Aug. 10, 2022 – We found that for most substances evaluated, the performance of these algorithms was perfect or near perfect. These models could be used to automate classification of unstructured free-text, thus avoiding the manual and time-consuming process of individually reading each entry and classifying them to a specific substance,” researchers reported in JAMA Network Open.

“Excellent performance was shown for multiple substances, including any opioid, heroin, fentanyl, methamphetamine, cocaine, and alcohol using models for general text. Yet for prescription opioids and benzodiazepines, there was a considerable performance gap.”

That “performance gap” is due in part to weaknesses in the drug classification system, which lumps many synthetic opioids under the same ICD-10 code, including fentanyl, fentanyl analogs, tramadol and buprenorphine – a semi-synthetic opioid used in the addiction treatment drug Suboxone. 

In the past, CDC has classified all drug deaths using that code as “prescription opioid overdoses” even though the drugs may have been illicit — which is the case for the vast majority of deaths involving fentanyl. This resulted in government estimates of prescription opioid overdoses being significantly inflated for many years.

Using the computer algorithms developed at UCLA, prescription opioids ranked far behind fentanyl, alcohol and other substances identified as the cause of death in 8,738 overdoses. 


[ninja-popup ID=12216]