A Probabilistic Differential Modeling for Capturing Health Disparities in Multivariate Data
For several decades, researchers have documented disparities across several domains in cancer risk, incidence, screening, diagnosis, treatment, survival and mortality. However, these efforts are unlikely to play a significant role in addressing disparities in risk, treatment, and mortality unless there is a better understanding and recognition of specific factors leading to disparities, an understanding of the populations most affected, and implementing targeted approaches to address these disparities. This clinical investigation seeks to address and overcome ethnicity-related survival disparities.In this project, we will build statistical models that simultaneously identify the key factors contributing to disparities using some of the traditional parametric methods as well as the advanced data mining techniques within multivariate data. Existing traditional and data mining methods do not ensure that the predictive models induced on different datasets can be compared to each other even if the data distribution on which the models were induced is similar.
In order to quantify the impact of various attributes that are critical for the differences in the data distributions, we propose a transition framework for the predictive modeling which will allow us to measure the similarities and the dissimilarities between two datasets. Significance: This is a novel approach to capture the disparity of group through a systemic comparison of multivariate models which cannot be done using any of existing methods available. This is a more comprehensive and accurate assessment of survivability risk factors contributing to health disparities using environmental, behavioral as well as genetic information, has potential to dictate more aggressive new therapeutic treatments for specific disparity subsets of high-risk patients, and can be used clinically to influence treatment decisions based on clinical pathology and treatment related factors.