Enhancing Cognitive Reserve Measurement with High-density lipoprotein as Biological Proxy through Machine Learning models: A Validation Study

Background: The role of cognitive reserve (CR) in the onset and course of schizophrenia spectrum disorders (SSD) is gaining ground. However, its estimation relies on heterogeneous proxies with limitations. The inclusion of biological measures, such as HDL levels, which are related to cognitive performance, holds promise to refine the assessment of CR. To validate that a CR measure comprising years of education and HDL levels offers a better-balanced explanatory power than prior CR metrics. Methods: We assessed 378 FEP patients and 149 healthy controls for sociodemographic, clinical, and neurocognitive data. The sample was classified into CR1 (years of education, occupation, and premorbid IQ) and CR2 (years of education and HDL levels). This classification enables comparative evaluations of both indices using machine learning models. Results: CR2 achieved an AUC of 0.725 (95% CI: 0.616-0.834) for FEP patients using Support Vector Machine. In comparison to CR1, CR2 achieved superior performance scores in metrics: accuracy (CR1: 62%; CR2: 66%), F-1 score (CR1: 59%; CR2: 69%), sensitivity (CR1: 54%; CR2: 70%), precision (CR1: 66%; CR2: 69%), and revealed an ROC curve characterized by higher discriminative capacity. Conclusions: Compared to literature-based CR index, biological approach CR index shows higher reliability and validity, as well as a more balanced relationship between sensitivity and specificity in data classification. This underlines that incorporating HDL levels into CR estimation improves accuracy of measurement.