Imagine you're an emergency dispatcher. The lines are ringing, but all your ambulances are already out. You have to decide which 'low-priority' caller gets the next available vehicle. Who is secretly the sickest? A new study tested whether a computer could help with that agonizing choice. Researchers in Sweden gave dispatchers a machine learning tool that calculated a risk score for patients. They compared this to the dispatchers' usual clinical judgment. When the next ambulance became free, dispatchers using the tool sent it to the patient with the highest clinical risk score 68.3% of the time. Without the tool, they got it right 62.5% of the time. The difference was small and the statistical finding was borderline, just barely crossing the line for significance. We need to be careful with this result. The study itself notes it was too small—'underpowered'—to reliably detect a difference. The confidence interval, which shows the range of possible true effects, includes the possibility of no benefit at all. It also only looked at patients already judged to be low-priority in two specific regions of Sweden. The tool wasn't tested on high-priority calls or in other countries. No safety issues were reported, but the study wasn't designed to track what happened to patients after the ambulance arrived. This is a first, cautious look at whether data can support human judgment in a high-stakes, resource-limited environment.
ML tool modestly improves dispatcher identification of high-risk patients in ambulance resource constraintsCan a computer help dispatchers spot hidden emergencies when ambulances are scarce?
AI-generated summary of the cited source, checked by automated accuracy review. How we work
This randomized controlled trial evaluated whether providing dispatch nurses with a machine learning (ML)-based risk assessment tool could improve their ability to identify the highest-risk patient during ambulance resource-constrained situations (RCS). The study included 1,245 RCS involving adult patients who had been assessed as requiring a low-priority ambulance response at two sites in Sweden. In the intervention arm, dispatchers were provided with an ML-based risk score; the control arm used standard clinical practice. The primary outcome was whether the first available ambulance was correctly sent to the patient with the highest National Early Warning Score (NEWS 2) based on subsequently collected vital signs.
The main result showed that in the intervention arm, 68.3% of RCS were assessed correctly, compared to 62.5% in the control group. This corresponded to an odds ratio of 1.28 (95% confidence interval 1.00 to 1.63, p = 0.047). The effect was modest, and the confidence interval includes the null value of 1.00. Safety and tolerability data were not reported.
Key limitations significantly constrain the interpretation of these results. The study was conducted only on patients pre-assessed as low-priority in two Swedish regions, limiting generalizability. Furthermore, the trial was underpowered for its primary outcome due to a smaller-than-expected sample size. The practice relevance for other emergency medical systems is unclear. While the findings suggest a potential signal for ML assistance in dispatch triage, the weak statistical significance and study limitations mean this should be viewed as preliminary evidence requiring confirmation in larger, more diverse settings.
View Original Abstract ↓
Meta-analysis reveals CMR and biomarkers key in ICI myocarditis
Heart MRI and Blood Tests Help Detect Immunotherapy Heart Inflammation
Related in Radiology & Imaging
From Other Specialties
