Randomized clinical trials frequently incorporate quality of life (QoL) measures in survival trials to evaluate whether patient QoL is maintained throughout the additional survival time. Patient QoL data is typically modeled using latent variable models from psychometrics, which treat responses to QoL items as observed manifestations of a latent construct. Limited research has been done on longitudinal item response theory (IRT) models in survival analysis context (i.e., when subject drop-out is assumed to occur under the MNAR mechanism). This research builds on existing methods referred to as IRT trees and/or partially-ordered sets. The work introduces a uniquely flexible IRT discrete hazard function, which allows for the simultaneous estimation of item response and MNAR drop-out. This approach reduces bias in the estimated parameters relating the latent variable to the treatment arm predictor, improving inference of separation of treatment arms on patient QoL in a survival analysis trial.