CLINICAL AND BIOLOGICAL PREDICTORS FOR COGNITIVE FRAILTY: A POPULATION PREDICTIVE MODEL

摘要

This study aims to create a population predictive model to gain a more in-depth understanding of the underlying biological mechanisms for cognitive frailty as currently defined by the International Consensus Group in 2013. Data were from the InCHIANTI study, collected at baseline from 1998–2000. This group is a representative sample (n=1,453) of a population of white European origin from two small towns in Tuscany, Italy. To build our model, we used biomarkers with implications for clinical research and practice; a total of 132 putative SNPs and 155 protein biomarkers were identified from a systematic review. We used a tree boosting model, Extreme Gradient Boosting (xgboost), a machine learning technique for supervised learning. We developed two predictive models with high accuracy, AUCs for Model I is 0.877 (95% CI 0.825–0.903) and 0.864 (95% CI 0.804–0.899) for Model II. Results provide clinical and biological evidence for the relationship between cognitive decline and physical frailty supporting findings of dysregulation across multiple systems, specifically depression, anticholinergic burden, inflammatory proteins, and elevated levels of circulating pro-inflammatory proteins (e.g., IL-6, TNF-alpha, IL-18, and IL-1-beta). Associated genetic variants that influenced the production of circulating proteins were also found to be predictive, specifically IL6 rs1800796, TNF rs1800629, IL-18 rs360722, and IL1-beta rs16944, thus suggesting that they are clinically relevant SNPs. The results from this study establish a foundation for an understanding of the underlying biological mechanisms for the relationship between cognitive decline and physical frailty.