Abstract

Objectives: Integration of Multimodal Data combining clinical evaluations and instrumental measurements (e.g., LC-OCT, HFUS, Cutometer®, skinFlex®) enhances the accuracy of predictive models for clinical scores. Significance of Variable Selection for identifying key parameters (VIPs) such as cellular metrics and collagen density is crucial for constructing effective predictive models. Inclusive Research Approach incorporating diverse ethnic groups ensures predictive models are representative and applicable to a wide range of populations. Introduction: Clinical evaluation of facial aging and instrumental measurements are complementary approaches. The goal is to propose a multivariate (chemometric) mathematical approach capable of constructing predictive models of clinical scores, establishing correlations, and explaining the appearance of clinical signs through modifications in the subsurface microstructures of the skin at structural, cellular, and biomechanical levels. Materials / method: Data were collected from 300 healthy female volunteers (Caucasian, Asian, Afro-American). More than 200 quantitative features, including clinical scores and metrics derived from 3D LC-OCT imaging, HFUS, Cutometer®, and skinFlex®, were analyzed. To account for possible non-linearity, each original variable was transformed into four splines, then processed through Sparse-Partial Least Square for variable selection using phototypes as dummy variables. The performance of prediction was assessed through cross-validation using 50 iterations, each using different training and test sets. Results: Predicting clinical scores using S-PLS yielded Root Mean Square Error Prediction between 1.4 and 1.8, underscoring the importance of integrating metrics from multiple techniques and the correlation between clinical and instrumental features. Variable Importance in the Projection identified key parameters. Cellular metrics from LC-OCT and collagen density from HFUS were as crucial as elasticity, tonicity, and firmness measured by Cutometer® and skinFlex®. The skinFlex®-LC-OCT block was predominant (73% of top 100 VIPs), while HFUS with biomechanical measurements represented 2%. Conclusion: Predicting clinical scores using multivariate analysis (S-PLS) provides a unique perspective on the links between visible features of aging and instrumental measurements while the approach integrates the notion of ethnic groups for an inclusive study of facial aging mechanisms.