A bayesian dynamic latent state model for predicting infant sleep-wake patterns under daily massage intervention

Sleep disturbances in infants present a persistent challenge for caregivers and healthcare providers. This study proposes a Bayesian Dynamic Latent State Model to predict infant sleep-wake patterns in response to daily massage, a non-pharmacological intervention. The model captures latent sleep propensity as a dynamic hidden process influenced by current and previous massages, individual random effects, and autoregressive components. Observed outcomes include nocturnal sleep duration and nighttime awakenings, modeled using Gaussian and Poisson distributions respectively. Through numerical simulations and a real-world case study, the model demonstrates clear benefits: average nocturnal sleep duration increased by approximately 1.2–1.5 hours, while nighttime awakenings decreased by about 35–40% on intervention days, with residual improvements on subsequent days. Compared to traditional static and time-series models, the proposed Bayesian approach provides greater flexibility in handling uncertainty, explicitly models carry-over effects, and integrates individual heterogeneity in sleep responses contributions that have not been fully addressed in prior infant sleep studies. This research thus advances the scientific understanding of dynamic, intervention-driven sleep processes, while also offering practical implications for evidence-based pediatric nursing and personalized infant care strategies. While promising, validation is currently limited to a small dataset and simplified assumptions. Future work will involve larger-scale testing, incorporation of additional external factors, and benchmarking against alternative machine learning models.