An accurate prediction of the remaining useful life (which is known as prognostics) is essential for effective health management of engineering systems. Model-based prognostics offers interpretability and robustness when only limited data is available. Nevertheless, it has limited accuracy when only sparse or early-stage degradation data are available for a component of interest. This paper presents a hierarchical Bayesian learning framework [1, 2] that enables integration of historical run-to-failure data from similar components into model-based prognostics. Instead of relying on fixed prior assumptions, the proposed approach learns the distribution of the probabilistic hyperparameters from historical data and uses it as an informative prior for inferring component-specific model parameters. It allows uncertainty originating from population-level variability to be explicitly quantified and propagated into remaining useful life predictions of the current component. The framework is demonstrated on two representative degradation processes: fatigue crack growth and lithium-ion battery degradation. Results show that incorporating historical data through hierarchical modeling leads to more stable RUL predictions when only limited degradation data are available for the component of interest, particularly in the early stages of degradation. By enabling information sharing across degradation datasets, the proposed framework explicitly captures population-level variability and propagates it into uncertainty-aware prognostic predictions.