Type-1 diabetes, also known as insulin-dependent diabetes, is a chronic illness brought on by the body’s inability to manufacture insulin. This study intends to comprehend the glucose levels of 12 patients from Chinese sources 572 times continuously throughout 143 hours, or once every 15 minutes. We used various Markov models, including the Markov model, hidden Markov model (HMM), and multilevel hidden Markov model, to comprehend the pattern in the data. Three states, hypoglycemia, normal blood sugar, and hyperglycemia, were used to create the probability mass function for the Markov model. For the HMM, emission states increase, decrease, and remain the same, whereas concealed states are hypoglycemia, normal, and hyperglycemia. We select data with a positively skewed distribution for the multilevel hidden Markov model, and the Akaike information criterion (AIC) score determines the hidden states. The models’ respective AIC values are Markov 3138.475 and MHMM 290.689. According to AIC values, the multilevel hidden Markov model is the best.

Markov, hidden Markov model, multilevel hidden Markov model, Type-1 diabetes.

Received: October 11, 2023;  Accepted: January 9, 2024; Published: February 5, 2024

How to cite this article: Tirupathi Rao Padi and Surnam Narendra, Multilevel hidden Markov models for studying Type-1 diabetes, JP Journal of Biostatistics 24(1) (2024), 161-176. http://dx.doi.org/10.17654/0973514324011

This Open Access Article is Licensed under Creative Commons Attribution 4.0 International License

References:
[1] E. Aarts, Multilevel hmm tutorial, 2019.
[2] F. Cartella, J. Lemeire, L. Dimiccoli and H. Sahli, Hidden semi-Markov models for predictive maintenance, Math. Probl. Eng. (2015), Art. ID 278120, 23 pp.
[3] D. F. de Carvalho, U. Kaymak, P. Van Gorp and N. van Riel, A Markov model for inferring event types on diabetes patients data, Healthcare Analytics 2 (2022), Article ID 100024. https://doi.org/10.1016/j.health.2022.100024.
[4] R. P. Dobrow, Introduction to Stochastic Processes with R, John Wiley & Sons, 2016.
[5] M. Dumitrescu and I. Popescu, On three Markov models for the clinical evolvement: a simulation study, Simulation Practice and Theory 2(4-5) (1995), 159-177.
[6] N. S. Gill and P. Mittal, A novel hybrid model for diabetic prediction using hidden Markov model, fuzzy based rule approach and neural network, Indian Journal of Science and Technology 9(35) (2016), 192-199.
[7] J. Ginn, S. M. Moraga and E. Aarts, Sample size considerations for Bayesian multilevel hidden Markov models: a simulation study on multivariate continuous data with highly overlapping component distributions based on sleep data, 2022. arXivPreprint arXiv:2201.09033.
[8] S. Kirchherr, S. Mildiner Moraga, G. Coudé, M. Bimbi, P. F. Ferrari, E. Aarts and J. J. Bonaiuto, Bayesian multilevel hidden Markov models identify stable state dynamics in longitudinal recordings from macaque primary motor cortex, European Journal of Neuroscience 58(3) (2023), 2787-2806.
[9] S. Mildiner Moraga and E. Aarts, Go multivariate: recommendations on Bayesian multilevel hidden Markov models with categorical data, Multivariate Behavioral Research (2023), 1-29. https://doi.org/10.1080/00273171.2023.2205392.
[10] S. Ruiz-Suarez, V. Leos-Barajas and J. M. Morales, Hidden Markov and semi-Markov models when and why are these models useful for classifying states in time series data? Journal of Agricultural, Biological and Environmental Statistics (2022), 1-25. arXiv:2105.11490v2 [stat.AP].
[11] M. K. Varshney, A. Sharma, K. Goel, V. Ravi and G. Grover, Estimation of transition probabilities for diabetic patients using hidden Markov model, International Journal of System Assurance Engineering and Management 11(S2) (2020), 329-334.
[12] N. Wang, S. Sun, Z. Cai, S. Zhang and C. Saygin, A hidden semi-Markov model with duration-dependent state transition probabilities for prognostics, Math. Probl. Eng. (2014), Art. ID 632702, 10 pp.