Background: Due to the class imbalance issue faced when Alzheimer’s disease (AD) develops from normal cognition (NC) to mild cognitive impairment (MCI), present clinical practice is met with challenges regarding the auxiliary diagnosis of AD using machine learning (ML). This leads to low diagnosis performance. We aimed to construct an interpretable framework, extreme gradient boosting-Shapley additive explanations (XGBoost-SHAP), to handle the imbalance among different AD progression statuses at the algorithmic level. We also sought to achieve multiclassification of NC, MCI, and AD. Methods: We obtained patient data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database, including clinical information, neuropsychological test results, neuroimaging-derived biomarkers, and APOE-ε4 gene statuses. First, three feature selection algorithms were applied, and they were then included in the XGBoost algorithm. Due to the imbalance among the three classes, we changed the sample weight distribution to achieve multiclassification of NC, MCI, and AD. Then, the SHAP method was linked to XGBoost to form an interpretable framework. This framework utilized attribution ideas that quantified the impacts of model predictions into numerical values and analysed them based on their directions and sizes. Subsequently, the top 10 features (optimal subset) were used to simplify the clinical decision-making process, and their performance was compared with that of a random forest (RF), Bagging, AdaBoost, and a naive Bayes (NB) classifier. Finally, the National Alzheimer’s Coordinating Center (NACC) dataset was employed to assess the impact path consistency of the features within the optimal subset. Results: Compared to the RF, Bagging, AdaBoost, NB and XGBoost (unweighted), the interpretable framework had higher classification performance with accuracy improvements of 0.74%, 0.74%, 1.46%, 13.18%, and 0.83%, respectively. The framework achieved high sensitivity (81.21%/74.85%), specificity (92.18%/89.86%), accuracy (87.57%/80.52%), area under the receiver operating characteristic curve (AUC) (0.91/0.88), positive clinical utility index (0.71/0.56), and negative clinical utility index (0.75/0.68) on the ADNI and NACC datasets, respectively. In the ADNI dataset, the top 10 features were found to have varying associations with the risk of AD onset based on their SHAP values. Specifically, the higher SHAP values of CDRSB, ADAS13, ADAS11, ventricle volume, ADASQ4, and FAQ were associated with higher risks of AD onset. Conversely, the higher SHAP values of LDELTOTAL, mPACCdigit, RAVLT_immediate, and MMSE were associated with lower risks of AD onset. Similar results were found for the NACC dataset. Conclusions: The proposed interpretable framework contributes to achieving excellent performance in imbalanced AD multiclassification tasks and provides scientific guidance (optimal subset) for clinical decision-making, thereby facilitating disease management and offering new research ideas for optimizing AD prevention and treatment programs.

Background and Objectives Various resources exist for treating mild cognitive impairment (MCI) or dementia separately as terminal events or for focusing solely on a 1-way path from MCI to dementia without taking into account heterogeneous transitions. Little is known about the trajectory of reversion from MCI to normal cognition (NC) or near-NC and patterns of postreversion, which refers to cognitive trajectories of patients who have reversed from MCI to NC. Our objectives were to (1) quantitatively predict bidirectional transitions of MCI (reversion and progression), (2) explore patterns of future cognitive trajectories for postreversion, and (3) estimate the effects of demographic characteristics, APOE, cognition, daily activity ability, depression, and neuropsychiatric symptoms on transition probabilities. Methods We constructed a retrospective cohort by reviewing patients with an MCI diagnosis at study entry and at least 2 follow-up visits between June 2005 and February 2021. Defining NC or near-NC and MCI as transient states and dementia as an absorbing state, we used continuous-time multistate Markov models to estimate instantaneous transition intensity between states, transition probabilities from one state to another at any given time during follow-up, and hazard ratios of reversion-related variables. Results Among 24,220 observations from 6,651 participants, there were 2,729 transitions to dementia and 1,785 reversions. As for postreversion, there were 630 and 73 transitions of progression to MCI and dementia, respectively. The transition intensity of progression to MCI for postreversion was 0.317 (2.48-fold greater than that for MCI progression or reversion). For postreversion participants, the probability of progressing to dementia increased by 2% yearly. Participants who progressed to MCI were likely to reverse again (probability of 40% over 15 years). Age, independence level, APOE, cognition, daily activity ability, depression, and neuropsychiatric symptoms were significant predictors of bidirectional transitions. Discussion The nature of bidirectional transitions cannot be ignored in multidimensional MCI research. We found that postreversion participants remained at an increased risk of progression to MCI or dementia over the longer term and experienced recurrent reversions. Our findings may serve as a valuable reference for future research and enable health care professionals to better develop proactive management plans and targeted interventions.

The detection of Parkinson’s disease (PD) in its early stages is of great importance for its treatment and management, but consensus is lacking on what information is necessary and what models should be used to best predict PD risk. In our study, we first grouped PD-associated factors based on their cost and accessibility, and then gradually incorporated them into risk predictions, which were built using eight commonly used machine learning models to allow for comprehensive assessment. Finally, the Shapley Additive Explanations (SHAP) method was used to investigate the contributions of each factor. We found that models built with demographic variables, hospital admission examinations, clinical assessment, and polygenic risk score achieved the best prediction performance, and the inclusion of invasive biomarkers could not further enhance its accuracy. Among the eight machine learning models considered, penalized logistic regression and XGBoost were the most accurate algorithms for assessing PD risk, with penalized logistic regression achieving an area under the curve of 0.94 and a Brier score of 0.08. Olfactory function and polygenic risk scores were the most important predictors for PD risk. Our research has offered a practical framework for PD risk assessment, where necessary information and efficient machine learning tools were highlighted.