-
PDF
- Split View
-
Views
-
Cite
Cite
Märt Vesinurm, Martial Ndeffo-Mbah, Dan Yamin, Margaret L Brandeau, Terminating pandemics with smartwatches, PNAS Nexus, Volume 4, Issue 3, March 2025, pgaf044, https://doi.org/10.1093/pnasnexus/pgaf044
- Share Icon Share
Abstract
Recent studies have demonstrated that wearable devices, such as smartwatches, can accurately detect infections in presymptomatic and asymptomatic individuals. Yet, the extent to which smartwatches can contribute to prevention and control of infectious diseases through a subsequent reduction in social contacts is not fully understood. We developed a multiscale modeling framework that integrates within-host viral dynamics and between-host interactions to estimate the risk of viral disease outbreaks within a given population. We used the model to evaluate the population-level effectiveness of smartwatch detection in reducing the transmission of three COVID-19 variants and seasonal and pandemic influenza. With a 66% reduction in contacts after smartwatch-based disease detection, we estimate that the reproduction number R would drop from 2.55 (interquartile range [IQR]: 2.09–2.97) to 1.37 (IQR: 1.00–1.55) for the ancestral COVID-19 variant; from 1.54 (IQR: 1.41–1.69) to 0.82 (IQR: 0.68–0.85) for the delta variant; from 4.15 (IQR: 3.38–4.91) to 2.20 (IQR: 1.57–2.52) for the omicron variant; from 1.55 (IQR: 1.34–1.74) to 0.81 (IQR: 0.63–0.87) for pandemic influenza; and from 1.28 (IQR: 1.18–1.35) to 0.74 (IQR: 0.64–0.79) for seasonal influenza. With a 75% reduction in contacts, R decreases below 1 for the delta variant and for pandemic and seasonal influenza. Sensitivity analyses across a wide array of parameter values confirm that self-isolation initiated shortly after smartwatch detection could significantly reduce R under diverse epidemiological conditions, different levels of smartwatch detection accuracy, and realistic self-isolation levels. Our study underscores the revolutionary potential of smartwatches to manage seasonal diseases and alter the course of future pandemics.
Recent studies have shown that smartwatches can be accurately used for early detection of diseases in both asymptomatic and presymptomatic individuals. However, the population-level effectiveness of smartwatch-initiated isolation strategies in controlling disease outbreaks remains unclear. We developed a multiscale model to evaluate the impact of such strategies on the transmission of various seasonal and pandemic strains of SARS-CoV-2 and influenza viruses. Model simulations demonstrate that initiating self-isolation shortly after smartwatch-based disease detection could lower the reproduction number of these diseases to below 1, across diverse epidemiological conditions, various smartwatch detection accuracies, and realistic self-isolation rates. These findings suggest a potential revolution in the control of seasonal diseases and future pandemics through the utilization of smartwatch-based detection systems.
Introduction
The COVID-19 pandemic underscored the rising potential for communicable diseases to spread rapidly and cause substantial morbidity and mortality worldwide. Of particular concern is the threat of emerging or reemerging (e.g. new variants of existing pathogens) transmissible diseases for which there is no effective vaccine. In response to such pandemics, the implementation of large-scale nonpharmaceutical interventions (NPIs) targeting the general population, such as lockdowns and social distancing measures, becomes a crucial strategy for mitigating disease transmission. However, such measures can inflict considerable economic strain and can have social effects such as worsened mental health and physical well-being (e.g. sleep disturbances, escalating depression and anxiety, reduced physical activity and social interaction, and intensified alcohol consumption) (1–3).
In the absence of vaccines, targeted NPIs such as social distancing or reducing social engagements among those infected can effectively curb the spread of infectious diseases (4). Specifically, minimizing social interactions directly influences the reproduction rate (R) of an infectious disease. The term R quantifies the average number of secondary infections generated by an infected individual (5). When R exceeds 1, the epidemic is likely to grow; conversely, if R falls below 1, the epidemic will subside. The value of R is inherently linked to the frequency of social contacts and the infectivity of the disease. By implementing early isolation and reducing social contacts, it is possible to decrease R, significantly slowing or potentially stopping the disease's transmission.
The ability to identify infections before symptoms emerge is crucial for lessening the severity of epidemics and halting pandemics (6). At the individual level, prompt diagnosis can prevent the progression to more severe disease states and enhance the efficacy of measures like case isolation and treatment (7, 8). Conversely, delays in diagnosis contribute to ongoing disease spread at the population level, undermining the effectiveness of control efforts such as social distancing.
For many infections, individuals naturally reduce their social interactions upon experiencing symptoms (9), which helps to reduce transmission. However, a significant challenge arises with diseases where individuals can be most contagious during the presymptomatic phase or even completely asymptomatic, a concern that intensifies with new pandemics. In such scenarios, the population typically lacks immunity, and the peak infectious viral load may precede the onset of symptoms (6). Viral load is the amount of virus contained in a given volume of bodily fluid specimens from infected individuals. For example, early models from the COVID-19 pandemic indicated that up to 44% of transmissions occurred in the presymptomatic stage, 1 to 2 days before symptoms appeared (10). This highlights the critical impact of early detection, even by a single day, on controlling transmission.
Smartwatches can play a pivotal role in identifying infections before symptom onset, so that infected individuals can socially isolate earlier than if they were only identified through clinical diagnosis or diagnostic tests. The onset of infection in otherwise healthy individuals is generally marked by subtle changes in physiological parameters in both presymptomatic and asymptomatic patients, and these changes are discernable by noninvasive wearable sensor devices such as smartwatches. Smartwatches can detect physiological factors that may be associated with infection such as changes in heart rate and heart rate variability, sleep patterns, activity levels, and skin temperature (11, 12). When integrated with machine learning models, these digital biomarkers of infection have been shown to be useful in detecting infections before symptom onset (11, 13).
