Long COVID undercoding and prediction from releasable electronic health record summaries in All of Us
- 1Vanderbilt University Medical Center
Abstract
ObjectiveTo estimate Long COVID undercoding in electronic health records (EHRs) against federal survey prevalence, and to evaluate a predictor of later coding that anyone can rebuild from counts a research enclave releases.
Materials and MethodsAmong 23,790 All of Us adults with COVID-19, we compared coded shares with Household Pulse Survey prevalence. Among 23,300 adults uncoded at day 90, we predicted a first U09.9 code by day 455. We fitted naive Bayes and linear discriminant models outside the enclave from released counts and compared them with an elastic net fitted inside to the same patients and features. Tests were a random 20% of patients (4,680, 50 coded) and 18 held-out EHR sites (5,760, 60 coded).
ResultsUnder survey-comparability assumptions, about 1 in 9 adults who ever had Long COVID carried its code, and the women-to-men coding ratio was 0.79 (95% sensitivity interval, 0.65-0.95). The released models areas under the receiver operating characteristic curve (AUROCs) were 0.68 (95% CI, 0.60-0.75) on the patient test and 0.64 (0.55-0.72) at held-out sites, 0.049 (0.001-0.104) above the elastic net. On the patient test, the elastic net had a higher area under the precision-recall curve (difference 0.062, 0.003-0.122). Released probabilities overestimated coding risk at held-out sites.
DiscussionFitting from released counts did not consistently lower discrimination. Coded outcomes inform disease discrimination only under stated coding assumptions.
ConclusionLong COVID is substantially and unevenly undercoded, and a model rebuilt from public counts predicts its later coding with moderate discrimination.
Related
- Dynamic Prediction to Predict Virologic Failure in Antiretroviral Treated People Living with HIV: Cohort Analysis of Routine Electronic Health Records Data in the Western Cape, South Africa
- Dynamic near-term risk prediction of clinically significant immune-related adverse events using longitudinal electronic health records
- Implementation of risk prediction models using electronic health data for early detection of pancreatic cancer: a prospective pilot feasibility study
- Cohort Versus Nested Case-Control Design for Prediction Models of Rare Outcomes Using Electronic Health Records
- Implementation of Risk Prediction Models Using Electronic Health Data for Early Detection of Pancreatic Cancer: A Prospective Pilot Feasibility Study.
BibTeX
@misc{asiaee2026long,
title = {{Long COVID undercoding and prediction from releasable electronic health record summaries in All of Us}},
author = {A. Asiaee and K. Aryan},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.18.26365858},
url = {https://www.medrxiv.org/content/10.64898/2026.10.18.26365858v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Long COVID undercoding and prediction from releasable electronic health record summaries in All of Us
AU - A. Asiaee
AU - K. Aryan
PY - 2026
DA - 2026/10/18
PB - medRxiv
DO - 10.64898/2026.10.18.26365858
UR - https://www.medrxiv.org/content/10.64898/2026.10.18.26365858v1
AB - ObjectiveTo estimate Long COVID undercoding in electronic health records (EHRs) against federal survey prevalence, and to evaluate a predictor of later coding that anyone can rebuild from counts a research enclave releases. Materials and MethodsAmong 23,790 All of Us adults with COVID-19, we compared coded shares with Household Pulse Survey prevalence. Among 23,300 adults uncoded at day 90, we predicted a first U09.9 code by day 455. We fitted naive Bayes and linear discriminant models outside the enclave from released counts and compared them with an elastic net fitted inside to the same patients and features. Tests were a random 20% of patients (4,680, 50 coded) and 18 held-out EHR sites (5,760, 60 coded). ResultsUnder survey-comparability assumptions, about 1 in 9 adults who ever had Long COVID carried its code, and the women-to-men coding ratio was 0.79 (95% sensitivity interval, 0.65-0.95). The released models areas under the receiver operating characteristic curve (AUROCs) were 0.68 (95% CI, 0.60-0.75) on the patient test and 0.64 (0.55-0.72) at held-out sites, 0.049 (0.001-0.104) above the elastic net. On the patient test, the elastic net had a higher area under the precision-recall curve (difference 0.062, 0.003-0.122). Released probabilities overestimated coding risk at held-out sites. DiscussionFitting from released counts did not consistently lower discrimination. Coded outcomes inform disease discrimination only under stated coding assumptions. ConclusionLong COVID is substantially and unevenly undercoded, and a model rebuilt from public counts predicts its later coding with moderate discrimination.
ER -