Beyond BMI: Linear Growth Stunting as a Predictor of Atherogenic Dyslipidemia in Pediatric Populations

Introduction

Pediatric metabolic health screening is structurally limited by a clinical dualism that segregates undernutrition—marked by linear growth stunting (height-for-age Z-score (HAZ) < -2 SD)-from overnutrition, characterized by obesity (BMI-for-age Z-score (BAZ) > +2 SD) [1,2]. While clinical guidelines focus almost exclusively on the overnutrition spectrum to mitigate adult cardiovascular disease (CVD), emerging evidence indicates that chronic undernutrition triggers severe metabolic and endocrine adaptations that disrupt the lipid profile [1,2]. In pediatric obesity, lipid dysregulation is driven by caloric excess, adipocyte hypertrophy, and chronic low-grade inflammation, which collectively induce systemic insulin resistance and yield the classic "atherogenic triad" of high triglycerides (TG), low High-density lipoprotein cholesterol (HDL-C), and borderline low-density lipoprotein cholesterol (LDL-C) [3-5]. Conversely, dyslipidemia in stunted children stems from bioenergetic adaptations designed to preserve central life processes during caloric deprivation at the expense of linear growth, which suppresses the growth hormone-insulin-like growth factor (GH-IGF) axis, downregulates hepatic low-density lipoprotein receptors (LDL-R), and reduces lipoprotein lipase (LPL) activity, resulting in profound HDL-C depression and hypertriglyceridemia [1,6].

This investigation is conceptually framed within the Developmental Origins of Health and Disease (DOHaD) hypothesis, which suggests that early-life nutritional deprivation may induce permanent endocrine and metabolic reprogramming [1,7]. While these adaptations optimize short-term survival in resource-poor environments, they restrict hepatic lipid clearance, predisposing individuals to metabolic syndrome and accelerated CVD when exposed to rapid nutritional transitions later in life [1,7]. Current cardiovascular screening guidelines from the National Heart, Lung, and Blood Institute (NHLBI) and the American Academy of Pediatrics (AAP) rely primarily on adiposity and family history, systematically overlooking the high-risk metabolic configurations embedded within stunted growth phenotypes [2,9]. To date, a knowledge gap remains in pediatric preventative cardiology: the comparative prevalence of dyslipidemia between stunted and obese children has not been directly evaluated using identical, standardized thresholds within a single matched cohort, and the multivariate clustering of lipid profiles in undernourished children remains unmapped [1,2,10]. Utilizing existing institutional medical records offers a valuable, non-invasive opportunity to analyze these established metabolic profiles across distinct growth phenotypes.

The primary aim of this hospital-based retrospective study is to establish an integrated, comparative metabolic framework to evaluate undernutrition- and overnutrition-related cardiovascular risk by analyzing historical laboratory data from parallel cohorts of stunted and overweight/obese pediatric patients. Specifically, the study will:

1. Quantify and Compare Prevalence: Extract and compare the prevalence of dyslipidemia between cohorts using standardized National Heart, Lung, and Blood Institute (NHLBI) pediatric thresholds.

2. Model Lipid Phenotypes: Apply unsupervised machine learning algorithms (K-means and agglomerative hierarchical clustering) to multivariate lipid profiles—incorporating TG, total cholesterol (TC), HDL-C, LDL-C, non-HDL-C, and the TG/HDL-C ratio—to identify distinct, growth-associated lipid phenotypes.

3. Evaluate Diagnostic Sensitivity: Assess the diagnostic sensitivity of current weight-centric screening guidelines in detecting atherogenic dyslipidemia within stunted populations.

Ultimately, this research aims to evaluate chronic linear growth retardation as a potential cardiovascular risk factor.

Materials & Methods

This investigation was executed as a hospital-based, quantitative, retrospective comparative cohort study over a two-year period, utilizing existing clinical data archives. The study setting comprised the Department of Pediatrics at a tertiary care teaching hospital, where we leveraged comprehensive data repositories from both the Outpatient Department (OPD) and Inpatient Department (IPD). Methodologically, we systematically extracted clinical data from the institutional Electronic Health Records (EHR) and Laboratory Information System (LIS) archives. This dual-source data pipeline captured concurrent anthropometric measurements and serum lipid panels, which enabled us to perform retrospective cohort matching and execute statistical modeling to contrast metabolic profiles between distinct pediatric growth phenotypes.

