npj health systemsSep 17, 2025

Identifying Long COVID Symptoms from Medical Notes Using Combined Language Processing Methods

Long Covid Weekly Brief ↗PubMed ↗DOI ↗OA ↗

Updated Jun 27, 2026

Abstract

Essence

A hybrid pipeline showed moderate-to-strong performance for extracting symptoms and assertion status from clinical notes.

Evidence

This multi-site model development and validation study used 160 intake progress notes from 11 RECOVER health systems for development and evaluation plus 47,654 progress notes for a prevalence study, achieving F1 scores of 0.82 internally and 0.76 externally for assertion detection.

Caveat

The abstract reports note-level NLP validation and prevalence processing, not direct evidence that the pipeline improves PASC diagnosis or patient outcomes.

Simplified

Accurately and efficiently diagnosing Post-Acute Sequelae of COVID-19 () remains challenging due to its myriad symptoms that evolve over long- and variable-time intervals. To address this issue, we developed a hybrid natural language processing pipeline that integrates rule-based named entity recognition with BERT-based assertion detection modules for PASC-symptom extraction and assertion detection from clinical notes. We developed a comprehensive PASC lexicon with clinical specialists. From 11 health systems of the RECOVER initiative network across the U.S., we curated 160 intake progress notes for model development and evaluation, and collected 47,654 progress notes for a population-level prevalence study. We achieved an average F1 score of 0.82 in one-site internal validation and 0.76 in 10-site external validation for assertion detection. Our pipeline processed each note at 2.448 ± 0.812 seconds on average. Spearman correlation tests showed ρ > 0.83 for positive mentions and ρ > 0.72 for negative ones, both with< 0.0001. These demonstrate the effectiveness and efficiency of our models and its potential for improving PASC diagnosis. P

Key numbers

0.82

Average F1 Score (Internal Validation)

Measured across all symptoms in the internal validation dataset.

2.448 ± 0.812 seconds

Average Processing Time per Note

Calculated across 11 health systems in the RECOVER initiative.

ρ > 0.83

Spearman Correlation for Positive Mentions

Based on symptom-mentioning patterns in the population-level prevalence study.

Full Text

We can’t show the full text here under this license.

View Full Text ↗

Funding

Competing interests

0%of authors

0 of 16

authors report competing interests

16 report none

PubMed

Featured in

Long CovidIssue #4

59% of long COVID patients still sick 6 months later + new brain fog treatments

↗

Identifying Long COVID Symptoms from Medical Notes Using Combined Language Processing Methods

Abstract

Key numbers

Full Text

Funding

Competing interests

Featured in

You found one interesting study. We’ll send the next 7.

What Lands in Your Inbox Each Week:

Recent issues from the long covid brief

Abstract

Key numbers

Full Text

Funding

Competing interests

Related papers

Featured in

You found one interesting study. We’ll send the next 7.

What Lands in Your Inbox Each Week:

Recent issues from the long covid brief