EHR surveillance uncovers wrong-side imaging orders across 355,000 studies
Wrong-side diagnostic imaging orders are the kind of error that can slip through conventional safety systems because they rarely announce themselves until a chart is reviewed line by line. In a six-facility health system, investigators built an automated electronic health record method to detect these events at scale, adapting the Retract-and-Reorder, or RAR, framework originally used for medication safety surveillance.
The practical value here is immediate: imaging orders are often created, corrected, co-signed, and re-entered across multiple clinicians and settings, so a single wrong-side order may be buried in the workflow rather than preserved in a formal incident report. By extending the detection window to 24 hours and looking for orders that switched from one side to the contralateral side, the team created a surveillance approach suited to the reality of imaging practice, where follow-up adjustments may reflect an error that was caught late rather than a clinically intended change.
A validated query found 1,667 potential events
Across 2021, the system screened 355,000 imaging orders with a side specified. The automated query identified 1,667 RAR events, a rate of 4.70 per 1,000 orders. A chart review of 100 randomly selected events showed a positive predictive value of 87%, with a 95% confidence interval of 79.0% to 92.2%. Using that validation step, the estimated rate of confirmed wrong-side errors was 4.09 per 1,000 orders.
One sentence captures the operational significance: the problem was frequent enough to measure, not merely anecdotal.
The strength of the approach lies in how it translates a known near-miss detection concept into a diagnostic setting. Imaging orders are not medication administrations, and the workflow is more fragmented, with orders moving through outpatient clinics, emergency departments, inpatient units, and administrative queues. Extending the RAR logic to capture contralateral switches within 24 hours made the method flexible enough to pick up likely wrong-side events without depending on voluntary reporting, which is often incomplete and unevenly used.
Outpatient ordering carried the strongest signal
Risk was not distributed evenly across care settings. Outpatient orders had significantly higher odds of an RAR event than inpatient orders, with an odds ratio of 4.53 and a 95% confidence interval of 3.80 to 5.42. Administrative staff also had higher odds of RAR events than attending physicians, with an odds ratio of 2.08 and a 95% confidence interval of 1.73 to 2.49. CT scans carried 79% higher odds of an RAR event than X-rays, reflected in an odds ratio of 1.79 and a 95% confidence interval of 1.34 to 2.39.
Those patterns make clinical sense. Outpatient care depends heavily on precise handoffs, pre-visit preparation, and order entry in a setting where the ordering clinician may have less immediate access to the patient and fewer opportunities for same-day correction. Administrative workflows can also introduce another layer of translation between clinical intent and the final order entered into the system, creating room for laterality mistakes if an order is routed through several hands. CT may be especially vulnerable because it is often used for more complex diagnostic questions and may involve more detailed anatomical laterality decisions than a plain radiograph.
The odds ratios do more than flag associations. They point to where the safety net is thinnest and where a screening tool can be most useful for quality improvement teams trying to identify vulnerable handoff points.
Why automated detection matters in diagnostic safety
Wrong-side imaging errors can delay diagnosis and expose patients to repeat procedures, added radiation, unnecessary follow-up, and avoidable anxiety. Their rarity makes them hard to study with traditional incident reporting alone, yet their preventability makes them a meaningful target for surveillance. An electronic health record-based method offers something qualitatively different from manual review: it can monitor large volumes continuously and identify patterns that would otherwise remain invisible.
The adapted RAR method is especially attractive because it uses data already present in routine clinical documentation. No new reporting workflow is required, and no specialized hardware is needed. In an environment where safety infrastructure often competes with clinical workload, a method that can be run continuously without demanding extra reporting from frontline staff has clear operational appeal.
Just as important, the validation step keeps the signal grounded in chart reality. A query that merely flags every side change would overcount noise. Here, the positive predictive value of 87% suggests the algorithm was capturing a high proportion of true events while remaining broad enough to support system-wide surveillance.
What the workflow may be missing
The findings also expose a familiar weakness in diagnostic safety: errors that occur before imaging begins often leave no durable trace unless they are actively sought. A wrong-side order can be corrected quietly, re-entered under time pressure, or absorbed into a chain of edits that obscures the original mistake. In that context, an automated detection strategy becomes less a retrospective research tool than a live safety monitor.
One sentence stands out from the data itself: the outpatient setting was the dominant risk environment.
That pattern suggests a concrete direction for intervention. Quality teams may need to focus on outpatient ordering interfaces, side-specific prompts, standardized laterality verification, and workflow checks for staff who place or process imaging orders. The study does not claim that any one intervention will eliminate these errors, and it does not measure clinical harm downstream from each event. It does show that the error signal can be detected reliably enough to support continuous surveillance and to evaluate whether local changes reduce the rate over time.
A scalable method for ongoing monitoring
The chief contribution of this work is methodological. By adapting RAR to imaging orders, the investigators created a scalable framework for hospitals and health systems that already possess the necessary EHR infrastructure. The model could support benchmarking across sites, track the effect of safety interventions, and help teams focus chart review on a narrower set of likely wrong-side events.
Because the study was conducted within a six-facility system, the method’s portability to other institutions will depend on local ordering workflows, documentation structure, and EHR design. The analysis also focuses on orders placed during a single year, so broader generalizability and the durability of performance across different practice environments remain open questions. Even so, the central message is concrete: wrong-side imaging order errors can be found systematically, and the electronic record already contains much of the information needed to find them.