Physicians and clinicians are often faced with tremendous gaps in necessary data while making diagnostic and clinical recommendations. With the rise of wearable devices enabling constant data collection, wearable sensors can leverage artificial intelligence tools such as big data and machine learning to fill in the gaps, finding unique trends and patterns in the process. By pioneering early research into the effects of physical movement and self-report data, we can see a glimpse into the future of constant, or “365” data collection for the assessment and treatment of ADHD. Recently Lindsay Ayearst and I gave a presentation on this topic at the 16th Annual ADHD Conference hosted by CADDRA. The following is an overview of that presentation.
Recognizing a need for an advancement in technology within the ADHD field
While it’s safe to say that the history of medicine is a story of technological advancement, it’s equally accurate to say innovation isn’t equally distributed. The mental health field is still reliant on techniques that have inherent gaps and limitations when it comes to diagnosis and treatment. For example:
- Patient observation is often limited to office visits
- Many rating scales, while strong validated tools, can be subjective, have bias, and inherent issues of recall
- Not enough patients receiving multimodal treatment
- Disparities regarding equal & accessible diagnosis and treatment
- Geographic variability in treatment
- Compliance/adherence is poor
- Discontinuation of medication – inconsistent administration, expensive, side effects, not working.
It’s clear there are a variety of limitations and this is where technology can help. Not to replace traditional forms of assessment and treatment – but to help fill some of the gaps that could lead to improved clinical decision making and more personalized treatment. If clinicians can know what is happening in real-time, the opportunity for improved diagnosis and treatment is within reach.
With artificial intelligence we gain access to data to resolve questions such as: How active is the patient — more than others of the same age/gender? We can also capture real-time characteristics regarding behavior: How much do they sleep? How well are they sleeping?
It’s easy to see the value of data that, rather than being self-reported is measured in real-time by a wearable and not in a lab. It’s the difference between the question, ”Are you getting enough sleep?” and having the respondent say “yes” vs. telling the patient “the data says you are typically getting up and staying up from midnight to 3 am". It’s the difference between objective data vs. subjective anecdotes.
What is AI and How Can it Help?
As the idea of artificial intelligence has moved into the mainstream, understanding of what it is and what it can do has been swamped beneath a wave of marketing jargon. Machine learning, deep learning, natural language processing (and the list goes on) are all common terms that refer to different types or subsets of artificial intelligence. A breakout of the specific differences between these terms is best left for another time, instead, let me share an example of how artificial intelligence can be used in the service of diagnosing ADHD.
Researchers at the University of Pennsylvania looked at the feasibility of predicting whether a Twitter user has ADHD or not based on their social media language. They studied more than one million tweets from 1,400 users with ADHD and were able to identify certain common characteristics. They concluded that social media language is predictive of ADHD (.836 AUC). While that’s not sufficient for diagnosis it could improve clinical decision making when integrated as part of available data to make a diagnosis.
Wearables and the Advantages of Continuous Data Collection
Beyond what artificial intelligence in and of itself can contribute to the ADHD community, things get interesting when we factor in the ability for wearables, a growing consumer market for the healthcare industry, to play a role in continuous data collection. Advances in microprocessors, batteries, and components opened new doors over the past five years, with on-board, off-board (cloud-based), and hybrid options now available. Advances in software network development allow for secure encryptions/transmissions of data to HIPAA-compliant cloud storage.
The opportunity is particularly strong for treating children with ADHD. The education system has shifted towards data-based decision making, yet current efforts to quantify focus via human efforts are too time-consuming to practically implement. Besides, key stakeholders — parents, clinicians, counselors — have limited visibility to children’s typical daily classroom behavioral/focus functioning.
Popular behavioral data collection metrics such as on-task behavior, class participation, physical activity/step count, and qualitative ratings can be collected faster, easier, and with less subjectivity when automated data collection, such as that done through wearables, is utilized.
There are two types of AI data collection, active and passive. Active requires manual user input and a concerted effort from the user/patient. Passive user input is automatically and seamlessly collected and requires no conscious effort on the part of the user/patient. We can add to this the idea of "365 Data Collection," which refers to an approach of collecting seamless, continuous data over consecutive minutes/hours/days/weeks/months/years. Such data sets allow for a truly comprehensive picture of a patient. 365 Data also allows for a detailed view at nearly any point in time, generating a fluid, systematic, 'living observation' of a patient.
It’s still relatively early in the development of artificial intelligence and consumer wearables in healthcare, but the potential to change the way diagnosis and treatment of ADHD is handled is clear to see.