Study Population

The study population comprised pediatric patients aged 2 to 19 years who received clinical care within the hospital network during the designated retrospective timeframe. To qualify for inclusion in the analytical database, each patient record required concurrent documentation of both a comprehensive fasting serum lipid profile and complete anthropometric measurements. Individuals outside this age spectrum or those with incomplete diagnostic entries were excluded from the final population to maintain data integrity for subsequent metabolic and growth phenotype comparative analyses.

Sample Size

The minimum sample size was determined using a two-sided power calculation for comparing two independent proportions. The parameters were set to assume a 1:1 allocation ratio between the cohorts, a statistical power (1 - β) of 80%, and a significance level (α) of 0.05. Based on ancestral clinical literature, the expected baseline prevalence of dyslipidemia was established at 25% (p1 = 0.25) in the overweight/obese cohort and projected at an estimated 45% (p2 = 0.45) in the stunted cohort. Applying Fleiss’s formula with continuity correction, the statistical calculation mandated a minimum sample size of 93 patients per cohort, yielding a total study population of N = 186 valid patient records.

Sampling Procedure

A purposive, non-probability consecutive sampling strategy was deployed to audit archived hospital medical records. Electronic records and laboratory entries that fulfilled all pre-specified eligibility criteria were extracted sequentially from the institutional database until the targeted sample size was achieved.

Eligibility Criteria

a. Case Definition

• Stunted Cohort (Undernutrition Spectrum): Defined by the presence of chronic linear growth failure, structurally characterized by a World Health Organization (WHO) height-for-age Z-score (HAZ) of less than -2 standard deviations (SD) below the international reference median [1].

• Overweight/Obese Cohort (Overnutrition Spectrum): Defined by the presence of excess adiposity, structurally characterized by a WHO BMI-for-age Z-score (BAZ) exceeding +1 SD for overweight status or exceeding +2 SD for obesity relative to the international reference median [2].

b. Inclusion Criteria

• Patient age ≥ 2 years and ≤19 years at the precise time of the initial clinical data logging.

• Availability of complete validated institutional anthropometric records, including absolute height, weight, age, and biological sex.

• Availability of complete, concurrent fasting serum lipid panel results registered within the LIS, containing absolute values for TG, TC, HDL-C, and LDL-C.

c. Exclusion Criteria

• Documented clinical history or laboratory confirmation of secondary causes of dyslipidemia, including type 1 or type 2 diabetes mellitus, chronic kidney disease, nephrotic syndrome, hypothyroidism, or confirmed familial hypercholesterolemia variants.

• Concurrent or recent systemic treatment with lipid-altering pharmacotherapies, systemic corticosteroids, or atypical antipsychotics known to metabolically distort serum lipid parameters.

• Documentation of mixed or overlapping malnutrition profiles that would confound the distinct cohorts, specifically including patients presenting with concurrent severe wasting (BAZ < -3 SD) and stunting, or patients presenting with simultaneous stunting (HAZ < -2 SD) and obesity (BAZ > +2 SD).

d. Details of Control Group

Because this investigation utilized a retrospective, analytical comparative cohort design, a healthy, normolipidemic control group was not employed. Instead, the overweight and obese pediatric cohort served as the active, well-characterized clinical comparative arm against the exposed stunted cohort. This design directly facilitated the head-to-head comparison of prevalence rates and enabled the computational isolation of divergent multivariate lipid phenotypes across the two nutritional extremes [1,2].

Study Variables

a. Independent Variables

• Anthropometric Status: HAZ and BAZ were utilized as the primary independent variables to define the cohorts, calculated retrospectively via WHO AnthroPlus software.

• Demographic Factors: Patient age, biological sex, and documented institutional socio-demographic indicators were extracted to serve as baseline covariates.

b. Dependent Variables

• Continuous Metabolic Parameters: Quantifiable biochemical markers extracted from patient records included serum TG, TC, HDL-C, LDL-C, non-HDL-C (TC - HDL-C), and the calculated TG/HDL-C ratio.

• Categorical Outcomes: Binary outcomes comprised the presence or absence of pediatric dyslipidemia as classified by the NHLBI thresholds. Multi-class categorical outcomes consisted of the specific growth-associated lipid phenotype cluster assignments generated by the unsupervised machine learning models.

