A0385
Title: Estimation and sequential forecast of disease progression in the absence of true disease state process
Authors: Yuanjia Wang - Columbia University (United States)
Zexi Cai - Columbia University (United States) [presenting]
Abstract: Forecasting future disease progression based on patients' evolving health information is challenging when limited by diagnostic capabilities. For example, the absence of gold-standard neurological diagnoses due to a lack of use of objective biomarkers hinders distinguishing Alzheimer's disease (AD) from related conditions such as AD-related dementias (ADRDs) and Lewy body disease (LBD). Despite the increasing use of biomarkers, not everyone has access to them, and some practitioners may not utilize them, resulting in less precise diagnoses. Borrowing information from a series of temporally dependent surrogate labels and health markers may improve the accuracy of future disease prediction. Integrating the hidden Markov model is proposed as a generative model to handle erroneous clinical diagnoses with a time-varying multinomial logistic regression as a discriminative model to identify features of disease progression. An adaptive forward-backwards algorithm is developed to facilitate parameter estimation with pseudo-expectation maximization, with a penalty introduced when many feature variables are present. Furthermore, the posterior rule and the Viterbi algorithm are developed to forecast disease progression. Asymptotic properties are established, and performance with finite samples is demonstrated via simulation studies. Analysis of the neuropathological dataset of the National Alzheimer's Coordinating Center (NACC) shows improved accuracy in distinguishing LBD from AD.
