New Body Shape Index Outperforms BMI for Screening Low Physical Fitness in Preschoolers: Evidence from 3,180 Children in Macao

Table of Contents

1. Key Highlights
2. Introduction
3. How the study was designed: sample, tests and the data-driven approach
4. Three distinct fitness phenotypes emerge in preschool years
5. Body shape mirrors fitness: which measurements mattered
6. Building a practical screening tool: the composite Body Shape Index (BSI)
7. How BSI compares to BMI — diagnostic performance and sex analysis
8. Practical applications: screening, early intervention and public health
9. Methodological strengths and caveats: what the study gets right and what to watch
10. Next steps: validation, refinement and policy translation
11. FAQ

Key Highlights

• A data-driven analysis of 3,180 preschool children in Macao identified three coherent physical-fitness phenotypes (low, moderate, high). Physical fitness clustered consistently across six standardized motor tests.
• A composite Body Shape Index (BSI) based on height, weight, waist circumference and pelvic width discriminated children at risk of low fitness far better than BMI (AUCBSI = 0.779 vs AUCBMI ≈ 0.54); sensitivity at the optimal cut-off reached 77.7% with 66.8% specificity.
• Waist circumference and pelvic width—measures of central adiposity and pelvic breadth—added independent predictive value beyond simple body size; the BSI shows promise as a rapid screening tool during routine checks but requires external and longitudinal validation before broader implementation.

Introduction

Preschool years set the foundation for movement competence, habitual activity and later health. Clinicians, educators and public-health practitioners routinely seek simple ways to flag children who may be falling behind in motor capacity and overall fitness. Historically, fitness screening in young children has relied on isolated performance tests and crude morphological metrics, most commonly body mass index (BMI). BMI is easy to measure but blurs critical differences in body composition and fat distribution that can matter for motor skills and functional capacity in rapidly growing preschoolers.

A recent study drawing on three waves of population surveillance in Macao (2010, 2015, 2020) used unsupervised machine learning to identify natural groupings in motor performance across six field tests and then linked those groupings to multidimensional anthropometry. The investigators derived a composite Body Shape Index (BSI) that integrates height, weight, waist circumference and pelvic width to screen for children likely to belong to a low-fitness phenotype. The index substantially outperformed BMI in discriminating risk in this population, suggesting a practical alternative for preliminary screening in clinical and school settings.

This article unpacks that study’s design, findings and implications, describes how the BSI works, and outlines what clinicians, public-health teams and researchers should consider before adopting or testing this approach more widely.

How the study was designed: sample, tests and the data-driven approach

The analysis pooled data from the Macao China Physical Fitness Surveillance and Physical Activity Survey across three survey years (2010, 2015, 2020). The combined sample included 3,180 children aged 3–6 years (1,903 boys; 1,277 girls). Sampling used stratified cluster methods, and measurements followed the China Physical Fitness Standards for Preschool Children (CPFS-preschool) protocol with trained examiners.

Six field-based fitness tests formed the basis of the phenotype discovery:

• Standing long jump (lower‑body explosive strength)
• Tennis ball throw (upper‑body power/coordination)
• 10‑m shuttle run (speed and agility)
• 15‑m obstacle run (complex motor coordination and agility)
• Sit-and-reach (flexibility)
• Balance beam walk (balance and motor control)

Raw scores were converted to age- and sex-specific Z-scores across the combined sample to remove typical developmental differences and make the tests comparable. The authors applied K-means clustering to the standardized test scores, running the algorithm with multiple random starts and assessing cluster stability via silhouette width and elbow/scree methods. Kaiser-Meyer-Olkin and Bartlett’s tests indicated multivariate structure was present and clustering was sensible.

After identifying clusters, the investigators compared nine anthropometric indicators across groups: height, sit height, weight, chest circumference, waist circumference, hip circumference, shoulder breadth, pelvic width and foot length. They then constructed a logistic regression model using the low‑fitness cluster as the outcome (coded 1 = low fitness) and all body-shape measures as candidate predictors. The final model produced the BSI from the coefficients of four independent predictors: height, weight, waist circumference and pelvic width.