Several empirical studies have evaluated the ability of wearable sensor-based approaches to detect and diagnose communicable diseases such as COVID-19 and influenza. These studies have shown wearable devices to be considerably accurate in detecting COVID-19 infection in patients (e.g. 88% accuracy at 4 days before symptom onset) (12, 13), especially when combined with data on symptoms (14) and used in conjunction with advanced machine learning methods (15, 16). For influenza, sensor-based wearable devices have been shown to accurately detect 90% of both symptomatic and asymptomatic infections up to 24 h before symptom onset (11, 17).
Given the rapid detection capability of wearable devices and the fact that for many infectious diseases a significant proportion of pathogen shedding leading to secondary infections takes place before and right after symptom onset, smartwatches can be a vital component of infectious disease management. Unlike standard diagnostic tests such as reverse transcriptase PCR or rapid testing that present significant logistical and compliance challenges for large-scale implementation and can at best be administered daily, wearable sensors operate on a continuous basis and with a potentially higher usage rate. The ability of wearable sensors to alert individuals and detect infection as opposed to current diagnostic approaches that require the patient to actively decide to undergo testing could provide individuals with earlier information about their infection status and could result in greater reductions in disease transmission risk through early initiation of control measures.
Few studies have examined the potential impact of smartwatch disease detection on control of communicable diseases. One study developed a compartmental model to illustrate the potential impact of wearable sensors for presymptomatic disease detection using data from the COVID-19 pandemic in Canada (18). Here, we model the impact of early disease detection via smartwatches on the reproduction rate, enabling assessment of the extent to which an epidemic can be slowed or stopped.
Using comprehensive data from previously published literature, we model the potential impact of rapid disease detection by smartwatches for the control of two viral pathogens: SARS-CoV-2, the causative agent of COVID-19 (ancestral strain, delta variant, and omicron variants), and influenza virus (pandemic and seasonal). We develop a multiscale model for estimating the risk of viral respiratory disease transmission that explicitly accounts for within-host viral dynamics and human social contact behavior. We use the model to evaluate the effectiveness of large-scale rapid disease detection through a smartwatch-based approach for the control and potential elimination of disease outbreaks.
Materials and methods
Model overview
We developed a multiscale transmission model that integrates within-host dynamics and between-host interactions. The model enables the exploration of the impact of early social withdrawal, facilitated by smartwatch detection, on disease spread (Fig. 1). The model incorporates pathogen-specific infectivity profiles for both symptomatic and asymptomatic cases, disease-specific information on diagnostic sensitivity of smartwatches, time lag between smartwatch alert and symptom onset, and reduction in social contacts after a smartwatch alert.

Schematic representation of the model. This figure illustrates the changes in infectivity associated with disease progression among both symptomatic (A and C) and asymptomatic (B and D) individuals throughout the infection period. The shaded gray area beneath each curve indicates the anticipated number of secondary infections that could stem from an infected person. In scenarios without smartwatch alerts, a decrease in infectivity follows the onset of symptoms occurs (A), due to a reduction in social interactions, but asymptomatic individuals have no behavior change (B). Decreased contacts are assumed to occur Td days after symptoms develop; in the above illustration, Td is set to 0. C and D) Depict situations with smartwatch alerts, where both symptomatic and asymptomatic individuals curtail their social interactions subsequent to receiving detection alerts from their smartwatches.
We used the model to explore the potential contribution of smartwatch detection in reducing transmission of seasonal and pandemic viral pathogens. We illustrate our model for SARS-CoV-2 (including its ancestral strain, delta, and omicron variants) and seasonal and pandemic influenza strains.
Reproduction number
For each disease, we calculate R, the average number of secondary infections produced by an individual case, by tallying the total number of people infected by one person throughout their infectious period, denoted as τ (19–21). Specifically, for a given disease d, an infected person may be symptomatic with probability , or asymptomatic with probability 1 − . The reproduction number for disease d is determined by the formula:
where represents the average number of infections caused by a symptomatic infected individual on day t of their infectious period and indicates the average number of infections caused by an asymptomatic individual on day t of their infectious period.
Consistent with previous studies (20, 21), we assume that the average number of infections generated by a symptomatic or asymptomatic individual on day t of their infectious period, denoted by or , respectively, is directly proportional to the interaction between their daily viral load and contact rate.
Viral load dynamics is a key factor in diagnosing viral infection and determining transmission risk. Influenza and SARS-CoV-2 viral load are estimated by using quantitative real-time reverse transcriptase polymerase chain reaction assays with the results reported as cycle threshold value and/or log10 (viral copies/mL) from respiratory or blood specimens (4, 22, 23). Empirical studies have shown that the likelihood of influenza and SARS-CoV-2 infection correlates with the logarithm of the viral load and is related to the dose of viral exposure (24–28). Therefore, we explicitly integrate the viral load in calculating the reproduction ratio of these pathogens using an age-of-infection framework (19), consistent with other modeling studies (21, 29).
This relationship can be expressed as the product of the logarithm of the daily infectious viral load (21, 30–33), or , and the number of daily contacts, or the individual engages in on that day. Incorporating this into Eq. 1 yields expression A that is proportional to the reproduction number:
Modeling the effect of smartwatch alerts
Smartwatches can achieve early detection of infections in presymptomatic and asymptomatic individuals. Most disease testing tools, such as PCR tests, are designed to detect viable pathogen materials as a proxy for the presence of infectious pathogen. As a result, the ability of traditional tools to detect early infection generally increases with disease progression (34, 35). To incorporate smartwatch detection in the model, we conservatively assume that from a certain day after infection a smartwatch can detect disease with a given accuracy (Fig. 1C and D). Prior to smartwatch detection, infected individuals transmit the pathogen at the same rate as if they were not wearing a smartwatch. However, once the disease is detected and an alert is issued, the individual may choose to social distance by reducing their number of contacts by a fraction for the remainder of their infectious period.
