Synthesized Blood Test Data for Dementia Risk Screening Using Generative Adversarial Networks

This study investigates the potential of Generative Adversarial Networks (GANs) to improve dementia risk prediction models through data augmentation with blood test data. Recent studies have shown that basic blood tests are cost-effective and fairly accurate in predicting cognitive function. However, the development of predictive models that can handle various conditions is challenging due to high costs, limited sample sizes, and missing data from tests not conducted in certain facilities. This study validates the use of conditional tabular GANs (CTGANs) for improving predictive models of dementia risk screening using blood test results. To evaluate the effectiveness of the synthesized data, we employed linear regression models and random forests. With the data from 202 inpatients and 137 external participants, we discuss the effectiveness of conditional tabular GANs in preserving data integrity and maintaining prediction accuracy, even with sparse input distributions.