Blog

Suicide Prevention with Modeling Tools

Read Time: 3 Minutes

Suicide is an important problem around the world, causing significant morbidity and mortality. It impacts family and friends, causing long-lasting wounds. Many people are interested in finding effective solutions to the problem, often looking to technology for answers.

Facts about Suicide

Before we can find effective solutions, we need to understand the problem. Figure 1 from the Centers for Disease Control and Prevention (CDC) Suicide Prevention website shows some of the relevant numbers related to suicide. The number of people committing suicide may be undercounted since there is a tendency to avoid calling a death at suicide. In addition, some forms of suicide such as death by cop can be missed.

Figure 1: Over 49,000 people died by suicide in 2022. That is one death every 11 minutes.

When we look at the numbers, the number of people seriously thinking about suicide is 4-5% of the population. Of this group, only 0.4% die from suicide. This suggests a heterogenous problem of varying severity that may require a complex strategy for different subpopulations.

It is often tempting when crafting a solution to make assumptions that simplify the task. In the case of suicide, it is very easy to recommend that everyone at risk be sent to the Emergency Department (ED) for assessment. However, this strategy introduces several problems. The number of people seriously thinking about suicide could overload the system. Many people who attempt suicide may have little or no health insurance, causing financial distress. Finally, because of EMTALA, an Emergency Department cannot discharge a person unless it is reasonably safe to do so. If no psychiatric hospital is willing to take a patient unable to pay, then the patient may be held in the ED for a long period, reducing the ability to see other emergencies. Any solution for the problem of suicide must address the entire spectrum of the disorder, only sending a patient to the ED when appropriate.

Screening for Suicide Using Natural Language Analysis

Several investigators have taken the approach of early detection of suicidal ideation, using natural language processing. They look for words and phrases in a patient’s communications that suggest depression or suicidal ideation. Figure 2 shows such a model that does monitoring a patient’s personal journal using Generative AI.

Figure 2: A model using natural language monitoring to detect suicidal thoughts.

More sophisticated systems can analyze responses over time, looking for trends and patterns. Once triggered, the model decides whether the risk is low, intermediate or high and triages the patient accordingly.

Orchestration

The patient spends the vast majority of her/his time out of touch with healthcare providers. Events that may trigger suicidal thoughts and the resources that can pull the patient back are in the home, workplace, and community. What the patient needs are interventions that prevent an escalation to crisis by optimizing personal resources. At the same time the patient needs to be able to access healthcare providers when necessary.

The patient and the patient’s care may need to be coordinated over months or years between multiple actors:

family and friends
primary care provider
mental health provider
suicide hot-line responder
emergency provider
psychiatric hospital provider

With orchestration modeling it is possible to control the interaction of all these participants over time, as shown in Figure 3.

Figure 3: Model showing interactions for the patient with mental health provider, emergency provider, crisis hot line and psychiatric hospital.

Hopefully, the patient can defuse the situation through interactions with community providers, family, and friends. However, if the issues escalate towards a crisis, then it is important to escalate the care according to need.

Conclusions

Suicide is an important problem around the world. Finding its solution is not simple. However, with the appropriate use of technology and modeling tools we should be able to find appropriate care for this emotionally vulnerable population.