Joint models of longitudinal and time-to-event data are essential for dynamic prediction in chronic disease research, but standard formulations typically assume linear biomarker–hazard associations and unidirectional influence from biomarkers to events. These assumptions are often violated in multimorbidity and oncology, where events alter subsequent biomarker trajectories and relationships are non-linear or threshold-like. We extend the joint modeling framework through the JMbdirect package by introducing semi-parametric association structures based on penalized splines and neural additive components, and by embedding explicit bidirectional feedback between events and biomarkers. Estimation is supported via penalized likelihood with Laplace approximation and a Bayesian alternative using Hamiltonian Monte Carlo. Simulation studies demonstrate that incorporating feedback and flexible links improves discrimination and calibration compared with classical linear specifications, particularly under non-linear risk mechanisms. Applications in oncology (locoregional control, progression-free survival, overall survival) and diabetes–cardiovascular multimorbidity illustrate how these methods provide more accurate and interpretable dynamic predictions. By combining methodological innovation with a publicly available dashboard interface, this work advances the translation of joint modeling into personalized risk prediction and clinical decision support.