In November 2021, the CDC used provisional data to estimate that U.S. deaths from opioid overdoses rose to 100,306 for the 12-month period that ended in April 2021. That number is a 28.5% increase from the 78,506 deals in the previous 12 months. Researchers from the University of Florida and University of Pittsburgh used artificial intelligence to create an algorithm to predict opioid overdoses. The Pittsburgh and Florida researchers tested whether a machine-learning algorithm developed from earlier Pennsylvania data could predict overdose risk for later years and for different states.
The study started with a training dataset from 8,641 Pennsylvania Medicaid beneficiaries who received one or more opioid prescriptions from 2013 to 2016. The machine-learning program tracked 284 potential factors from health-care and pharmaceutical claims to create subgroups of beneficiaries at high-risk of opioid overdose. Once the model was constructed, the researchers ran datasets from two other groups through the algorithm. The test groups included 2,705 Pennsylvania Medicaid beneficiaries in 2017 to 2018 to validate the algorithm for different years. The second group comprised 2,410 Arizona Medicaid beneficiaries from 2015 to 2017 to validate the algorithm for different states.
The validation test results captured 73% of the overdoses in the high-risk subgroups in the Pennsylvania test group and 55% of the overdoses in the high-risk subgroups in the Arizona test group. The researchers state that the algorithm could be valuable in stratifying risk and predicting opioid overdoses in Medicaid beneficiaries. The group also noted that the study did not include overdoses with subjects who paid cash for prescriptions, for people who used only illicit opioids, or for overdoses that didn’t have medical attention.
Using machine-learning dataset analysis and predictions to take action with people may seem creepy, but the argument in support is based in the potential to save lives.