If we denote by and the time at which the smartwatch detects infection in infected individuals who become symptomatic or remain asymptomatic, respectively, the adjusted daily contact rates for such individuals, and can be written as:
where is the binary delta function.
The total reduction in the reproduction number due to early smartwatch detection is proportional to expression B:
Combining expressions A and B, the reproduction number with smartwatch capabilities, , is
Note that the relationship between A and B indicates that if the percentage reduction in contacts with disease progression is known, the actual number of contacts is not required in the model and can be normalized to 1. However, for a heterogeneous model, such as one with age-specific settings, an explicit contact matrix is necessary and can be informed by previous literature (36).
Model parameters
Our models include epidemiological and behavioral parameters as well as parameters relating to smartwatch sensitivity in detecting diseases. Parameter values, sourced from the available literature, are shown in Tables 1 and S1. Each parameter is disease-specific.
Pathogen . | Reproduction number under status quo, . | Proportion of individuals who remain asymptomatic, . | Mean day of symptom onset for individuals who become symptomatic . | Time from symptom onset to testing . | Smartwatch sensitivitya . | Difference between smartwatch alert and symptom onset . | % Reduction in contacts given smartwatch alert . |
---|---|---|---|---|---|---|---|
COVID-19 (ancestral) | 2.68 (IQR: 2.00–2.90) (37, 38) | 15.6% (39) | 5 (35) | 53 (95% CI: 49–58) hours (40) | 88% (13) | 3 days (13) | (66–75%) (9) |
COVID-19 (delta) | 1.55 (95% CI: 1.17, 1.93) (41) | 8.4% (42) | 4 (35) | 53 (95% CI: 49–58) hours (40) | 88% (13) | 3 days (13) | (66–75%) (9) |
COVID-19 (omicron) | 4.20 (95% CI: 2.05, 6.35) (41) | 25.5% (42) | 4 (35) | 53 (95% CI: 49–58) hours (40) | 88% (13) | 3 days (13) | (66–75%) (9) |
Influenza (pandemic, H1N1) | 1.65 (IQR: 1.30–1.70) (43) | 22.7% (44) | 2 (45) | 53 (95% CI: 49–58) hours (40) | 90% (11) | 36 h (11) | (66–75%) (9) |
Influenza (seasonal) | 1.28 (IQR: 1.19–1.37) (43) | 21.0% (44) | 2 (45) | 39 (95% CI: 34–45) hours (40) | 94% (17) | 23 h (17) | (66–75%) (9) |
Pathogen . | Reproduction number under status quo, . | Proportion of individuals who remain asymptomatic, . | Mean day of symptom onset for individuals who become symptomatic . | Time from symptom onset to testing . | Smartwatch sensitivitya . | Difference between smartwatch alert and symptom onset . | % Reduction in contacts given smartwatch alert . |
---|---|---|---|---|---|---|---|
COVID-19 (ancestral) | 2.68 (IQR: 2.00–2.90) (37, 38) | 15.6% (39) | 5 (35) | 53 (95% CI: 49–58) hours (40) | 88% (13) | 3 days (13) | (66–75%) (9) |
COVID-19 (delta) | 1.55 (95% CI: 1.17, 1.93) (41) | 8.4% (42) | 4 (35) | 53 (95% CI: 49–58) hours (40) | 88% (13) | 3 days (13) | (66–75%) (9) |
COVID-19 (omicron) | 4.20 (95% CI: 2.05, 6.35) (41) | 25.5% (42) | 4 (35) | 53 (95% CI: 49–58) hours (40) | 88% (13) | 3 days (13) | (66–75%) (9) |
Influenza (pandemic, H1N1) | 1.65 (IQR: 1.30–1.70) (43) | 22.7% (44) | 2 (45) | 53 (95% CI: 49–58) hours (40) | 90% (11) | 36 h (11) | (66–75%) (9) |
Influenza (seasonal) | 1.28 (IQR: 1.19–1.37) (43) | 21.0% (44) | 2 (45) | 39 (95% CI: 34–45) hours (40) | 94% (17) | 23 h (17) | (66–75%) (9) |
CI, confidence interval; IQR, interquartile range.
aAvailable data on smartwatch sensitivity did not distinguish between COVID-19 variants.