Key analytical safeguards included standardized inputs for clustering, multiple initializations to stabilize K‑means, testing for heteroscedasticity and choosing Welch’s ANOVA where appropriate, multicollinearity checks (VIFs), Hosmer-Lemeshow goodness-of-fit testing, and internal validation via 10‑fold cross‑validation and bootstrap confidence intervals for ROC estimates.

Three distinct fitness phenotypes emerge in preschool years

The clustering procedure produced a clear, replicable three-cluster solution that explained about 39% of the variance in the six-test profile. The clusters were interpreted as:

• Low Fitness (Cluster 1; n = 412)
• Moderate Fitness (Cluster 2; n = 1,200)
• High Fitness (Cluster 3; n = 1,568)

Cluster profiles showed a gradient across motor skills. Children in the Low Fitness cluster scored markedly below age- and sex-specific norms on agility and coordination measures (e.g., shuttle run, obstacle run) and performed less well on balance and continuous jump. Sit-and-reach (flexibility) had modest differences across clusters, suggesting flexibility alone is a poor discriminator of overall integrated fitness at this age.

Why does a phenotype approach matter? Single-test comparisons can misclassify children who perform well in one domain (for example, flexibility) but poorly in others (balance, coordination). A multi-test clustering approach recognizes that integrated physical fitness in early childhood is a multidimensional latent construct; identifying coherent profiles reveals children whose overall motor repertoire is lower across domains, not just in one isolated skill.

Real-world parallel: kindergarten fitness screenings often record a handful of separate test scores for each child. A simple cluster-derived typology lets teachers or school health staff interpret those scores as a pattern—identifying children who show consistently low performance across multiple motor skills—so that interventions can target general motor competence rather than isolated skills.

Body shape mirrors fitness: which measurements mattered

All nine anthropometric indicators differed significantly across the three fitness clusters, with a consistent gradient: children in the High Fitness cluster tended to be taller, heavier and to have larger circumferential and skeletal breadth measures than those in the Low Fitness cluster. Effect sizes ranged from small to moderate-large depending on the indicator; height showed the largest effect (Cohen’s f ≈ 0.19), while waist circumference showed a notably smaller but still statistically significant effect (Cohen’s f ≈ 0.025).

The multivariate model distilled these nine measures into four independent predictors of low-fitness status:

• Height (negative association): taller children were less likely to belong to the low-fitness cluster.
• Weight (negative association): heavier children—within the preschool range—were, counterintuitively, less likely to be in the low-fitness cluster after accounting for other measures.
• Waist circumference (positive association): greater central girth increased the odds of low fitness.
• Pelvic width (positive association): wider pelvic breadth increased the odds of low fitness.

Interpretation requires nuance. Height and overall weight acted as indicators of general body size and maturity; within the preschool window, children who are relatively taller tend to display more advanced motor skills. Waist circumference and pelvic width appear to capture dimensions of shape and fat distribution—central adiposity or specific body proportions—that negatively associate with integrated motor performance when present beyond what would be expected for size.

A practical implication: waist circumference—a simple tape measure taken at the level specified by the measurement protocol—can add predictive value for fitness beyond BMI. Pelvic width, a skeletal breadth measure, is less commonly recorded in routine checks but showed independent association in this sample. Where pelvic breadth is not routinely measured, chest or hip circumferences may offer partial substitutes, but their independent predictive value was smaller in the final model.

Building a practical screening tool: the composite Body Shape Index (BSI)

The logistic regression model produced a linear predictor—the Body Shape Index (BSI)—combining the four anthropometric variables. The formula reported in the study is:

BSI = −2.376 − 1.156 × Height (cm) − 0.412 × Weight (kg) + 0.493 × Waist Circumference (cm) + 0.225 × Pelvic Width (cm)

Directionality: positive coefficients for waist circumference and pelvic width indicate that larger values of those measures raise the BSI and therefore the predicted odds of low fitness. Negative coefficients for height and weight indicate that greater overall size reduces the predicted odds of low fitness.

