Data Stream and Hospital-Level Variation in Intensive Care Unit Vital-Sign Record Density Across Medical Information Mart for Intensive Care IV and eICU Collaborative Research Database: Retrospective Data Quality Study.

Authors

Yngve Mikkelsen

Published in

JMIR medical informatics. Volume 14. Pages e99638. Oct 29, 2026. Epub Oct 29, 2026.

Abstract

Electronic health record-derived process measures such as vital-sign record density are used for quality monitoring, risk prediction, and cross-system benchmarking, but whether they retain consistent meaning across sites and data streams is rarely tested.
This study aimed to determine whether vital-sign record density varies primarily by data stream, database, contributing site, or care unit, and what follows for comparative use.
We retrospectively analyzed the Medical Information Mart for Intensive Care IV (MIMIC-IV) version 3.1 and the eICU Collaborative Research Database (eICU-CRD) version 2.0. Record density and timing metrics were computed for the first 24 intensive care unit hours from nurse-charted streams and, in eICU-CRD, from the monitor stream. Heart rate was primary; oxygen saturation and respiratory rate were replication variables. Variance was decomposed by database, hospital, and unit using η² with hospital-clustered bootstrap inference and mixed-effects sensitivity analyses. A hospital-level plausibility floor of 12 nurse-charted heart rate observations per 24 hours was applied for the principal analysis; unrestricted and alternative-floor analyses are reported as sensitivities.
A pooled-item MIMIC-IV heart rate count included alarm-limit documentation events, so label-based pooling mixes the two. In 117,572 eICU-CRD stays present in both streams, nurse- and monitor-derived counts were almost unrelated (Spearman ρ=0.024), with bottom-decile overlap at chance. Between-database η² for record count was 0.960 against the eICU monitor stream and 0.026 against the nurse stream. Eight of 68 eICU-CRD hospitals fell below the plausibility floor and were excluded from principal estimates, a restriction introduced during review. Among the remaining 60 hospitals (88,560 stays), hospital η² for record count was 0.354 (95% CI 0.259-0.450) against unit-within-hospital η² of 0.043; for the interval-structure metrics, the hospital and unit components were not separable. For record count, the hospital component fell from 0.441 unrestricted to 0.321 at an hourly floor. A negative-binomial mixed model gave latent-scale hospital and unit variance-partition coefficients of 0.348 (95% CI 0.252-0.457) and 0.112 (0.078-0.163), equivalent to 0.318 and 0.110 observed. A pooled bottom-decile rule agreed with hospital-specific rules on 52.5% (4726/9006) of flagged stays, with 6 hospitals supplying half the pooled decile. The hospital component for record count replicated at 0.322 for oxygen saturation and 0.333 for respiratory rate.
Within a multisite nurse-charted stream, vital-sign record volume varied more across contributing sites than across care units within hospitals; for the interval-structure metrics, the site and unit components were not separable at this cohort size. Apparent cross-database differences were reproduced by source-table choice, and site-level variation remained after excluding the most evident contribution-completeness outliers. Comparative use therefore requires specifying the generating stream, enforcing the analytic window, and explicitly modeling site-level variation; the residual site component should not be interpreted as documentation practice alone.

PMID:
42956367
Bibliographic data and abstract were imported from PubMed on 30 Oct 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.