We conducted an international, multicenter, prospective observational study using the flash mob study design [28] across 34 hospitals in four countries in May 2024: the Netherlands (n=17), the United Kingdom (n=7), Denmark (n=9), and Switzerland (n=1). This study was previously described in a protocol paper [29]. Each site recruited patients over a single day between 8 AM and 10 PM.

Ethical approval was obtained from the Medical Ethical Assessment Committee (MECC-2022-0795) of the Erasmus University Medical Center, Rotterdam, the Netherlands, and the London–Harrow Research Ethics Committee, Bristol, United Kingdom (IRAS 321129). Approval for all other study sites was acquired in accordance with national and local regulations. All patients gave written informed consent to participate in the study. All collected data were pseudonymized before data analysis.

All patients presenting to ED, AMU, or SDEC were screened for inclusion and exclusion criteria and asked about their use of monitoring devices. Inclusion criteria were ≥18 years, having a device (smartphone or smartwatch) capable of measuring resting heart rate or step count, and the ability to provide informed consent. Patients were excluded for presentation with trauma or clinical instability, as determined by the treating physician. Written informed consent was obtained by trained investigators. Researchers collaborated with patients to complete questionnaires, after which physiological data were extracted from the patients’ devices.

Patient variables included gender, age groups (18‐30, 31‐50, 51‐65, and 65+ years), educational attainment, digital literacy, and device brand and type. The resting, maximum, and minimum heart rates were collected from the patient’s device for 4 time points: 30 days, 7 days, 1 day before the hospital visit, and on the day of admission. In addition, the heart rate measured at presentation to hospital was extracted from the patient’s clinical record. Step count data were collected from the patient’s device for 9 time periods: 30 days, the preceding 7 days before the hospital visit, and the day of the visit. As changes in step count fluctuate more, we chose to collect these on more time points than heart rate. For either heart rate or step count, a minimum of two time points had to be available for data analysis. Additionally, the value of the NEWS, the Clinical Frailty Scale (CFS) [30], and a standardized assessment of gait were recorded. Follow-up data collected up to 7 days after presentation included the date of visit, disposition (admission to hospital or discharge), admission to intensive care or high care areas, death, and discharge, if applicable. Patients were followed up over a period of up to 7 days.

Due to the lack of published data on the frequency distribution of the relevant parameters in the hospitalized population, a formal sample size calculation was not feasible. We estimated that around 200 patients with smart devices and the corresponding data would be recruited based on the number of sites, the average amount of patients visiting the ED daily, and the percentage of people with a smart device [18,19,31]. This cohort size was considered representative in providing data with high external validity.

The primary outcome was feasibility, which was defined as having a smart device with heart rate or step count data recorded on at least twice between 30 days (baseline) before and on the day of admission, enabling trend analysis, and measured as the percentage of patients with an acute care visit. Patient characteristics were reported as absolute numbers per category or subcategory for the total patient population, discharged patients, and admitted patients. Variables were assessed for normal distribution. The heart rate and step count data were compared for their respective mean or median for the different time points. Heart rate changes were also assessed relative to baseline. Both were tested for significant interaction between time, disposition, and change in either heart rate or step count using a generalized linear model (GLM). To further assess the effect of time on change in heart rate and step count, a generalized linear mixed model (GLMM) with binomial distribution and random effects for country and hospital was conducted. Data were collected and safely stored using Castor EDC [32] and analyzed using SPSS (version 28, IBM Statistics). Statistical significance was defined at α=.05.

A total of 1137 patient cases were screened for this study (Figure 1). Of these, 685 were excluded for the following reasons: not having a smart device, not having their device with them during presentation to the hospital, not measuring health data with the smart device, and not having worn their device in the previous days due to illness. A total of 243 patients were excluded because of having insufficient amount of data, and 209 (18%) patients formed the core group for further analysis. Among these patients, 89 (43%) used a smartwatch (in combination with a smartphone) and 120 (57%) only used their smartphone (Multimedia Appendix 1). Heart rate data were recorded in 84 (40%) patients, and step count data were available in 207 (99%) patients.

We included a total of 209 patients who had either a smartwatch or smartphone that measured heart rate or step count in the week prior to their visit (Table 1). Distribution between male and female patients was equal. Most included patients were aged under 65 years, had attained higher education, and were confident in the usage of internet and applications. The overwhelming majority of the patients had low NEWS scores, low frailty scores, and normal gait. In terms of patient disposition, 75 (36%) patients were admitted and 126 (60%) were discharged, and for 8 (4%) patients, the outcome was unknown. Follow-up data showed that no patients died, 3 were admitted to the intensive care unit, and 122 were discharged within 7 days.

