name: word_alerts
fullname: Word Alerts Report
description: >
The word alerts template summarizes daily chief complaints
and discharge diagnosis term alerts for a selected syndrome definition.
Original Code By: Michael Sheppard.
Adapted to the Rnssp Package by: Gbedegnon Roseric Azondekon.
National Syndromic Surveillance Program (NSSP).
Centers for Disease Control and Prevention (CDC).
create_dir: true
The Word Alerts Report template (word_alerts) summarizes daily chief complaints and discharge diagnosis term alerts for a selected syndrome definition.
The purpose of the word alert algorithm is to seek anomalous chief complaint free-text terms and discharge diagnosis codes relative to a 28-day sliding baseline.
The current implementation of word alerts in NSSP-ESSENCE is limited to data from the past 7 days. This template lets users run the word alert algorithm on any selected date range spanning up to 90 days.
The ESSENCE word alert algorithm uses 28-day baselines over which unigram and bigram frequencies are computed. As with the univariate temporal anomaly detection algorithms, a 2-day guardband is used to separate the baseline and test date. For each term that occurred in the testing block (test date), contingency tables are formed to count the number of emergency department encounters with the term/code occurring in the chief complaint/discharge diagnosis and the number of emergency department encounters without the term/code in the chief complaint/discharge diagnosis for both the testing and baseline blocks.
| Block | Visits with Term | Visits without Term |
|---|---|---|
| Test Date (Last 24 Hours) | A: 136 | B: 5,111 |
| Baseline (28 Days) | C: 2,851 | D: 149,477 |
To test for anomalous increases in term occurrence on the test date relative to the baseline, either Fisher’s Exact Test or a Chi-squared Test is applied. Fisher’s Exact Test is applied for moderate counts, whereas a Chi-squared Test is applied for large counts. Terms are determined to have moderate counts if the number of visits in the baseline with the term is less than 1,000 or if the number of visits on the test date without the term is less than 1,000, i.e., min(B,C) < 1,000. Unigrams and bigrams are considered anomalous if the resulting p-value is less than 0.001.
Pre-processing of chief complaint free text includes removal of punctuation, discharge diagnosis codes, uninformative patterns, medical processing terms, and common first names. Duplicate terms in the same chief complaint and discharge diagnosis are also removed. For discharge diagnosis bigrams, pairings of codes are arranged in alphanumeric order so that permutations of the same codes are not counted separately.
This HTML report includes interactive visualizations and tables that accommodate hovering to display additional information. ICD-9 and ICD-10 code descriptions can be displayed by hovering over discharge diagnosis codes. Note that descriptions for SNOMED codes are not available.
When knit with parameters, this template generates a report in HTML format.
# Add `word_alerts` to my existing Rnssp installation
Rnssp::add_rmd_template("word_alerts")
# Remove `word_alerts` from my existing Rnssp installation
Rnssp::remove_rmd_template("word_alerts")
For questions, ideas for improvement/collaboration, or attribution, please submit an issue here.