مقالات پذیرفته شده در نهمین کنگره بین المللی زیست پزشکی
Association of Non-Alcoholic Fatty Liver Disease (NAFLD) and Hypothyroidism in Iranian Adolescents: The Potential Role of Artificial Intelligence in Diagnosis
Association of Non-Alcoholic Fatty Liver Disease (NAFLD) and Hypothyroidism in Iranian Adolescents: The Potential Role of Artificial Intelligence in Diagnosis
Introduction: Non-alcoholic fatty liver disease (NAFLD) is increasingly recognized as a major pediatric health challenge worldwide, with a reported prevalence of up to 36.1% in obese children. Its pathogenesis is complex, involving genetic predispositions, insulin resistance, and obesity-related metabolic disturbances. NAFLD is not only a hepatic disorder but also a condition linked to increased cardiovascular risk, insulin resistance, and long-term morbidity.
Thyroid hormones play a central role in regulating lipid and glucose metabolism. Consequently, thyroid dysfunction, particularly subclinical hypothyroidism (SH), may contribute to the development and progression of NAFLD. Several studies in adults have highlighted this relationship, yet data in children and adolescents remain limited, particularly in Middle Eastern populations with a high burden of obesity.
The present cross-sectional study aimed to investigate the association between NAFLD and thyroid dysfunction in obese Iranian children and adolescents. In addition to biochemical and imaging evaluations, we also considered the potential role of artificial intelligence (AI) tools to enhance risk prediction and diagnostic accuracy, given the multifactorial nature of these disorders.
Methods: A total of 155 obese children and adolescents (age range: 6–18 years) were enrolled in this cross-sectional study. Participants with a history of alcohol consumption, chronic liver diseases of other etiologies, or ongoing hepatotoxic medication use were excluded.
Clinical and anthropometric data included age, sex, blood pressure (systolic and diastolic standardized scores), and BMI standard deviation score (BMI-SDS).
Biochemical analyses involved fasting blood samples to measure thyroid-stimulating hormone (TSH), free triiodothyronine (fT3), free thyroxine (fT4), alanine aminotransferase (ALT), aspartate aminotransferase (AST), fasting glucose, lipid profile, and insulin resistance using the homeostasis model assessment (HOMA-IR). Subclinical hypothyroidism (SH) was defined as elevated TSH with normal fT4 levels.
Imaging was performed using abdominal ultrasonography to diagnose and grade hepatic steatosis.
Genetic analysis of PNPLA3 and TM6SF2 variants was also included to assess their modifying role on thyroid-liver interactions.
Artificial intelligence component: Anthropometric, biochemical, and genetic data were fed into a machine learning pipeline (random forest, logistic regression, and support vector machines) to explore predictive models for NAFLD and thyroid dysfunction.
Results: Of the 155 participants, 77 (49.7%) had NAFLD confirmed by ultrasonography, while 78 (50.3%) were classified as obese without NAFLD.
Thyroid function: Mean TSH levels were significantly higher in NAFLD patients compared with controls (4.25 ± 1.57 vs. 2.12 ± 1.45 µUI/ml, p < 0.0001). The prevalence of SH was also greater in NAFLD patients (15.7%) compared to non-NAFLD (7.1%, p < 0.001). No significant differences were observed in fT3 and fT4 between groups.
Liver enzymes and metabolic markers: ALT, AST, and HOMA-IR were significantly elevated in NAFLD patients (p < 0.001 for all), indicating hepatic and metabolic involvement. Triglycerides were also higher in the NAFLD group (102.5 ± 48.7 vs. 83.9 ± 44.4 mg/dL, p < 0.001).
Obesity and comorbidities: Children with combined obesity, SH, and NAFLD demonstrated the highest BMI-SDS and HOMA-IR values, as well as longer duration of obesity.
Genetic associations: TM6SF2 genotype showed an inverse correlation with TSH levels, but this association was significant only in the NAFLD group (p = 0.02). PNPLA3 polymorphisms were not significantly associated with thyroid parameters.
AI model performance: The random forest model outperformed other algorithms, achieving an AUC of 0.87, sensitivity of 78%, and specificity of 84% in predicting NAFLD based on combined thyroid, metabolic, and anthropometric features. TSH, BMI-SDS, and triglycerides were identified as the most influential predictors.
Conclusion: Our study demonstrates a significant association between NAFLD and elevated TSH levels, as well as an increased prevalence of subclinical hypothyroidism among obese Iranian adolescents. The coexistence of SH and NAFLD was linked with more severe obesity, higher insulin resistance, and longer duration of obesity, suggesting a synergistic effect of thyroid dysfunction and hepatic steatosis on metabolic risk.
Genetic findings further indicate that TM6SF2 variants may modulate thyroid function in the context of NAFLD, although this requires validation in larger cohorts.
Importantly, our AI-assisted approach highlighted the potential of machine learning models to enhance early risk detection of NAFLD using simple, routinely available clinical and biochemical data. Such models may provide valuable support for clinicians in resource-limited settings by identifying high-risk adolescents who may benefit from early intervention.
In conclusion, the interplay between thyroid function and NAFLD in children warrants routine screening, particularly in obese populations. Integrating traditional clinical evaluation with AI-based prediction tools offers a promising pathway toward personalized prevention and treatment strategies in pediatric metabolic and liver disorders.