Disease management and latent choices

Sean Michael Murphy

doi:10.7273/000005616

This dissertation consists of three independent essays in the field of health economics. The first essay analyzes factors influencing the initial treatment choice of inpatients with severe hypertension. A thorough analysis of treatment choice has been largely overlooked in the hypertension literature, and few studies of any disease have conducted comprehensive multivariate analyses on treatment choices using such a diverse array of socioeconomic variables and hospital locations. According to the results, characteristics other than morbidity affect the type of treatment received; indicating public policy could improve care. The second essay analyzes the effect that a patient's reference point has on her perceived effectiveness of subsequent treatment. One commonly used measure of treatment effectiveness for conditions where treatments are palliative, and clear objective symptoms do not exist, is self-reported changes in disease status. Factors such as treatment history provide a reference point that may influence patients' expectations of how effective further treatment might be. Therefore, decisions about whether to proceed with additional treatment, as well as perceptions of how effective that treatment is, may be influenced by this point. Although there is an extensive literature on how patient expectations influence treatment outcomes, work testing how expectations depend on these reference point factors appears to be missing. The results indicate that these factors influence perceived treatment effectiveness. The final essay focuses on missing and ambiguous observations in a dataset with binary dependent variables. The ability to reallocate these responses could aid in the correction of potentially biased estimates. Using the "latent-choice multinomial logit model" (LCMNL), it is possible to determine whether these incomplete responses are more likely to belong to another outcome. Simulations of this model are performed to determine whether the estimated conditional probabilities are accurate enough to evaluate the likelihood that any given observation belongs to a particular outcome, and whether doing so improves parametric estimation. Tests imply that the reclassifications indicated by the LCMNL's conditional probabilities are accurate. The best method for dealing with ambiguous observations in empirical analysis is also assessed. Results indicate that the best method depends on the source of the ambiguity

Disease management and latent choices

Files and links (1)

Abstract

Metrics

Details