Diagnostic threshold: the authors identified an optimal cut-off of −2.003 on the BSI using Youden’s index. At this threshold, the index achieved 77.7% sensitivity and 66.8% specificity for classifying children as belonging to the low‑fitness phenotype in the study sample. Internal validation returned a cross-validated AUC of ≈0.765, indicating stable predictive performance within the dataset.

How to use the BSI in a routine setting (conceptual steps):

1. Collect the four measures following the CPFS-preschool protocol: height, weight, waist circumference and pelvic width.
2. Compute the BSI as the linear combination above using the original units (cm, kg).
3. Compare the score to the cut-off (−2.003): scores above the threshold indicate elevated risk of low integrated fitness and a recommendation for more thorough assessment or intervention referral.

Operational caveat: the BSI was derived from measurements taken under standardized procedures in Macao; real-world use requires the same measurement conventions and training to ensure comparable results. Before adopting the BSI for clinical decisions, institutions should consider local validation and staff training in anthropometric technique.

Practical example (qualitative): A public-health nurse conducting preschool health checks who measures a child’s waist circumference and pelvic width alongside routine height and weight can calculate the BSI. If the score exceeds the threshold, the child could be flagged for a targeted motor competence assessment, referral to a physiotherapist or enrolment in structured movement play sessions at preschool.

How BSI compares to BMI — diagnostic performance and sex analysis

The study contrasted the BSI with BMI-only models. BMI—calculated from height and weight and standardized for age and sex—showed minimal discriminative ability for the low-fitness phenotype in this preschool sample. Reported AUCs for BMI were close to chance (boys AUC = 0.52; girls AUC = 0.51; total sample AUC ≈ 0.54). DeLong’s test confirmed that the BSI’s AUC (0.779) was significantly larger than BMI’s (difference ≈ 0.239; Z = 10.15; p < 0.001).

Sex-stratified analysis for the BSI showed strong and consistent performance in both boys and girls: AUC = 0.87 for boys and 0.88 for girls within the study sample. Those high AUCs suggest that the combined morphological signals captured by the BSI generalize across sexes in the Macao preschool population, at least internally. The investigators noted, however, that a sex-specific model could yield different coefficients and might be explored in future work.

Why BMI underperformed here

• BMI conflates lean mass and fat mass. Two children with identical BMI can have very different body compositions and fat distribution—factors that plausibly influence motor performance.
• In early childhood, rapid, heterogeneous growth changes make BMI a noisy proxy for adiposity and physique. Waist circumference and pelvic breadth capture central shape and skeletal proportioning that BMI masks.
• Fitness reflects both neuromotor development and body mechanics. Shape measures that influence movement biomechanics (e.g., central adiposity that shifts the center of mass) plausibly relate more directly to functional motor performance than a single mass-to-height index.

Implication: BMI retains value for population surveillance of weight-for-height and metabolic risk, but it does not reliably identify preschool children whose integrated motor competence is low and who may benefit from early movement interventions.

Practical applications: screening, early intervention and public health

Where could this index fit into practice?

1. Routine well-child visits and preschool health checks

• Child health clinics often conduct height and weight measurements; adding waist circumference requires minimal additional time and equipment. Pelvic width requires one additional anthropometric measure and modest training.
• The BSI could serve as a first-line, low-cost screener to triage children for more comprehensive motor competence evaluation (e.g., direct observation, standardized motor skill batteries).

2. School and daycare screening programs

• Kindergarten teachers and school nurses can incorporate the index into periodic fitness check-ups to identify children who might need targeted movement activities or adapted physical education.

3. Public-health surveillance

• Community-level surveillance programs can include a BSI-based flag alongside other developmental screens to estimate population burden of low integrated fitness and to prioritize resources.

