The hopes and hazards of using personal health technologies in the diagnosis and prognosis of infections

Abstract

Work from numerous groups has shown the potential of using data from wearable devices to characterise each individual's unique baseline, identify deviations from that baseline suggestive of a viral infection, and to aggregate that data to better inform population surveillance trends. With a growing global population of connected wearable users, this could potentially help improve the earlier diagnosis and management of infectious individuals and improve timeliness and precision of tracking infectious disease outbreaks. However, despite these possibilities, there are important considerations when interpreting wearable data, including generalisability of user populations, sensor accuracy and comparability, and overfitting of models. Additionally, before deploying such tools as a global health solution, the proactive integration of community engagement through a user-centred design is, at a minimum, required to begin mitigating the risk of digital health illiteracy, structural inequality, and social marginality. Regional frameworks establishing transparent standards for participant data protection as data privacy, data use, and data rights are required to truly protect participant's rights, empower participants, and minimise risk of harm. As these sensors continue to evolve, standardised data and reporting, as well as collaboration and data sharing across study and technology groups will also be necessary. Personal health technologies—technologies (consumer or medical grade) used by an individual that generate data from that individual—offer an unprecedented opportunity to completely alter how we currently detect and manage infectious diseases at an individual as well as population level. Currently, management decisions in the setting of possible infection are made based on symptoms and objective physiological measurements, whether the person primarily making those decisions is the patient or a health-care provider. Because symptoms are often non-specific, the recognition of abnormalities in physiological parameters have an important role in diagnosing and guiding therapeutic interventions. However, abnormality has historically been defined by what is normal for a healthy population, rather than what is normal for an individual. For example, an elevated temperature, which has been recognised as a hallmark of infection since the beginning of recorded history, is commonly (but not consistently) classified as a temperature of 38°C or higher.1Atkins E Fever: its history, cause, and function.Yale J Biol Med. 1982; 55: 283-289PubMed Google Scholar Similarly, a respiratory rate of more than 20 breaths per min or a heart rate of more than 100 beats per min in an adult would also be considered abnormal and might indicate a person who requires expedited, or a higher level of care. However, what is actually a person's "normal" varies substantially between individuals. For oral temperature, that range falls between 35·7°C and 37·4°C,2Geneva II Cuzzo B Fazili T Javaid W Normal body temperature: a systematic review.Open Forum Infect Dis. 2019; 6ofz032Crossref PubMed Scopus (49) Google Scholar respiratory rate between 12 and 20 breaths per min,3Chourpiliadis C Bhardwaj A Physiology. Respiratory Rate.https://www.ncbi.nlm.nih.gov/books/NBK537306/Date: 2019Date accessed: May 6, 2021Google Scholar and resting pulse rate between 40 and 109 beats per min.4Quer G Gouda P Galarnyk M Topol EJ Steinhubl SR Inter- and intraindividual variability in daily resting heart rate and its associations with age, sex, sleep, BMI, and time of year: retrospective, longitudinal cohort study of 92,457 adults.PLoS One. 2020; 15e0227709Crossref PubMed Scopus (22) Google Scholar Therefore, these broad population ranges are imprecise when applied to a specific individual, especially for identifying an early change in their physiological status. Individuals experience daily, weekly, and seasonal fluctuations unique to them in a range of physiological parameters and activities.4Quer G Gouda P Galarnyk M Topol EJ Steinhubl SR Inter- and intraindividual variability in daily resting heart rate and its associations with age, sex, sleep, BMI, and time of year: retrospective, longitudinal cohort study of 92,457 adults.PLoS One. 2020; 15e0227709Crossref PubMed Scopus (22) Google Scholar, 5Jaiswal SJ Quer G Galarnyk M Steinhubl SR Topol EJ Owens RL Association of sleep duration and variability with body mass index: sleep measurements in a large US population of wearable sensor users.JAMA Intern Med. 2020; 1801694Crossref PubMed Scopus (5) Google Scholar Only by knowing what is normal for an