Analysis of longitudinal data using a three-state continuous time Markov model with misclassification
From the clinical setting, the capability for clinicians to predict prognosis of disease progression will allow them to predict how long a patient will be in a mild stage of severity of disease before reaching the moderate stage would be valuable in tailoring a treatment plan that parallels the progression, better manage diseases. However, determining the progression rate of unobservable diseases is challenging. Neuropsychological evaluations are often the primary variables used in identifying stage of disease severity. Although these variables provide a good indication of disease severity, they are not considered to be "gold standard" for measuring the stage of disease and may be misclassifying the true stage of disease. This research proposes examine the dynamics characteristics of disease as it changes over time. Several approaches exist to examine misclassification in longitudinal studies where the outcome is binary, in both continuous and discrete time settings. When the outcome has more than two categories, hidden Markov models have been used to investigate misclassification but with discrete time. We have developed a likelihood based procedure that describes the dynamic characteristics of change over time in disease severity and allows for possible misclassification of stage of disease based on surrogate variables. We have shown the severe implications of using conventional transitional methods when the data is truly misclassified. The methodology that was developed has several valuable implications. First, the ability to model the severity of diseases is necessary in designing treatment plans which better manage the disease as well as drug and behavior intervention studies that aim to prevent onset or progression of diseases. Second, identifying stages of severity of disease will allow researchers to measure disease progression and obtain sensitivity and specificity of neuropsychological variables on true disease severity. Third, the ideas in this proposed research can be applied to other longitudinal studies where the stage of disease is difficult to measure, such as Alzheimer's disease and Parkinson's disease.
Benoit, Julia Sanders, "Analysis of longitudinal data using a three-state continuous time Markov model with misclassification" (2014). Texas Medical Center Dissertations (via ProQuest). AAI3639280.