A Review of Suicide-Related Risks through Propensity Score Methods using the Maryland Suicide Data Warehouse

Predicting who might take their own life is one of the most difficult challenges in medicine. Doctors and researchers have long known that certain conditions, like severe depression or substance use, make a person more likely to die by suicide. However, knowing which factors are the most important is complicated because people rarely have just one problem. A person might be struggling with anxiety, losing their job, and dealing with chronic pain all at once. When researchers try to study these situations using computer models, the sheer number of overlapping factors can confuse the results. It becomes hard to tell if a specific diagnosis is the cause of the risk, or if it is just a side effect of the person's difficult life circumstances. To solve this, scientists have developed statistical tools that act like a scale, allowing them to weigh different groups of people so they can compare them fairly, as if they were identical in every way except for the one factor being tested.

A team of researchers at Johns Hopkins University decided to test these tools on a massive collection of real-world data to see if they could clarify what truly drives suicide risk. They worked with a database from the state of Maryland that links medical records from hospitals and clinics with official death records. This allowed them to look at nearly one million living patients and compare them to almost two thousand people who had died by suicide between 2016 and 2020. The researchers wanted to see if using these advanced weighing methods would change the story the data told, particularly regarding how often people visited doctors and what specific diagnoses they carried in the year leading up to their death.

The researchers started by looking at the raw numbers without any adjustments. Unsurprisingly, they found that people who died by suicide were much more likely to have been diagnosed with mental health conditions like bipolar disorder, post-traumatic stress, or psychotic disorders compared to the living patients. They were also more likely to have struggled with drug or alcohol use. In the unadjusted data, having a diagnosis of bipolar disorder made a person appear about twenty-six times more likely to die by suicide than someone without that diagnosis. Similarly, having a history of substance use or social struggles like housing instability showed strong links to the outcome. The data also showed that these individuals visited doctors, emergency rooms, and hospitals far more often than the general population, especially in the days and weeks before they died.

However, the story changed significantly once the researchers applied the statistical weighing methods to balance the groups. When they adjusted for the fact that people with mental illness often have other overlapping problems, the strength of the connection between certain factors and suicide risk shifted. For some conditions, the risk appeared even higher than the initial numbers suggested. For example, after balancing the groups, the link between opioid use and suicide death became stronger, with the adjusted analysis showing an odds ratio of 3.08 compared to 2.1 in the initial model. The same was true for people with multiple substance use disorders; the adjusted analysis showed their risk was far greater than a simple count of hospital visits would indicate.

Perhaps the most surprising shift occurred with social needs. In the unadjusted analysis, having coded social needs like housing instability or food insecurity actually appeared to lower the risk of suicide, which seemed counterintuitive. But once the researchers used the weighing methods to account for the complex web of medical and social factors these patients faced, the picture flipped completely. The adjusted results showed that these social struggles were actually strong risk factors, increasing the likelihood of suicide death by more than three times. This suggests that the initial models were missing the full picture, likely because people with severe social needs were also receiving different types of care that masked the true danger.

The study also looked closely at how often people saw a doctor and how close to their death those visits happened. The researchers found that the frequency of contact was a powerful signal. People who died by suicide had a much higher number of medical encounters in the year before they died compared to living patients. This difference became most dramatic in the final week of life. In the seven days before death, 2.6% of the people who died had six or more separate medical visits, whereas only a tiny fraction of the living control group had that many visits. When the researchers calculated the risk based on this high frequency of contact, the difference was stark: having six or more visits in a week was associated with a risk nearly one hundred times higher than having fewer visits. This pattern held true across all the different statistical methods they used, confirming that a sudden spike in medical activity is a critical warning sign.

The researchers concluded that these statistical weighing tools are essential for getting the right answer. They found that traditional methods often smooth over the details, making it hard to see the true strength of a risk factor. By using these tools, they could see that the relationship between social struggles and suicide is much stronger than previously thought, and that the sheer volume of medical care a person receives in the final days of their life is a massive indicator of risk. The study does not claim to have solved the problem of suicide prediction, but it offers a clearer way to look at the data. It suggests that when clinicians see a patient with a history of social hardship or a sudden increase in hospital visits, they are looking at a much higher level of danger than standard models might indicate. The work highlights that understanding suicide risk requires looking at the whole person and their entire history of care, not just isolated diagnoses.