individual when they are well is it possible to identify the earliest possible deviations from that normal. This is what personal health technologies make possible, in the real world outside of a health-care setting, and in a nearly passive manner. Because of the emergence of COVID-19, numerous studies examining personal sensor data have found digital signals supporting the potential benefit of using these technologies to identify and track viral illness. Continuously evolving personal health technologies can be harnessed to collect unique baseline data on individuals and populations, which allows for earlier and more precise detection of viral illness; but for these tools to be successful, biases related to health inequity, loss to follow-up, an absence of data harmonisation, and others need to be addressed. It is relatively recent that personal health technologies have become available that enable investigators to even consider addressing the potential value of identifying individual changes in physiological parameters in large populations. Wearable sensors available to individuals can now track not only activity, but also body position, heart rate and rhythm, skin temperature, oxygen saturation, an electrocardiogram, and electrodermal activity and sound (figure 1). From these metrics respiratory rate, heart rate variability, heart arrhythmias, sleep, and sleep stages are derived. Other medical-grade wearables, not yet designed for a large consumer market, can also continuously track blood pressure as well as derive systemic vascular resistance and cardiac output.6Nachman D Gepner Y Goldstein N et al.Comparing blood pressure measurements between a photoplethysmography-based and a standard cuff-based manometry device.Sci Rep. 2020; 1016116Crossref PubMed Scopus (13) Google Scholar Wearable sensor technologies are available in a wide range of form factors beyond the most commonly used wrist-worn sensors. There are options for rings, arm bands, earbuds, adhesive patches, and clothing that all are capable of tracking multiple physiological parameters. Each offers certain advantages and disadvantages, but the most effective will be the one that provides the most reliable information that the intended user will be willing and able to use as frequently and for as long as needed. There are little data available on long-term use of any of these wearables, with one survey7Macridis S Johnston N Johnson S Vallance JK Consumer physical activity tracking device ownership and use among a population-based sample of adults.PLoS One. 2018; 13e0189298Crossref PubMed Scopus (25) Google Scholar in Canada finding that of people who purchased their own activity tracker, just 55% were still using it, but those that were still wearing it wore it an average of 23 days in the previous month. In another study8Galarnyk M Quer G McLaughlin K Ariniello L Steinhubl SR Usability of a wrist-worn smartwatch in a direct-to-participant randomized pragmatic clinical trial.Digit Biomark. 2019; 3: 176-184Crossref PubMed Scopus (2) Google Scholar that gave participants a wrist wearable, approximately 25% wore the sensor for the majority of the 4-month requested monitoring period.8Galarnyk M Quer G McLaughlin K Ariniello L Steinhubl SR Usability of a wrist-worn smartwatch in a direct-to-participant randomized pragmatic clinical trial.Digit Biomark. 2019; 3: 176-184Crossref PubMed Scopus (2) Google Scholar A study of nearly 4·5 million insurance plan members who were offered financial incentives for achieving activity goals found that only 1·2% activated a device, but of those that did, 80% were still using the device at 6 months.9Patel MS Foschini L Kurtzman GW et al.Using wearable devices and smartphones to track physical activity: initial activation, sustained use, and step counts across sociodemographic characteristics in a national sample.Ann Intern Med. 2017; 167: 755-757Crossref PubMed Scopus (39) Google Scholar A meta-analysis10Pratap A Neto EC Snyder P et al.Indicators of retention in remote digital health studies: a cross-study evaluation of 100,000 participants.NPJ Digit Med. 2020; 3: 21Crossref PubMed Scopus (41) Google Scholar found that median participant retention across eight studies was only 5·5 days, and that most studies did not include a population representative of the ethnicities and diversity of the USA. For many, a non-wearable, passive multiparametric sensor can provide the best option for longitudinal data. For example, under-mattress pads that can monitor heart rate, respiratory rate, and various sleep parameters have been found in preliminary work to identify early physiological decompensation.11Bennett MK Shao M Gorodeski EZ Home monitoring of heart failure patients at risk for hospital readmission using a novel under-the-mattress piezoelectric sensor: a preliminary single centre experience.J Telemed Telecare. 