Joint modeling of time-to-event data and multiple ratings of a discrete diagnostic test without a gold standardWon, Seung Hyun (2014) Joint modeling of time-to-event data and multiple ratings of a discrete diagnostic test without a gold standard. Doctoral Dissertation, University of Pittsburgh. (Unpublished)
AbstractHistologic tumor grade is a strong predictor of risk of recurrence in breast cancer. Nevertheless, tumor grade readings by pathologists are susceptible to intra- and inter-observer variability due to its subject nature. Because of this limitation, histologic tumor grade is not included in the breast cancer stating system. Latent class models have been considered for analysis of such discrete diagnostic tests with regarding the underlying truth as a latent variable. However, the model parameters in latent class models are only locally identifiable, that is, any permutation on the categories of the underlying truth can lead to the same likelihood value. In many clinical practices, the underlying truth is known associated with the risk of a certain event in a trend. Here, we proposed a joint model with a Cox proportional hazards model for time-to-event data where the underlying truth is a latent predictor and a latent class model for multiple ratings of a discrete diagnostic test without a gold standard. With the known association between the underlying truth and the risk of an event in a trend, the proposed joint model not only fully identifies all model parameters but also provides valid assessment of the association between the diagnostic test result and the risk of an event. The modified EM algorithm was used for estimation with employing the survey-weighted Cox model in the M-step. To test whether the known trend imposed on model parameters can be assumed, we applied the Union-Intersection principle for the proposed joint model. The proposed method is illustrated in the analysis of data from the National Surgical Adjuvant Breast and Bowel Project (NSABP) B-14 sub-study and through simulation studies. The proposed method is relevant to public health fields, such as chronic diseases and psychiatry, where some components of the initial diagnostics are subjective but have important implications in patient management. Application of our method leads to accurate assessment on the association between the diagnostic tests and the clinical outcomes and subsequently significant improvement in decision-making on treatment or patient management. Share
Details
MetricsMonthly Views for the past 3 yearsPlum AnalyticsActions (login required)
|
![[feed]](/images/feed-icon-32x32.png){kind=icon}