Tech Reads Distress Between the Lines

A computational tool developed at Western to analyse text message counselling to identify those at risk of a dangerous decline in mental health could transform mental health service delivery to rural and remote communities.

The algorithm was developed by a team led by Associate Professor Mark Antoniou, who studies language, cognition and their interface with technology at the MARCS Institute for Brain, Behaviour and Development. 

Antoniou’s work was part of a federally-funded pilot of Virtual Psychologist, which provided SMS-based counselling with qualified psychologists to Australian farmers. “It’s a service for people who find telephone or face to face counselling too confronting, too embarrassing, too inconvenient, or simply unavailable,” explains Dervla Loughnane, CEO of Virtual Psychologist and a collaborator with Western. 

“Fire, flood and drought are just some of the stressful factors that regularly disrupt the lives of Australians in rural and remote regions”, says Antoniou. In addition, fewer mental health practitioners are in regional centres than in cities, and the distance from a remote property to the nearest service can be vast. 

“Farmers also face obstacles of entrenched stoicism and the stigma around mental health, so it’s a perfect storm of increased need, fewer services, and cultural and behavioural barriers,” Antoniou says.

“If the person is willing to engage, the evidence shows that text-based counselling is as effective as traditional forms of counselling,” he adds. For him, the text-based nature of the service brings with it the possibility of developing smart tools. People who are anxious, stressed or depressed are known to display particular patterns of language use, such as using more first-person pronouns like I, me and my, he explains. So his team developed a tool to detect these patterns in SMS counselling transcripts. “We used advanced computational linguistic analysis on the language use patterns, to glean patients’ current mental health status and make predictions as to where they are headed,” Antoniou says. “The goal is to be able to flag someone who is very distressed and may be having suicidal thoughts.”

“So far, we are able to identify anxiety, stress and depression with about 80% accuracy,” he says. “We are also able look at a person’s language use and predict how they are going to feel weeks or even months away, with an accuracy of 80 to 90%.”The ability to rapidly diagnose treatable conditions by scanning SMS data is extremely powerful, says Loughnane. “Within Australia and globally there is a shortage of mental health practitioners. The data Mark and the team produced is ground-breaking.”