ML Analysis of Handgun Purchases Can Forecast Suicide Risk
New research shows that administrative data on handgun transactions, analyzed using machine-learning techniques, may predict firearm suicide risk and provide insights to inform prevention strategies.
Source: Getty Images
By - A new study published in JAMA Network Open indicates that machine-learning (ML) models designed to analyze handgun transaction data can accurately forecast firearm suicide risk, resulting in insights that could inform targeted interventions for suicide prevention.
According to the study, research suggests that limiting access to firearms among individuals at high risk for suicide is an effective method for suicide prevention, but accurately identifying them to initiate an intervention is a challenge. The authors of the study hypothesized that firearm purchasing records might offer a large-scale and objective data source for developing tools to predict firearm suicide risk.
To test their hypothesis, the researchers used California’s Dealer’s Record of Sale (DROS) database, which consisted of 4.9 million handgun transaction records from 1.9 million individuals between Jan. 1, 1996, and Oct. 6, 2015. The DROS records include purchaser identifiers, addresses, dates and times of the transaction, identifiers for the seller, and firearm calibers, types, makes, and models.
They fed this data into an ML-based model designed to predict suicide risk based on 41 factors pulled from the DROS data. The researchers also pulled California death records from 1996 to 2016 to measure rates of firearm suicide within one year of firearm transactions, as defined by the ICD-9 and ICD-10 code sets. They identified firearm suicide deaths by linking purchasers in DROS to the California Department of Public Health’s Death Statistical Master File, using probabilistic matching on name, date of birth, and gender/sex.
Overall, 2,614 unique individual purchasers representing 3,278, or 0.07 percent, of total transactions died by firearm suicide within one year of purchase. Generally, transactions involving people who died by firearm suicide included a larger fraction of revolver purchases, a higher average age of the purchaser, a higher proportion of White purchasers and female purchasers, and fewer average prior firearm purchases in the previous 10 years.
Those who died by firearm suicide had fewer handgun transactions. Further, approximately 69.5 percent of those purchasers who died by firearm suicide, compared to 39.3 percent of other purchasers, had no prior transactions.
The ML model’s firearm suicide risk analysis utilized random forest classification, in which a higher random forest score correlated with higher risk. Among the top 5 percent of risky transactions, close to 40 percent were associated with a purchaser who died by firearm suicide within one year. Among the small number of transactions with a random forest score of 0.95 and above, more than two-thirds were affiliated with a purchaser who died by firearm suicide within one year.
Fifteen predictive variables were found to be most important for firearm suicide risk forecasting, including handgun category (revolver, semiautomatic, or other), purchaser race and ethnicity, purchaser age, the month of the transaction, the number of transactions within the past 10 years, the distance between the purchaser’s address and the dealer, and the census tract percentage of the population younger than 18 years.
These findings show that passively collected individual-level handgun purchase data can be used to create moderately informative predictive algorithms of firearm suicide risk, the researchers stated.
These predictive algorithms have the potential to generate insights that could be used to inform future firearm suicide prevention strategies and targeted interventions, they added.
