Multi-state Markov Model In Outcome Of Mild Cognitive Impairment To Alzheimer's Disease

Objective: The aim of this study was to introduce multi-state Markov model in outcome prediction from mild cognitive impairment to Alzheimer's disease and to find out related factors in order to provide theory basis for AD prevention and early intervention among elderly people. By comparing with the analysis results of logistic model, the strength of multi-state Markov model for longitudinal data analysis was discussed. From the point of application, problems of using multi-state Markov model and suggested method for exploring influencing factors for various progressive stages of other chronic disease were provided.Methods: MCI, moderate to severe cognitive impairment, and AD were defined as state 1, 2 and 3, respectively. Multi-state Markov model was applied and a three state homogeneous model with discrete states and discrete times from six visits follow-up data was constructed to explore factors for various progressive stages from MCI to AD. Logistic models were also made to find out factors between two kinds of transitions, state 1 to state 2, state 2 to state 3, with the transition outcome as dependent variable and other variables as independent variables.Results: Multi-state Markov model was fitted well. At the level of 0.10, univariate analysis showed that gender, age, occupation, hypertension, diabetes and ApoEε4 were statistically significant for transition from state 1 to state 2; Gender, age, education level, marital status, diabetes and reading were statistically significant for transition from state 2 to state 1; Gender, age, marital status, education level, hypertension, high cholesterol in the serum, diabetes, SBP and ApoEε4 were statistically significant for transition from state 2 to state 3; At the level of 0.05, multivariate analysis showed that gender, age, hypertension were statistically significant for transition from state 1 to state 2; Age, education level and reading were statistically significant for transition from state 2 to state 1; Gender, age, hypertension, diabetes, ApoEε4 were statistically significant for transition from state 2 to state 3; Combined with the result of logistic regression models, we can conclude that women, older, hypertension, diabetes, ApoEε4 were risk factors for MCI to AD; While mental work, higher-income, reading were protecting factors for MCI to AD. According to the fitted multi-state Markov model, survival curves, transition intensities, three years transition probabilities during each state were also estimated. Conclusion: To delay the disease progression of MCI, prevention measures phase by phase can be taken based on the main factors of each stage. Compared with logistic regression model,the results of multi-state Markov model is more all-around,and the transition intensity and transition probability can also be estimated. As an effective tool for dealing with longitudinal data, multi-state Markov model can imitate the natural history of disease and have a great advantage in dynamically evaluating the development of chronic diseases with multi-states and multi-factors.