Machine-Learning Approach to Identify Tissue Inhibitors of Metalloproteinases (TIMP) and Clinical Variables Predicting Executive Phenotypes in HIV-infected Adults.

Background: Despite effective combination antiretroviral therapy (cART), executive function impairment remains prevalent among people living with HIV (PLWH). The pathogenesis of HIV-associated neurocognitive disorder is multifactorial, involving chronic immune activation, metabolic alterations, and neuroinflammatory processes. Matrix metalloproteinases and their tissue inhibitors (TIMPs) contribute to neuroinflammation and blood-brain barrier disruption, but their relationship with executive dysfunction in HIV remains unclear. Methods: We enrolled 169 middle-aged Taiwanese PLWH who underwent comprehensive clinical and laboratory assessments. Executive function was evaluated using Wisconsin Card Sorting Test, and participants were classified into executive impaired and unimpaired group. Six variables-years since HIV diagnosis, BMI classification, nadir CD4 count, and plasma levels of TIMP-1, TIMP-2, and TIMP-4-were used to develop predictive models. Logistic regression, support vector machine, random forest, and Extreme Gradient Boosting (XGBoost) algorithms were applied. Model performance was evaluated using the area under the receiver operating characteristic curve with 10-fold cross-validation, and SHAP analysis was used to interpret feature contributions. Results: Executive impairment was identified in 9.5% of participants. Among single predictors, TIMP-1 demonstrated the highest discriminative performance (AUC = 0.76 by XGBoost), whereas TIMP-2 and TIMP-4 showed lower predictive performance. Combining clinical and laboratory variables markedly improved model performance, with XGBoost achieving an AUC of 0.89. Ten-fold cross-validation yielded a mean AUC of 0.77, supporting model robustness. Conclusions: Executive dysfunction remains an important neurocognitive phenotype in middle-aged PLWH. Integrating clinical and biological variables through machine-learning supports early identification of individuals at risk and may facilitate timely neuropsychological evaluation and intervention.