4. Program evaluation and targeted interventions

• Programs aiming to boost motor competence (structured play, fundamental movement skills curricula) can use the BSI to identify eligible participants and to monitor change over time, provided measurements and cut-offs remain consistent.

Implementation considerations

• Measurement standardization: waist and pelvic measures must follow explicit landmarks. The CPFS-preschool protocol used in the study provides such definitions; adopting them reduces measurement error.
• Training and quality control: brief training sessions, measurement demonstration and periodic inter-rater checks can sustain reliable data collection.
• Communication: flags should prompt supportive, constructive communication with families—focus on activity opportunities and motor skill development, not on weight stigma.
• Referral pathways: screen-positive children require a clear pathway—motor skills assessment by qualified staff, physical activity programming, and follow-up monitoring.

Real-world example: a municipal early-childhood program in a mid-sized city could pilot the BSI during the annual preschool health check, train nurses on waist and pelvic measurement, and offer a 12-week structured motor-skills program for children flagged by the index. Program success could be evaluated by pre/post direct motor assessments and parent-reported activity measures.

Methodological strengths and caveats: what the study gets right and what to watch

Strengths

• Large, population-based sample across three survey waves increased statistical power and internal representativeness for Macao.
• Use of multiple, standardized field tests and Z-score standardization controlled for age and sex differences, enabling phenotype detection independent of expected developmental variation.
• Clear, reproducible analytical pipeline: K-means clustering for phenotype discovery, rigorous assumption checks, purposeful variable selection, multicollinearity control and internal validation (10‑fold cross-validation; bootstrap CIs).
• Direct comparison with BMI using DeLong’s test provided a formal statistical basis for claims of superiority.

Caveats and limitations

• Cross-sectional design: associations between morphology and fitness phenotypes do not establish directionality. Low fitness could contribute to changes in body shape over time, and body shape could influence the acquisition of motor skills.
• Measurement protocol specifics: pelvic width and waist circumference may be captured differently across studies and settings. Results depend on strict adherence to measurement landmarks and technique.
• Potential cohort effects: combining data from three survey years improves sample size but may introduce unmeasured period- or cohort-specific influences (dietary patterns, physical activity opportunities, early-childhood programming differences).
• External validity: the BSI was developed and validated internally within the Macao preschool population. Generalizability to other populations—different ethnicities, socio-economic contexts, or measurement practices—remains untested.
• Model interpretability and scale: the reported BSI coefficients are based on the study’s measurement units and sample. Users must apply the formula with identical units and measurement conventions. Where measurement protocols differ, recalibration is necessary.
• Clinical thresholds: the optimal cut-off identified in-sample (−2.003) balances sensitivity and specificity for that dataset. Different populations will yield different trade-offs and may require context-specific thresholds.

Researchers and implementers should therefore treat the BSI as a promising screening concept that requires careful external validation and implementation planning rather than an off-the-shelf clinical diagnostic.

Next steps: validation, refinement and policy translation

Research priorities

• External validation: prototyping the BSI in independent cohorts across geographic regions and different ethnic groups to quantify transportability.
• Longitudinal studies: tracking children over time to determine whether the BSI predicts future motor development, sustained physical activity patterns, school-readiness outcomes or cardiometabolic trajectories.
• Sex- and age-specific models: exploring whether stratified models improve performance, especially in the narrow developmental window of preschool years where growth rates vary.
• Alternative predictors: integrating direct measures of body composition (bioelectrical impedance, DXA where feasible) or functional tests to refine predictive models and understand underlying mechanisms.
• Implementation science: evaluating feasibility, acceptability and cost-effectiveness of routine BSI screening in primary care and school settings, including training needs and referral mechanisms.

