Background: Diagnostic errors (DEs), or the failure to accurately identify or provide timely explanations of a patient’s health problem, are a significant source of patient harm. DEs occur in up to 23% of adult inpatients who transfer to intensive care units (ICU) or die. Few studies have examined how clinician team composition impacts DE risk.
Methods: We conducted a secondary analysis of an observational study of hospitalized patients who died or transferred to the ICU after hospital day two at 29 US hospitals from 1/1/2019 to 12/31/2019. Patient charts underwent structured two-physician review to determine the occurrence of DEs, their causes, and care team models. There were three care team models of interest for this secondary analysis: teaching models (resident with attending); advanced practice provider (APP) models (APP with or without an attending); and direct care models (hospitalist without an APP or resident). APP team structure was further explored through a site survey (response rate 72%). Among responding sites who used APP models, 87% (N= 13) used collaborative APP-attending models (attending and APP saw or verbally discussed each patient) and 13% (N= 2) used an independent model (no mandatory attending involvement). Cases from the original observational study (N= 2428) were excluded in this secondary analysis if the patient was never admitted to general medicine (N= 663), if DE timing was missing (N= 9), and if the care team model was unclear (N= 212), leaving a final cohort of 1544 cases. Bivariable analyses compared patient characteristics between the 3 staffing models of interest. There were no statistical differences in age, sex, race, ethnicity, or comorbidity burden (Table 1). Adjusted rate ratios (ARRs) for DE and harmful DE were calculated from negative binomial GEE models or, if those did not converge, negative binomial regression models with cluster-robust variances (Table 2). The heterogeneity and limited sample sizes in the APP models precluded subset analyses within APP model subtype.
Results: Of 2428 cases, 1544 met criteria for inclusion. 969 cases (62.8%) were on teaching services, 133 (8.6%) were on heterogenous APP services, and 442 (28.6%) were on direct care services.After adjusting for patient characteristics, direct care had a significantly higher ARR of DEs compared to teaching models (ARR 1.54; CI 1.18-2.01) and heterogenous APP models (ARR 1.34; CI 1.05-1.71). No significant difference was found between direct care and APP models (ARR 0.87; CI 0.70-1.08). For harmful DEs, APP models had a significantly lower ARR than teaching (ARR 0.80; CI 0.70-0.93) and direct care models (ARR 0.72; CI 0.57-0.91), with no significant difference found between teaching and direct care models (ARR 1.11; CI 0.94-1.32) (Table 2).
Conclusions: Clinician team composition may influence the risk of DE. Teaching and APP models, many of which are team-based and involve opportunities for regular collective discussions, demonstrate lower ARRs of DEs than direct care services. Our results should be viewed with some caution. We had limited power to fully examine the diversity of care models, particularly within the APP group. We also have not yet accounted for how sites assign patients to care teams, including possible preferential placement of more or less complex patients. Despite these limitations, our results give initial insights into the importance of understanding and optimizing care team models with an explicit eye toward improving diagnostic processes and patient safety.
