AI tools speed up data collection on overdose deaths

LOS ANGELES (CNS) – An automated process based on computer algorithms that can read the text of medical examiners’ death certificates can dramatically speed up the collection of data on overdose deaths – which, in turn, can guarantee a time faster public health response than the system currently in use, according to a UCLA study released Monday.


What do you want to know

  • The UCLA research, published in the peer-reviewed JAMA Network Open, used artificial intelligence tools to quickly identify substances that caused overdose deaths
  • The automated process based on computer algorithms that can read the text of medical examiners’ death certificates can ensure a faster public health response time than the system currently in use, according to the analysis released Monday.
  • They found that of the 8,738 overdose deaths recorded that year, the most common specific substances were fentanyl (54%), alcohol (33%), cocaine (26%), methamphetamine (21 %), heroin (18%), prescription opioids (14%), and any benzodiazepine (12%)
  • The researchers noted some limitations to the study, the main one being that the system has not been tested on less common substances such as anti-epileptic drugs or other designer drugs, so it is not known if it would work for those drugs.

The analysis, published in the peer-reviewed JAMA Network Open, used artificial intelligence tools to quickly identify substances that caused overdose deaths.

“America’s overdose crisis is the number one killer of young adults, but we don’t know the true number of overdose deaths until months after the fact,” said Dr. David Goodman-Meza, assistant professor of medicine. . in the division of infectious diseases at the David Geffen School of Medicine at UCLA.

“We also don’t know the number of overdoses in our communities, because the data released quickly is only available at the state level, at best,” he said. “We need systems that disseminate this data quickly and locally so that public health can respond. Machine learning and natural language processing can help bridge this gap.

Recording overdose data currently involves several steps, starting with medical examiners and coroners, who determine the cause of death and record suspected drug overdoses on death certificates, including the drugs that caused the death. The certificates, which include unstructured text, are then sent to local jurisdictions or the Centers for Disease Control and Prevention, which codes them according to a World Health Organization classification of diseases and related health problems.

According to the UCLA researchers, the coding process is time-consuming because it can be done manually. Consequently, there is a significant lag between the date of death and the reporting of these deaths, which slows the dissemination of surveillance data. This, in turn, slows the public health response.

To complicate matters further, under this system, different drugs with different uses and effects are grouped together under the same code – for example, buprenorphine, a partial opioid used to treat drug use disorders. opioids, and fentanyl, a synthetic opioid, are listed under the same code, the UCLA analysis found.

For the new study, researchers used artificial intelligence to analyze nearly 35,500 death records for all of 2020 from Connecticut and nine US counties, including Los Angeles and San Diego. The scientists described how the combination of AI, which uses computer algorithms to understand text, and machine learning can automate the deciphering of large amounts of data with precision and accuracy.

They found that of the 8,738 overdose deaths recorded that year, the most common specific substances were fentanyl (4,758, 54%), alcohol (2,866, 33%), cocaine (2,247 , 26%), methamphetamine (1,876, 21%), heroin (1,613, 18%), prescription opioids (1,197, 14%) and any benzodiazepine (1,076, 12%). Of these, only the classification of benzodiazepines was suboptimal by this method and the others were perfect or nearly perfect.

More recently, the CDC released preliminary overdose data no earlier than four months after deaths, Goodman-Meza said.

“If these algorithms are integrated into the medical examiner’s offices, the time could be reduced as soon as the toxicology tests are completed, which could be around three weeks after death,” he said.

The researchers noted some limitations to the study, the main one being that the system has not been tested on less common substances such as anti-epileptic drugs or other designer drugs, so it is not known if it would work for those. Also, since the models must be trained to rely on a large volume of data to make predictions, the system may be unable to detect emerging trends, the researchers said.

Comments are closed.