0.23 and 0.24), whereas the BMI at age 3 years accounts for more (^{2} = 0.32) of the variability. Furthermore, rebound age provides almost no independent information on BMI in adolescence if the BMI at ages 6 or 7 years is known. In addition, among children with a BMI rebound of age 2 years, the BMI value at age 2 years was a strong predictor of obesity after age 14 years.

Williams et al. (^{2} heavier in early adulthood than children with a rebound age ≥ 7 years. However, the rebound age conveyed no information on adult BMI if the BMI at age 7 years was known. A larger study of 17,000 children (

Because the estimation of rebound age requires at least three BMI values throughout childhood, it is uncertain why childhood BMI has received little attention in previous studies of adiposity rebound. Although rebound age was originally defined as the age corresponding to the lowest BMI before the increase in adiposity (

In order to address the difficulties in determining rebound age, some investigators have used various regression models to identify rebound age (^{2}. However, this latter approach would substantially increase the rebound age for many children in

We used lowess to determine rebound age but obtained very similar results with other techniques. However, minor errors in measuring weight and height, particularly if height is recorded in whole inches, can lead to relatively large BMI differences (^{2}. However, if this child’s weight and height were recorded as 39.5 lb and 41 in, the BMI will be 16.5, a difference of 1 kg/m^{2}. Measurement errors such as measuring height with or without shoes and the day-to-day variability in weight would further complicate the determination of the rebound age in clinical practice. In contrast to the difficulty in determining rebound age, we found that a single BMI measurement at age 2 years provided almost as much information, whereas BMI at age 3 years provided more information than did the age at BMI rebound on BMI levels after age 14 years.

Although the prevalence of obesity in PEDSnet data agrees well with National Health and Nutrition Examination Study (NHANES) (

Imposing data sufficiency requirements such as follow-up time or the number of measurements could further increase bias (

The PEDSnet data used in the current study reflect a reasonably broad geographic distribution of the United States, but there are gaps, and children in rural areas are underrepresented. In the entire PEDSnet data set, about 4% of the children appear to have received care at multiple institutions and are represented as different individuals from each site.

Our results confirm that an early BMI rebound is associated with higher BMI levels in adolescence. However, as compared with the information provided by rebound age (^{2} = 0.24) on BMI levels after age 14 years, a child’s BMI at age 2 years provides almost as much information (0.23), whereas the BMI at age 3 years conveys more information (0.32). Because it can be challenging to determine a child’s rebound age, particularly from a small number of BMI values, it may be more practical to identify subsequent obesity at a very early age by a child’s BMI at ages 2 or 3 years.

This project was funded in part by a grant (award number RI-CRN-2020-007) from the Patient-Centered Outcomes Research Institute (PCORI). This report’s findings and conclusions are those of the authors and do not represent the official position of the Centers for Disease Control and Prevention.

CONFLICT OF INTEREST

The authors declared no conflict of interest.

Nomogram for predicted probability of obesity after age 14 years (bottom row) based on sex, race/ethnicity, rebound age, and BMI at age 3 years. For each predictor variable, the number of points (top line) would be calculated. These values would then be summed to derive a child’s total points and probability of obesity (final two rows). For the race/ethnicity categories, “W” indicates White non-Hispanic children, “H” indicates Hispanic children, and “Oth” indicates a race coded as other. Black children had nine points

Predicted probability of obesity after age 14 years by BMI level at a rebound age of 2, 4, 5, or 7 years. Predicted probabilities, based on logistic regression, are for a 15-year-old individual. BMI values represent the BMI or mean BMI at the rebound age

BMI values of 16 children who were randomly selected from children who had 10 to 12 visits between ages 2 years and <8 years

Relation of age at rebound to various characteristics at the time of the rebound and to values at the final visit among 12,228 children (PEDSnet)^{a}

