Harmful diagnostic errors may occur in 1 in every 14 general medical hospital patients

Harmful diagnostic errors may be occurring in as many as 1 in every 14 (7%) hospital patients—at least those receiving general medical care—suggest the findings of a single centre study in the US, published online in the journal BMJ Quality & Safety.

Most (85%) of these errors are likely preventable and underscore the need for new approaches to improving surveillance to avoid these mistakes from happening in the first place, say the researchers.

Previously published reports suggest that current trigger tools for picking up medical mistakes aren’t good enough to detect harmful diagnostic errors, including those with less severe outcomes, suggest the researchers.

They therefore developed and validated a structured case review process to enable clinicians to interrogate the electronic health record (EHR) to evaluate the diagnostic process for hospital patients, assess the likelihood of a diagnostic error, and characterise the impact and severity of harm.

They used the process to estimate retrospectively the prevalence of harmful diagnostic errors in a randomly selected sample of 675 hospital patients out of a total of 9147 in receipt of general medical care between July 2019 and September 2021, excluding the height of the COVID-19 pandemic (April-December 2020).

Cases deemed to be at high risk of diagnostic error were categorised as: transfer to intensive care 24 or more hours after admission (130; 100%); death within 90 days of admission either in hospital or after discharge (141; 38.5%); complex clinical issues, but no transfer to intensive care or death within 90 days of admission (298; 7%).  

Complex clinical issues included clinical deterioration; treatment by several different medical teams; unexpected events, such as cancelled surgery; unclear or discrepant diagnostic information recorded in the medical notes. 

Cases deemed to be at low risk were those fulfilling none of the high risk criteria (106; 2.5%).

Each case was reviewed by two adjudicators trained to judge the likelihood of diagnostic error and identify associated process failures using the Diagnostic Error Evaluation and Research Taxonomy modified for acute care. 

Harm was classified as minor, moderate, severe, and fatal, and whether the diagnostic error contributed to the harm and whether it was preventable were also assessed.

Cases with discrepancies or uncertainty about the diagnostic error or its impact were further reviewed by an expert panel.

Among all the cases reviewed, diagnostic errors were found in 160 cases (154 patients). These comprised: intensive care transfer (54); death within 90 days (34); complex clinical issues (52); low risk patients (20).

Harmful diagnostic errors were assessed to have occurred in 84 cases (82 patients), of which 37 (28.5%) occurred among intensive care transfers; 18 (13%) among those who died within 90 days; 23 (8%) among those with complex clinical issues; and 6 (6%) in low risk cases.  

Harm severity was characterised as minor in 5 (6%), moderate in 36 (43%), major in 25 (30%) and fatal in 18 (21.5%). 

In all, an estimated 85% of harmful diagnostic errors were preventable, with older, White, non-Hispanic, non-privately insured and high-risk patients most at risk.

Weighted to take account of the population, the researchers estimated the proportion of harmful, preventable, and severely harmful diagnostic errors in general medical hospital patients to be just over 7%, 6%, and 1%, respectively.

Process failures were significantly associated with diagnostic errors, particularly uncertainty in initial assessments and complex diagnostic testing and interpretation (4 times the risk), suboptimal subspecialty consultation (3 times the risk), concerns reported by patients (3 times the risk) and history taking (2.5 times the risk).

Forty (48%) diagnostic errors were related to the primary diagnosis at admission or discharge and 44 (52.5%) to a secondary diagnosis. Fifty two (62%) were characterised as delays. Errors associated with major or fatal harm were frequent in the high risk group (55%, 43/78) and rare in the low risk group (0/6).

The most frequent diagnoses associated with diagnostic errors included heart failure, acute kidney failure, sepsis, pneumonia, respiratory failure, altered mental state, abdominal pain and hypoxaemia (low blood oxygen levels).

Careful analysis of the errors and integrating AI tools into the workflow should help to minimise their prevalence, by improving monitoring and triggering timely interventions, suggest the researchers.

This is an observational study, based on estimates, drawn from data on patients receiving general medical care at one single centre, and should be interpreted in that context, caution the researchers.

They also acknowledge that the sample was restricted to patients with a length of hospital stay under 21 days, and that the study relied on information captured in the electronic health record, which is prone to inaccurate recording of deaths within 90 days.

Nevertheless, they conclude: “We estimate that a harmful [diagnostic error] occurred in 1 of every 14 patients hospitalised on general medicine, the majority of which were preventable. Our findings underscore the need for novel approaches for adverse [diagnostic error] surveillance.”