Background: Diagnostic errors (DE) are common in patients who die or go to the ICU and are caused by gaps in diagnostic processes. Few data describe whether this observation holds true among patients with sepsis, a disease that progresses quickly and requires a range of clinical information to diagnose correctly. The objective of our study was to determine whether patients with sepsis had different risks for DE or in underlying diagnostic process faults.
Methods: We analyzed data from a multicenter cohort study of 2,428 medicine patients hospitalized at 29 academic medical centers between January 1-December 31, 2019 and who either died or were transferred to the intensive care unit (ICU). Patients included in the UPSIDE study were randomly selected using the Vizient Clinical Data Base. Each participant’s medical record was reviewed to determine whether a DE occurred and errors were mapped to Diagnostic Error Evaluation and Research (DEER) diagnostic process fault domains, such as access/presentation, history taking, physical exam, testing, patient monitoring, obtaining referrals, teamwork, communication, and assessment. Patients were identified with sepsis based on the Clinical Classifications Software ICD-10 discharge code grouper. We used modified Poisson regression modeling, using generalized estimating equations to account for clustering, to estimate adjusted risk ratios (aRR) associated with sepsis and DEER factors. In order to determine differences in diagnostic process faults in patients with and without sepsis, we tested interaction terms between sepsis and each DEER factor.
Results: Slightly more than half (1,218, 50.1%) of patients in our cohort had a primary discharge diagnosis of sepsis, with 272 (22.3% of sepsis patients) experiencing a DE. Those with sepsis had a number of statistically significant differences in baseline characteristics – they were more likely to be older, female, admitted through the emergency department, or to have died during hospitalization. Patients with sepsis also had more chronic illnesses, barriers to communication other than language, and altered mental status on presentation. After controlling for DEER factors and other characteristics, patients with sepsis did not have higher risk for DE (aRR 1.01, 95% CI 0.90-1.14). Among individual DEER proceses (Table 1), patients with sepsis had a lower adjusted risk of DE if there was a gap in history-taking (i.e., failure or delay in providing or eliciting a critical piece of medical history)(aRR 1.34, 95% CI 1.07-1.68 in sepsis patients compared with aRR 2.31, 95% CI1.43, 3.75 without, p for differences in aRR =0.02). Sepsis patients with gaps in referrals had lower aRR than those without sepsis (aRR 1,23, 95% CI 0.93,1.65 in patients with sepsis, aRR 2.08, 95% CI 1.44, 3.02 in patients without, p for differences in aRR =0.0047). No other interactions in observed aRRs for DEER features met tests of statistical significance.
Conclusions: Although diagnostic errors were common in patients with sepsis who died or went to the ICU, sepsis was not associated with higher risk for DE. While association between risk for error and history taking or referral problems is worth further exploration, lack of association between other DEER features and risk of DE may indicate that errors in sepsis patients are driven by broader challenges to the diagnostic process.
