Chuchu Qiu

BACKGROUND Vital sign measurement is essential to assess a patient’s physiological status to guide clinical decisions on triage and resuscitation in the emergency department (ED). Manual measurement contributes to bottlenecks and delays in the triage and care processes when a surge in service demand or manpower shortage occurs. Remote photoplethysmography (rPPG) enables contactless vital sign estimation by analyzing changes in light reflected from the skin in smartphone-captured facial videos caused by pulsatile changes in tissue blood volume and light absorption. rPPG has the potential to automate vital sign measurement. However, its accuracy in the ED remains unclear. OBJECTIVE We aimed to compare the accuracy of rPPG–based contactless heart rate (HR) estimation with manual measurement in the ED and assess patient satisfaction and comfort with different measurement methods. METHODS This was a prospective cross-sectional study on a convenience sample of ambulatory adult patients in a tertiary ED in Hong Kong from October to November 2024. The reference standards were manual HR measurements by a research nurse using a standard hospital device. Simultaneously, 25-second facial videos were recorded with an iPhone for contactless HR estimation using a proprietary convolutional neural network algorithm. The accuracy of the contactless method was assessed using the intraclass correlation coefficient (ICC), mean absolute error (MAE), root-mean-square error (RMSE), and Bland–Altman plot, with ± 5 beats per minute (bpm) predefined as clinically acceptable levels of agreement. Patient satisfaction and comfort, rated on a 0-100 mm visual analog scale (VAS), were compared between the two methods using the Wilcoxon signed-rank test. RESULTS We analyzed 478 videos from 161 patients, including 97 women and 64 men. The mean patient age was 55.0 (SD 16.6) years, and 73.9% were Chinese (119/161) with Fitzpatrick skin types of 3 (61.5%, 99/161) and 4 (34.8%, 56/161). The contactless method had an ICC of 0.991 (95% CI 0.987-0.993), a MAE of 1.38 beats per minute (bpm), and a RMSE of 1.82 bpm. The Bland–Altman plot showed a bias of 0.54 bpm (95% CI –2.89 to 3.97 bpm), which falls within the predefined clinically acceptable levels of agreement. Patient satisfaction (contactless VAS 95.4 mm vs. manual VAS 90.3 mm, P<.001) and comfort (contactless VAS 98.0 mm vs. manual VAS 87.0 mm, P<.001) were significantly higher with the contactless method compared to manual measurement. CONCLUSIONS rPPG–based HR estimation from smartphone-captured facial videos is accurate in the ED. Further validation studies involving broader patient populations with a wider range of vital signs and skin colors are warranted. Future research should focus on extending the application of rPPG technology to other vital signs to fully realize its potential in automating vital sign measurements. CLINICALTRIAL ClinicalTrials.gov NCT06536647; https://clinicaltrials.gov/study/NCT06536647