Tracking Cognitive Trajectories in Mild Cognitive Impairment Using a Machine Learning Technique of Subtype and Stage Inference
	    		
		   		
		   			
		   		
	    	
    	 
    	10.12779/dnd.2025.24.1.44
   		
        
        	
        	
        	
        		- Author:
	        		
		        		
		        		
			        		Hui Jin RYU
			        		
			        		
			        		
			        			1
			        			
			        		
			        		
			        		
			        		
			        		;
		        		
		        		
		        		
			        		Kyoung Ja KWON
			        		
			        		;
		        		
		        		
		        		
			        		Yeonsil MOON
			        		
			        		
		        		
		        		
		        		
		        		
		        			
			        		
			        		Author Information
			        		
		        		
		        		
			        		
			        		
			        			1. Department of Neurology, Konkuk University Medical Center, Seoul, Korea
			        		
		        		
	        		
        		 
        	
        	
        	
        		- Publication Type:Original Article
 
        	
        	
            
            
            	- From:Dementia and Neurocognitive Disorders
	            		
	            		 2025;24(1):44-53
	            	
            	
 
            
            
            	- CountryRepublic of Korea
 
            
            
            	- Language:English
 
            
            
            	- 
		        	Abstract:
			       	
			       		
				        
				        	 Background:and Purpose: Recognizing cognitive decline patterns in mild cognitive impairment (MCI) is crucial for early screening and preventive interventions. However, studies on the trajectory of individual cognitive functions in MCI are limited. Thus, the purpose of this study was to identify subtypes and stages of cognitive decline in MCI using a machine learning method. 
				        	
				        
				        	Methods:A total of 944 subjects consisting of those who were cognitively normal and those with MCI were enrolled. Fifteen neuropsychological tasks were used in the analysis.The optimal number of subtypes was determined based on the cross-validation information criterion. Fifteen stages of cognitive trajectory were estimated for each subtype. 
				        	
				        
				        	Results:The following three subtypes were identified: amnestic-verbal subtype, dysexecutive subtype, and amnestic-visual subtype. Of 723 (76.6%) subjects who had reached stage 1 at least, amnestic-verbal subtype accounted for the most (n=340, 47.0%), followed by dysexecutive subtype (n=253, 35.0%) and amnestic-visual subtype (n=130, 18%). The amnestic-verbal subtype had significantly more males (amnestic-verbal: 41.8%, dysexecutive: 31.2%, and amnestic-visual: 28.5%), younger subjects (amnestic-verbal: 72.01 years, dysexecutive: 74.43 years, and amnestic-visual: 75.06 years), higher educational years (amnestic-verbal: 11.06 years, dysexecutive: 9.53 years, and amnestic-visual: 9.79 years), lower Clinical Dementia Rating sum of boxes (amnestic-verbal: 1.40, dysexecutive: 1.61, and amnestic-visual: 1.71), and lower Korean-Instrumental Activities of Daily Living score (amnestic-verbal: 0.20, dysexecutive: 0.27, and amnestic-visual: 0.26). 
				        	
				        
				        	Conclusions:Three types of MCIs were extracted using SuStaIn. Pathways of MCI deterioration could be different. The amnestic type could be bisected based on whether episodic verbal or visual memory is degraded first.