Researchers at UCLA have developed an innovative AI tool aimed at identifying undiagnosed Alzheimer’s disease by utilizing electronic health records, targeting significant underdiagnosis, especially in underrepresented communities. Published in npj Digital Medicine, the study highlights the stark disparities in Alzheimer’s diagnoses: African Americans have nearly double the prevalence but a significantly lower diagnosis rate, while Hispanic and Latino populations also face similar challenges. The UCLA model employs semi-supervised positive unlabeled learning, enhancing prediction accuracy and fairness in diagnosis. Analyzing health records of over 97,000 patients, it achieved sensitivity rates of 77 to 81%, surpassing traditional models. The AI tool identifies key risk factors, such as neurological indicators and unexpected symptoms. The researchers aim to expand validation in diverse health systems to enhance its clinical utility, ultimately working towards equitable Alzheimer’s care and addressing diagnostic disparities for better patient outcomes. Early identification is crucial as new therapies emerge.