Pathogen . | Reproduction number under status quo, . | Proportion of individuals who remain asymptomatic, . | Mean day of symptom onset for individuals who become symptomatic . | Time from symptom onset to testing . | Smartwatch sensitivitya . | Difference between smartwatch alert and symptom onset . | % Reduction in contacts given smartwatch alert . |
---|---|---|---|---|---|---|---|
COVID-19 (ancestral) | 2.68 (IQR: 2.00–2.90) (37, 38) | 15.6% (39) | 5 (35) | 53 (95% CI: 49–58) hours (40) | 88% (13) | 3 days (13) | (66–75%) (9) |
COVID-19 (delta) | 1.55 (95% CI: 1.17, 1.93) (41) | 8.4% (42) | 4 (35) | 53 (95% CI: 49–58) hours (40) | 88% (13) | 3 days (13) | (66–75%) (9) |
COVID-19 (omicron) | 4.20 (95% CI: 2.05, 6.35) (41) | 25.5% (42) | 4 (35) | 53 (95% CI: 49–58) hours (40) | 88% (13) | 3 days (13) | (66–75%) (9) |
Influenza (pandemic, H1N1) | 1.65 (IQR: 1.30–1.70) (43) | 22.7% (44) | 2 (45) | 53 (95% CI: 49–58) hours (40) | 90% (11) | 36 h (11) | (66–75%) (9) |
Influenza (seasonal) | 1.28 (IQR: 1.19–1.37) (43) | 21.0% (44) | 2 (45) | 39 (95% CI: 34–45) hours (40) | 94% (17) | 23 h (17) | (66–75%) (9) |
Pathogen . | Reproduction number under status quo, . | Proportion of individuals who remain asymptomatic, . | Mean day of symptom onset for individuals who become symptomatic . | Time from symptom onset to testing . | Smartwatch sensitivitya . | Difference between smartwatch alert and symptom onset . | % Reduction in contacts given smartwatch alert . |
---|---|---|---|---|---|---|---|
COVID-19 (ancestral) | 2.68 (IQR: 2.00–2.90) (37, 38) | 15.6% (39) | 5 (35) | 53 (95% CI: 49–58) hours (40) | 88% (13) | 3 days (13) | (66–75%) (9) |
COVID-19 (delta) | 1.55 (95% CI: 1.17, 1.93) (41) | 8.4% (42) | 4 (35) | 53 (95% CI: 49–58) hours (40) | 88% (13) | 3 days (13) | (66–75%) (9) |
COVID-19 (omicron) | 4.20 (95% CI: 2.05, 6.35) (41) | 25.5% (42) | 4 (35) | 53 (95% CI: 49–58) hours (40) | 88% (13) | 3 days (13) | (66–75%) (9) |
Influenza (pandemic, H1N1) | 1.65 (IQR: 1.30–1.70) (43) | 22.7% (44) | 2 (45) | 53 (95% CI: 49–58) hours (40) | 90% (11) | 36 h (11) | (66–75%) (9) |
Influenza (seasonal) | 1.28 (IQR: 1.19–1.37) (43) | 21.0% (44) | 2 (45) | 39 (95% CI: 34–45) hours (40) | 94% (17) | 23 h (17) | (66–75%) (9) |
CI, confidence interval; IQR, interquartile range.
aAvailable data on smartwatch sensitivity did not distinguish between COVID-19 variants.
To assess the impact of early social distancing, enabled by smartwatch detection, on controlling disease outbreaks, we compared the effective reproduction number with smartwatch capabilities, denoted in the model as , against the reported baseline reproduction number in the existing literature, . The value is influenced by factors such as preexisting population immunity levels from vaccination and/or prior exposure. For pandemics, the population is assumed to be largely susceptible (6) and is considered to be equivalent to the basic reproduction number (R0). R0 is the average number of people an infectious individual will infect in a completely susceptible population. Thus, for the ancestral COVID-19 variants and pandemic H1N1 influenza, we assumed the reproduction number was equal to R0. For the other pathogens, the reproduction number was set equal to the effective reproduction number reported at the start of an epidemic season or immediately after a new strain's emergence (5).
For each disease, the infectious viral load evolves with disease progression and can differ between asymptomatic and symptomatic individuals (21, 35, 45, 46). To accurately capture these variations, we calibrated the daily log viral load for each disease and presentation type (symptomatic or asymptomatic) by fitting it to the available viral load data using a gamma distribution (Table S1) (35, 45, 47).
We estimated that 15.6, 8.4, and 25.5% of individuals infected with the COVID-19 ancestral strain, delta variant, and omicron variant, respectively, would remain asymptomatic (39, 42); and that 22.7 and 21.0% of individuals infected with pandemic and seasonal influenza, respectively, would remain asymptomatic (44). For individuals who become symptomatic, the timing of symptom onset varies by disease: 5 days after infection for COVID-19 ancestral variant (35), 4 days for COVID-19 delta and omicron variants (35), and 2 days for pandemic and seasonal influenza (45). Based on a recent prospective study, we estimated that time from symptom onset to testing was 53 (95% confidence interval [CI]: 49–58) hours for COVID-19 variants, 52 (95% CI: 48–56) hours for pandemic influenza, and 39 (95% CI: 34–45) hours for seasonal influenza (40).
The smartwatch's day of disease detection and accuracy vary between pathogens and were informed using data from randomized clinical trials (11, 13, 17). We estimated that smartwatches would have 88% sensitivity in detecting COVID-19 variants (13), 90% sensitivity in detecting pandemic influenza (11), and 94% sensitivity in detecting seasonal influenza (17) and that smartwatches could detect COVID-19 infection 3 days before symptom onset (13), pandemic influenza 36 h before symptom onset (11), and seasonal influenza 23 h before symptom onset (17).
In the absence of smartwatches, we assumed that infected symptomatic individuals would reduce their social contacts Td days after symptom onset, which corresponds to the average time from symptom onset to seeking care/diagnosis for each disease, but asymptomatic individuals would not reduce their social contacts. We assumed that symptomatic individuals would reduce their contact rates by 66–75% and that this reduction would continue throughout their infectious period. This range of values is consistent with empirical estimates of reduction in social contacts for symptomatic seasonal influenza cases (9).
With smartwatches, we tested scenarios where both symptomatic and asymptomatic individuals who learn of their infection status through smartwatch alerts would immediately reduce social effective contacts by 66 and 75%. This effective reduction in contacts can be regarded as the product of two factors: the probability that individuals decide to self-isolate (which can also be referred to as compliance with smartwatch detection) and the degree of contact reduction among those who choose to comply (i.e. the level of isolation). Our assumptions are supported by a recent prospective study of COVID-19, which found that individuals reduced their contacts by 68% after testing (40). Additionally, a survey conducted in early 2020 indicated that providing compensation for lost wages could increase compliance with self-quarantine from 57 to 94% (48), suggesting that these reductions may be even more significant under the right conditions. Given the uncertainty, we conducted a sensitivity analysis for contact reduction following an alert, with reductions ranging from 50 to 100%.
