Several psychiatric disorders including major depression, bipolar disorder and schizophrenia are accompanied by changes in vocal patterns and use of language. These disorders are also characterized by fluctuations in symptomatology that are not well captured using standard clinical assessments. More fine grain monitoring of symptoms could be very valuable in early identification of relapse and improve clinical care in these conditions. In this webinar, we discuss how psychiatric symptoms map to speech pattern changes and how this may be leveraged in research and clinical settings.  

Speakers: 

Bill Simpson, Head of Partnerships at Winterlight Labs (Moderator) 

Michael Spilka, Scientist Psychiatry at Winterlight Labs 

Sunny Tang, Researcher and Physician at Northwell Health  

Michael Birnbaum, Researcher and Physician at Northwell Health  

Watch the full recording:

In CNS clinical trials the gold standard primary outcome measures are often clinician rated instruments. While these assessments are capable of capturing changes in disease they can easily be confounded by gaps in rater training or knowledge and errors in administration. To combat this, manual review by experts is often deployed within a trial. This can be costly and does not allow for all assessments to be evaluated. During this webinar, we highlight the potential for automated speech analysis to identify these inconsistencies and scale the review process.

Speakers:

Bill Simpson, Head of Partnerships at Winterlight Labs (Moderator)

Celia Fidalgo, Head of Product at Winterlight Labs

Barb Enger, Clinical Trial Consultant at AVP Clinical and Signant Health

John Harrison, CSO at Scottish Brain Sciences

Watch the full recording

Changes in speech and language are hallmark characteristics of Alzheimer’s disease. While these changes are core to clinical evaluation, we have historically lacked objective tools for measuring these changes in the context of clinical trials. During this webinar, we review advances in speech biomarker validation, using both short, focused speech tasks as well as leveraging passively collected speech from clinical interviews.

Speakers:

Bill Simpson, Head of Partnerships at Winterlight Labs (Moderator)

Jessica Robin, Director of Clinical Research at Winterlight Labs

Mike Detke, CMO at Lighthouse Pharma

Jennifer Sorinas Nerin, Data Scientist at Novartis Institute for Biomedical Research (NIBR)

Watch the full recording:

During the COVID-19 pandemic, healthcare workers have been exposed to substantial amounts of stress, with a negative impact on their mental and physical health. The 4YouandMe Stress and Recovery in Frontline Healthcare Covid-19 Workers Study aimed to understand how different facets of acute or sustained stress could be quantified using remote digital health technologies.      

In this webinar, R&D Scientist at Cambridge Cognition, Dr Alex Anwyl-Irvine was joined by Dr Sarah Goodday and Dr Stephen Friend from 4YouandMe. The panellists discussed a sub-analysis of the Stress and Recovery in Frontline Healthcare COVID-19 Workers study focusing specifically on the impact of the pandemic on mood, sleep and cognition.

Study design

First, Dr Goodday discussed the rationale behind the Stress and Recovery in Frontline Healthcare Covid-19 Workers Study. Targeting a frontline healthcare worker population during the first stages of the pandemic offered a chance to follow an uniquely stressed population. Stress is challenging to measure accurately in the real world, partially due to gaps in how we traditionally measure stress from self-reported measures during infrequent periods.  Further challenges are in disentangling subjective perceptions of stress from objective measures of stress and other associated factors such as mood, cognition and sleep. Better longitudinal measures of real-world, objectively measured stress and characterising temporal aspects of wellbeing, stress, sleep and activity over time may help to broaden knowledge of stress and what makes it risky for some individuals.    

Participants in the study were all US frontline healthcare COVID-19 workers and were mostly nurses. They were asked to wear an Oura smart ring while off-shift and to complete regular assessments on multiple lifestyle and psychological factors using a smartphone app. Follow up lasted between four to six months.

The study was designed in a participant-centric way with feedback loops created through a series of regular phone check-ins and Zoom calls. Feedback was incorporated into the study in real-time. Using this study design methodology, the stress and recovery project had a much higher retention rate than other digital health studies – more than 80% were retained among the original enrolled study sample. 

Analysing the data

Next, Dr Anwyl-Irvine discussed a sub-analysis that focused on a subset of participants who regularly completed questionnaires on mood assessments, including PROMIS, PHQ-9 and GAD. They had also completed cognitive assessments using the Cognition Kit N-back, which measures attention and working memory. By conducting an analysis using data from several time points, we can capture information that would otherwise have been missed.

Participants were grouped through unsupervised machine learning. Patterns were also identified across timeseries variables and a general linear model was constructed to compare these timeseries between groups.  Two clear groups were identified, one of people who had increased mental health difficulties and the other with decreased mental health difficulties. The group with high mental health difficulties scored above a mild threshold for either anxiety or depression and were two times more likely to self-report a diagnosis of a mental health disorder.

Next, temporal patterns were characterised by Multivariate Vector Autoregressive Models using a custom Python library. The data showed that the group with the lowest scores for mental health displayed a tighter coupling between stress, low mood, and less physical activity. A novel result was that the group with more mental health difficulties showed increased N-back performance with less sleep, indicating a possible role of hypervigilance.

Through this analysis, it was shown that there are discrete groups based on baseline health, which were primarily weighted on mental health. Common temporal patterns were found connecting Mood, Stress, and Activity, N-back and Stress and also Sleep and Stress. The patterns vary by mental health grouping, with differences in the Cognition-Sleep and Stress-Activity relationships.