2017; 23: 60-67Crossref PubMed Scopus (7) Google Scholar A wide range of contactless sensors for in-home use that use computer-vision, infrared thermography, radar, and audio can track individual vital signs, as well as cough frequency and quality, in some cases, even when in different rooms.12Adib F Mao H Kabelac Z Katabi D Miller RC Smart homes that monitor breathing and heart rate.in: Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems. Association for Computing Machinery, Seoul, Republic of Korea2015: 837-846Crossref Scopus (366) Google Scholar, 13Negishi T Abe S Matsui T et al.Contactless vital signs measurement system using RGB-thermal image sensors and its clinical screening test on patients with seasonal influenza.Sensors. 2020; 20E2171Crossref PubMed Scopus (20) Google Scholar, 14Antink CH Lyra S Paul M Yu X Leonhardt S A broader look: camera-based vital sign estimation across the spectrum.Yearb Med Inform. 2019; 28: 102-114Crossref PubMed Scopus (22) Google Scholar, 15Hall JI Lozano M Estrada-Petrocelli L Birring S Turner R The present and future of cough counting tools.J Thorac Dis. 2020; 12: 5207-5223Crossref PubMed Scopus (5) Google Scholar As remarkable as progress has been over the last several years in the availability of consumer technologies that monitor and report physiological variables, it is important to recognise that the accuracy of the information provided is variable and should not be considered clinically dependable unless appropriate validation evidence exists, and ideally regulatory agency approval.16Manta C Jain SS Coravos A Mendelsohn D Izmailova ES An evaluation of biometric monitoring technologies for vital signs in the era of COVID-19.Clin Transl Sci. 2020; 13: 1034-1044Crossref PubMed Scopus (13) Google Scholar For example, there have been recent concerns raised of the accuracy of oxygen saturation determined not only from the wrist with consumer sensors,17Bent B Goldstein BA Kibbe WA Dunn JP Investigating sources of inaccuracy in wearable optical heart rate sensors.NPJ Digit Med. 2020; 3: 18Crossref PubMed Scopus (74) Google Scholar but also from hospital-based finger-tip devices due to skin pigmentation differences.18Perry TS Should you trust Apple's new blood oxygen sensor? View from the valley. IEEE Spectrum.https://spectrum.ieee.org/view-from-the-valley/biomedical/devices/should-you-trust-apples-new-blood-oxygen-sensorDate accessed: May 6, 2021Google Scholar, 19Sjoding MW Dickson RP Iwashyna TJ Gay SE Valley TS Racial bias in pulse oximetry measurement.N Engl J Med. 2020; 383: 2477-2478Crossref PubMed Scopus (69) Google Scholar To aggregate findings from multiple sensors it will also be important to understand how sensor-specific and usually proprietary algorithms for calculating metrics such as daily resting heart rate and respiration rate can differ across devices. There is typically a delay from the time someone gets sick to when they develop symptoms, seek care, get tested, and finally receive a test result for COVID-19 and other viral infections. It then takes an additional 1–3 weeks before test results are collected and aggregated into a central surveillance system, which often relies on outdated reporting methods such as fax machines.20Kliff S Sanger-Katz M Bottleneck for U.S. Coronavirus response: the fax machine. The New York Times, July 13, 2020https://www.nytimes.com/2020/07/13/upshot/coronavirus-response-fax-machines.htmlDate accessed: May 6, 2021Google Scholar When wearable data is aggregated for a population, it is possible to so called nowcast or track viral activity in real time. Previous studies have shown that identifying resting heart rate and sleep data outside of an individual's normal levels, can be used to improve real-time predictions for influenza-like illness at the state level.21Quer G Radin JM Gadaleta M et al.Wearable sensor data and self-reported symptoms for COVID-19 detection.Nat Med. 2021; 27: 73-77Crossref PubMed Scopus (61) Google Scholar Similarly, models to predict COVID-19 anomalies in China and south-central Europe using wearable data from 1·3 million users who wore Huami smartwatches (Hefei, China) also showed promise (table 1).32Zhu G Li J Meng Z et al.Learning from large-scale wearable device data for predicting epidemics trend of COVID-19.Discrete Dyn Nat Soc. 2020; 20206152041Crossref Scopus (28) Google Scholar The Robert Koch institute, Berlin, Germany, has launched a similar fever trend tracker based on resting heart rate and activity data from more than 500 000 participants.31Corona