The mean resting heart rate increased over the 30 days preceding hospital attendance in all patient groups, including both admitted and discharged patients (Figure 2, Multimedia Appendix 2). Patients who were admitted had a higher mean resting heart rate than those who were discharged. The GLM showed a significant change in the mean resting heart rate over time (P<.001), but there was no significant interaction between time and patient disposition (P=.23). The GLMM correctly classified 85% (51/60) of cases, with higher accuracy for discharge (38/40, 95%) than for admission (13/20, 65%). Among the predictors, only the mean resting heart rate on the day prior to presentation was significantly associated with patient disposition (β=–.083, SE=.038, P=.035). A higher resting heart rate on the day before an acute care visit was linked to a lower likelihood of admission (odds ratio 0.92 per beats per minute [bpm], 95% CI 0.85‐0.99). Mean resting heart rate at 1 month (P=.66), 1 week (P=.23), on the day of presentation (P>.99), and during admission (P=.92) was not significantly associated with disposition. The percentage increase in the mean resting heart rate from baseline prior to hospital admittance was greater in admitted patients than in those who were discharged (Figure 3). In contrast, patients discharged after an acute care visit exhibited the highest percentile change in the mean resting heart rate.

The median step count decreased over the 30 days leading up to hospital presentation in both patient groups (Figure 4, Multimedia Appendix 3). The GLM showed a significant change in median step count over time (P<.001) but no significant interaction between time and disposition (P=.10). The GLMM correctly classified 82% (112/136) of cases, with higher accuracy for discharge (83/88, 94%) than admission (29/48, 60%) patients. Among the predictors, only median step count on the day prior to presentation was significantly associated with disposition (β≈.0001, SE≈0.00009, P=.04). A lower median step count on the day before hospital presentation predicted an increased likelihood of admission (odds ratio=0.90 per 1000 steps, 95% CI 0.82‐1.00). Median step counts at 30 (P=.43), 7 (P=.78), 6 (P=.74), 5 (P=.84), 4 (P=.22), 3 (P=.11), and 2 days (P=.18) prior to hospital presentation were not significantly associated with disposition.

This study found that 40% of patients with an acute care visit had a smart device. In only 50% of the included patients, heart rate and step count data were available for trend analysis, resulting in data available in 20% of the total cohort. Within these patients, a significant increase in heart rate and a decrease in step count were observed several days before they sought emergency care.

The major strength is the generalizability of this point prevalence study into the use of smart devices in Northwestern Europe, including several countries, regions, and hospital settings. Additionally, our study highlights the clinical value of patient-held sensors and understanding patients’ own reference value. This could be used to track a patient’s health remotely and proactively (ie, monitoring individuals before they have an acute illness and become patients) by both patients and health care providers.

Our study had several limitations. First, by only including patients who had a smart device and sufficient data, we created a selection bias. We selected patients with higher education and most likely a higher socioeconomic status, representing a healthier population. However, we expect the underlying pathophysiology to be unaffected by this selection. Second, because of the flash mob research design and because we wanted to minimize the workload for the participating centers, we chose to collect only a limited number of parameters. Resting heart rate between 6 and 2 days before an acute care visit, reason for attendance, medication use, medical history, psychosocial aspects, and ethnicity were not considered even though these parameters could have provided more insight into which groups of patients are being admitted. Finally, the sample size was relatively small, making the subgroup analysis (eg, device type) not feasible. The small sample size and selection bias toward individuals with less compromised physiology may lead to the underestimation of the changes before and on the day of contact with the emergency services.

Proactive monitoring of health data in populations at high risk of acute illness could enable earlier identification of deteriorating health. Monitoring on a daily basis may help detect subtle changes in vital signs or physical activity, allowing for timely medical assessment in primary or ambulatory care settings and potentially preventing a hospital visit. This strategy is already recommended for atrial fibrillation detection by the European Society of Cardiology guidelines 2024 [55]. This benefit is potentially greatest for frail patients, but they are less likely to use a smart device to monitor their health. In addition, changes in vital signs may be more subtle in frail patients, making it more difficult to detect changes. In fitter patients, changes in heart rate and step count may be steeper, but as shown in our study population, the theoretical benefit in a nonselected population is limited. Overall, patients might be encouraged to bring their smart devices with them to an acute care visit, and clinicians might consider asking patients to look at these smart devices, as information on recent heart rate and step counts may be helpful, which can be used as a valid measurement of resting heart rate, as shown by several studies [46,47]. Additionally, both users and manufacturers of smart devices can help health care providers by expanding their reach. Currently, users tend to be younger, more educated, and digitally literate patients, while devices could show clinically valuable trends for all. Both patients and professionals should be wary of misdiagnosis or overdiagnosis. This study provides a foundation for further research to help patients and clinicians pick up early deterioration in health, which might attract interest from tech companies and mobile application developers. Furthermore, a targeted information campaign could be considered, which should be aimed at high-risk patient groups.

Our study showed that it is feasible to use a patient’s own smart device to measure vital signs in the days preceding an acute care visit. We found a significant change in heart rate and step count prior to presentation to hospital, where disposition can be predicted using a patient’s own smart device data. These smart devices are mostly used by younger and healthier patients with higher educational attainment. The use of a patient’s own smart devices for health monitoring in high-risk patient groups is very limited.