Background: The COVID-19 pandemic required physicians to care for a new disease with uncertain and evolving characteristics while also adhering to physical and social distancing practices, and under conditions of extreme duress – all factors likely to lead to missed or delayed diagnoses among patients admitted ‘under investigation’ for COVID-19.

Methods: We carried out a retrospective study of patients admitted under investigation (PUI) for COVID-19 to one of 8 medical centers affiliated with the Hospital Medicine Reengineering Network (HOMERuN). We used local Infection Control data tracking PUI admissions to General Medicine services taking place between February and June 2020 to identify cases. Patients were included if they had high-risk features and were awaiting a COVID-19 test, had negative test prior to hospitalization but persistent symptoms prompting an additional test, or had a known positive test on admission. We excluded patients whose tests were obtained under universal screening programs. Up to 4 cases per site per month underwent structured 2-physician adjudications using a rigorous process established for a 31-hospital study of diagnostic error in non-COVID patients. Adjudications determined whether a diagnostic error (DE, defined as a missed or delayed diagnosis) took place, and whether any diagnostic process faults were associated with DEs. We used bivariable statistics to compare patients with and without DE, and multivariable models to determine processes or patient factors associated with DE.

Results: Two hundred and fifty-eight patients were selected, of which 36 (14%) had a diagnostic error during hospitalization. Patients with and without errors were statistically similar in terms age, gender, race, ethnicity, and presence of comorbidities such as chronic lung disease, diabetes, chronic kidney disease, and obesity. Patients with and without errors were similar in terms of housing instability, not speaking English as a first language, or use of tobacco or alcohol, as well as risk factors for COVID (such as living in a congregate setting), presenting symptoms, COVID-19 test turnaround times, and eventual COVID-19 test positivity. In unadjusted analyses, diagnostic errors were most associated with problems with clinical assessment, testing choices, history taking gaps, and physical examination (all p< 0.001); teamwork problems were less frequent but were associated with errors (p=0.005). COVID-19 was implicated as a contributor to the diagnostic error in 22 (61%) of patients with errors, and crossed several dimensions, including seeking care (13 cases), history-taking (6), physical exam (16), test interpretation (5), and anchoring on COVID-19 as the diagnosis (19). After adjustment in multivariable models accounting for clustering at the site level, similar findings (though with wide confidence intervals due to relatively small sample size) were seen. A rise in diagnostic error rates over time was just short of statistical significance (p=0.05).

Conclusions: In this study of diagnostic errors among PUI patients, COVID-19 errors were common and associated with factors (such as physical examination or history-taking) likely affected by physical distancing. A rise in diagnostic errors over time is limited by small sample sizes but coincides with hospital re-opening and return of non-COVID patients, suggesting potential overlapping contributions of burnout, workload, and health system changes to diagnostic errors.

IMAGE 1: Table: Association between any DEER process faults and diagnostic errors (N = 258)

IMAGE 2: Table: Diagnostic errors in COVID PUI patients by month of pandemic