Age (y) at BMI rebound | |||||||
---|---|---|---|---|---|---|---|

<3 | 3–4.9 | 5–6.9 | 7 | Correlation with rebound age | Correlation with BMI at rebound age | ||

| 3,233 | 2,764 | 3,627 | 2,604 | |||

Girls (%) | 55% | 50% | 45% | 43% | |||

At BMI rebound | Age | 2.2 ± 0.3 | 4.0 ± 0.5 | 5.7 ± 0.5 | 7.5 ± 0.3 | ||

BMI | 16.1 ± 1.9 | 15.8 ± 1.5 | 15.3 ± 1.3 | 15.1 ± 1.2 | ---- | ---- | |

BMIz | −0.25 ± 1.3 | 0.17 ± 1.2 | −0.10 ± 1.0 | −0.44 ± 0.9 | ---- | ---- | |

Height-for-age | 0.57 ± 1.17 | 0.44 ± 1.11 | 0.16 ± 1.06 | −0.23 ± 1.09 | --- | --- | |

At last visit | Age | 15.5 ± 1.2 | 15.5 ± 1.2 | 15.5 ± 1.2 | 15.4 ± 1.2 | ---- | ---- |

BMI | 27.4 ± 7.5 | 24.9 ± 5.8 | 21.9 ± 4.1 | 20.2 ± 3.4 | −0.44 (−0.43 to −0.46) | 0.57 (0.56 to 0.58) | |

BMIz | 1.20 ± 1.1 | 0.84 ± 1.0 | 0.27 ± 1.0 | −0.17 ± 1.0 | −0.46 (−0.47 to −0.44) | 0.57 (0.56 to 0.59) | |

Obesity (%) | 36% | 23% | 6.9% | 2.5% | --- | --- |

Values are mean ± SD or correlation coefficient (95% CI).

Mean follow-up time: 13.3 years (range: 11.1–17.9 years).

Stratified analyses of longitudinal correlations between BMIz at last visit and both age at rebound and BMI at age at rebound

Correlation with BMIz at last visit | |||||
---|---|---|---|---|---|

| Age at rebound | BMI at rebound | |||

Sex | |||||

Boys | 6,348 | −0.44 | 0.58 | ||

Girls | 5,880 | −0.47 | 0.60 | ||

Race/ethnicity | |||||

White | 6,562 | −0.43 | 0.57 | ||

Black | 3,743 | −0.48 | 0.57 | ||

Hispanic | 964 | −0.47 | 0.59 | ||

Asian | 293 | −0.43 | 0.55 | ||

Multiple | 133 | −0.30 | 0.55 | ||

Other | 149 | −0.50 | 0.50 | ||

Unknown | 384 | −0.45 | 0.61 |

Abbreviation: BMIz, BMI

Multiple ^{2}s for various regression models predicting final BMI from covariates^{a}, age_{rebound}, and BMI value at various ages (PEDSnet)

Multiple ^{2} | ||
---|---|---|

BMI at specified age (y) | BMI^{b} | BMI and rebound age |

Age 2 | 0.23 | 0.42 |

Age 3 | 0.32 | 0.43 |

Age 4 | 0.41 | 0.48 |

Age 5 | 0.50 | 0.54 |

Age 6 | 0.57 | 0.58 |

Age 7 | 0.62 | 0.62 |

Linear regression models include sex, childhood height-for-age ^{2} of a model with these three covariates alone was 0.08, whereas the ^{2} of a model with the covariates and rebound_{age} was 0.24.

For children who had more than one BMI value per year of age, we used the mean BMI in the analyses. Additional models that included one randomly selected BMI for a given age yielded virtually identical results.

An early adiposity rebound (e.g., <4 years) increases the risk for obesity in adolescence and adulthood.

A child’s BMI at age 3 years provides more information on BMI after age 14 years than does the child’s rebound age (multiple ^{2}s of 0.32 vs. 0.24).

Information on age at adiposity rebound provides no information on BMI levels after age 14 years if the BMI at age 6 years is known.

Because it can be challenging to determine adiposity rebound, it may be best to use the BMI at age 3 years to predict subsequent obesity.