Writing analysis may help predict suicides
Hamilton researchers hoping to create app to alert caregivers of need for intervention
In the two months leading up to the suicide of British author Virginia Woolf, her letters and daily diary entries became increasingly forlorn.
She used negative words such as “nothing,” “last” and “never” more frequently as her bipolar disorder took her down a darkening path.
That trail ultimately led her to wade into the River Ouse on March 28, 1941, her pockets filled with stones, a suicide note left for her husband. “I feel certain that I am going mad again,” she wrote to Leonard. “I feel we can’t go through another of those terrible times. And I shan’t recover this time.”
Her drowning death, combined with her prolific writings, have inspired a team of researchers to try to predict suicide from subtle changes in a person’s writing. Their hope is to create an app that will analyze texts, emails and social media posts of at-risk patients who have consented to participate, so their circle of caregivers can be alerted when intervention is needed.
The research is a collaboration between researchers from St. Joseph’s Healthcare Hamilton, McMaster University and the University of Rio Grande do Sul in Brazil.
“We want to be able to extract the suicidality from the behaviour,” says Dr. Flavio Kapczinski, the lead psychiatrist on the project who works with McMaster and St. Joseph’s. “We could notify the circle of trust that a risk is emerging.”
The research team’s study was published Wednesday in PLOS One, a peerreviewed open access scientific journal. Kapczinski says it is the first step in a project the team hopes will result in a practical application for patients at risk of suicide.
Woolf, whose books include A Room of One’s Own, Mrs. Dalloway and To The Lighthouse, was one of the brightest members of the Bloomsbury Group, a set of British writers, philosophers and artists in the early half of the 20th century. Her vast achievements are even more remarkable considering the struggles of her life. She was sexually abused as a child, was conflicted over her bisexuality and had bipolar disorder (evidenced by periods of mania and of depression) that led to several suicide attempts.
For those reasons, along with the fact that she wrote every day of her life, the researchers chose to focus on Woolf’s words from her letters and diary, believing they would give insight into her “mood states.”
“She was so bright and so productive and then she gave so many warnings,” says Kapczinski.
Clouds created from words frequently used by Woolf in 46 documents written in her final two months were compared with clouds created from random samplings from 54 of her letter and diary entries prior to that, says Dr. Diego LibrenzaGarcia, a post-doctoral fellowship at the university in Brazil. The contrast is stark.
In the cloud compiled from happier times, frequently used words include: love, tomorrow, nice, hope and good.
In the cloud created from her final months, the words include: little, miss, war, nothing, never, can’t and don’t. The researchers write that these “negative words” may indicate Woolf’s “thoughts of lack of efficacy, self-criticism, worthlessness, nostalgia, melancholy and mainly hopelessness.”
In that final period, Woolf’s vivid prose often describes her writing frustrations: “I have written you three separate letters, and torn each of them up”; “I have been trying to write this letter in hand writing, but my hand is like the cramped claw of an aged fowl.”
The researchers created a “text classification algorithm” unique to Woolf’s vocabulary and concluded it would have been able to predict her suicide with 80.45 per cent accuracy.
“She tried to seek treatment,” says Kapczinski, but “treatment was at its beginning.” At one point, Woolf went to Dr. Sigmund Freud’s protégés for help. One treatment prescribed to her was to stop writing.
An app that would build an algorithm for each individual patient, based on their word choice as well as when they write, who they write to and in what format they write, could help predict suicide in a way that has always eluded the medical community, says Kapczinski. “This is addressing a very practical need.”
The approach to estimating the probability of an event occurrence is called “machinelearning.”
A machine, able to unlock the mind?
“My own brain is to me the most unaccountable of machinery,” wrote Woolf. “Always buzzing, humming, soaring, roaring, diving, and then buried in mud.”
“We want to be able to extract the suicidality from the behaviour. We could notify the circle of trust that a risk is emerging.” DR. FLAVIO KAPCZINSKI LEAD PSYCHIATRIST