In 2016, Canada Health Infoway, a Canadian government funded non-profit organization launched the Data Impact Challenge II as part of an ongoing effort to explore how digital health can be better developed and adopted nationally. Following its inaugural year, the Challenge called again those with legitimate access to the country’s digital information to address some of the pressing healthcare concerns and in turn, reinterpret how big data can be righteously leveraged for related evidence-based decision making in the near future.
SAS, a prominent North Carolina based global analytics software developer, was one of the challengers. They chose to look at how social media could provide additional information to identify young adults aged between 15 and 25 with self-harm and/or suicidal thoughts. The company formed a four-person team making up of healthcare professionals and data scientists and extracted 2.3 million Tweets over 64 days. They then used a machine learning-driven text mining tool to sift out 1.1 million that were perceived to be written by teenagers living in Canada.
The team highlighted certain vocabularies and created topics around these Tweets to estimate the percentage of individuals who might be talking about depression or suicide. Overall, they believe the results fill a gap in existing data, most of which are gathered from old-school written surveys. They trust the analyses shed lights on patterns and trends: to tell whether a specific region in Canada; a particular school, or time of the year, is facing or about to face a mental health crisis. This will create room for more targeted campaigns and render priorities to those who are most at risk. The team went on to win the $10,000 Best Answer Award that they donated to two mental health charities.
The role of analytics in mental health
Recently, AIMed spoke with Greg Horne, SAS Global Principal for Health and the prime motivator behind the company’s participation in the Challenge. Horne said the work that he and his team had done not only pinpointed hotspots where mental health issues may be a problem, but also reflects how many people are openly communicating about the matter but are not actively seeking help; which is a big worry.
Although Horne expressed the Challenge was a springboard for them to know more about the role of analytics in mental health, SAS had in fact been working with The Centre for Addiction and Mental Health (CAMH), the largest mental health teaching hospital in Canada for years, assisting them in their digital transformation and attainment of HIMSS AMAM (Adoption Model for Analytics Maturity) Stage 6 status.
At the moment, SAS is still sharing its analytics expertise with CAMH; in the form of a predictive model, most probably to be used upon inpatient admission which optimizes care for ALC patients (i.e., alternate level of care: patients who are occupying an acute care hospital bed but are not in need of inpatient medical service). By predicting whether a patient is of ALC nature upon admission, CAMH is hoping to be able to move them through the system at a faster pace. Furthermore, SAS and CAMH are also looking at supporting risk flagging through the use of analytics.
Horne is confident that analytics will become the key to reduce and remove stigma in mental health. He cited that last year SAS was working with a US organization to find out why a very high proportion of patients with mental health issues were not attending their first appointment and that’s when they began to realize the importance of virtual care.
“These are the people who had sought help and were given a referral, yet they had chosen to miss their first appointment… What we found was, if I live in a small town and drive my car down the street to the psychiatry office, my neighbors, my friends, and those who know me are likely to question why I was there and it invades my privacy immediately. I may not want others to know that I am seeking mental health support and that’s when virtual care will be helpful,” he explains.
Challenges and vision for the new decade
By pairing analytics and virtual care, a more personalized action plan can be developed for an individual who had come forward to seek help and questions such as “what’s the next best action for this patient”, “what would we advise the patient to do next” and so on, can be answered more readily. “We want to know the background of these patients; we want to have their records and social determinant data available so that at the end of the day, it will make it easier for one to get access to mental health services,” said Horne.
Nevertheless, mental health does pose some unique challenges to analytics. According to Horne, physical health is a relatively quantitative domain since most measures can be captured and recorded numerically. Analytics can be easily run by comparing discrete values. However, when it comes to mental health, most of the data around patients are unstructured; they are text-based, opiniated, and written in a way that’s not easy for analytics to ingest. SAS is tapping onto natural language processing (NLP) to build its own taxonomy of words around mental health to locate information that may enhance its analytics or indicate the outcome that they are searching for.
Besides, Horne noted in the healthcare space particularly, people who are creating the data may not necessarily have been taught to appreciate the value of these data. Speaking from his personal experience, Horne said he used to perform a lot of data collection during his days as a radiographer in London, but most of the time, nobody would explain to him and fellow clinicians the purpose of the collection. As such, there was no absolute commitment or personal connection to ensure the collected data was accurate.
He believes educating clinical staff who are gathering data the value of the data and why the practice of good collection is important will be a grand step towards overcoming the problem of data inadequacy. At the same time, it may also change the “our data is not very good, so we will not do analytics” attitude. “I think sometimes we need to understand that while data doesn’t always have everything we need, it will certainly have elements that get us started. By having that pragmatic approach, we will understand that analytics is very much an iterative process: no one will ever finish developing their models, change their outcomes, and finish acquiring new data. So, don’t seek perfection.”
Horne believes we will witness a big take-off of analytics within healthcare in this new decade primarily driven by the rise of connected devices. Yet, our population remains divided as of now because individuals are very prepared to share their health data openly on social media platforms, but they become anxious when they are asked to share the same piece of information with healthcare providers or the government.
“I think we are going to see more personal data being shared through connected devices and through a change in the way we interact with healthcare as well… But we also have to have a discussion or debate in the next decade on appropriate uses of data and how to get people’s consent. Ultimately, we need to think how this will drive us towards better health outcomes.”