How to conduct a successful real-world study

Dr Friend ended the webinar by discussing the opportunities that come with conducting real-world studies involving several temporal measurements. These include understanding how a disease trajectory unfolds over time and rebuilding taxonomies of disease. It also helps to avoid aggregating disease trajectories and instead building a picture for an individual’s journey with their condition. For example, no two people with Parkinson’s disease are the same, and an individual’s symptoms of the condition change over time. Digital measures that are taken over several timepoints can also help to untangle causes and consequences, for example the complex relationships between mood, stress, sleep and cognition.

Dr Friend concluded by emphasising the elements that are needed for a successful study using digital health and wearables. These include creating a community and building trust, providing tailored and regular support and ensuring data privacy.  

Watch the full recording

Purposeful consideration of statistical analysis can be hugely beneficial for clinical studies by increasing the potential to detect an efficacy signal and reducing the signal-to-noise ratio. Pre-screening and patient stratification are two methods that can be used to help drive efficiencies and boost the statistical power of clinical studies.

In this talk, Dr Elizabeth Baker, Director of Statistical Sciences at Cambridge Cognition, discussed “Successful statistical strategies for pre-screening and patient stratification”. The talk particularly focused on clinical studies with cognitive endpoints.

Dr Baker began by discussing how best to integrate statistical planning into a cognitive study design. The most important place to start is making sure the research question is well understood. This can then be used to derive the study objective, design, and statistical analysis. Dr Baker recommended using the ESTIMAND framework, which helps to elucidate the research question, fully derive the objective, and create a link between the estimate and analysis methods.

It is also important to account for sources of variability, which may be found within the population, the disorder being studied or between trial sites. With cognitive endpoint selection, objective measures of cognition must be used to ensure consistency between different people, sites and raters. The CANTAB^TM Spatial Working Memory task is an example of an objective tool that will reduce variability.

The second part of the talk focused on patient stratification.

For cognitive studies, stratification is of most benefit when there is significant heterogeneity in the population that is caused by a known factor. Stratification can help to ensure balance across the study design, for example by balancing recruitment across centres in multicentre trials and accounting for any between-centre differences in a stratified analysis.  

The EBBINGHAUS study investigated potential cognitive effects of cholesterol-lowering therapies. To ensure cognitive safety was appropriately assessed, it was important to stratify individuals by cholesterol level at screening to tangibly adjust for the risk of a cardiovascular event.

Another condition where patient stratification can be useful is schizophrenia, which has high variability in cognitive symptoms. One way to adjust for this variability is to stratify individuals by cognitive impairment. A recent post-hoc analysis has shown that people with a low level of cognitive impairment in schizophrenia responded differently to treatment than people with a high level of impairment. Stratification can therefore help to identify individuals who are more likely to respond.

The final part of the talk focused on pre-screening and included a deep dive into a study that had successfully used pre-screening: the Models of Patient Engagement for Alzheimer’s Disease (MOPEAD) study.

Pre-screening can help to increase chances of success in recruiting a particular patient population. It involves asking individuals to take an assessment prior to screening, for example a cognitive assessment. This helps to confirm that the individual meets the requirements for the target population before going through the longer screening process. In conditions where there is high heterogeneity in cognitive impairment such as Alzheimer’s disease, pre-screening can help to ensure that study participants are at a similar stage of disease. When partnered with remote testing, pre-screening can also reach participants from more diverse backgrounds.

Watch the full recording:

As we move towards decentralized and hybrid trial designs, it is increasingly important to understand the feasibility and reliability of digital measurement tools.

Here, experts from Otsuka and Cambridge Cognition shared insights from a recent collaborative study, discussing the value of remote monitoring of health and behavior.

Dr. Corey Reuteman-Fowler, Director of Global Clinical Development in CNS and Digital Medicine at Otsuka, presented results from a 4-week non-interventional study in patients with Major Depressive Disorder (MDD), which aimed to evaluate health professional-assisted versus unassisted methods of decentralized trial design. We learned that assistance had no real impact on subjects and despite some participants’ preference for a particular method, data collection was not affected.

Methods of obtaining remote psychological and physiological assessments were explored, including the use of RMDAS, which allows clinicians to observe patients’ movements over video, to assess extra pyramidal symptoms. Challenges and benefits of recruitment via social media were discussed, particularly in relation to patient demographics.

Dr. Francesca Cormack, Director of Research and Innovation at Cambridge Cognition, demonstrated the use of voice-based assessments to capture extra-pyramidal symptoms. She spoke to the scientific rationale of using speech as a digital biomarker, including the overlap of brain regions affected by extra pyramidal symptoms and brain regions controlling speech movements.

Two tasks, namely sustained phonation and PaTaKa, were used to analyze vocal acoustics and characteristics of speech. A machine learning algorithm to classify patients experiencing extra pyramidal symptoms showed high accuracy for both global extra pyramidal symptoms and specific subdomains such as slowness or abnormal and involuntary movement.

Drs. Reuteman-Fowler and Cormack summarized take home messages from this study:

• Clinical efficacy measures can be collected remotely for clinical interpretation.
• Decentralized trials, with or without healthcare professional assistance, show promising operational feasibility.

A thoughtful Q&A session explored the broader implications of this study for psychiatry, the promise and challenges associated with regulating digital measurement, and the impact of decentralized trials on increasing diversity in clinical trial populations.

Watch the full recording: