Researchers Attempt To Predict & Prevent Suicide Using Deep Learning And Math
For years scientists have focused on the causes behind veteran suicide. Now the U.S. Department of Energy (DOE) and several national labs are teaming up to use deep learning and mathematics to intervene before it happens.
And like many scientific advancements birthed by military necessity, the attempt will ultimately serve all facets of the population, including older Americans.
Deep learning and the Internet itself were born from military research.
Consider that it was 1862 when U.S. Surgeon General William Hammond put out a call to medical field officers in the Union Army to send “any specimens of morbid anatomy that might be valuable to military medicine and surgery.”
And it was during the First World War—better known as the first mass killing of the 20th century, with an estimated 10 million military deaths alone— from which ambulances, antiseptic and anesthesia emerged.
About half of the United States’ annual budget is spent on defense. And ironically, some of the greatest advances in life-saving science in human history have come about as a direct result of military innovation.
It’s no secret that suicide is a leading cause of death in the U.S. According to the Centers for Disease Control and Prevention (CDC) 2016 WISQARS Leading Causes of Death Reports, suicide was the tenth leading cause of death overall in the U.S., claiming the lives of nearly 45,000 people. It was the second leading cause of death among people between the ages of 10 and 34 and the fourth leading cause of death among individuals between the ages of 35 and 54.
In fact, based on death certificate information compiled by the CDC, there were more than twice as many suicides (44,965) in the U.S. as there were homicides (19,362) in 2016. And among females, the suicide rate was highest for those aged 45-54 (10.3 per 100,000). Among males, it was highest for those aged 65 and older (32.3 per 100,000).
Though older adults make up only 12 percent of the U.S. population, they account for 18 percent of suicide deaths, according to the American Association for Marriage and Family Therapy (AAMFT). The risk increases with age, as 75- to 85-year-olds having higher rates of suicide than those who are between 65 and 75, and individuals 85 or older have the highest risk of all.
And elderly suicide rates are estimated by the AAMFT to be under reported by 40 percent or more due to “silent suicides”—overdoses, self-starvation, self-dehydration and “accidents.” The AAMFT reports that the elderly also tend to have higher double-suicide rates, wherein both partners take their own lives at the same time.
In the wake of these sobering statistics, including the fact that someone in the U.S. dies by suicide every 12 minutes, a leading team of U.S. researchers is working to save lives using an unlikely method—math.
The multi-faceted effort spans eight U.S. laboratories and multiple universities and colleges. Its goal is to gain insight into patterns associated with high-risk populations—such as war veterans—and develop an artificial intelligence tool that can predict when a suicide attempt is looming, enabling health practitioners to intervene before it’s too late.
“Suicide is an urgent public health crisis and yet preventing it remains a largely unsolvable problem,” said Xinlian Liu, an associate professor in the Computer Science and Information Technology department Hood College in Frederick, Maryland, and affiliate with the Lawrence Berkeley National Laboratory (Berkeley Lab) in Berkeley, California. “Our research approach is challenging because we’re analyzing many risk factors at once, beyond just mental health, and identifying complicated patterns of human behavior that can then be used like an early warning system. Math and computer science are essential for this analysis.”
Researchers in the Computational Research Division (CRD) at the Berkeley Lab are applying deep learning and analytics to the enormous amount of data held in the electronic health record (EHR) of the Veterans Administration (VA) to help the VA address a host of medical and psychological challenges affecting many of the nation’s 700,000 military veterans.
Previous suicide research has focused on biology, attempting to identify genetic and environmental factors that contribute to suicide, Liu said. But the recent availability of large EHR datasets combined with deep learning techniques have fueled a new wave of research that looks at tackling the problem using healthcare information. “Whereas biologists are more interested in genomes and pinpointing specific physiological factors that put people at a higher risk of suicide, we are focused on analyzing thousands of complex factors in order to establish and recognize patterns of behavior,” Liu explained. “Many studies have focused on finding a specific cause and they have not been successful. We are less focused on isolating a particular cause, but rather more interested in finding a pattern from all factors combined including genetic and environmental from the vast data provided by the VA.”
For example, the most common means of attempted suicide is drug overdose, which often takes multiple unsuccessful tries that end in a hospital visit. Assuming that an unplanned hospitalization therefore correlates strongly with a suicide attempt, the researchers are training their model to—through deep learning algorithms—recognize the pattern.
Liu said researchers hope these patterns will ultimately teach them how to distinguish between suicidal ideation and the person who will actually follow through with killing themselves. “How do I know this person really wants to commit suicide,” Liu said. “Many seniors say they don’t want to live. Is that the same as wanting to commit suicide versus someone who says, ‘I don’t want to live and I want to kill myself?’”
“We’re trying to predict when a patient may have an unplanned readmission to the hospital a month before it happens. We’re working to provide a predictive tool for doctors and clinics so they can quickly identify a high-risk group and prioritize resources to reach out to them,” Liu said, noting that if clinicians can be provided with a warning, they will have time to intervene. “Even if they simply call and ask if the patient is doing okay, it will make a difference, since we already know that reaching out to patients is critical for suicide prevention. We’re using high-end math to solve a complicated problem that will ultimately save lives.”
Part of a collaboration between the DOE and the VA, the project combines the VA’s EHR system with DOE’s high performance computing, artificial intelligence and data analytics resources. According to the Berkeley Lab, through the “Million Veteran Program-Computational Health Analytics for Medical Precision to Improve Outcomes Now” (MVP-CHAMPION) partnership, announced in May 2017, the DOE and VA are working to apply supercomputing, networking and software development resources at several national laboratories to medical data sets collected by the VA from some 700,000 veterans and EHR data from another 22 million veterans. The initial focus of the program is on suicide prevention, prostate cancer and cardiovascular disease.
