| Sumario: | Purpose: This study aimed to examine cognitive trajectories and to identify predictors of cognitive decline in middle-aged and older cancer survivors using longitudinal data and machine learning models. Methods: Data from 399 cancer survivors aged 45 years and older were analyzed from the Korean Longitudinal Study of Aging (KLoSA). Latent class growth analysis was used to identify cognitive trajectories, while logistic regression, random forest, neural network, and support vector machine algorithms were employed to predict trajectory group membership based on sociodemographic, health-related, and cancer-related variables. Results: Two distinct cognitive trajectories were identified: maintenance (85.2%) and decline (14.8%). For predicting the decline trajectory, the random forest model achieved the best performance (accuracy=0.92, AUC=0.93), followed by logistic regression and support vector machine (accuracy=0.86, AUC=0.86), whereas the neural network demonstrated lower performance (accuracy=0.82, AUC=0.78). Key predictors included age, education, physical activity, BMI, time since cancer diagnosis, symptom progression, and functional limitations related to cancer. Conclusion: These findings highlight the importance of proactive cognitive monitoring and the integration of targeted, personalized interventions into survivorship care for aging cancer survivors. Future studies should incorporate comprehensive clinical data and conduct external validation to enhance model reliability and clinical applicability.
|
|---|