A reproducible analytics workflow for longitudinal healthcare digital twins

Healthcare digital twins are increasingly explored as mechanisms for personalised monitoring, simulation and decision support, with longitudinal health data forming a central analytical component. While existing research emphasises predictive models and domain-specific implementations, less attention has been given to the reliability of analytics workflows required for continuous updating in real-world settings. This gap is particularly relevant for high-frequency data collected through mobile health platforms and ecological momentary assessment (EMA), which are often heterogeneous, irregularly sampled and incomplete, placing sustained demands on analytical pipelines as data accumulate. This paper examines reproducible longitudinal analytics as a foundational layer for healthcare digital twins, rather than proposing a complete twin system. We describe a system-oriented workflow that integrates EMA data, clinical records and application-level interaction logs into stable longitudinal representations that support repeated updating. The workflow prioritises transparent preprocessing, principled handling of missing data and deliberately constrained longitudinal modelling to enhance stability, interpretability and reproducibility under incremental data growth. Using a real-world digital health dataset as a motivating case study, we demonstrate how workflow design decisions influence reliability and scalability as longitudinal complexity increases. The paper therefore shifts focus from isolated model optimisation to infrastructure-level properties that determine whether longitudinal analytics can sustainably support healthcare digital twin applications in practice.