Overview of Participant Selection and RF Model Performance
a. The participant selection process and the performance of the random forest (RF) model were outlined in this study. Data from the UCSF Electronic Health Records (EHRs) and the UCSF Memory and Aging Center (MAC) database were extracted and prepared for analysis. Clinical information was filtered and encoded for training the RF models to predict the risk of Alzheimer’s disease (AD) diagnosis. The models were assessed using AUROC/AUPRC on a 30% evaluation set and interpreted based on feature importance and a knowledge network (SPOKE). Top features were validated in external databases.
b. A consistent set of individuals with AD and controls from the UCSF EHR were filtered for model training and testing. The filtered participant cohorts were split into a 30% held-out set for testing.
c. The performance of the RF models was evaluated through bootstrapping on the held-out evaluation set. The AUROC performance for models trained and tested on female and male strata was showcased. The results were illustrated using quartiles and outliers to demonstrate the model’s predictive power.
Credit: Nature Aging (2024). DOI: 10.1038/s43587-024-00573-8
UC San Francisco researchers have developed a method to predict Alzheimer’s disease up to seven years before symptoms manifest by leveraging machine learning algorithms.
The primary factors influencing Alzheimer’s prediction were high cholesterol levels, and in the case of women, osteoporosis.
This study underscores the potential of artificial intelligence (AI) in identifying patterns within clinical data, enabling researchers to explore large genetic databases to ascertain the underlying risk factors. The ultimate goal is to expedite the diagnosis and treatment of Alzheimer’s and other complex diseases.
Lead author Alice Tang, an MD/Ph.D. student at UCSF’s Sirota Lab, emphasized, “This marks an initial stride towards utilizing AI on routine clinical data not only for early risk detection but also for unraveling its biological underpinnings. The strength of this AI approach lies in identifying risks based on disease combinations.”
The research outcomes were published in Nature Aging.
Insights from Clinical Data and Predictive Capabilities
Researchers have long been dedicated to uncovering the biological determinants and early indicators of Alzheimer’s disease, a degenerative form of dementia that impacts memory retention. Alzheimer’s affects approximately 6.7 million Americans, with a higher prevalence among women. While age is a significant risk factor for the disease and women generally outlive men, this alone does not explain the higher incidence among women.
By analyzing UCSF’s clinical database comprising over 5 million patients, researchers identified co-occurring conditions in Alzheimer’s patients from the Memory and Aging Center compared to non-AD individuals. They achieved a 72% predictive accuracy in forecasting disease onset up to seven years in advance.
Several factors such as hypertension, high cholesterol, and vitamin D deficiency were predictive for both genders. Additionally, erectile dysfunction and enlarged prostate were indicative for men, whereas osteoporosis emerged as a crucial predictor for women.
However, the presence of osteoporosis does not guarantee Alzheimer’s development. Tang clarified, “Our model’s predictive ability stems from the synergy of multiple diseases. The identification of osteoporosis as a predictive factor for females underscores the intricate relationship between bone health and dementia risk.”
Advancing Precision Medicine
To delve into the biological basis of the model’s predictive prowess, researchers delved into public molecular databases and utilized SPOKE (Scalable Precision Medicine Oriented Knowledge Engine), a specialized tool developed at UCSF. SPOKE, a comprehensive repository of interconnected databases, aids researchers in identifying patterns and potential therapeutic targets. It highlighted the well-established association between Alzheimer’s and high cholesterol, particularly through the APOE4 variant of the apolipoprotein E gene. Furthermore, in conjunction with genetic databases, SPOKE unveiled a link between osteoporosis and Alzheimer’s in women, attributed to a variant in the less-known gene, MS4A6A.
The researchers envision extending this approach to address other challenging-to-diagnose conditions such as lupus and endometriosis.
Marina Sirota, Ph.D., the senior author of the study and an associate professor at the Bakar Computational Health Sciences Institute at UCSF, remarked, “This study exemplifies the potential of integrating patient data with machine learning to forecast the likelihood of Alzheimer’s onset while unraveling the underlying mechanisms.”
More information:
Alice S. Tang et al, Leveraging electronic health records and knowledge networks for Alzheimer’s disease prediction and sex-specific biological insights, Nature Aging (2024). DOI: 10.1038/s43587-024-00573-8
Provided by University of California, San Francisco
Citation:
How AI can help spot early risk factors for Alzheimer’s disease (2024, February 21)
retrieved 21 February 2024
from https://medicalxpress.com/news/2024-02-ai-early-factors-alzheimer-disease.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.
