fe5_anaphylaxis

The FE5 Anaphylaxis pipeline.

Overview

This Python-based pipeline is used for identifying and extracting anaphylaxis cases based on diagnosis and procedure codes, encounter note types (inpatient, outpatient, ED visits), and patient EHR notes.

Pipeline Structure

This pipeline involves three main steps:

A. Pre-processing

1. Reads four SAS files that contains the disgnosis, procedure codes, and encounter note types for 2019 and 2020

2. Classifies the encounter type into Inpatient, ED, Outpatient, or Other.

3. Implements three criteria (“paths”) for anaphylaxis (for details, please refer to Here):

   • Path 1 (“Wash criterion A”): For an inpatient/ED row with a List 1 code, check that no prior encounter (within 60 days) for the same patient had the same List 1 code.

   • Path 2 (“Walsh criterion B”): For an outpatient row with a List 1 code, check that on the same calendar day the patient has at least one List 2 code (Group I or II).

   • Path 3 (“Walsh criterion C”): For an ED or inpatient row with a List 3 code, check that on the same day (for ED) or within the same encounter (for inpatient) there is at least one List 2 code from Group I and one from Group II.

4. The target patient cohort is defined as patients with at least one encounter that meets any of the three criteria.

5. The output of this step is a CSV file that contains the patient_id, note_id, encounter_id, note_date, provider_id, and the note_context

B. Assign feature status for anaphylaxis using cTAKES

Based on the CSV file generated through step A, the pipeline uses the fe5_cTAKES pipeline, to extract anaphylaxis-related features from the EHR notes of the target patients. The CUIs used for anaphylaxis is available Here. We collected anaphylaxis-related CUIs from UMLS Metathesaurus, filtering results by searching “Anaphylaxis.” After manually removing unrelated CUIs, we retained 123 relevant CUIs.

C. Post-processing

Based on the output of step B, this step generates an encounter-level result that aggregates the note-level features extracted from the EHR notes of the target patients. The reason of using a standalone post-processing is that for anaphylaxis, we only have two feature status code (A or U), which is different from the original fe5_cTAKES pipeline, so any non-A feature status needs be reassigned to U before aggregtion.

How the encounter type is determined

*Below are the rules of classifying the encounter type:

• Emergency department encounter includes: ED

• Inpatient encounter includes: EI, IP

• Outpatient encounter includes: AV, IS, OS, IC, TH, OA

• All other types are classified as "Other"

The abbreviations are defined as follows:

• AV=Ambulatory Visit

• ED=Emergency Department

• EI=Emergency Department Admit to Inpatient Hospital Stay (permissible substitution)

• IP=Inpatient Hospital Stay

• IS=Non-Acute Institutional Stay

• OS=Observation Stay

• IC=Institutional Professional Consult (permissible substitution)

• TH=Telehealth

• OA=Other Ambulatory Visit

• NI=No information

• UN=Unknown

• OT=Other