Quantitative Nurse

Vanderbilt University Medical Center · Department of Biomedical Informatics

We develop clinical NLP and EHR phenotyping methods that identify substance use disorder in health records, and release them as open-source tools.

The Quantitative Nurse lab is directed by Alvin D. Jeffery, PhD, RN. A second line of work asks how nurses use clinical decision support, and how to design it from evidence rather than intuition.

Selected finding

JAMA Psychiatry, 2025

Clinical notes find problematic opioid use that diagnostic codes miss.

Among 8,063 patients with chronic pain, automating the Addiction Behaviors Checklist with regular expressions identified problematic opioid use far more accurately than ICD diagnostic codes, and the advantage held at an independent validation site.

Each dot is an F1 score (0 to 1, higher is better); the line shows its 95% confidence interval.

F1 balances sensitivity (cases found) and positive predictive value (flags that were correct). Source: Chatham AH, Bradley ED, et al. Automating the Addiction Behaviors Checklist for problematic opioid use identification. JAMA Psychiatry. 2025;82(6):591.
View as table
F1 scores by site and method
SiteMethodF1 (95% CI)
Vanderbilt (n = 8,063)Automated checklist0.73 (0.62–0.83)
Vanderbilt (n = 8,063)ICD diagnostic codes0.08 (0.00–0.19)
Geisinger (n = 100)Automated checklist0.70 (0.50–0.85)
Geisinger (n = 100)ICD diagnostic codes0.29 (0.07–0.50)
01

Research

  • Active

    Substance use disorder phenotyping

    Problem

    Diagnostic codes are an unreliable record of substance use disorder: studies report sensitivity for opioid use disorder codes as low as 0.17. Genetic studies of SUD need data pooled across many institutions, and there is no robust, shareable way to say who has an SUD and who doesn’t.

    Approach

    We build probabilistic phenotypes from billing codes, medications, and clinical notes, using interpretable NLP that clinicians can audit. We also study how to make the manual chart review behind these labels faster without losing accuracy.

    Funding
    NIH/NIDA DP1DA056667, Avenir Award (PI: Jeffery)
    NIH/NIDA R03 (PI: Niarchou)
    Software
    sudregex on PyPI
    abc_regex on GitHub

  • Completed

    SPECTACULAR

    Systematic Process for the Equitable Customization of Technology Applied to Contexts and Users for Learning and Adapting Rapidly

    Problem

    Clinical decision support rarely fits nurses’ cognitive and physical workflows, and redesigning it is slow, expensive, and guided by heuristics rather than evidence.

    Approach

    SPECTACULAR combines a factorial research design with a Bayesian adaptive trial to evaluate many design options rapidly and empirically, with respect to both nurses’ preferences and their performance.

    Findings have not been published yet. To learn more, contact us.

    Learn more
    Contact us

All research →

02

Software

We release the code behind our papers so other teams can reproduce and extend the work.

  • sudregex

    Python package · MIT · v0.1.8, July 2026

    Regex-driven extraction for clinical text, with configurable negation scope, substance-context gating, and false-positive pruning. Reports match counts, not just flags. Runs on pandas locally or distributed on Spark and Databricks.

    pip install sudregex

  • abc_regex

    Python · research code

    Automates the Addiction Behaviors Checklist over clinical notes with regular expressions. The published code for our 2025 JAMA Psychiatry study.

  • Snorkel Interactive

    Not publicly released · available on request

    Research teams interested in using it can contact us for access.

03

Recent work

  1. Commentary

    Ambient AI’s Value for Nursing: Will “Saving Time” Be a “Staffing Trap”?

    Lee SY, Wyse RJ, Jeffery AD

    Applied Clinical Informatics 17(4):763–766

  2. Software

    sudregex 0.1.8

    Quantitative Nurse Lab

    Python Package Index

  3. Review

    Natural Language Processing for Substance Use Disorder Information Extraction: A Systematic Literature Review

    Wyse RJ, Samuels DC, Sanchez-Roige S, Schirle L, Rhoten BA, Lee SY, Jeffery AD

    Current Addiction Reports 13(1):34

  4. Article

    Automating the Addiction Behaviors Checklist for Problematic Opioid Use Identification

    Chatham AH, Bradley ED, Troiani V, Beiler DL, Christy P, Schirle L, … Jeffery AD

    JAMA Psychiatry 82(6):591

  5. Article

    Use of noisy labels as weak learners to identify incompletely ascertainable outcomes: A Feasibility study with opioid-induced respiratory depression

    Jeffery AD, Fabbri D, Reeves RM, Matheny ME

    Heliyon 10(5):e26434

  6. Article

    Inpatient nurses’ preferences and decisions with risk information visualization

    Jeffery AD, Reale C, Faiman J, Borkowski V, Beebe R, Matheny ME, Anders S

    Journal of the American Medical Informatics Association 31(1):61–69

All publications →

04

Work with us

We collaborate with clinicians, informaticians, and health systems. If you want to use our software, validate a phenotype at your institution, or ask about joining the lab, email us.

quantitativenurse@gmail.com