Laboratory Investigations - Parameters and Procedures

Because this investigation was designed as a retrospective database study, no prospective biological samples were drawn, and no direct experimental laboratory manipulations were performed. All biochemical metrics were retrieved as historical data fields from the institutional Laboratory Information System (LIS) database. The analyzed parameters relied on standardized, validated analytical procedures previously executed by the hospital’s central pathology unit. TC and TG were quantified using automated enzymatic colorimetric methods via the cholesterol oxidase/peroxidase and glycerol phosphate oxidase pathways, respectively. HDL-C was determined through a direct, homogenous enzymatic assay. LDL-C was primarily calculated using the Friedewald equation (LDL-C = TC - HDL-C – TG/5) for records with TG < 400 mg/dL. While the Friedewald equation is limited in states of severe hypertriglyceridemia, we maintained high analytical accuracy in our most severe metabolic outliers by utilizing direct homogenous enzymatic assays for all patient records presenting with serum TG ≥ 400 mg/dL.

Data extraction was executed using an encrypted, structured digital proforma developed in Microsoft Excel. The extraction architecture captured unique, anonymized alphanumeric patient identifiers, alongside chronological age, biological sex, absolute weight, and absolute height. Derived data fields included calculated anthropometric Z-scores (HAZ and BAZ) and absolute biochemical values for TG, TC, HDL-C, LDL-C, non-HDL-C, and the TG/HDL-C ratio.

Statistical Analysis

Statistical analyses were performed using SPSS 26.0 and Python 3.10. Normality was assessed via the Shapiro-Wilk test; normally distributed data are presented as mean ± SD, and skewed data as median (IQR). Group proportions were compared using Chi-square or Fisher’s exact tests. Lipid profiles were analyzed using K-means and agglomerative hierarchical clustering on min-max normalized data, with the optimal cluster number (k) determined by Silhouette coefficients and the Elbow method. Cluster characteristics were compared using one-way ANOVA or the Kruskal-Wallis test. Multivariable logistic regression calculated adjusted odds ratios (aOR) for stunting (HAZ) relative to atherogenic phenotypes, controlling for confounders. Significance was defined as p < 0.05.

Results

An analysis of the demographic and anthropometric data (N = 214) revealed highly significant differences between the stunted (HAZ < -2SD), n = 107 and overweight/obese (BAZ > +1 SD, n = 107) cohorts. The overweight/obese cohort was significantly older than the stunted cohort (11.3 ±2.9 vs. 8.5 ±3.2 years; p < 0.001), suggesting age-related shifts in malnutrition phenotypes. While biological sex distribution remained statistically uniform across groups (p = 0.785), SES demonstrated a stark, highly significant divergence (p < 0.001). Specifically, a vast majority of the stunted cohort belonged to households Below the Poverty Line (BPL; 84.1%), whereas the overweight/obese cohort was predominantly comprised of families Above the Poverty Line (APL; 83.2%). Expectedly, baseline anthropometric metrics confirmed profound deviations between the two groups (p < 0.001), with the stunted cohort presenting severe linear growth deficits (HAZ = -2.59 ±0.40) and the overweight/obese cohort exhibiting severe adiposity (BAZ = +2.45 ±0.4). Taken together, these findings highlight a clear socioeconomic and age-based polarization within this population's double burden of malnutrition (Table 1).

Comparative lipid analysis revealed an atherogenic phenotype dominant in the overweight/obese cohort alongside distinct hypoalphalipoproteinemia (low HDL) vulnerability in the stunted cohort. The overweight/obese group displayed widespread lipid accumulation, presenting significantly elevated levels of total cholesterol (188.6 ±26.1 vs. 148.2 ±18.9 mg/dL; p < 0.001), LDL-C (118.9 ±21.8 vs. 89.4 ±16.1 mg/dL; p < 0.001), and non-HDL-C (147.3 ±24 vs. 114.7 ±18.3mg/dL; p < 0.001), which culminated in a higher overall dyslipidemia burden (72.9% vs. 56.1%; p = 0.01). This overnutrition-driven profile was further marked by hypertriglyceridemia, with higher median triglycerides (142.0 vs. 126.5mg/dL; p = 0.01) and a higher prevalence of elevated triglycerides (78.5% vs. 66.4%; p = 0.048). Conversely, the stunted cohort exhibited an inverted risk profile; despite maintaining lower absolute atherogenic fractions and a higher median TG/HDL-C ratio (3.78 vs. 3.44; p = 0.035), they demonstrated a critical systemic deficit in cardioprotective lipoproteins, with a vastly disproportionate prevalence of critically low HDL-C (76.6% vs. 44.9%; p < 0.001) notwithstanding their lower absolute mean HDL-C value (33.5 ±4.1 vs. 41.3 ±5.9 mg/dL; p < 0.001) (Table 2).