“The Energy Department is notably helping our veterans by using the world-class artificial intelligence and supercomputing capabilities at our National Labs to address suicide risks, traumatic brain injury, opioid addiction and a number of other critical areas,” said U.S. Secretary of Energy Rick Perry in a statement, “I am truly pleased that DOE is providing such strong support to President Trump’s ‘call to action’ to ‘empower veterans and end the national tragedy of veteran suicide.’”
Last month, President Trump signed an executive order titled “National Initiative to Empower Veterans and End Veterans Suicide,” focusing on improving the quality of life for America’s Veterans and ending Veteran suicide.
According to the VA, the executive order “mandates the establishment of the Veteran Wellness, Empowerment and Suicide Prevention Task Force, which will include the Secretaries of Defense, Health and Human Services, Energy, Homeland Security, Labor, Education and Housing and Urban Development, as well as the Director of the Office of Management and Budget, Assistant to the President for National Security Affairs, and Director of the Office of Science and Technology Policy.”
Liu said the Berkeley Lab team’s goal is to improve identification of patients at risk for suicide through new patient-specific algorithms that can provide tailored and dynamic suicide risk scores—such as whether a person who has been in the hospital for a suicide attempt will attempt it again within 30 days—and make these resources available to VA caregivers and patients.
Five college students—Rafael Zamora-Resendiz, Shirley Wang, Shahzeb Khan, Cheng Ding and Ryan Kingery—spent their summer at the lab through the Computing Sciences (CS) summer internship program working with Silvia Crivelli, a computational biologist in CRD who has been spearheading Berkeley Lab’s involvement in the suicide prevention project and Liu. The students developed algorithms to do statistical analysis of EHRs to look for key factors related to suicide risks and apply deep learning methods to these large and complex datasets. “Working with a publicly available dataset (MIMIC-III) that contains medical record information on about 40,000 patients from one Boston hospital intensive care unit, they searched for patterns that might point to suicide risk,” Berkeley Lab reported.
Liu presented his research team’s project at a Society for Industrial and Applied Mathematics (SIAM) conference last month. The work was also featured in an article on the March 1, 2019 SIAM News blog. Headquartered in Philadelphia, Pennsylvania, SIAM is an international society of more than 14,000 individual, academic and corporate members from 85 countries. SIAM seeks to build cooperation between mathematics and the worlds of science and technology to solve real-world problems.
Since EHR datasets contain both structured data (such as demographics, prescribed medications, lab work and procedures) and unstructured data (such as handwritten doctors’ notes and discharge notes), the team’s early efforts focused primarily on finding patterns in the EHR’s diverse and complex information. For example, Zamora-Resendiz was charged with developing a deep learning network that can distinguish and classify patients at high risk for suicide from discharge notes and physicians’ notes in these datasets.
“We first trained the neural network to classify between patients who are at high risk for suicide and those who are not to find patterns in the doctors’ language,” said Zamora-Resendiz. “We then applied some techniques on the trained network to find which words contributed most to the final prediction. The real challenge is figuring out a way of tracing how these words are combined internally within the network. This will help provide better insight on common motifs found between suicidal patients.”
Though the performance of these neural networks is impressive, they are hard to interpret, Liu said. Researchers hope that incorporating the physician’s knowledge will improve the accuracy and defensibility of scientific deep learning models.
Liu said he will head back to Berkeley Lab this summer with more students to focus again on the project, as researchers are expected to gain approval to use the Million Veteran Program (MVP) dataset—which contains even more data than the MIMIC dataset—to further extend their research. The MVP is a national, voluntary research program funded by the VA’s Office of Research & Development with the goal of building one of the world’s largest medical databases by collecting blood samples and health information from one million Veteran volunteers.
Data-driven scientific discovery is poised to deliver breakthroughs across many disciplines, and deep learning methods represent a promising approach for analytics in science for discovering subtle patterns in very complex scientific data of all kinds, Liu said. According to the Berkeley Lab, researchers hope “the use of these methods for natural language processing will have a significant impact on energy applications that involve human interaction, in addition to veterans’ healthcare.”
Berkeley Lab addresses the world’s most urgent scientific challenges by advancing sustainable energy, protecting human health, creating new materials and revealing the origin and fate of the universe. Founded in 1931, Berkeley Lab’s scientific expertise has been recognized with 13 Nobel prizes. The University of California (UC) manages Berkeley Lab for the DOE’s Office of Science, the U.S.’s single largest supporter of basic research in the physical sciences.
Liu said the Berkeley team expects to begin working on the MVP data by the end of April. This data has required a considerable approval process, he said, including specific security clearances and additional training. “The most powerful computers are within the DOE. Their specialty is in super computing, so we are talking about huge data, over 400 terabytes. No one else has the capacity to process that kind of data.”
Liu said saving people from taking their own lives is one of the best ways he can think of for using deep learning. “We feel tremendously like we’re doing something meaningful,” he said. “The power of deep learning is that it can go through large data and detect a very faint signal; and that is what has been making suicide prevention so difficult. We are very optimistic about making some progress in the field. The window of action between when someone thinks about suicide or does it is 48 hours. We want to utilize the power of super computers to intercept them in that time. And if our algorithm can identify and stop just one or two, we will feel really good about that.”