Policy and practice pathways

• Pilot implementation: health departments can run small pilots in selected clinics to assess logistical feasibility, train staff and generate local calibration data.
• Toolkit development: measurement guides, quick calculators (secure apps or spreadsheets) and training resources will facilitate standardized use.
• Integration with existing checks: the BSI is most useful where anthropometry is already collected; embedding it into routine electronic health records can automate flagging and support decision-making.
• Equity considerations: ensure follow-up supports are accessible to families irrespective of income or language; screening without access to interventions risks widening disparities.

FAQ

Q: What does the BSI add beyond BMI? A: The BSI combines height and weight with waist circumference and pelvic width. Waist circumference captures central adiposity and pelvic width captures skeletal breadth—dimensions that BMI cannot distinguish because BMI reduces body form to a single mass-to-height ratio. In this study, those additional measures improved discrimination of a low integrated-fitness phenotype substantially.

Q: Is the BSI ready for clinical use? A: The BSI shows promise as a rapid screening tool, but it was derived and validated internally in one population. Clinics or schools interested in adopting it should pilot the index locally, ensure measurement protocols match those used in the study, and consider local recalibration. The index should be used for preliminary screening and referral to more detailed assessment or interventions—not as a definitive diagnostic.

Q: How do I measure pelvic width reliably? A: Pelvic width was measured according to the CPFS-preschool anthropometric protocol used in the study. That involves identifying consistent bony landmarks and measuring breadth with a sliding caliper or anthropometer. Staff require training and periodic retraining in landmark identification and measurement technique. If pelvic width cannot be measured reliably, programs may test alternative circumferential or breadth measures but should expect some loss of predictive precision.

Q: Does a higher BSI score mean a child is overweight or obese? A: No. The BSI is a predictor of belonging to a low-fitness phenotype, not a direct measure of adiposity or obesity. Higher waist circumference contributes positively to BSI, but height and weight contribute negatively in the model used. Interpretation should focus on the child’s risk of integrated low motor fitness and the need for assessment or activity support, not on weight labeling.

Q: Will the BSI identify every child with poor motor skills? A: No screening tool is perfect. At the study’s optimal cut-off, the BSI achieved about 78% sensitivity and 67% specificity—meaning it correctly identified about three-quarters of children in the low-fitness cluster and correctly excluded about two-thirds of children not in that cluster. It is designed to triage children for further assessment, not to replace formal motor competence testing.

Q: Can the BSI be used longitudinally to monitor change? A: The current evidence is cross-sectional. Using the BSI longitudinally to monitor change or response to intervention is conceptually attractive but requires longitudinal validation to confirm that changes in the index correspond to meaningful changes in motor competence and function.

Q: How does the BSI perform across sexes? A: In the Macao sample, the BSI performed similarly and robustly in both boys and girls (AUCs ≈ 0.87–0.88). The investigators noted that sex-specific models may refine performance further, and such analyses represent a logical next step.

Q: What are practical next steps for a preschool program that wants to use the BSI? A: Start with a pilot: train staff on the CPFS-preschool measurement protocol, collect triage data alongside existing checks, calculate BSI scores for a subset of children, and compare flags to a small-scale, standardized motor competence assessment. Use pilot findings to tune measurement technique, local cut-offs and referral pathways.

Q: Could technology simplify BSI use? A: Yes. Secure mobile apps or spreadsheets with built-in calculators can reduce calculation errors and speed triage decisions. Integration into electronic health records could automate flags and follow-up prompts. Any technological solution must enforce standardized units and clear measurement instructions.

Q: Does this study suggest policy changes? A: The study supports reconsidering reliance on BMI alone for early screening of motor-related fitness risk. Policymakers should consider piloting multidimensional anthropometric screening that includes waist circumference and, where feasible, pelvic breadth; but rollout should follow local validation and resource planning for follow-up services.

The Macao study translates a multidimensional view of early physical development into a pragmatic screening approach. It identifies a consistent pattern: integrated preschool fitness can be described through coherent phenotypes and linked to body shape in ways that single indicators like BMI fail to capture. The BSI provides a concrete starting point for screening, but responsible adoption requires standardization, validation and clear pathways to assessment and remediation for children who are flagged.