We’ve written about numerous advances in medical diagnostics using artificial intelligence. In one of the most exciting AI-related achievements to date, IBM and Pfizer recently announced the results of a jointly-developed AI model that predicts Alzheimer’s disease with significantly greater accuracy than current clinical scales.

The IBM/Pfizer model analyzes one-to-two minute speech samples from a clinically administered cognitive test with natural language processing. The researchers employed language data speech samples from the Framingham Heart Study, a longitudinal study that has tracked the health of more than 5,000 families since 1948.

The joint research team analyzed language samples of the same people before and after changes in cognition. Due to the richness of the Framingham data, the researchers were able to check the accuracy of the model’s predictions with a high degree of precision. In one example cited by IBM, the model predicted that one participant would develop Alzheimer’s by age 85 based on a sample recorded when that person was 65.

According to an IBM blog post, their model was able to predict Alzheimer’s Disease with an overall 0.70 accuracy and 0.72 accuracy under the curve (AUC, which accounts for true positives and false positives). The model isn’t perfect, but much more accurate than clinical scales based on patient biomedical data which averages overall accuracy of 0.59.

This is an exciting development in the ability to predict the onset of Alzheimer’s. The disease is the sixth-leading cause of death in the U.S. for all ages and the fifth most common cause among people who are 65 or older. There is no current cure for Alzheimer’s, but early intervention can delay the disease’s onset and slow its progress, so a valid predictive tool has significant societal value.