Longitudinal Examination of Wandering in Children and Adolescents With Autism.

Dementia is a clinical condition characterized by a gradual decline in cognitive function that affects memory, reasoning, communication, and the ability to perform daily tasks. This decline is typically the result of brain cell damage from conditions such as Alzheimer’s disease (AD). Early diagnosis can play a crucial role in customizing treatment plans and managing the disease more effectively. In recent years, artificial intelligence (AI) has shown promising potential in improving the accuracy of dementia diagnoses by applying advanced machine learning (ML) techniques to medical imaging data, including magnetic resonance imaging (MRI). These models are capable of detecting subtle, often subclinical, structural brain changes, enabling earlier and more reliable diagnoses than traditional clinical approaches. Nevertheless, AI-driven diagnostic systems may not generalize equally across racial and ethnic groups.. In dementia classification, such differences may arise from unequal representation in training data, population-level variation in brain structure and disease expression, and broader clinical and social determinants of health.. These factors can lead to underdiagnosis or overdiagnosis, and may amplify existing healthcare disparities. A prior epidemiologic study reported significant variations in dementia incidence across six population groups over 14 years. Similarly, Gilsanz et al. explored dementia rates in individuals aged 90 and above in a broadly representative cohort, offering insight into how risk varies across diverse backgrounds. Prior work has also shown that racial and ethnic differences in Alzheimer’s disease risk, biomarker associations, and disease presentation may influence both predictive modeling and interpretation. For example, although the -4 allele is a recognized risk factor among individuals of White descent, African American and Hispanic populations have been reported to exhibit elevated Alzheimer’s risk regardless of genotype. In addition, ancestry-related genetic and epigenetic variation, shaped by social and environmental exposures, may influence brain morphology and dementia pathology across populations. For machine learning models, such variation can alter the meaning of imaging-derived features, introduce hidden confounding, and reduce the transferability of attribution patterns across groups. Consistent with this concern, prior work has noted that differences in brain morphology, socioeconomic conditions, and access to care may contribute to these disparities, and models trained predominantly on White American populations often show reduced accuracy in other groups. More broadly, race-neutral healthcare algorithms can still perpetuate diagnostic imbalance; for example, higher false-positive rates have been reported for Hispanic and Asian patients in colorectal cancer prediction models that did not account for ethnicity.