The potential impact of smartwatch detection on disease transmission depends on two key factors: the timing of the smartwatch alert and the degree of contact reduction, which is influenced by compliance with self-isolation and the accuracy of the alerts. These parameters are dynamic and may evolve with advancements in smartwatch technology. To assess this, we conducted a two-way sensitivity analysis for each pathogen, examining both the time from exposure to smartwatch detection (ranging from 1 to 7 days) and the percentage contact reduction in response to the alert (ranging from 50 to 100%).
Model simulation
For each disease, we simulated the model to evaluate the effectiveness of early social distancing prompted by smartwatch detection. Parameters for each disease were independently sampled, and the projected reproduction number after smartwatch alert was calculated based on Eq. 5. For each disease, we simulated 1,000 parameter set realizations and for each run evaluated the reproduction number. We used these outcomes to calculate reductions in the effective reproduction number due to earlier smartwatch detection (and associated 95% CIs) as well as the probability that the reproduction number will be below 1 (a necessary condition for preventing a disease outbreak or eliminating an ongoing outbreak). There are no primary data underlying this work, and all data are publicly available from the references cited in this study. For complete transparency, the model code and all associated data are publicly available at https://github.com/MartVesinurm/TerminatingPandemicsWithSmartwatches.
Results
In a baseline scenario (Table 2, top panels of Fig. 2), if smartwatch detection results in a 66% reduction in social contacts, and smartwatch sensitivity is as reported in the current literature, then for COVID-19 we estimate that R would be reduced from 2.55 (interquartile range [IQR]: 2.09–2.97) to a mean of 1.37 (IQR: 1.00–1.55) for the ancestral variant, corresponding to a 46% reduction in transmission risk; from 1.54 (IQR: 1.41–1.69) to 0.82 (IQR: 0.68–0.85) for the delta variant, corresponding to a 47% reduction in transmission risk; and from 4.15 (IQR: 3.38–4.91) to 2.20 (IQR: 1.57–2.52) for the omicron variant, corresponding to a 47% reduction in transmission risk. For pandemic influenza, we estimate that R would be reduced from 1.55 (IQR: 1.34–1.74) to 0.81 (IQR: 0.63–0.87), corresponding to a 48% reduction in transmission risk; for seasonal influenza the reduction would be from 1.28 (IQR: 1.18–1.35) to 0.74 (IQR: 0.64–0.79), corresponding to a 42% reduction in transmission risk.

Estimated reproduction number with and without smartwatch-facilitated early disease detection for ancestral, delta, and omicron variants of COVID-19 (left panels), and pandemic and seasonal influenza (right panels). It is assumed that infection is detected 72 h before symptom onset for COVID-19 variants, 36 h for pandemic influenza, and 23 h before symptom onset for seasonal influenza, followed by immediate withdrawal from social activities. The upper panels assume 66% withdrawal from social activities, and the lower panels assume 75% withdrawal.
Change in reproduction number due to smartwatch detection, with 66 or 75% reduction in social contacts.
Pathogen . | Reproduction number under status quo, mean (IQR)a . | Reproduction number with smartwatch detection, mean (IQR) . | Reduction in reproduction number, mean (95% CI) . |
---|---|---|---|
66% reduction in social contacts | |||
COVID-19 (ancestral) | 2.55 (2.09–2.97) | 1.37 (1.00–1.55) | 1.18 (1.14–1.21) |
COVID-19 (delta) | 1.54 (1.41–1.69) | 0.82 (0.68–0.85) | 0.72 (0.70–0.74) |
COVID-19 (omicron) | 4.15 (3.38–4.91) | 2.20 (1.57–2.52) | 1.95 (1.89–2.01) |
Influenza (pandemic, H1N1) | 1.55 (1.34–1.74) | 0.81 (0.63–0.88) | 0.74 (0.72–0.76) |
Influenza (seasonal) | 1.28 (1.18–1.35) | 0.74 (0.64–0.79) | 0.54 (0.53–0.55) |
75% reduction in social contacts | |||
COVID-19 (ancestral) | 2.55 (2.09–2.97) | 1.16 (0.80–1.25) | 1.39 (1.35–1.43) |
COVID-19 (delta) | 1.54 (1.41–1.69) | 0.69 (0.54–0.67) | 0.86 (0.84–0.88) |
COVID-19 (omicron) | 4.15 (3.38–4.91) | 1.87 (1.24–2.05) | 2.27 (2.21–2.35) |
Influenza (pandemic, H1N1) | 1.55 (1.34–1.74) | 0.68 (0.50–0.70) | 0.87 (0.85–0.89) |
Influenza (seasonal) | 1.28 (1.18–1.35) | 0.64 (0.54–0.67) | 0.64 (0.63–0.65) |
Pathogen . | Reproduction number under status quo, mean (IQR)a . | Reproduction number with smartwatch detection, mean (IQR) . | Reduction in reproduction number, mean (95% CI) . |
---|---|---|---|
66% reduction in social contacts | |||
COVID-19 (ancestral) | 2.55 (2.09–2.97) | 1.37 (1.00–1.55) | 1.18 (1.14–1.21) |
COVID-19 (delta) | 1.54 (1.41–1.69) | 0.82 (0.68–0.85) | 0.72 (0.70–0.74) |
COVID-19 (omicron) | 4.15 (3.38–4.91) | 2.20 (1.57–2.52) | 1.95 (1.89–2.01) |
Influenza (pandemic, H1N1) | 1.55 (1.34–1.74) | 0.81 (0.63–0.88) | 0.74 (0.72–0.76) |
Influenza (seasonal) | 1.28 (1.18–1.35) | 0.74 (0.64–0.79) | 0.54 (0.53–0.55) |
75% reduction in social contacts | |||
COVID-19 (ancestral) | 2.55 (2.09–2.97) | 1.16 (0.80–1.25) | 1.39 (1.35–1.43) |
COVID-19 (delta) | 1.54 (1.41–1.69) | 0.69 (0.54–0.67) | 0.86 (0.84–0.88) |
COVID-19 (omicron) | 4.15 (3.38–4.91) | 1.87 (1.24–2.05) | 2.27 (2.21–2.35) |
Influenza (pandemic, H1N1) | 1.55 (1.34–1.74) | 0.68 (0.50–0.70) | 0.87 (0.85–0.89) |
Influenza (seasonal) | 1.28 (1.18–1.35) | 0.64 (0.54–0.67) | 0.64 (0.63–0.65) |
CI, confidence interval; IQR, interquartile range.
aStatus quo reproduction number values were obtained via 1,000 simulation runs using the values given in Table 1.
