Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Intimate Partner Violence and Injury Prediction From Radiology Reports

Irene Y. Chen, Emily Alsentzer, Hyesun Park, Richard Thomas, Babina Gosangi, Rahul Gujrathi, Bharti Khurana

TL;DR

Intimate partner violence (IPV) is a critical, under-detected public health issue; this study leverages radiology reports to predict IPV and associated injuries using ML and NLP methods. The authors compare multiple models, including logistic regression, random forests, gradient boosting, Bag-of-Words neural nets, and ClinicalBERT-based representations, on 34,642 radiology reports from 1,479 patients, achieving a mean AUC of $0.852$ for IPV (sensitivity $0.64$, specificity $0.95$) and $0.887$ for injury. They further demonstrate that IPV can be predicted a median of $3.08$ years before program entry and discuss error analyses across demographic subgroups, as well as deployment challenges and future directions such as incorporating clinical history to improve accuracy and generalizability. Overall, the work shows the feasibility of automated IPV risk stratification from radiology text and highlights important considerations for real-world integration.

Abstract

Intimate partner violence (IPV) is an urgent, prevalent, and under-detected public health issue. We present machine learning models to assess patients for IPV and injury. We train the predictive algorithms on radiology reports with 1) IPV labels based on entry to a violence prevention program and 2) injury labels provided by emergency radiology fellowship-trained physicians. Our dataset includes 34,642 radiology reports and 1479 patients of IPV victims and control patients. Our best model predicts IPV a median of 3.08 years before violence prevention program entry with a sensitivity of 64% and a specificity of 95%. We conduct error analysis to determine for which patients our model has especially high or low performance and discuss next steps for a deployed clinical risk model.

Intimate Partner Violence and Injury Prediction From Radiology Reports

TL;DR

Intimate partner violence (IPV) is a critical, under-detected public health issue; this study leverages radiology reports to predict IPV and associated injuries using ML and NLP methods. The authors compare multiple models, including logistic regression, random forests, gradient boosting, Bag-of-Words neural nets, and ClinicalBERT-based representations, on 34,642 radiology reports from 1,479 patients, achieving a mean AUC of for IPV (sensitivity , specificity ) and for injury. They further demonstrate that IPV can be predicted a median of years before program entry and discuss error analyses across demographic subgroups, as well as deployment challenges and future directions such as incorporating clinical history to improve accuracy and generalizability. Overall, the work shows the feasibility of automated IPV risk stratification from radiology text and highlights important considerations for real-world integration.

Abstract

Intimate partner violence (IPV) is an urgent, prevalent, and under-detected public health issue. We present machine learning models to assess patients for IPV and injury. We train the predictive algorithms on radiology reports with 1) IPV labels based on entry to a violence prevention program and 2) injury labels provided by emergency radiology fellowship-trained physicians. Our dataset includes 34,642 radiology reports and 1479 patients of IPV victims and control patients. Our best model predicts IPV a median of 3.08 years before violence prevention program entry with a sensitivity of 64% and a specificity of 95%. We conduct error analysis to determine for which patients our model has especially high or low performance and discuss next steps for a deployed clinical risk model.

Paper Structure

This paper contains 23 sections, 1 figure, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Intimate partner violence
  4. Clinical prediction
  5. Natural language processing
  6. Dataset
  7. IPV patient selection
  8. Control group selection
  9. Injury labels
  10. Data cleaning
  11. Demographic data
  12. Methodology
  13. Experiment setup
  14. Models
  15. Evaluation
  16. ...and 8 more sections

Figures (1)

  • Figure 1: Scatterplots and marginal histograms for random forest classifier for IPV prediction. Left: Earliest possible report-program date gap per patient ($x$-axis) compared to earliest predicted date gap ($y$-axis) with sensitivity of 64% and specificity of 95%. Right: Report-program date gap ($x$-axis) and IPV prediction probability ($y$-axis) for all radiology reports of IPV victims.