Associations of Multidimensional Image-Derived Facial Skin Scores with Adiposity Measures in a Large Screening Cohort in Japan: A Cross-Sectional Study with Internal Temporal Evaluation

Background/Objectives: Proprietary, age-referenced, image-derived facial skin scores may contain adiposity-related information, but their validity is uncertain. We tested whether 14 scores added predictive information for body mass index (BMI), bioelectrical impedance analysis (BIA)-derived body fat percentage, and a proprietary BIA-derived visceral fat level score beyond age category and sex, after redness, and, exploratorily, after BMI. Methods: This study included 387,859 deduplicated account IDs from a Japanese screening program, a 95.5% nested subset of the previous report’s unique-account cohort. Gradient-boosting models were developed using April–July data and evaluated without retuning on August–October data. Performance was assessed with R2, mean absolute error (MAE), and paired bootstrap confidence intervals. Results: In the internal temporal evaluation cohort (159,726 account IDs), the facial skin-score panel provided incremental predictive information beyond age category and sex and beyond redness. After BMI inclusion, however, the additional explained variance was negligible (ΔR2 = 0.00165 for body fat percentage and 0.00017 for the visceral fat level score). Conclusions: The panel provided modest, order-dependent information in internal temporal evaluation but negligible additional information after BMI inclusion. These unvalidated proprietary scores are hypothesis-generating and do not support diagnostic or body-compartment-specific use.