Performance of Phenotype Algorithms for the Identification of Opioid-Exposed Infants

Area of Science:

• Perinatal Health
• Data Science in Healthcare
• Neonatal Research

Background:

• Observational studies on opioid-exposed infants are challenged by algorithms that may undercount infants without neonatal opioid withdrawal syndrome (NOWS).
• Accurate identification of opioid-exposed infants is crucial for understanding their health outcomes.
• Electronic health records (EHRs) offer a potential data source for identifying these infants.

Purpose of the Study:

• To develop and validate phenotype algorithms for identifying opioid-exposed infants using EHR data.
• To improve the accuracy of identifying opioid-exposed infants, particularly those without NOWS.
• To provide reliable tools for research on the outcomes of opioid-exposed infants.

Main Methods:

• Phenotype algorithms were developed using combinations of six indicators from infant and birthing person records (2010-2022).
• Indicators included diagnoses, toxicology results, and medication records related to opioid exposure.
• Positive predictive values (PPVs) were determined using medical record review as the gold standard.

Main Results:

• Among 41,047 dyads, 3.80% had evidence of opioid exposure; 0.08% met all six indicators.
• A phenotype algorithm using a single indicator achieved a PPV of 95.4% (CI: 93.3-96.8).
• Algorithms combining multiple indicators showed high PPVs ranging from 95.4% to 100.0%.

Conclusions:

• Opioid-exposed infants can be accurately identified through EHR data analysis.
• The developed phenotype algorithms are validated and publicly available.
• These algorithms can facilitate research on opioid-exposed infants, with or without NOWS.