Cluster analysis partitioned the study population into three highly distinct metabolic phenotypes (p < 0.001), revealing a stark alignment between nutritional status and dyslipidemic profiles. While Cluster 1 represented a mild/normolipidemic baseline (n = 60), the remaining cohorts exposed a striking double burden of cardiovascular risk driven by distinct pathophysiologies. Cluster 2 (Obese Combined Atherogenic; n = 70) captured a classic hyperlipidemic state composed entirely of overweight/obese individuals with severe adiposity (BAZ = +2.59 ±0.33), exhibiting peak concentrations of atherogenic fractions including TC (201.8 ±20.9mg/dL), LDL-C (128.1 ±18.2mg/dL), and TG (168.9 ±31.8mg/dL) to produce the highest overall risk (TG/HDL-C} = 4.23 ±0.85). Crucially, Cluster 3 (Stunted Low-HDL Phenotype; n = 84) demonstrated that severely stunted individuals (HAZ = -2.65 ±0.38) face a parallel, severe cardiovascular threat; despite lower absolute atherogenic fractions, this undernutrition phenotype was uniquely characterized by severe hypoalphalipoproteinemia with the lowest systemic HDL-C levels (32.9 ±3.4mg/dL), driving an elevated TG/HDL-C ratio (4.01 ±0.72) that directly rivaled the obese cluster (Table 3).

Multivariable logistic regression analysis demonstrated that linear stunting status (HAZ < -2) is a profound, independent driver of critical low-HDL susceptibility, remaining highly significant even after controlling for demographic and socioeconomic confounders. Individuals experiencing linear stunting exhibited a nearly fourfold increase in the likelihood of presenting with severe hypoalphalipoproteinemia (aOR = 3.85, 95%CI: 2.12- 7.01; Wald chi² = 21.43, p < 0.001). Conversely, phenotypic expressions of this metabolic risk were not significantly influenced by demographic or structural gradients, as chronological age (OR = 1.03,95%CI:0.92--1.15; p = 0.571), biological sex (aOR = 0.89, 95% CI:0.51-1.55;p = 0.689), and socioeconomic status (aOR = 1.3,95%CI:0.65-2.62;p = 0.475) failed to demonstrate independent predictive value. These findings isolate linear growth restriction as the primary clinical marker for this specific lipoprotein deficit, independent of age, sex, or poverty lines (Table 4).

Discussion

The coexistence of undernutrition and overnutrition, known as the double burden of malnutrition, represents a critical epidemiological challenge in transitioning societies. Analysis of the study population (N = 214) reveals a demographic polarization where younger, impoverished children (84.1% BPL) are more prone to stunting, while older, more affluent children (83.2% APL) exhibit higher rates of overweight and obesity. This reflects a broader shift where, while higher wealth initially correlates with pediatric adiposity, late-stage transitions often see obesity burdens migrate toward lower socioeconomic groups due to the availability of inexpensive, energy-dense foods [10,11]. Sex distribution remains uniform across both cohorts, aligning with national-level findings such as the Comprehensive National Nutrition Survey (CNNS) of India [12].

Cardiovascular risk pathways are starkly divergent between these groups. The overweight/obese cohort presents with a classic atherogenic phenotype, characterized by elevated total cholesterol, LDL-C, and triglycerides, consistent with established pediatric metabolic syndrome guidelines [13]. In contrast, the stunted cohort exhibits "metabolic obesity," characterized by severe, systemic hypoalphalipoproteinemia (HDL-C < 40 mg/dL in 76.6% of the group) and an elevated TG/HDL-C ratio—a highly sensitive marker for insulin resistance and cardiovascular risk [1,12,13]. This "silent" dyslipidemia in undernourished children is often overlooked by BMI-centric screening protocols that prioritize overweight populations [2].

