AI model predicts PTSD risk from ER patients' own words

Daylight saving time linked to 13% more phishing clicks

AI model predicts PTSD risk from emergency patients’ own words

Many emergency departments lack time for an extra PTSD risk assessment, and stigma around mental health can keep patients from saying much. In the Short Wave science roundup on NPR published Nov. 12, hosts Emily Kwong and Regina Barber described a model meant to close that gap.

Researchers took trauma patients’ answers to a routine screening question — what brought you to the emergency department today — and fed those answers into an AI model designed to predict the patient’s risk of developing PTSD a month later. The model got it right around 95% of the cases, Barber said.

Schultebraucks said the results are exciting because they use a patient’s story in their own words. “Some people talk, like, very elaborative of what happens. Some are in such high distress they may only say a few words and a few sentences,” she said. “And we are looking not on PTSD symptoms right now, but we’re looking in what happens in the future.” The findings were published in the journal Nature Communications.

Kwong said trained clinicians who screen for PTSD risk do tend to find it, but the assessment is an extra step many emergency departments have no time for, and stigma around mental health might make some patients clam up. She said using the model could streamline the process so medical professionals can act sooner to prevent PTSD from developing.

What is unique about this work, Barber said, is that the model gave its reasoning at each step of the analysis. Munmun De Choudhury, a computer scientist not involved in the work, said the work shows AI does not have to be a black box. Barber said De Choudhury believes the work has real potential to guide clinicians to the right kind of interventions.

On privacy, Kwong said Schultebraucks reported that the model did not share data on a cloud service and was HIPAA-compliant, and added that Schultebraucks considers it important that any patient consents to having their information recorded and used by an AI model.

Daylight saving time loss preceded more phishing clicks

Barber noted that nearly one-third of U.S. adults don’t get enough sleep. Kwong said Eitan Schechtman, an assistant professor at the University of California, Irvine, wanted to know how bad a modest amount of lost sleep is for decision-making, and tested it with phishing attacks. The university regularly sends fake phishing emails to staff as cybersecurity training, containing suspicious links employees are not supposed to click.

Schechtman worked with the university’s security team to divide 11,000 participants into two groups: some received the phishing email the Monday before daylight saving time and some the Monday immediately after, when some may have lost an hour of sleep. People emailed after daylight saving time were 13% more likely to fall for the phishing attack the following morning and click the link, Barber said. The study was published in the Proceedings of National Academy of Sciences this week.

Muhammad Adeel Rishi, a sleep specialist at Indiana University School of Medicine not involved in the study, found the finding compelling and said it adds to growing evidence about the consequences of daylight saving. Kwong said spring forward has been linked to an increase in heart attacks and fatal traffic accidents.

Schechtman said he wants cybersecurity companies aware of the finding so they can strengthen security around moments of widespread sleep loss, such as holidays or major sporting events. “There might be more of these — more susceptibility to click these links or generally to make impulsive decisions, whether it’s on the road, in the stock market or cybersecurity-related decision,” he said.

Scott Detrow, who hosted the science roundup, noted that some places have gotten rid of time changes, including most of Arizona and Hawaii, and Barber agreed. Kwong said multiple studies have shown that standard time is healthier for our bodies, because standard time has more natural light in the morning, which better aligns to circadian rhythms.

Technique subtracts starlight to reveal a faint exoplanet

The second science story concerned a new technique for finding faint planets beyond our solar system. The exoplanet, 51 Eridani b, was already known; a team studied existing images of the star it orbits to see if light from the planet could be found, as a test of the technique, Barber said. She said the test matters because often it’s hard to see exoplanets in images.

Kwong said exoplanets are hard to see because the light they reflect from the closest star is often much fainter than the light the star gives off — the star, in effect, hides the planet. The researchers subtracted light from the image until they could clearly see the planet and not the stars. The paper was published in Nature last week.

Barber said the technique corrects that problem and millions of other issues that make finding an exoplanet harder, such as the telescope being unstable, instrumental errors or temperature fluctuations. She added that the work was done with physics knowledge and algorithms and no AI.

Because the images were taken in infrared by the James Webb Space Telescope, Kwong said, they can tell researchers more about what these exoplanets are made of. Rodrigo Ferrer-Chavez, lead author and a Ph.D. candidate at Northwestern University, said: “With this, it is the first time that we can actually get this specific information-rich image of this planet that might tell us a lot about how it formed.”

Ferrer-Chavez also said the tool can be used to find even more exoplanets hiding in James Webb Space Telescope images. Barber said the technique could help future telescopes that are already supposed to be more powerful, but added that other astronomers she talked to described it as just one cool technique of many that scientists are developing for finding faint, distant worlds.