Childhood overweight and obesity are significantly associated with adverse health outcomes in adulthood, including diabetes, hypertension, and coronary heart disease. The global prevalence of childhood obesity has dramatically increased, raising concerns about its long-term health implications. In England, the prevalence of overweight and obesity among children aged 10–11 years is high and has been steadily increasing, despite relatively stable rates in younger children. Children from deprived areas are disproportionately affected. While the National Child Measurement Programme (NCMP) provides weight status feedback to parents, this alone has not resulted in significant lifestyle changes. Early identification of at-risk children is crucial for effective interventions. Previous research focused on predicting overweight at age 4–5 years. This study aims to develop and validate prediction models for overweight and obesity at age 10–11 years, utilizing routinely collected weight and height measurements at age 4–5 years, along with maternal and early-life health data.

The literature extensively documents the adverse health consequences of childhood obesity, highlighting its persistence into adulthood. Studies show a significant increase in childhood obesity prevalence globally and in England. Research using the Millennium Birth Cohort and NCMP data reveals the trajectory of weight status, showing a high likelihood of maintaining or developing obesity from age 5 to 11. While NCMP feedback aims to increase parental awareness, its effectiveness in promoting lifestyle changes remains limited. School-based interventions combining diet and physical activity show promise, emphasizing the need to identify at-risk children early. Previous work by the authors developed prediction models for overweight at age 4–5 years using maternal and early-life predictors. However, limited research exists on predicting overweight and obesity at age 10–11 using routinely collected data in the UK. Existing prediction models from other countries often utilize predictors not readily available in the UK healthcare system.

This study utilized two cohorts: the Studying Lifecourse Obesity Predictors (SLOPE) dataset for model development and the Born in Bradford (BiB) cohort for external validation. SLOPE is a population-based, anonymized linked cohort of maternal antenatal and birth records and child health records from University Hospital Southampton and surrounding areas. BiB is a longitudinal multi-ethnic birth cohort study from Bradford. The outcome was childhood overweight/obesity (BMI ≥ 91st centile) at age 10–11 years, using NCMP data. Predictor variables included: **Year R (age 4–5 years):** weight, height, sex. **First antenatal appointment:** maternal age, BMI, smoking status, education, ethnicity, employment status, folic acid supplement use, parity, first language, partnership status. **Birth:** birthweight, gestational age, mode of birth. Logistic regression models and multivariable fractional polynomials were used to select predictors and identify optimal transformations of continuous variables. Stepwise backward elimination with an AIC-based significance level was employed for variable selection. Models were developed in stages, starting with Year R data and then adding pregnancy data. Internal validation used bootstrapping, and external validation used the BiB cohort. Model performance was assessed using AUC (discrimination) and calibration plots. Missing data in pregnancy predictors were imputed using multiple imputation by chained equations (MICE). Shrinkage factors were calculated to adjust for optimism.

The SLOPE dataset included 6566 children with valid weight and height measurements at both ages 4–5 and 10–11 years. 14.6% were overweight at age 4–5 and 26.1% at age 10–11. The model incorporating only BMI at age 4–5 years and sex achieved an AUC of 0.82 during development and 0.83 during external validation. Adding maternal predictors from the first antenatal appointment increased the AUC to 0.84 (development) and 0.85 (external validation). Both models demonstrated good calibration. The inclusion of maternal data slightly improved prediction accuracy. Several maternal factors (age, BMI, smoking status, education, employment status, ethnicity, and parity) were significant predictors. Birth-related variables were not retained in the final models. Using a 30% risk threshold, the Year R model identified 31.3% of children at risk, with a sensitivity of 65.8%, specificity of 80.9%, PPV of 55.0%, and NPV of 87.0%. The model with pregnancy factors identified 37.7% at risk with higher sensitivity (71.3%) but slightly lower specificity (74.2%).

This study successfully developed and validated prediction models for childhood overweight and obesity at age 10–11 using routinely collected data. The findings demonstrate the feasibility of identifying at-risk children as early as age 4–5 years, using readily available information. The inclusion of maternal data slightly improved predictive accuracy, highlighting the importance of considering the influence of the prenatal and early postnatal period on later weight status. These models can inform the allocation of resources for preventive interventions, targeting those most at risk. The relatively high negative predictive value suggests that the models are effective at identifying children unlikely to become overweight or obese, allowing for efficient allocation of resources. However, the positive predictive values are modest, indicating that many children identified as high-risk will not develop obesity. The current practice of providing NCMP feedback to parents remains insufficient to induce significant changes in behavior. Targeted interventions accounting for socioeconomic factors are needed.

This research provides robust prediction models for childhood overweight and obesity using routinely collected data, enabling early identification of at-risk children. The models' good discrimination and calibration make them suitable for implementation in healthcare settings to facilitate targeted preventive interventions. Future research could investigate the cost-effectiveness of using these models to guide resource allocation and explore the impact of culturally tailored interventions on obesity prevention.

The study's limitations include the potential for outcome data missingness due to changes in recording practices or children leaving the healthcare system. Early-life variables were not included due to high missing data rates. The external validation cohort, while diverse, might not fully represent the entire UK population. The lack of standardized risk thresholds for childhood obesity made determining optimal risk cut-off points challenging. The reliance on self-reported data for some maternal variables introduces potential bias.