Research suggests the optimal method for identifying hospital adverse events (AEs) is a two stage approach using manual chart abstraction. First, clinicians abstract the entire medical record to identify potential AEs using screening criteria. Second, narrative summaries of potential AEs are reviewed by a physician to assess presence, preventability, and severity. We sought to compare the performance of manual abstraction to data warehouse facilitated abstraction using programmed queries to screen for potential AEs.
This study was a retrospective chart review involving general medical patients admitted to an urban 897bed teaching hospital. Data from the hospital’s fully integrated electronic health record and computerized provider order entry system, incident reporting system, admission/discharge/transfer system, and billing system populate an enterprise data warehouse (EDW), which is updated nightly. We created 51 automated queries based on 32 manual screening criteria used in prior research. Examples of queries include discrete laboratory values beyond a specified threshold occurring during the hospitalization (e.g., INR > 6 after hospital day 2 and excluding patients with INR > 4 on day 1), administration of certain medications (e.g., naloxone), incident reports in certain categories (e.g., falls), ICD9 codes suggesting hospital acquired conditions, and text searches using natural language processing to identify potential adverse events (e.g., “perforation” not preceded by “no”). We randomly selected 250 patients hospitalized from 9/1/09 to 8/31/10 and assigned 1/2 to each of two hospitalists for manual abstraction. After completing manual abstractions, each of the two hospitalists performed targeted abstractions for patients with positive EDW queries in the complementary half of the sample. Narrative summaries of potential AEs were rated by a third hospitalist for presence, preventability, and severity of AEs.
Manual and EDW facilitated abstraction identified 65 (26%) and 52 (21%) patients with one or more AE. Overall, 140 patients (56%) had one or more positive EDW queries (total 366 positive queries). Of the 145 AEs detected by at least one method, 100 (69%) were detected by manual abstraction, 94 (65%) by EDW faciliated abstraction, and 49 (34%) by both methods. Of the 10 total preventable AEs, 6 (60%) were detected by manual abstraction, 6 (60%) by EDW facilitated abstraction, and 2 (20%) by both methods. Of the 44 total serious AEs, 31 (70%) were detected by manual abstraction, 27 (61%) by the EDW abstraction, and 14 (32%) by both methods.
Relatively little overlap existed between AEs detected by data warehouse facilitated abstraction as compared to manual abstraction. Results were consistent across designations of preventability and severity. A combination of methods may be the optimal approach to detecting AEs among hospitalized patients.
To cite this abstract:Patel A, Barnard C, Malkenson D, O'Leary K, Williams M, Landler M, Sama P, Devisetty V, Thompson W. Comparison of Manual Abstraction to Data Warehouse Facilitated Abstraction for Identifying Hospital Adverse Events. Abstract published at Hospital Medicine 2012, April 1-4, San Diego, Calif. Abstract 97590. Journal of Hospital Medicine. 2012; 7 (suppl 2). https://www.shmabstracts.com/abstract/comparison-of-manual-abstraction-to-data-warehouse-facilitated-abstraction-for-identifying-hospital-adverse-events/. Accessed September 18, 2019.