Machine learning model forecasts Alzheimer’s-driven cognitive decline

By Nick Taylor

- Last updated on GMT

(Image: Getty/wildpixel)
(Image: Getty/wildpixel)

Related tags: machine learning, Alzheimer, Alzheimer's disease, cognitive decline, Clinical trials, trial recruitment

Computer model predicts worsening cognitive scores, opening door to the use of machine learning in the selection of participants in Alzheimer’s disease trials.

One of the many challenges facing developers of Alzheimer’s drugs is how to identify the people most likely to respond to a therapy. The conventional wisdom is that many clinical trials have enrolled people whose disease is too advanced to treat.

The recent failure of Amgen and Novartis’ BACE inhibitor in healthy people at high risk of developing Alzheimer’s shows disease progression is not the sole barrier to the development of drugs that treat the disease but there remains an interest in improving subject selection.

Researchers discussed the factors underpinning that interest in a paper on their use of a computer model to predict changes in scores on ADAS-Cog13, a scale that measures cognitive performance.

The authors wrote, “Accurate prediction of symptom onset in the time window of 6 to 24 months is critical to participant selection and formation of clinical trials. Thus, having access to accurate future estimates of the progression of cognitive scores such as ADAS-Cog13 within this time frame is of great importance​.”

At the Machine Learning for Health Care conference, the researchers provided an update on their ongoing effort to use a computer model to predict ADAS-Cog13. 

In developing the latest version of the model, a previous iteration of which was discussed in a paper last year, the researchers used the concept of meta-learning to try to improve on the “suboptimal​” results achieved in the past.

Meta-learning entails observing how well machine learning approaches perform on learning tasks and using these insights to identify new ways to do the tasks. The decision to turn to meta-learning was driven by the limitations of the models developed up to that point. 

Oggi Rudovic, a MIT Media Lab researcher and co-author of the paper, said, "We couldn't find a single model or fixed combination of models that could give us the best prediction. So, we wanted to learn how to learn with this meta-learning scheme. It's like a model on top of a model that acts as a selector, trained using metaknowledge to decide which model is better to deploy​."

The change led to “large improvements in accurately forecasting future ADAS-Cog13 scores​,” the researchers wrote, that resulted in the model outperforming existing ways of predicting disease progression.

There remains scope to improve the forecasts, though. In doing so, the researchers plan to assess their models in a larger cohort of patients but that will require the use of techniques to fill in missing information “as the existing data is highly sparse​,” according to the authors of the study.

Related topics: Clinical Development

Related news

Show more

Follow us

Products

View more

Webinars