Change in reproduction number due to smartwatch detection, with 66 or 75% reduction in social contacts.
Pathogen . | Reproduction number under status quo, mean (IQR)a . | Reproduction number with smartwatch detection, mean (IQR) . | Reduction in reproduction number, mean (95% CI) . |
---|---|---|---|
66% reduction in social contacts | |||
COVID-19 (ancestral) | 2.55 (2.09–2.97) | 1.37 (1.00–1.55) | 1.18 (1.14–1.21) |
COVID-19 (delta) | 1.54 (1.41–1.69) | 0.82 (0.68–0.85) | 0.72 (0.70–0.74) |
COVID-19 (omicron) | 4.15 (3.38–4.91) | 2.20 (1.57–2.52) | 1.95 (1.89–2.01) |
Influenza (pandemic, H1N1) | 1.55 (1.34–1.74) | 0.81 (0.63–0.88) | 0.74 (0.72–0.76) |
Influenza (seasonal) | 1.28 (1.18–1.35) | 0.74 (0.64–0.79) | 0.54 (0.53–0.55) |
75% reduction in social contacts | |||
COVID-19 (ancestral) | 2.55 (2.09–2.97) | 1.16 (0.80–1.25) | 1.39 (1.35–1.43) |
COVID-19 (delta) | 1.54 (1.41–1.69) | 0.69 (0.54–0.67) | 0.86 (0.84–0.88) |
COVID-19 (omicron) | 4.15 (3.38–4.91) | 1.87 (1.24–2.05) | 2.27 (2.21–2.35) |
Influenza (pandemic, H1N1) | 1.55 (1.34–1.74) | 0.68 (0.50–0.70) | 0.87 (0.85–0.89) |
Influenza (seasonal) | 1.28 (1.18–1.35) | 0.64 (0.54–0.67) | 0.64 (0.63–0.65) |
Pathogen . | Reproduction number under status quo, mean (IQR)a . | Reproduction number with smartwatch detection, mean (IQR) . | Reduction in reproduction number, mean (95% CI) . |
---|---|---|---|
66% reduction in social contacts | |||
COVID-19 (ancestral) | 2.55 (2.09–2.97) | 1.37 (1.00–1.55) | 1.18 (1.14–1.21) |
COVID-19 (delta) | 1.54 (1.41–1.69) | 0.82 (0.68–0.85) | 0.72 (0.70–0.74) |
COVID-19 (omicron) | 4.15 (3.38–4.91) | 2.20 (1.57–2.52) | 1.95 (1.89–2.01) |
Influenza (pandemic, H1N1) | 1.55 (1.34–1.74) | 0.81 (0.63–0.88) | 0.74 (0.72–0.76) |
Influenza (seasonal) | 1.28 (1.18–1.35) | 0.74 (0.64–0.79) | 0.54 (0.53–0.55) |
75% reduction in social contacts | |||
COVID-19 (ancestral) | 2.55 (2.09–2.97) | 1.16 (0.80–1.25) | 1.39 (1.35–1.43) |
COVID-19 (delta) | 1.54 (1.41–1.69) | 0.69 (0.54–0.67) | 0.86 (0.84–0.88) |
COVID-19 (omicron) | 4.15 (3.38–4.91) | 1.87 (1.24–2.05) | 2.27 (2.21–2.35) |
Influenza (pandemic, H1N1) | 1.55 (1.34–1.74) | 0.68 (0.50–0.70) | 0.87 (0.85–0.89) |
Influenza (seasonal) | 1.28 (1.18–1.35) | 0.64 (0.54–0.67) | 0.64 (0.63–0.65) |
CI, confidence interval; IQR, interquartile range.
aStatus quo reproduction number values were obtained via 1,000 simulation runs using the values given in Table 1.
We considered a more optimistic scenario in which smartwatch detection results in a 75% reduction in social contacts (Table 2, bottom panels of Fig. 2). In this case, we estimate that R for COVID-19 would be reduced to a mean of 1.16 (IQR: 0.80–1.25) for the ancestral variant, to 0.69 (IQR: 0.54–0.67) for the delta variant, and to 1.87 (IQR: 1.24–2.05) for the omicron variant, corresponding to 55, 56, and 55% reductions in transmission risk, and that R would be reduced to 0.68 (IQR: 0.50–0.70) for pandemic influenza and to 0.64 (IQR: 0.54–0.67) for seasonal influenza, corresponding to 56 and 50% reductions in transmission risk.
To evaluate the sensitivity of our results to parameter variability, we conducted an uncertainty analysis by sampling all model parameters over their distribution (Tables 1 and S1) while considering a uniform distribution unif(0.66,0.75) for the reduction in social contacts following smartwatch alerts. We then evaluated the probability that early disease detection through smartwatch alerts with resulting self-isolation can reduce the effective reproduction number R below 1. Figure 3 shows the probability that R will be below 1—and thus the probability of early epidemic termination—as a function of the time from exposure (pathogen infection) to smartwatch detection.

