|
Abstract:
|
We propose an initial severity-dependent longitudinal model to account for the influence of initial severity of depression on the subsequently, post-treatment measured severity of depression. The proposed model has the flexibility of nonparametric modeling since it allows coefficients to vary with initial severity of depression. In addition, the model provides interesting and practical patient-specific interpretation of initial severity-dependent coefficients. As a result, the proposed model enables patient-specific modeling and treatment recommendations consistent with assessment of a patient's initial severity and thus it can be used as a decision support tool for clinicians. The empirical likelihood method is used for efficient estimation and statistical inference about the initial severity-dependent coefficients. In contrast to the literature on the marginal regression models, the proposed estimation procedure allows for the nuisance parameters associated with the working correlation matrix and the error variances to vary smoothly with initial severity. The effectiveness of the proposed procedure is demonstrated via simulation studies.
|
