A data-driven prospective study of dementia among older adults in the United States

Abstract

Background Studies examining risk factors for dementia have typically focused on testing a priori hypotheses within specific risk factor domains, leaving unanswered the question of what risk factors across broad and diverse research fields may be most important to predicting dementia. We examined the relative importance of 65 sociodemographic, early-life, economic, health and behavioral, social, and genetic risk factors across the life course in predicting incident dementia and how these rankings may vary across racial/ethnic (non-Hispanic white and black) and gender (men and women) groups. Methods and findings We conducted a prospective analysis of dementia and its association with 65 risk factors in a sample of 7,908 adults aged 51 years and older from the nationally representative USbased Health and Retirement Study. We used traditional survival analysis methods (Fine and Gray models) and a data-driven approach (random survival forests for competing risks) which allowed us to account for the semi-competing risk of death with up to 14 years of follow-up. Overall, the top five predictors across all groups were lower education, loneliness, lower wealth and income, and lower self-reported health. However, we observed variation in the leading predictors of dementia across racial/ethnic and gender groups such that at most four risk factors were consistently observed in the top ten predictors across the four demographic strata (non-Hispanic white men, non-Hispanic white women, non-Hispanic black men, non-Hispanic black women). Conclusions We identified leading risk factors across racial/ethnic and gender groups that predict incident dementia over a 14-year period among a nationally representative sample of US aged 51 years and older. Our ranked lists may be useful for guiding future observational and

Improved Alzheimer’s Disease Detection by MRI Using Multimodal Machine Learning Algorithms

“…Most health informatics researchers prefer data-centric machine learning approaches in the diagnosis of early-stage AD [35][36][37]. In data-centric approaches, data are systematically changed or preprocessed for the datasets for enhancing the performance of ML models.…”

Mapping of machine learning approaches for description, prediction, and causal inference in the social and health sciences

“…Death as a competing risk needs to be accounted for in investigations on aging-associated diseases, e.g., with random survival forests, which also allow modeling of time-varying risk factors ( 41 ). If a focus is on the short-term consequences of time-varying treatment, i.e., if causal conclusions are intended, then it may be better to use established methods in the potential outcomes framework such as marginal structural models or the g-formula ( 42 ).…”

“…BART is also an ensemble learning algorithm; however, they are often used for causal research questions as relevant software packages offer numerous settings for causal analysis based on domain knowledge ( 55 , 56 ) and will be presented in the “ML for causal inference” section. First, because of their interpretability, ensemble methods such as Random Forests ( 57 ), have been used in numerous studies in the social and health sciences such as ( 41 , 58 ).…”

“…Possible improvements of model performance should be tested by building additional models, adding more information (variables) to the model, or by accounting for higher-order interactions or nonlinear relationships. Typical research questions have been to test improved predictive accuracy of ML compared to traditional modeling, e.g., regarding the social determinants of health ( 58 ), the prediction of dementia ( 41 ), and diabetes and diabetes complications ( 69 ), with ML approaches typically not markedly outperforming traditional regression-based modeling. Other studies have however shown that, for instance, estimation of remaining life expectancy with a long short-term memory recurrent neural network using electronic medical records outperformed human (doctor) estimations; remaining life expectancy is a relevant patient-oriented measure for timely start of advance care (end-of-life) planning ( 70 ).…”

“…An overview of different papers that use ML for inverse probability weighting and propensity score matching is given in ( 102 ). Particularly in aging research, we recommend systematically assessing bias coming from selective attrition and competing risk of death, e.g., with random survival forests ( 119 ) that have been applied in studies on dementia ( 41 ).…”