Where Do We Go Wrong?

Auerbach AD, Lee TM, Hubbard CC, et al. Diagnostic errors in hospitalized adults who died or were transferred to intensive care. JAMA Intern Med. 2024;184(2):164-173. In hospitalized adults who were transferred to an ICU or died, how many instances of diagnostic error were identified, and what were their underlying causes and associated harms? Design: Retrospective cohort study. Setting: Twenty-nine academic medical centers in the United States. Population: A convenience sample of adult patients admitted to participating sites. Exposure: Patients who were transferred to an ICU or died during their inpatient stay. Primary and Secondary Outcomes: The primary outcome was the presence or absence of diagnostic error during the index hospitalization. Secondary outcomes included harmful diagnostic errors and error classification. Sponsor: Agency for Healthcare Research and Quality. The authors identified a cohort of 487,532 patients from participating sites during calendar year 2019, of whom 24,591 (5.0%) were transferred to an ICU or died during their hospitalization. A subset of 4,371 cases were randomly selected and reviewed until each site had reviewed at least 100 cases, resulting in a final cohort of 2,428. Of those reviewed, 550 (23.0%) patients experienced a diagnostic error, a rate varying widely across sides. Among patients with diagnostic errors, these errors were judged to have contributed to temporary harm, permanent harm, or death in 436 patients (77.1%). The diagnostic process faults most commonly identified with errors were problems with assessment (eg, delay in considering diagnosis and failure to recognize complications), access and presentation problems (eg, incorrect triage and failure or delay in seeking care), and testing-related issues (eg, delay in ordering or performing needed tests and erroneous clinician interpretation of test). After multivariate adjustment, assessment and testing faults were most highly associated with diagnostic error, corresponding to proportional attributable fractions of 21.4% and 19.9%, respectively. The authors conclude, in their cohort of patients who were transferred to the ICU or died, diagnostic errors were common, harmful, and associated with factors potentially amenable to interventions. This study does not describe the frequency with which errors occur but does broadly describe the most frequently harmful dimensions of diagnostic error. Identifying the common harmful error subtypes assists patient safety administrators and researchers in designing safer processes and procedures within the health care delivery ecosystem. 1.These authors used a "look back" approach in this descriptive report on diagnostic error. What benefits and limitations are inherent to this approach? A "look back" approach constructs a cohort for analysis by starting with the condition or outcome of interest.1Committee on Diagnostic Error in Health Care; Board on Health Care ServicesInstitute of Medicine; The National Academies of Sciences, Engineering, and Medicine.in: Balogh E.P. Miller B.T. Ball J.R. Improving Diagnosis in Health Care. National Academies Press, 2015Google Scholar In this study, patients were included based on being transferred to the ICU, death in the hospital, or both. From this cohort, structured chart review was then performed to identify contributing factors or classify outcomes within an established framework. The primary benefit for this approach is the potential enrichment of the cohort of interest by associated features. Harmful diagnostic errors are not solely present in patients who are transferred to an ICU or die. However, it is reasonable to suggest that harmful errors are more likely to be present in those who do suffer unanticipated deterioration. The present study, part of the "Utility of Predictive Systems in Diagnostic Errors" work, aims to advance machine-learning methods to improve diagnostic performance.2Dalal A.K. Schnipper J.L. Raffel K. et al.Identifying and classifying diagnostic errors in acute care across hospitals: Early lessons from the Utility of Predictive Systems in Diagnostic Errors (Upside) study.J Hosp Med. 2024; 19: 140-145Crossref Scopus (3) Google Scholar The success of their future models depends on efficiently identifying the largest cohort of clinically important diagnostic errors. The limitation of "look back," or retrospective, methods is an inability to quantify the magnitude of importance of any specific feature as a potential contributor to diagnostic error. For example, in patients identified as having experienced a missed opportunity to diagnose stroke, "dizziness" is represented as a major feature.3Newman-Toker D.E. Missed stroke in acute vertigo and dizziness: It is time for action, not debate.Ann Neurol. 2016; 79: 27-31Crossref PubMed Scopus (92) Google Scholar However, the frequency with which patients presenting with a complaint of dizziness are having a stroke, and the frequency with which strokes are missed in patients complaining of dizziness, is orders of magnitude smaller.4Atzema C.L. Grewal K. Lu H. et al.Outcomes among patients discharged from the emergency department with a diagnosis of peripheral vertigo.Ann Neurol. 2016; 79: 32-41Crossref PubMed Scopus (43) Google Scholar Additional work is required to determine the relative yield of factors contributing to missed diagnoses.2.The authors conclude the process faults identified are potential opportunities for interventions. How is it determined which faults are amenable to intervention, and which are not? A common visualization deployed in patient safety analyses is the so-called "Swiss cheese" model, showing how patient harms stem from multiple failed opportunities in prevention or detection.5Wiegmann D.A. Wood L.J. Cohen T.N. et al.Understanding the "Swiss cheese model" and its application to patient safety.J Patient Saf. 2022; 18: 119-123Crossref Scopus (26) Google Scholar Any individual harm event is the result of transmission to the patient through a variety of intermediary steps, and tools such as a "root-cause analysis" can be valuable approaches to identify opportunities for intervention. Determining which of these steps are related to individual or system factors may inform potential solutions to reduce recurrent medical errors. Additionally, diagnostic errors can be classified using frameworks such as the Diagnosis Error Evaluation and Research taxonomy, as used in this work, or the Safer Dx framework.6Schiff G.D. Hasan O. Kim S. Diagnostic error in medicine: analysis of 583 physician-reported errors.Arch Intern Med. 2009; 169: 1881-1887Crossref PubMed Scopus (457) Google Scholar,7Singh H. Sittig D.F. Advancing the science of measurement of diagnostic errors in healthcare: the Safer Dx framework.BMJ Qual Saf. 2015; 24: 103-110Crossref PubMed Scopus (130) Google Scholar These frameworks classify errors into broad categories of shared attributes, potentially informing the scope and expertise required for interventions. Interventions addressing the Diagnosis Error Evaluation and Research dimensions of "history taking" and "physical examination" may take the form of contextual nudges or cognitive debiasing tools. In contrast, those in a "patient follow-up and monitoring" dimension may benefit from an examination of cultural and system-level processes and procedures. Finally, even after such analyses are performed, substantial work is required to determine the yield and acceptability of interventions. The detection of patients with sepsis, for example, is a significant priority to both individual clinicians and health systems. However, interventions in the electronic health record aimed at improving detection have demonstrated poor clinical utility and are likewise reviled by clinicians.8Kamran F. Tjandra D. Heiler A. et al.Evaluation of sepsis prediction models before onset of treatment.NEJM Ai. 2024; 1Crossref Google Scholar,9Knack S.K.S. Scott N. Driver B.E. et al.Early physician gestalt versus usual screening tools for the prediction of sepsis in critically ill emergency patients.Ann Emerg Med. 2024; (S0196-0644(24)00099-4)Abstract Full Text Full Text PDF Scopus (1) Google Scholar There remains substantial unmet need in measurement and validation of interventions otherwise intended to improve health care quality and to evaluate the potential associated costs and burdens.10Saraswathula A. Merck S.J. Bai G. et al.The volume and cost of quality metric reporting.JAMA. 2023; 329: 1840-1847Crossref PubMed Scopus (11) Google Scholar