IBM and Pfizer claim that AI can predict the onset of Alzheimer’s disease with 71% accuracy

Pfizer and IBM researchers claim to have developed a machine learning technology that can predict Alzheimer’s disease years before symptoms appear. By analyzing a small sample of linguistic data from clinical language tests, the team says their approach achieved 71% accuracy when tested against a group of cognitively healthy people. I will.

Alzheimer’s disease begins with vague, often misunderstood signs of mild memory loss, followed by a slow, gradual and severe decline in cognitive ability and quality of life. According to the non-profit Alzheimer’s Disease Association, more than 5 million Americans of all ages have Alzheimer’s disease, with at least 14% prevalence of Alzheimer’s disease in all states between 2017 and 2025. Expected to rise. The nature of Alzheimer’s disease and how it settles in the brain, it is likely that the best way to delay its onset is through early intervention. However, it is often too late to prevent the disease from accelerating.

IBM had previously considered using AI to identify proteins that could predict the concentration of amyloid beta, a peptide that changes before memory problems associated with Alzheimer’s disease become apparent. And beyond IBM, others have investigated the ability of AI to characterize Alzheimer’s disease and dementia. For example Unlearn.AINewcomers designing software tools for clinical research have recently paper Lay out a system that can predict the progression of the disease and predict the symptoms that the patient will experience.another paper Co-authored by researchers at the University of California, Berkeley AI system It can superficially predict Alzheimer’s disease from brain scans up to 6 years before clinical diagnosis.

However, IBM and Pfizer argue that this latest study is “significantly” different from previous studies and the application of AI to help predict Alzheimer’s disease. In contrast to studies that focused on subjects with signs of cognitive impairment to predict onset, researchers used samples collected before the subjects in the study experienced the first signs of disability. They also assessed the risk of Alzheimer’s disease in the general population, rather than just targeting high-risk groups, and captured a wide range of people, including those without a family history of the disease or other risk factors.

The study included 703 samples from 270 participants, half of whom developed Alzheimer’s disease before the age of 85 (mean time to diagnosis of mild Alzheimer’s disease was about seven and a half years. did). From a linguistic point of view, researchers have over 87 variables such as misspellings, punctuation, capital letters, verbosity, vocabulary richness, and repeatability. In addition to this, they examined test results from age, gender, education, visuospatial and management reasoning, object naming, memory, attention, abstraction, and Montreal’s cognitive assessment. mocha..

A team of IBM and Pfizer analyzed the transcription of a sample of participants in natural language processing. This allowed participants to use AI to spot subtle discourse changes that they might otherwise have missed. Data from the original subjects (and their descendants and spouses) of the Framingham Heart Study, a population-based study conducted by the US Public Health Service after consent and approval from the Boston University Institutional Review Board. I used. Epidemiology and risk of cardiovascular disease. In the Framingham study, enrolled people were evaluated by a 2-minute Mini-Mental State Examination speech test every 4 years and a neuropsychological test annually if family members reported potential cognitive decline. Will be done.

These steps provided a larger dataset than those used in other studies, allowing us to validate our predictions with real-world results. For example, if a model developed by co-authors of IBM and Pfizer predicts that a 65-year-old Framingham subject will develop Alzheimer’s disease by the age of 85, they will check that person’s records and the subject will be diagnosed with the disease. Whether and when the diagnosis was made.

It should be noted that Framingham visited only thousands of adults in the city of Framingham, Massachusetts. Studies show that much of the data used to train algorithms for diagnosing illness can perpetuate inequality.Recently, a team of British scientists found Since almost all eye disease datasets are from patients in North America, Europe, and China, it is uncertain whether eye disease diagnostic algorithms will work well in racial groups in underestimated countries. In another study, researchers at Stanford University claimed that most of the US data on studies, including medical use of AI, came from California, New York, and Massachusetts.

In fact, researchers have found evidence of bias within their own model that more accurately predicted the development of Alzheimer’s disease in non-university degree participants than in university-degree participants. (76% vs. 70%). They also achieved higher accuracy in females compared to males (83% vs. 64%), averaging 2.61 times better performance in female subjects compared to males.

Recognizing this, IBM and Pfizer researchers plan to use datasets that expand the geographic, socio-economic, and racial diversity of the subject as their work continues. Is called. “This wide range of data is often very difficult to obtain in terms of disease prediction, and accessing the data allowed us to accurately train these models,” they wrote in a blog post. writing. “we [will] Continue training your algorithms, always respecting the core principles of privacy, transparency and consent. “

Teams have their own work Clinical Medicine lancet — Eventually it was supposed to reach the production system, but it could help physicians determine the need for more complex and demanding psychiatric assessments, tests, and surveillance. It may also open the door to more successful clinical trials, as people who are likely to develop the disease may participate in prophylactic treatment trials.

“Our vision is to allow clinicians to one day use multiple AI and machine learning tools to identify whether an individual is at risk of developing Alzheimer’s disease. [Our model’s accuracy] Is significantly higher than the clinical scale prediction (59%). This is a prediction based on other biomedical data available from patients, “the team continued. “One day, doctors may be able to use a combination of voice tests and blood tests to use AI to predict the risk of Alzheimer’s disease and lay the foundation for preventative measures.”