Innovative AI Tool from UCLA Tackles Alzheimer’s Underdiagnosis
Researchers at UCLA have made a significant breakthrough in addressing a critical gap in Alzheimer’s care. They’ve developed an artificial intelligence (AI) tool that utilizes electronic health records (EHRs) to identify patients with undiagnosed Alzheimer’s disease. This innovative approach is particularly vital given the longstanding issue of underdiagnosis, especially in underrepresented communities.
Background: The Alzheimer’s Diagnosis Gap
Alzheimer’s disease, the sixth leading cause of death in the United States, affects approximately 1 in 9 Americans aged 65 and older. Alarmingly, disparities exist in the diagnosis of Alzheimer’s, particularly among African American and Hispanic communities. Research indicates that African Americans are nearly twice as likely to have the disease compared to non-Hispanic whites, yet they are only 1.34 times more likely to receive a diagnosis. Similarly, Hispanic and Latino populations face a similar challenge, being 1.5 times more likely to have the disease but only 1.18 times more likely to be diagnosed.
Dr. Timothy Chang, the study’s corresponding author from UCLA Health, emphasizes the magnitude of the diagnostic gap. “The gap between who actually has the disease and who gets diagnosed is substantial, and it’s particularly significant in underrepresented communities,” he stated, highlighting the urgent need for equitable solutions.
The New AI Model: A Paradigm Shift
While previous research has explored machine learning models to predict Alzheimer’s disease using EHRs, most were based on traditional frameworks that often overlook diagnostic biases. The UCLA team took a pioneering step forward by employing a method known as semi-supervised positive unlabeled learning. This distinct approach is tailored to promote fairness while ensuring high accuracy in predicting undiagnosed cases.
The model utilized data from over 97,000 patients at UCLA Health, incorporating both confirmed Alzheimer’s diagnoses and patients with uncertain status. This broad dataset enabled the model to better understand the complexities of Alzheimer’s disease.
Enhanced Sensitivity in Diagnosis
The UCLA AI model demonstrates remarkable improvements in sensitivity rates. It achieved sensitivity rates of 77% to 81% across various demographic groups, including non-Hispanic white, non-Hispanic African American, Hispanic/Latino, and East Asian populations. By comparison, traditional supervised models only reached sensitivity rates of 39% to 53%. This enhanced ability to detect undiagnosed Alzheimer’s disease marks a significant advancement in diagnostic capabilities.
Understanding Predictive Features
One of the critical components of the UCLA tool is its ability to analyze various patterns in health records. The model examines factors such as prior diagnoses, age, and other clinical indicators. Interestingly, it identifies not only common neurological indicators like memory loss but also unexpected patterns, such as the occurrence of decubitus ulcers and heart palpitations, which may hint at undiagnosed cases.
Fairness at the Core
A unique aspect of the UCLA model is its learning methodology. Unlike traditional approaches that depend solely on confirmed diagnoses for training data, the UCLA AI tool learns from both confirmed cases and patients with unknown Alzheimer’s status. Throughout the development process, fairness measures were integrated, ensuring that the model addresses and mitigates diagnostic disparities.
Validation and Genetic Insights
The researchers validated the model through various methods, including the analysis of genetic data. Patients predicted to have undiagnosed Alzheimer’s exhibited significantly higher polygenic risk scores and notable genetic markers, such as the APOE ε4 allele counts. Dr. Chang believes this tool could help clinicians pinpoint high-risk patients who may require additional evaluation or screening for Alzheimer’s. Early identification is essential, especially as emerging Alzheimer’s treatments and lifestyle interventions hold promise in slowing disease progression.
Future Directions
Looking ahead, the UCLA research team plans to validate the AI model prospectively in collaboration with other health systems. This step aims to assess the model’s generalizability and clinical utility before considering its implementation in routine healthcare practices.
Dr. Chang’s vision is clear: “By ensuring equitable predictions across populations, our model can help remedy significant underdiagnosis in underrepresented populations.” The potential of this AI tool to bridge the gap in Alzheimer’s diagnosis reflects a hopeful stride toward equitable healthcare for all communities.
For those interested in a deeper dive into the research, the findings were published in the journal npj Digital Medicine, shedding light on this transformative approach.