Background: Diagnostic errors (DE) often occur in the hospital and can lead to preventable harm. As part of our AHRQ-funded Patient Safety Learning Laboratory, we previously estimated the prevalence of DE to be 22.7% in a stratified cohort of patients hospitalized on the general medicine service. Using a structured electronic health record (EHR) chart review process, we used the Safer Dx Instrument to determine the likelihood of DE in these cases and the modified Diagnostic Error Evaluation and Research (DEER) Taxonomy adapted for acute care (41 possible diagnostic process failures) to identify key failures across eight diagnostic process domains. While little is known about which types of factors lead to DE in hospitalized patients, an analysis of diagnostic process failures can help identify high-yield targets for intervention. The purpose of this study is to determine the extent to which certain diagnostic process failures were associated with DE.

Methods: We conducted a secondary analysis of 113 DE positive and 236 DE negative cases of patients hospitalized on general medicine for at least 24 hours with lengths of stay of 21 days or less at a large academic medical center in Boston, MA from July 2019 to March 2020. Cases were previously sampled from 4715 available cases using the following EHR trigger criteria (number, % of available cases in trigger cohort sampled): transfer to the ICU after 24 hours (n=97, 99%); death during and within 90-days of index hospitalization (n=72, 28.8%); various clinical triggers (multiple transfers, multiple consultants, new oxygen requirement, rapid response / code, new acute kidney injury; n=127, 2.7%); and no triggers (n=53, 2.2%). We enumerated individual failures within each diagnostic process domain for all cases within each trigger subgroup. Next, we conducted a bivariate analysis by comparing the mean number of failures within each diagnostic process domain in DE positive versus DE negative cases within each trigger cohort. We then calculated the population estimate and weighted odds ratio (wOR) for the overall cohort. Lastly, we ranked the top failures within each diagnostic process dimension.

Results: Our cohort of 349 patients had the following baseline characteristics: mean (SD) age of 65.6 (18.1) years; 55.9% female; 87.4% English speaking; mean (SD) Elixhauser index of 5.0 (2.5). The mean number of diagnostic process failures were significantly associated with DE within each subgroup across all domains (wOR [95% CI] 2.68 [1.91, 3.77], p < 0.01). The mean number of diagnostic process failures (Table 1) were significantly associated with DE within each subgroup in 6 of 8 diagnostic process domains in descending magnitude of effect: physical exam and assessment; history; diagnostic test ordering, performance, and interpretation; subspecialty consultation; diagnostic information and patient follow up; access and presentation. When ranking failures within each diagnostic process domain (Table 2), suboptimal weighing of history was the most frequent diagnostic process failure.

Conclusions: Most domains of the diagnostic process defined by the modified DEER Taxonomy adapted for acute care were associated with DE in a stratified cohort of cases at our institution. While the diagnostic process is complex, our analysis provides direction for identifying high-yield intervention targets for mitigating risk of DE in acute care. Potential targets include improving gaps in physical exam assessment, history, and diagnostic testing and interpretation.


IMAGE 2: *Example: In a patient with Von Willebrand Disease (and increased risk of bleeding) and recent hospitalization for necrotizing pancreatitis, suboptimal weighing of chronic abdominal pain resulted in delayed recognition of spontaneous splenic rupture and subsequent hemorrhagic shock as etiology for observed drop in hemoglobin during hospitalization.