ReHouSED NLP

Overview

This package is a medspaCy implementation of an NLP system for identifying patient housing stability in clinical texts. This system was originally developed in the Department of Veterans Affairs to study housing outcomes of Veterans participating in the Supportive Service for Veteran Families (SSVF) program. The development and validation of this system is described in "ReHouSED: A Novel Measurement of Veteran Housing Stability Using Natural Language Processing" by Chapman et al. (published Journal of Biomedical Informatics).

This system attempts to classify housing stability at two levels:

1. Document-level: Each document processed by the NLP is classified as either "STABLY_HOUSED", "UNSTABLY_HOUSED", or "UNKNOWN"
2. Patient-level: A set of documents over a period of time are processed and aggregated to a patient level. This is a numeric score ranging from 0-1 called "Relative Housing Stability in Electronic Documentation" (ReHouSED)

Detailed examples and explanations of the logic are provided in notebooks/

Disclaimer

This system is an approximation of the system described in the manuscript and has been modified to exclude logic specific to VA documentation. It is far from perfect and will certainly make mistakes!

Installation

You can install rehoused_nlp using pip:

pip install rehoused-nlp

Or the source code found in this repository:

python setup.py install

rehoused_nlp requires Python 3.7 or 3.8, medspaCy, and spaCy 2.2.X. spaCy 3 is not currently supported.

Quick start

Document-level example

from rehoused_nlp import build_nlp, visualize_doc_classification

nlp = build_nlp()

text = """
History of present illness: The patient was evicted from her apartment two months ago. 
Since then she has lived in a shelter while looking for an apartment.

Past medical history:
1. Pneumonia
2. Afib
3. Homelessness

Housing Status: Stably Housed

Assessment/Plan: The patient was accepted to an apartment and signed the lease last week. 
"""

doc = nlp(text)

visualize_doc_classification(doc)

Patient-level example

from rehoused_nlp import calculate_rehoused
import pandas as pd

df = pd.read_csv("path/to/data.tsv", sep="\t")
print("Input:")
df.head()

print("Output:")
rehoused = calculate_rehoused(df)
rehoused.head()

Input:

Output: