By Jessica Robin, director, clinical research, Winterlight Labs
Like many other cognitive conditions, Alzheimer’s disease (AD) can cause changes to a person’s speech and language. Difficulty in retrieving words, pauses, word choice, sentence complexity, and
the amount of information conveyed are all signals that can help identify the disease and tell us how far it has advanced. While some of these changes can be obvious in conversation, others are more challenging to identify, even for trained professionals.
Artificial intelligence (AI) and machine learning scientists are working on automating speech assessments to detect the disease earlier and improve understanding of the effectiveness of novel treatments. Incorporating these measures into today’s AD clinical trials may provide a more clinically relevant way to assess disease progression and patient response to interventions.
Speech As A Clinical Assessment Tool
Research into AD can incorporate several different assessment tools and various clinical endpoints. Most standard primary endpoints are clinical measures of cognition and function, such as the Clinical Dementia Rating (CDR) scale or the Alzheimer’s Disease Assessment Scale–Cognitive Subscale (ADAS-Cog). While few would dispute the ability of these measures to offer a clinical judgment on disease severity, their ability to adequately demonstrate meaningful changes in the prodromal stage of AD has come under increasing scrutiny. This limitation may even factor in the number of early-stage interventions that have failed to demonstrate efficacy in Phase 2 and Phase 3 clinical trials.
Standard approaches to testing are complex and often burdensome to patients. Single tests can take up to an hour to complete, and most need to be supervised in a clinic by a clinician or psychometrist.
While assessing patient speech is often part of clinical practice, developments in natural language processing and machine learning are leading to the development of tools to measure speech in more accessible ways. Emerging technology can automatically extract hundreds of acoustic and linguistic properties of natural speech. This fine-grained data may help characterize speech patterns associated with early-stage AD before other symptoms manifest.
Advancements in speech biomarkers are helping to establish a better picture of the common aspects of speech that can decline with the progression of certain dementias. A 1-minute recording can be used to detect hundreds of speech characteristics. These include lexical diversity, syntactic complexity, semantic content, and acoustic properties of the voice. Measurements can offer clinicians and investigators rapid insights into the severity of a patient’s condition that is not compromised by subjectivity, offering a detailed understanding of the subtle ways patients’ symptoms improve or deteriorate in response to novel treatments.
Taking Speech Assessments To Patients
Advancements in speech-based assessments are emerging as research moves toward decentralized study designs that demand improved ways of measuring patients remotely. Via the microphone of a patient’s tablet or smartphone, speech assessments can collect rich patient data through various tasks that take 1 to 2 minutes. These include picture description, paragraph reading, or simply asking a patient to express how they are feeling that day. Recordings pass through a speech analysis platform, are assigned a score, and, subsequently, these outcome measures are provided to a trial sponsor as an exploratory endpoint. The approach is scalable and repeatable and, crucially, enables patients to integrate the short assessments into their daily lives at their convenience and at high frequency.
With enthusiasm growing for the use of speech in AD trials, there are several points that those working in AD research might want to consider.
Considerations In Incorporating Speech As An Endpoint
The acoustic and linguistic properties of speech should both factor in assessments. While previous research identified changes to acoustic properties (those relating to the sound wave) of the voice relating to disease, the latest research found that the linguistic components, including word choice, sentence structure, and the content of what is said, are equally if not more critical in Alzheimer’s disease and other forms of dementia.
Any technology choice should be backed by scientific rigor, validation, and published results to ensure transparency. There ideally needs to be a rich background of scientific evidence to show that the technology consistently and accurately measures a clinical outcome.
Investigators will also want to consider the point at which they include speech assessments in their research. Starting early, with proof of concept and methodology studies can generate evidence to motivate inclusion as an endpoint in Phase 3 trials. That said, where pre-existing data are available, sponsors have successfully brought in speech assessments at later stages.
The assessment of speech is complementary to most study designs. While they may not form part of an official regulatory submission, including these assessments along the development pathways will build up an additional evidence base for investigators. Several large pharmaceutical companies have included speech assessments as an exploratory endpoint in intervention trials. Some companies have already moved past pilot studies to more extensive Phase 2 or 3 dementia trials, including speech as an experimental endpoint.
The Future Of Speech-Based Digital Biomarkers In AD Research
As command of speech and language are ecologically valid and functionally relevant clinical markers of disease progression, the advancement of objective digital tools that can measure changes are helping to establish more sensitive endpoints administered remotely and unobtrusively to patients. While speech-based biomarkers are currently being used as an exploratory endpoint, they hold promise for detecting subtle signals in language that can point to early indicators of AD or response to effective therapies. As the industry builds the evidence and validation of these measures, we will see more trials using them as primary and secondary endpoints.
Jessica Robin will present new data on the characterization of progressive speech changes in early-stage AD at the Alzheimer’s Association International Conference on July 31, 2022.
About The Author:
Jessica Robin is the director of clinical research at Winterlight Labs. She leads Winterlight’s clinical research program, including internal validation and development research on speech-based biomarkers for neurodegenerative and psychiatric diseases and disorders. She completed a Ph.D. in cognitive neuroscience from the University of Toronto and a postdoctoral fellowship at the Rotman Research Institute.