Probability of the reproduction number falling below 1 for ancestral, delta, and omicron variants of COVID-19 (left panel), and pandemic and seasonal influenza (right panel). Smartwatch detection under our base case assumptions (see Table 1) is marked by a light blue dot. Detection at symptom onset is marked by a dark red dot. It is assumed that level of withdrawal from social activities is uniformly distributed between 66 and 75%.
For COVID-19, if a smartwatch detects infection 3 days before symptom onset, as assumed in our baseline scenario, the probability of reducing R below 1 is 34, 90, and 5.4% for the ancestral, delta, and omicron variants, respectively (Fig. 3, left panel). If the infection is instead detected 1 day prior to onset of symptoms, the probability of interrupting transmission through smartwatch alerts and self-isolation is reduced to 8.0, 62, and 1.1% for the ancestral, delta, and omicron variants, respectively, and if the infection is detected on the day of symptom onset, the probability of reducing R below 1 is 3.1, 20, and 0.8% for the ancestral, delta, and omicron variants, respectively.
For pandemic influenza, if a smartwatch detects infection 36 h before symptom onset, as assumed in our baseline scenario, the probability of reducing R below 1 is 90% (Fig. 3, right panel). For seasonal influenza, if a smartwatch detects infection 23 h before symptom onset, as assumed in our baseline scenario, the probability of reducing R below 1 is 94%. If these infections are instead detected at symptom onset, the probability of reducing R below 1 remains 90% for pandemic influenza and 94% for seasonal influenza.
We evaluated the minimum required contact reduction to achieve a 50 or 75% probability of disease elimination (R < 1) (Fig. 4). For 50% probability of disease elimination, given baseline smartwatch detection times reported in Table 1, a 72, 55, and 83% reduction in contacts would be required for the ancestral, delta, and omicron COVID-19 variants, respectively, and a 57 and 47% reduction in contacts would be required for pandemic and seasonal influenza, respectively. For 75% probability of disease elimination, and baseline smartwatch detection times, a 78, 60, and 87% reduction in contacts would be required for the ancestral, delta, and omicron COVID-19 variants, respectively, and a 64 and 52% reduction in contacts would be required for pandemic and seasonal influenza, respectively.

Required reduction in social contacts for the probability of the reproduction number falling below 1 being 50% (black lines) and 75% (gray lines) for ancestral, delta, and omicron variants of COVID-19 (left panel), and pandemic and seasonal influenza (right panel). Smartwatch detection under our base case assumptions (see Table 1) is marked by a blue dot. Detection at symptom onset is marked by a red dot; these dots are not shown for ancestral and omicron COVID-19 variants because no amount of social contact reduction at symptom onset would be sufficient for disease elimination (R < 1).
We conducted a two-way sensitivity analysis for each pathogen, varying the time from exposure to smartwatch detection and the percentage contact reduction in response to the alert (Fig. 5). The analysis shows that high population-level compliance is crucial for significantly reducing disease transmission. For example, when the smartwatch alerts an infected individual within 2 days of exposure and there is a 75% reduction in contacts, R < 1 is achievable for the COVID-19 delta variant, pandemic influenza, and seasonal influenza. The analysis also shows that earlier detection plays a more critical role in reducing disease spread than contact reduction, as indicated by the more vertical contours in the plot. For instance, an alert after 4 days, even with 75% contact reduction, is unlikely to bring R below 1. However, an alert after 2 days with 66% contact reduction could reduce R below 1 for the COVID-19 delta variant and for pandemic and seasonal influenza. Even with lower compliance, significant transmission reduction is still possible with early detection: for example, an alert 3 days after exposure with only a 50% reduction in contacts could still lower transmission by 20–35%.

Two-way sensitivity analysis illustrating the relationship between time from exposure to smartwatch detection (x-axis, in days) and effective reduction in social contacts (y-axis, as a percentage) for COVID-19 (ancestral, delta, and omicron variants), pandemic influenza, and seasonal influenza. Contour lines indicate the percentage of transmission reduction achievable under varying detection times and levels of contact reduction. The dark solid line in each panel represents the threshold where the reproduction number R can be reduced below 1 under specific conditions of detection time and contact reduction.
Discussion
This study demonstrates the potential population-level impact of smartwatches in managing seasonal diseases and future pandemics. Our findings, observed across a range of realistic scenarios, show that a 66–75% reduction in social contacts soon after detection by smartwatches can lead to a 40–65% decrease in disease transmission. This reduction can even tip the scale toward controlling a pandemic or epidemic, with the reproduction number falling below 1. A two-way sensitivity analysis on the time of detection and the level of reduction in contacts shows that earlier detection plays a bigger role than reduction in contacts in reducing disease spread.
Technological advancements that utilize wearables have been previously introduced to reduce transmission (49). For example, during the COVID-19 pandemic, smartphone Bluetooth technology, with citizens’ consent, was used to alert individuals who had come into proximity to an infected person. While this approach provides valuable early information, it often leads to a high number of false alerts, mistakenly alerting individuals who were never infected. To illustrate this, consider a fairly contagious disease with an effective reproduction ratio of 2. Although an infected individual may come into contact with 15–60 people during their illness (40), they are likely to transmit the disease to only two people. In contrast, a smartwatch detection approach suggests a more precise strategy, as it targets individuals who are already infected, focusing on the time when their risk of transmission is highest.
Our findings are especially relevant in addressing pandemics. For many infections, individuals naturally reduce their social interactions upon experiencing symptoms (9) and even more so after testing (40), which contributes to reducing transmission. However, a significant challenge arises with diseases where individuals can be most contagious during the presymptomatic phase or even completely asymptomatic, a concern that intensifies with new pandemics. In such scenarios, the population typically lacks immunity, and the peak infectious viral load may precede the onset of symptoms (6). For example, early models from the COVID-19 pandemic indicated that up to 44% of transmissions occurred in the presymptomatic stage, 1 to 2 days before symptoms appeared (10). This underscores the crucial role of early detection, even by a single day, in controlling transmission.
