Current code-based models for concrete creep and shrinkage present significant challenges when applied to large structures such as balanced cantilever box girder bridges, due to their complex geometry and staged construction. Case studies on such structures with multi-decade deflection measurements, though scarce in literature, are essential for evaluating the long-term accuracy of these models. This paper presents a unique case study of a three-span balanced cantilever box girder bridge in the Netherlands with 44 years of deflection measurements, revealing excessive and ongoing deflections. A detailed finite element model incorporating construction stages was developed to study the impact of four creep and shrinkage models on the predictions of multi-decade deflections and prestress losses. All models significantly underestimated long-term deflections, by about 75% for Eurocode 2 (2004), Eurocode 2 (2023), and fib Model Code 2010, and 35% for RILEM B4. The choice of the creep and shrinkage model also strongly affects the predicted prestress losses, with RILEM B4 estimating losses up to 1.7 times higher than Eurocode 2 (2004). Since the working prestress level is a key factor for structural assessment of large-span concrete bridges, these findings have important implications for engineering practice. Supported by rare long-term validation data, this study reveals the performance and limitations of current creep and shrinkage models, the possible consequences of their underestimations, and the need for more reliable long-term prediction models for the design and assessment of box girder bridges.

Aging infrastructure in the Netherlands presents a significant challenge, particularly with precast girder bridges made continuous, which exhibit inadequate shear reinforcement per current design codes. To address shear capacity and accuracy of current assessment practices, a research program including full-scale shear tests is underway at Delft University of Technology. As a part of the research, a blind pre-diction contest with two specimens has been organized. The experiments showed that the loss of composite action at the interface is the primary failure mechanism, and generally an accurate model of the interface behaviour in such composite members is missing. In this paper, a further review of the available interface models and previous tests is conducted. This review leads to the challenges in the accurate evaluation of the interface behaviour, as well as the next steps to address them: a new set of full-scale specimens, and a small-scale test setup reflecting a realistic stress distribution.