UTILITY OF ENVIRONMENTAL COMPLEXITY AS A PREDICTOR OF ALZHEIMER’S DISEASE DIAGNOSIS: A BIG-DATA MACHINE LEARNING APPROACH
M. Yuan, K.M. Kennedy
J Prev Alz Dis 2023;2(10):223-235
Background: Rural-urban differences and spatial navigation deficits have received much attention in Alzheimer’s Disease research. While individual environmental and neighborhood factors have been independently investigated, their integrative, multifactorial effects on Alzheimer’s diagnosis have not. Here we explore this “environmental complexity” for predictive power in classifying Alzheimer’s from cognitively-normal status.
Methods: We utilized data from the National Alzheimer’s Coordinating Center (NACC) uniform data set containing annual visits since 2005 and selected individuals with multiple visits and who remained in their zipcode (N = 22,553). We georeferenced each subject with 3-digit zipcodes of their residences since entering the program. We calculated environmental complexity measures using geospatial tools from street networks and landmarks for spatial navigation in subjects’ zipcode zones. Zipcode zones were grouped into two cognitive classes (Cognitively-Normal and Alzheimer’s-inclined) based on the ratios of AD and dementia subjects to all subjects in an individual zipcode zone. We randomly selected 80% of the data to train a neural network classifier model on environmental complexity measures to predict the cognitive class for each zone, controlling for salient demographic variables. The remaining 20% served as the test set for performance evaluation.
Results: Our proposed model reached excellent classification ability on the testing data: 83.87% accuracy, 95.23% precision, 83.33% recall, and 0.8889 F1-score (F1-score=1 for perfect prediction). The most salient features of “Alzheimer’s-inclined” zipcode zones included longer street-length average, higher circuity, and slightly fewer points of interest. Most “cognitively-normal” zipcode zones appeared in or near urban areas with high environmental complexity measures.
Conclusion: Environmental complexity, reflected in frequency and density of street networks and landmarks features, predicted with high precision the cognitive status of 3-digit zipcode zones based on the etiologic diagnoses and observed cognitive impairment of NACC subjects residing in these zones. The zipcode zones vary widely in size (1.6 km2 to 35,241 km2), and large zipcode zones suffer high spatial heterogeneity. Other proven AD risk factors, such as PM2.5, disperse across zones, and so do individual’s activities, leading to spatial uncertainty. Nevertheless, the model classifies diagnosis well, establishing the need for prospective experiments to quantify effects of environmental complexity on Alzheimer’s development.
M. Yuan ; K.M. Kennedy ; (2023): Utility of Environmental Complexity as a Predictor of Alzheimer’s Disease Diagnosis: A Big-Data Machine Learning Approach. The Journal of Prevention of Alzheimer’s Disease (JPAD). http://dx.doi.org/10.14283/jpad.2023.18