Currently, the most prevalent tools for early detection are rapid home tests, first deployed on a large scale during the COVID-19 pandemic (50). While these tests are essential, they present several logistical challenges ranging from cost to the need for mass, repeated testing (50, 51). In addition, the use of rapid tests is often prompted by symptoms, which smartwatch detection can significantly precede. Together, these modes of disease detection could provide a powerful tool for public health practitioners with prompt smartwatch detection leading to timely confirmation by rapid tests.
Our study focuses on the effectiveness of early disease detection via smartwatches in reducing transmission. Another critical aspect of early detection is that it facilitates earlier treatment, which is pivotal in reducing disease complications. For instance, a recent study suggested that treating high-risk individuals with antiviral therapy within 48 h of influenza symptom onset, while maintaining current treatment coverage levels, could prevent 5.5–7.1% of all hospitalizations associated with seasonal influenza (29). Similarly, for COVID-19, Paxlovid has been found to have optimal effectiveness when administered within 5 days of symptom onset, highlighting the importance of prompt treatment to improve outcomes (52).
Smartwatch detection could help reduce the spread of other communicable diseases. For example, HIV is a relevant disease to explore in the context of our work, given its viral load patterns and their impact on transmission. The infectious viral load of HIV typically peaks between 3 and 6 weeks after exposure, often coinciding with flu-like symptoms in some patients (53). During this relatively short period, concentrations of HIV in blood and semen are the highest and transmission risk is therefore the greatest (54, 55). Smartwatch detection of HIV through changes in physiological measures could have a profound impact, at both the individual and population levels, by facilitating earlier treatment that reduces an individual's viral load. This proactive approach could markedly reduce the spread of HIV.
Our analysis has several limitations. Because data regarding factors such as pathogen transmissibility, time of detection, and degree of reduction in social contacts are uncertain, we performed stochastic sensitivity analyses and reported ranges and CIs. Further knowledge of uncertain parameters would allow us to make more accurate predictions about the degree of reduction in disease transmission with smartwatch detection. Additionally, we assumed that smartwatches would be 88 to 94% sensitive in detecting physiological changes associated with infection. However, smartwatches may also lead to false alerts (13). In this context, it is uncertain to what extent individuals will reduce their contacts following smartwatch alerts. The degree of reduction in social contacts likely depends on factors such as public perception, which may be affected by how contagious or virulent the disease is. While it is not clear what level of reduction in contacts will occur following smartwatch alerts, the level of reduction we considered was based on observations from prospective studies which indicated that individuals tend to reduce their contacts after initially experiencing symptoms and even more so after undergoing a diagnostic test (9, 40). Detection by a smartwatch might be perceived by individuals either as a sign of initial symptoms or as a form of diagnostic confirmation, leading us to believe that our assumptions are reasonable.
While recent advances in smartwatch detection capabilities offer a promising direction for identifying infections, the challenge of improving diagnostic accuracy remains. In the context of pandemics, it is possible to enhance detection using a multilayered approach that also accounts for the spatiotemporal dynamics of disease (e.g. (56)). For example, in areas and times when influenza is circulating, an anomaly in physiological measures from the smartwatch is more likely to be linked to influenza than other diseases. Smartwatch alerts can also trigger rapid tests for the most likely causes. For instance, one study found that participants underwent COVID-19, adenovirus, and influenza tests following smartwatch alerts (57). Regardless of the specific pathogen causing the infection, an early alert will raise awareness and likely lead to a reduction in an individual's contacts with others, thereby decreasing transmission. Providing individuals with more accurate and targeted information empowers them to make informed decisions to reduce their contacts early in the onset of infection, which helps curb transmission. As detection becomes more precise, compliance is likely to increase, reducing the need for more stringent interventions such as lockdowns.
Though early infection detection is pivotal for decreasing disease transmission, this cannot be achieved without substantial population-level compliance with contact reduction, self-isolation, or quarantine. At the onset of a large-scale disease outbreak or pandemic, people generally have a favorable view of public health recommendations such as mass testing and self-isolation. Despite this favorable view and likely willingness to comply, there may be obstacles to compliance such as financial and mental health concerns (48, 58). However, these concerns can be alleviated through public health education campaigns and provision of adequate incentives (48, 59, 60). For example, a 2020 survey study in Israel found that providing compensation for lost wages would increase compliance with self-isolation from 57 to 94% (48).
Our study suggests that smartwatch-based detection could substantially aid in controlling infectious diseases. Although our study focused on two viral respiratory pathogens, our modeling framework and results are applicable to other communicable diseases. Our findings point to a potential revolution in the management of seasonal diseases and future pandemics through the utilization of smartwatch-based detection systems. This integration not only promises to enhance disease surveillance but also supports proactive health management on a global scale, advocating for systematic inclusion of smartwatch technology in the fight against epidemics.
Supplementary Material
Supplementary material is available at PNAS Nexus online.
Funding
This work was supported by the European Research Council, project #949850, and a Koret Foundation gift for Smart Cities and Digital Living. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Author Contributions
D.Y. was involved in conceptualization, supervision, funding acquisition, validation, methodology, writing—original draft, and writing—review and editing. M.V. was involved in data curation, software, formal analysis, and writing—review and editing. M.N.-M. was involved in conceptualization, data curation, validation, methodology, and writing—review and editing. M.L.B. was involved in supervision, funding acquisition, and writing—review and editing.
Data Availability
There are no primary data underlying this work, and all data are publicly available from the references cited in this study. For complete transparency in our analysis, the model code and all associated data are publicly available at https://github.com/MartVesinurm/TerminatingPandemicsWithSmartwatches.
References
Author notes
Dan Yamin and Margaret L Brandeau contributed equally to this work.
Competing Interest: The authors declare no competing interests.