Departamento de Matemáticas, Facultad de Ciencias, Universidad Nacional Autónoma de México, Circuito Exterior s/n, Ciudad Universitaria, Del. Coyoacán, C.P. 04510 Ciudad de México, Mexico.
Departamento de Matemáticas, Facultad de Veterinaria, Universidad de Extremadura, Avda. de la Universidad s/n, C.P. 10003 Cáceres, Spain.
Biostatistics. 2020 Oct 1;21(4):743-757. doi: 10.1093/biostatistics/kxz004.
Motivated by a study tracking the progression of Parkinson's disease (PD) based on features extracted from voice recordings, an inhomogeneous hidden Markov model with continuous state-space is proposed. The approach addresses the measurement error in the response, the within-subject variability of the replicated covariates and presumed nondecreasing response. A Bayesian framework is described and an efficient Markov chain Monte Carlo method is developed. The model performance is evaluated through a simulation-based example and the analysis of a PD tracking progression dataset is presented. Although the approach was motivated by a PD tracking progression problem, it can be applied to any monotonic nondecreasing process whose continuous response variable is subject to measurement errors and where replicated covariates play a key role.
受一项基于从语音记录中提取的特征来跟踪帕金森病（PD）进展的研究的启发，提出了一种具有连续状态空间的非齐次隐马尔可夫模型。该方法解决了响应中的测量误差、重复协变量的个体内可变性以及假定的非递减响应问题。描述了一个贝叶斯框架，并开发了一种有效的马尔可夫链蒙特卡罗方法。通过基于模拟的示例评估了模型性能，并呈现了对 PD 跟踪进展数据集的分析。虽然该方法是受 PD 跟踪进展问题的启发，但它可以应用于任何单调递增的过程，其连续响应变量受到测量误差的影响，并且重复协变量起着关键作用。
