AbstractBackgroundMany clinical biomarkers have been associated with dementia. This paper uses a novel data set from an understudied population – a national sample of older Indians ‐ to assess the relative importance of 41 biomarkers using a variety of approaches. The value of these data is the extensiveness of the biomarker collection, the validated classification of dementia, and the uniqueness of the population with low average education. Many approaches to both summarizing and reducing data complexity have been suggested. Some of these are based on theory about biological mechanisms and some are agnostic and based on machine‐learning approaches. The results are informative for others collecting and analyzing biomarker data in population samples.MethodWe use 6 approaches to assessing the value of the biomarkers in explaining variance in dementia diagnosis and cognitive functioning – some traditional social science approaches and some based on machine learning. The aim of the different approaches is to determine how to best characterize the influence of biology as well as how to trim and combine the data. The five approaches to measuring biological risk: 41 individual biomarkers; identification of subsets of biomarkers with elastic net; support vector machine learning; factors developed with factor analysis; a principal components approach; and a factor classification based on a theoretical approach.ResultsResults indicate if the aim is explain variance in dementia, all the biomarkers or a reduced set of biomarkers identified by elastic net do the best job at explaining variability. However, biological markers chosen as most important do not fit well our understanding of biological mechanisms.ConclusionIf the aim is understanding the relative importance of biological systems, traditional social science approaches, e.g. the factor analysis and principal components approach, allow better understanding and interpretation of the biomarkers associated with cognitive performance.
Read full abstract