Contributed Commentary by Jessica Robin, Winterlight Labs 

September 8, 2023 | As experimental drugs to treat Alzheimer’s disease increasingly target the disease’s very initial stages, monitoring the impact of novel treatments on cognition has become a complex undertaking for trial sponsors. The challenge lies in monitoring the effects of these treatments during the crucial early onset, where several recent interventions have fallen short of showcasing positive outcomes. This difficulty may be attributed, in part, to the lack of sufficiently sensitive tools capable of detecting the subtle changes in cognitive function that are so critical at this stage.  

A pressing need for more sensitive evaluation tools has spurred the emergence of digital biomarkers, among which speech analysis powered by natural language processing (NLP) has gained significant traction. By monitoring and analyzing changes in speech patterns over time, this approach holds promise for gaining profound insights into the impact of novel treatments during the earliest phases of Alzheimer’s disease. This advancement could offer a more precise assessment of disease progression and, critically determine whether a new drug can effectively impede or halt its progress.  

Natural Language Processing 

Speech offers a rich source of information for gauging disease severity and progression in many neurological conditions, including Alzheimer’s disease. Scientific literature extensively documents that alterations in speech and language patterns can be detected in dementia populations. These changes to speech may be detectable years before other noticeable symptoms, serving as one of the earliest signs of the condition. Patients may have trouble finding the right words and exhibit more frequent pauses in their speech. Additionally, they may display what clinicians describe as ‘empty speech,’ characterized by a lack of information content.  

These deficits in language result in measurable changes that make speech analysis a valuable tool for quantification and novel measurement forms. Using speech assessments could help researchers to unlock profound insights into the intricate cognitive changes associated with Alzheimer’s disease and pave the way for more comprehensive diagnostic tools and a better understanding of the effect of therapeutic interventions.  

Automated Speech Assessments 

Automated speech analysis offers researchers a unique opportunity to assess various speech and language domains simultaneously. Emerging technology can delve into the acoustics and characteristics of a person’s voice and evaluate vocabulary usage, sentence construction, and the meaning and content of spoken words. By exploring these different aspects, researchers can identify disease-related changes that provide meaningful signals for clinical research.  

Metrics can now be derived from as little as a 30-second voice recording, which can be gathered on a person’s device through various mechanisms, such as digital health apps, digital therapeutics, telemedicine platforms, or customized data collection apps. Once the recordings are collected, they are directed through an acoustic and linguistic processing pipeline. Signal processing techniques break down the acoustic signals, while natural language processing algorithms decode the spoken content. This process generates nearly 800 metrics, providing valuable insights into different aspects of speech and language. These metrics can serve as inputs for feature composites, machine learning models, or AI algorithms, enabling accurate disease progression and diagnosis predictions.  

Extensive validation efforts are underway to explore these measures’ complete potential in detecting Alzheimer’s disease. Classification models have been developed to predict the presence of the condition based on vocal and linguistic signals, achieving an accuracy rate of over 83% based on a few minutes of speech. Changes to speech have additionally undergone rigorous validation against existing disease biomarkers and clinical assessments to ensure their reliability. A critical objective, especially valuable for clinical trials, is to track changes over time. Specific metrics have been identified and constructed to monitor disease progression in untreated individuals, with the hope that these metrics will prove sensitive to the effectiveness of treatments.  

Automation Advantages For Clinical Trials 

Speech, as a fundamental component of daily life, provides an ecologically valid and functionally relevant measure. One notable advantage of speech assessments is their low burden on patients, typically requiring less than five minutes to complete. This makes them particularly well-suited for remote assessments conducted more frequently than clinic visits. These measures are also automated and objective, complementing existing speech and language evaluation methods that rely on expert clinical administration or subjective rating. Integrating speech assessments can mitigate biases associated with subjective interpretations, leading to more reliable and standardized measurements.  

Speech assessments are now used as an exploratory endpoint in Alzheimer’s disease trials alongside traditional measures. Genentech, in a recent study focusing on prodromal-to-mild Alzheimer’s disease, employed an automated speech processing pipeline to identify progressive changes in speech and language patterns among participants over 18 months. The study findings, published in Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring, a journal of the Alzheimer’s Association, demonstrated that automated measures could effectively characterize longitudinal decline, showing comparable sensitivity to traditional clinical-administered neuropsychological assessments.  

Notably, the speech scores derived from automated speech samples were achieved in approximately two minutes, in contrast to clinical interviews that can take up to an hour to administer and score. These findings highlight the potential of automated, objective speech-based scores as sensitive measures of disease progression and treatment response for speech-related symptoms in Alzheimer’s.  

Clinical trials exploring novel treatments for Alzheimer’s disease encounter distinct challenges. Integrating automated speech assessments could address some of the limitations of traditional endpoints. These evaluations may offer enhanced sensitivity and the potential for earlier detection of changes while supporting remote and decentralized study designs. Leveraging automated speech assessments may enable researchers to gather more comprehensive data, ultimately strengthening the evidence for the effectiveness of interventions.  

Jessica Robin is the Director of Clinical Research at Winterlight Labs. Jessica leads Winterlight’s clinical research program, including internal validation and development research on speech-based biomarkers for neurodegenerative and psychiatric diseases and disorders. Jessica completed a PhD in cognitive neuroscience from the University of Toronto and a postdoctoral fellowship at the Rotman Research Institute. She can be reached at jessica@winterlightlabs.com.