Skip to content
PaperFren

U09.9 coding reveals long COVID clusters and gaps

Open paper intelligence

Among 33,782 U09.9-coded US EHR patients, co-occurring diagnoses clustered into organ-system groups while coded cases skewed toward advantaged, White, female patients.

Source

Coding long COVID: characterizing a new disease through an ICD-10 lens

Pfaff ER · BMC medicine · 2023

doi.org/10.1186/s12916-023-02737-6Read the full paper ↗148 citationscc by

Study at a glance

Design
Cohort — N3C EHR cohort with ICD-10-CM U09.9 from 34 health systems (Oct 2021–May 2022)
N
N=33782 · After excluding inpatient index codes; demographics and 60-day co-occurring care
Population
Patients coded with U09.9 (Post COVID-19 condition) in US health systems
Outcome
Demographics, co-occurring diagnoses, and care patterns around U09.9 coding

Structured fields used in claim comparison tables when every cited study has a complete layer.

What they did

Using NIH N3C harmonized EHR data from 34 health systems, researchers identified patients with ICD-10-CM U09.9 (Oct 2021–May 2022), excluded inpatient index codes, and analyzed demographics, social determinants, co-occurring diagnoses (Louvain clustering), and procedures/medications within 60 days of coding, stratified by age.

What they found

The final cohort was 33,782 patients. U09.9 rapidly replaced the placeholder B94.8 after release. Co-occurring conditions clustered into cardiopulmonary, neurological, gastrointestinal, and comorbid themes that shifted by age. Imaging was common (22.8%). Yet U09.9 capture skewed toward female, White, non-Hispanic individuals and lower-poverty areas—suggesting under-diagnosis in more deprived populations despite higher acute COVID burden there.

The limits

What it doesn't show

U09.9 was new and inconsistently adopted, so absence of the code does not rule out long COVID; billing vs clinical coding cannot be separated, and clusters are hypothesis-generating—not proven causal subtypes.

Key terms

U09.9
ICD-10-CM code for post COVID-19 condition, unspecified, released October 2021.
N3C
NIH National COVID Cohort Collaborative—large harmonized US EHR enclave used for this study.
Louvain clustering
Network community-detection algorithm grouping diagnoses that co-occur within the same patients.
Social determinants of health (SDoH)
Area-level measures such as poverty and unemployment linked to who receives a U09.9 diagnosis.
Computable phenotype
Rule-based EHR definition of a condition beyond a single diagnosis code.
B94.8
Non-specific sequelae code used as a long COVID placeholder before U09.9 existed.

Flashcards

1 / 10

Research intelligence for this paper

See its role on concept claims, tensions it is part of, placement history, and related discoveries.

Open paper intelligence

Quiz yourself

1 / 6

Final analyzed U09.9 cohort size was:

Common questions

When did U09.9 become available?

The ICD-10-CM U09.9 code was released for clinical use in October 2021—nearly two years after patients began reporting long COVID.

Did severe acute COVID predict who got U09.9?

Not strongly—only 9.7% of U09.9 patients had been hospitalized for acute COVID-19, lower than in some claims-based reports.

What symptom clusters emerged?

Diagnoses co-occurring with U09.9 grouped into cardiopulmonary, neurological, gastrointestinal, and comorbid clusters varying by age.

Can researchers rely on U09.9 alone?

Authors advise combining U09.9 with richer computable phenotypes because many affected patients remain uncoded.

More on Health equity