Skip to main content
ICPE 2021 Main banner
Back
ViewAttendees 1

Methods in Pharmacoepidemiology

Validation and Algorithms: Getting the Present and Future Right

OP-27B - Death Documentation Lag in COVID-19 Electronic Health Record (EHR) Suggests Caution in Near-real-time Analysis

Tuesday, August 24, 2021
1:45 PM – 3:30 PM US Eastern Time
Location: OP-27
Background: High volumes of near-real-time EHR data are transforming the nature of health research. To avoid publishing erroneous data, researchers must understand and plan for the limitations of such data. Mortality is an important outcome in COVID-19 research, but the lag time for documentation of death data in a near-real-time EHR-based dataset is not known.

Objectives: To determine the time lag of in-hospital death reporting in the COVID-19 version of a major commercially available EHR database.

Methods: The data source is a near-real-time version of an established commercial EHR, launched to support COVID-19 research. The study population included patients from the 2021-01-06 data release who died during a COVID-19-associated hospitalization. The completeness of death documentation of each of four earlier releases (2020-09-02, 2020-09-16, 2020-10-14, and 2020-11-11) were analyzed for patients who died before that earlier release and who had data records in that earlier release. The completeness was calculated as the proportion, of those patients, whose deaths were documented in the earlier release, stratified by week of death. Week of death was designated relative to the end-of-data date of the release being examined (e.g., week 1 of 2020-09-16 release spanned the period from 9th through 15th of September).

Results: A total of 6,247 distinct in-hospital deaths in February-November 2020 were included. The completeness of death documentation in the 2020-09-02 data release was 67%, 74%, and 75% for 5th, 6th, and 7th week prior to the end of the data release. For the 2020-09-16 data release, the corresponding completeness was 75%, 79%, and 97%. For the 2020-10-14 data release, completeness was 79%, 94%, and 97%. For the 2020-11-11 release, completeness was 84%, 95%, and 100%.

Conclusions: Death data recording in this dataset is incomplete during the weeks immediately following the patient death. Assessment of mortality in recent patients may appear falsely decreased relative to earlier patients, leading to mis-estimation of changes in death rate over time. The improving completeness in the successive data releases underscores the difficulty of adapting pre-existing data processes to new near-real-time demands. Researchers using near-real-time data should evaluate the lags of all outcomes of interest prior to use.