How your Twitter feed can predict postnatal depression

"Discrimination in relation to pregnancy and return to work after parental leave is a continuing problem in Australia" ...
"Discrimination in relation to pregnancy and return to work after parental leave is a continuing problem in Australia" ... Fair Work Ombudsman Natalie James 

The risk of a new mother suffering from postnatal depression could be predicted weeks before the birth of her child simply by monitoring her Twitter feed, computer scientists have found.

Microsoft labs discovered it is possible to spot which pregnant women will struggle with motherhood based on the language they use before the birth.

Intriguingly, the algorithm does not depend on the mother talking about the pregnancy or her baby, but picks up subtle verbal cues which reveal her underlying unhappiness or anxiety.

General negativity in language, with a rise in the number of words such as "hate", "miserable", "disappointed", increased use of the word "I" and a jump in the number of expletives are all clues that a new mum will suffer postnatal depression.

"We saw several patterns in the language of women with postnatal depression, said Eric Horvitz, co-director of Microsoft Research in Washington.

"Then we wondered if we could go back in time and see if this trend could be spotted before the birth. And we found we could.

"We found that two to three weeks before the birth the same clues were there in around 80 per cent of cases.

"Psychologists have found in strong work that shifts to higher frequencies of the use of first-person pronoun can indicate onset of depression, as people become more self-focused.

"You really get a feeling of what is going on in the heads of those people who were struggling," he said.


The study looked at the language of several hundred women three months before their birth and three months afterwards.

They noticed the 15 per cent of women who went on to show a severe change consistent with postnatal depression asked more questions, had lower levels of positivity and increased levels of anger and anxiety.

Horvitz believes an app could be designed which picks up on these clues and could direct new mothers towards help.

"Postnatal depression is known to be under reported because of the stigma attached," he said.

"It's not one for Microsoft, but a welfare group could create an app that women could run on a smartphone which warns them of the onset of depression and points them to resources to help them deal with it."

Professor James Pennebaker of the Department of Psychology at the University of Texas has found the content of what people say online is not as important as how they say it through their use of "function words" such as pronouns, conjunctions and prepositions.

"Function words tell us how a person is analysing their world and about their mental state. We can get a sense of how psychologically well they are doing," said Pennebaker.

He believes in future the same algorithms could be applied to the work of historical characters to gain a greater insight into their mental state when they wrote diaries or letters.

Horvitz is one of several scientists who believe big data stores such as Twitter, Facebook and Google could all be used to spot general trends in the population.

Recently it was discovered that unknown side-effects of taking two drugs together could be predicted by looking at search terms related to the medication.

For example many people searched for cholesterol-lowering drug pravachol and antidepressant paroxetine alongside such words as ''tired", "thirsty", "dizzy", "itchy", and "out of breath", exposing a link to raised blood sugar that can come with mixing the drugs which had not been picked up in clinical trials.

It is hoped such data analysis could eventually complement drug databases.

A separate study showed an increase in admissions to Washington hospitals for heart problems could be predicted by looking at how many people had Googled salty food recipes.

Therefore, monitoring online recipe trends could be a simple way for hospitals to improve staffing levels.

A team at Cornell University discovered Twitter posts could be used to predict the general mood of the population, discovering that people are most happy when they wake up.

"We found that people are happiest around breakfast time in the morning and it's downhill from there," said Michael Macy, professor of arts and sciences in the Department of Sociology at Cornell University.

"But it wasn't about work because we found the same pattern on the weekend but delayed by an hour and a half and we think people are sleeping in.

"Baseline happiness was higher at the weekend and we think that being able to wake naturally rather than with an alarm clock was one of the key factors," he said.

Macy's team has also disproved the theory that the cyberspace is borderless.

He demonstrated that alliances on Twitter and Facebook are founded along the same lines as eight traditional cultural and religious divides identified by Samuel Huntingdon, such as Western, Islamic and Orthodox.

"Looking at the digital records of social interactions really supports the idea that the world is aligned by these families of culture," he said.

The academics and scientists were speaking at the American Association for the Advancement of Science meeting in Chicago.

Telegraph, London