Machine learning (k-means clustering) supports these findings by identifying a distinct "Stunted Low-HDL Phenotype" that demonstrates a lipid profile comparable to that of obese individuals [14]. This metabolic pathology is not merely an acute response to diet but a consequence of DOHaD, where early-life nutritional deprivation—characterized by reduced IGF-1 and altered cortisol levels—impairs hepatic protein synthesis and reverse cholesterol transport [15-17]. Experimental models confirm that chronic protein-energy restriction disrupts the functional assembly of cardioprotective HDL particles, a condition that may be further exacerbated by rapid, carbohydrate-driven catch-up growth [6,18].

Multivariable logistic regression identifies a clinical association between linear stunting (HAZ < -2) and low-HDL susceptibility (aOR = 3.85; 95% CI: 2.12–7.01), independent of age, sex, or socioeconomic status. While these findings suggest a potential linkage between early-life stunting and long-term cardiometabolic risk, the retrospective design limits our ability to establish definitive temporal causality, requiring future longitudinal validation [7,10,12,18,19].

Limitations

Several limitations warrant consideration. First, the retrospective design identifies clinical associations rather than definitive temporal causality, necessitating future longitudinal research to map the trajectory of these metabolic phenotypes. Second, while the Friedewald equation was used for most lipid calculations, it has recognized limitations in states of severe hypertriglyceridemia; to mitigate this, we employed direct enzymatic assays for all patients with TG ≥ 400 mg/dL to ensure analytical precision. Third, unmeasured factors—such as dietary macronutrient composition, specific micronutrient status, and physical activity—were not captured in this retrospective EHR-based design. Finally, the single-center nature of the data collection may limit the absolute generalizability of the findings, requiring cautious interpretation.

Future Directions and Mechanistic Hypotheses

We hypothesize that the "Stunted Low-HDL Phenotype" is driven by a state of hepatic metabolic inflexibility, where early-life nutritional deprivation induces stable epigenetic modifications in genes governing reverse cholesterol transport and apolipoprotein synthesis, effectively "programming" the liver to favor storage over excretion even when systemic energy availability later increases. This leads to a crucial research question: Do specific DNA methylation patterns in hepatic lipid-metabolism genes—such as those involved in ApoA-I or LCAT expression—serve as stable biomarkers for future atherosclerotic risk in stunted children, and can these markers be reversed by targeted, micronutrient-dense nutritional re-priming? Furthermore, future studies must investigate whether this stunted metabolic phenotype is permanently fixed or if it exhibits therapeutic plasticity during specific developmental windows. If proven, these implications would shift the pediatric clinical standard from reactive, weight-based management to proactive, biomarker-driven metabolic interventions, potentially preventing the looming epidemic of early-onset cardiovascular disease in post-stunting populations by targeting the underlying hepatic pathophysiology rather than merely treating the symptoms of metabolic syndrome.

Conclusion

This study demonstrates a clinical association between pediatric undernutrition and metabolic profiles potentially linked to cardiovascular risk. Through unsupervised machine learning, we identified a distinct 'Stunted Low-HDL Phenotype' characterized by severe hypoalphalipoproteinemia and elevated TG/HDL-C ratios, independent of BMI and socioeconomic status. Given these findings, we suggest that linear growth retardation (HAZ < -2 SD) warrants further evaluation as an indication for targeted pediatric lipid screening. Public health interventions should prioritize shifting from simple calorie-dense supplementation toward balanced, micronutrient-dense strategies to mitigate long-term metabolic programming and prevent future non-communicable disease burdens.

Declaration

Ethical Considerations

This study posed negligible risk, as it utilized a retrospective review of de-identified electronic health records (EHR) without patient contact or experimental interventions. Data confidentiality was ensured by removing all direct identifiers and assigning anonymous alphanumeric keys to each record prior to analysis.

Data Availability

The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.

Conflicts of Interest

The authors declare that they have no competing interests.

Funding Statement

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Acknowledgments

None

Author Contributions

All authors contributed equally to the conceptualization, data collection, drafting of the manuscript, critical revision, and final approval of the version to be published.