BIOLOGICAL RISK FACTORS FOR DEMENTIA AND COGNITIVE FUNCTION AMONG OLDER INDIANS: FINDINGS FROM LASIDAD

Abstract

Abstract A number of clinical and exam-generated biomarkers have been associated with dementia. We assess the relative importance of 41 biomarkers using a novel dataset from an understudied population – a national sample of older Indians. The value of our data includes the biomarker extensiveness, the validated classification of dementia, and the relatively lower average education of the population. We use both traditional social science methods based on biological theories and agnostic machine learning algorithms to examine how biomarkers explain variance in dementia diagnosis and cognitive functioning. Comparing different approaches shows how to best characterize the influence of biology and how to trim and combine biomarkers. The six approaches used in our study include: (1) 41 individual biomarkers; (2) identification of subsets of biomarkers with elastic net; (3) support vector machine learning; (4) factor analysis; (5) principal component analysis; and (6) factor classification based on a theoretical approach. Preliminary results show that all the biomarkers or a reduced set of biomarkers identified by elastic net do the best job at explaining variability in dementia outcome, but the biomarkers chosen as most important by elastic net do not match well our understanding of biological mechanisms. Traditional social science approaches (e.g. factor analysis and principal components approach) provide better understanding and interpretation of the relative importance of biological systems as well as the association between biomarkers and cognition. These results are informative for others collecting and analyzing biomarker data in population samples.