Item bank development in computerized adaptive testing: a lifecycle governance synthesis for defensible measurement

Computerized adaptive testing (CAT) enhances measurement precision by dynamically selecting items matched to an examinee’s latent trait level, yet its validity depends on the structural integrity of the underlying item bank. Despite this reliance, item bank development remains methodologically fragmented, with limited integrative guidance on how blueprinting, calibration, and monitoring interact in operational settings. This systematic review synthesised methodological practices in CAT item bank development across 84 peer-reviewed studies, drawn from a primary lifecycle search and a targeted computational update, using PRISMA-aligned qualitative synthesis. Findings indicate three sequentially dependent components: structural blueprinting, psychometric calibration, and operational quality surveillance. Early design decisions were found to constrain calibration accuracy and influence subsequent monitoring demands, challenging the view of item bank development as a one-time preparatory process. Emerging machine-learning and artificial-intelligence approaches were found to redistribute rather than displace psychometric functions across the lifecycle, extending established practices in item generation, calibration, and monitoring while remaining reliant on psychometric validation for construct definition and post-deployment surveillance. The evidence supports a lifecycle governance model requiring concurrent maintenance of all three components, with clear specification of where automation can be trusted and where independent psychometric oversight remains necessary, to sustain score interpretability over time. This synthesis provides a structured framework for adaptive test development across interdisciplinary measurement contexts.