AI sped pulmonary-embolism triage at one German hospital—but did patients benefit?
A retrospective before-and-after study linked an AI-supported response-team workflow to faster diagnosis and treatment, but reductions in hospital stay and 30-day mortality were not statistically significant.
The short version
- What happened
- Researchers retrospectively screened 2,500 presentations with symptomatic central or segmental pulmonary embolism at University Hospital Essen; 424 patients met the study's inclusion criteria.
- Who it affects
- For a person with a dangerous clot, faster coordination could shorten the wait between imaging and a specialist decision. The present evidence cannot yet say that this reduces death, complications or time in hospital.
- What remains unproven
- This was a retrospective before-and-after study at one German university hospital, not a randomised or contemporaneously controlled trial. Secular changes, staffing and greater use of the response team could explain part of the association.
Keep in mind
This was a retrospective before-and-after study at one German university hospital, not a randomised or contemporaneously controlled trial. Secular changes, staffing and greater use of the response team could explain part of the association.
Only 424 of 2,500 screened presentations met the inclusion criteria. The accessible abstract does not provide the pre/post group sizes, baseline balance, false-alert rate or algorithm performance.
Hospital stay, intensive-care stay and 30-day mortality did not improve significantly. The cost result came from an exploratory model and is not evidence of realised savings.
The journal page exposed the peer-reviewed accepted-manuscript abstract but gated the full PDF at review time. Funding and conflict declarations were therefore not verifiable from the accessible record and should be checked when the final article appears.
Source analysis · 1 linked source. Primary analysed source date: 22 October 2026.
Read the full report · 4 min The evidence, methods, implications and limitations
The clearest result is a faster pathway, not a clinical cure
A peer-reviewed study published on 22 October reports what happened after one German university hospital added an algorithm to its pulmonary-embolism pathway. The software analysed CT scans, flagged suspected clots in the lungs and supported triage by the hospital's pulmonary embolism response team, or PERT. These multidisciplinary teams decide quickly whether a patient needs treatment beyond anticoagulant medicine.
The researchers retrospectively screened 2,500 people who presented with symptomatic central or segmental pulmonary embolism from August 2022 to July 2024. A total of 424 met the study's inclusion criteria. Outcomes in the one-year period before implementation were compared with those in the year after it; the accessible abstract does not report the number in each period.[1]
Team activation and treatment intervals changed sharply
PERT activation increased from 9.2% before implementation to 84.5% afterwards. The time between completing a CT scan and recording the pulmonary-embolism diagnosis fell from 2.1 hours to 1.3 hours. Among patients receiving an intervention, the interval from diagnosis to treatment was halved. The reported statistical tests supported all three process differences.
Those numbers matter because high-risk pulmonary embolism can deteriorate rapidly. A reliable alert may help the right clinicians assemble sooner and reduce time spent waiting between scan, diagnosis and a treatment decision. But the intervention was a package: an algorithm, an alert and a team workflow. This design cannot isolate how much of the change came from the model rather than more consistent PERT use, staff learning or other changes over time.[1]
The distinction that matters
What changed—and what the study did not establish
The intervention linked image analysis to a human response team. The strongest evidence concerns the speed and consistency of that pathway.
CT scan
Algorithm flags a suspected pulmonary embolism
Earlier case visibility
PERT triage
Multidisciplinary team reviews the flagged patient
Activation rose from 9.2% to 84.5%
Patient outcomes
Compare stays and 30-day mortality before and after
No statistically significant improvement established
The outcomes patients feel most remained uncertain
Hospital and intensive-care stays and 30-day mortality moved in a favourable direction, but none of those differences reached statistical significance. That does not show the workflow had no benefit; it means this cohort did not establish one. A model-based budget analysis also suggested lower per-patient costs and annual savings, mainly from reduced resource use. It was exploratory rather than an observed financial result.
The next evidential step is a prospective, adequately powered comparison across several hospitals. It should prespecify patient outcomes, separate alert performance from team activation, report false alerts and missed cases, and measure whether any savings actually occur. Until then, the study supports a promising way to organise urgent care—not a claim that AI itself saves lives or hospital budgets.[1]