Predicting progression of mild cognitive impairment patients through four distinctive subgroups obtained by unsupervised learning algorithms

Background Mild cognitive impairment (MCI) exhibits considerable heterogeneity, requiring accurate characterization through classification and prognostic models. In clinical research, data-driven models offer valuable insights for classification, stratification, and predicting progression to dementia. Objective We implemented computational techniques to characterize MCI patients and develop multistate progression models for Alzheimer's disease (AD). Methods Datasets comprising 544 MCI patients from Coimbra University Hospital and 497 from the ADNI, were processed using machine learning techniques, including dimensionality reduction and partition clustering algorithms. For longitudinal measures (n = 351), multistate non-Markov was applied to generate transition probability estimates. Results Our analyses gave 4 possible subgroups of MCI patients: 1) increased cognitive reserve, 2) suspected AD pathology, 3) psychological manifestations, and 4) cardiovascular risk factors. Progression within these subgroups showed variations. The likelihood of progressing to AD dementia was estimated over a range of 5 months for those with suspected AD pathology and 66 months for those with psychological manifestations. Conclusions Our findings support the significance of computational methods to improve the characterization and prognosis of MCI patients. We suggest that these four MCI subgroups should be considered for clinical monitoring.