When a drug trial fails, many factors could be involved, including the way the study was designed, the treatment’s efficacy or safety risks. Some of these are easier to fix than others. Innovative diagnostics based on cognitive tests, lab tests and imaging agents, for example, have enabled more efficient selection of patients into clinical trials. This could help with showing whether a drug is safe and effective.
One area where trial failure is particularly notable is Alzheimer’s disease research, where between 2002 and 2012, an astonishing 99.6 percent of trials failed. Patient-selection criteria have been a significant factor, as have safety aspects. These recent high-profile failures prompted the use of molecular imaging techniques to screen for suitable patients at study entry.
However, researchers can now use brain imaging with positron emission tomography (PET) as an adjunct to other clinical evaluations for selecting patients who have amyloid deposits in their brains. These specific protein deposits have been associated with Alzheimer’s.
This technique has improved the process of matching suitable patients for a clinical trial, and has been adopted by pharmaceutical companies. However, significant challenges still exist, and for the past 18 months a team of researchers at GE has been attempting to address these with a digital solution.
The GE team, made up of imaging scientists and software engineers, is developing an algorithm that could further improve the patient-selection process by identifying those who suffer from a form of the disease that is progressing faster.
Current scientific evidence shows that with Alzheimer’s, beta-amyloid accumulates in the brain over time before neurodegeneration develops and cognitive decline begins. The challenge facing pharmaceutical researchers is that each individual’s journey along the disease continuum is different. So when subjects diagnosed with preclinical or mild cognitive impairment due to Alzheimer’s are randomized in a clinical trial, the trial will be made up of a mixed group of patients, each at a slightly different disease stage and trajectory toward dementia. This makes it difficult to gather enough evidence of drug efficacy, particularly as some patients will have significant levels of amyloid present in the brain, but may not develop symptoms of Alzheimer’s for many years or even at all.
The new GE app could help researchers identify which patients have the hallmark amyloid plaques present in their brains and will go on to develop Alzheimer’s within three years, a time frame commensurate with Alzheimer’s disease treatment regimens in most clinical trials.
The app relies on predictive analytics based on machine learning algorithms. It determines a risk score for the likelihood of progression to Alzheimer’s disease for each trial candidate. The software works with a wide range of clinical, genetic and imaging data collected during the diagnostic and trial selection process for each patient. It uses machine learning techniques to combine the data in a new and insightful way. The result is a probability score that makes it possible to select a more homogeneous group of patients for testing new drugs that could modify the disease.
Currently, the app is able to predict which patients will progress rapidly with 86 percent accuracy. This is a 24 percent improvement over using PET imaging alone and suggests an improved ability to select patients for a particular trial. Early feedback has been promising. GE shared the app with pharma companies at the 10th Clinical Trials on Alzheimer’s Disease meeting held late last year in Boston.
The hope is that the app may help bring disease-modifying drugs for Alzheimer’s to the market sooner. The app also possibly could be used in routine clinical practice to identify patients who may respond to new therapies.
One day, patients could even use apps to collect data about their own disease progression, delivering on the promise of personalized medicine.