DatenspendeNews.https://corona-datenspende.de/science/en/Date: Dec 27, 2020Date accessed: December 27, 2020Google Scholar Kinsa smart thermometers (San Francisco, CA, USA) have also shown utility in predicting influenza-like illness activity29Miller AC Singh I Koehler E Polgreen PM A smartphone-driven thermometer application for real-time population- and individual-level influenza surveillance.Clin Infect Dis. 2018; 67: 388-397Crossref PubMed Scopus (19) Google Scholar, 33Ackley SF Pilewski S Petrovic VS Worden L Murray E Porco TC Assessing the utility of a smart thermometer and mobile application as a surveillance tool for influenza and influenza-like illness.Health Informatics J. 2020; 26: 2148-2158Crossref PubMed Scopus (7) Google Scholar, 34Miller AC Peterson RA Singh I Pilewski S Polgreen PM Improving state-level influenza surveillance by incorporating real-time smartphone-connected thermometer readings across different geographic domains.Open Forum Infect Dis. 2019; 6ofz455Crossref Google Scholar and potentially COVID-19.30Chamberlain SD Singh I Ariza CA Daitch AL Philips PB Dalziel BD Real-time detection of COVID-19 epicenters within the United States using a network of smart thermometers.medRxiv. 2020; (published online April 10.) (preprint).https://doi.org/10.1101/2020.04.06.20039909Google Scholar Novel data streams from sensors can offer key insight into trends, timing of outbreaks, and identifying specific geographical hotspots of infection. They might prove especially useful when integrated with both traditional clinical and laboratory surveillance and other novel surveillance data, including those from wastewater, internet search terms, social media, and mobility data (figure 2).Table 1Observational studies of wearables in the prediction of viral illnessesParameters analysedWearable sensors includedStudy populationKey findingTotal participants, NCOVID-19 positive participants, nIndividual-level studiesTemPredict22Smarr BL Aschbacher K Fisher SM et al.Feasibility of continuous fever monitoring using wearable devices.Sci Rep. 2020; 1021640Crossref PubMed Scopus (13) Google ScholarSkin temperature, heart rate, respiratory rate, heart rate variabilityOura ring sensor5050Peripheral temperature elevations can be captured by wearable sensors and correlate with self-reported feverStanford consumer smartwatches23Mishra T Wang M Metwally AA et al.Pre-symptomatic detection of COVID-19 from smartwatch data.Nat Biomed Eng. 2020; 4: 1208-1220Crossref PubMed Scopus (59) Google ScholarHeart rate, sleep, physical activityFitbit, Apple Watch, Garmin, and others52623226 (81%) COVID-19 individuals had changes in their heart rate, steps, or sleep; retrospectively, 20 (63%) of COVID-19 cases could be detected pre-symptoms onset using extreme elevations in resting heart rateFitbit Study24Natarajan A Su HW Heneghan C Assessment of physiological signs associated with COVID-19 measured using wearable devices.NPJ Digit Med. 2020; 3: 156Crossref PubMed Scopus (36) Google ScholarResting heart rate, activity, respiration rate, heart rate variability, sleepFitbit187 5732745 (PCR) and 1117 (serology)Physiological data could predict illness on a specific day with an AUC of 0·77DETECT21Quer G Radin JM Gadaleta M et al.Wearable sensor data and self-reported symptoms for COVID-19 detection.Nat Med. 2021; 27: 73-77Crossref PubMed Scopus (61) Google ScholarResting heart rate, sleep, physical activityData from Fitbits and any sensors connected with HealthKit or GoogleFit30 52954Wearable sensors data can significantly improve symptom only based models to distinguish COVID-19 positive versus negative symptomatic infections (AUC 0·80 [95% CI 0·73–0·86])WHOOP system25Miller DJ Capodilupo JV Lastella M et al.Analyzing changes in respiratory rate to predict the risk of COVID-19 infection.PLoS One. 2020; 15e0243693Crossref PubMed Scopus (18) Google ScholarRespiratory rate, resting heart rate, heart rate variabilityWHOOP; wrist-worn strap27181Model identified two (20%) of COVID-19 positive cases 2 days before symptom onset and eight (80%) positive cases by day 3 of symptomsEvidation26Shapiro A Marinsek N Clay I et al.Characterizing COVID-19 and influenza illnesses in the real world via person-generated health data.Patterns. 2021; 2100188Summary Full Text Full Text PDF PubMed Scopus Google ScholarResting heart rate, physical activity, and sensor data showed similar in daily changes of and heart rate measurements for both influenza and COVID-19 RP M L et of physiological data from a wearable device to identify infection and symptoms and predict COVID-19 Med 2021; PubMed Scopus Google ScholarHeart rate collected heart rate metrics can identify the diagnosis of COVID-19 and related JM Topol EJ Steinhubl SR wearable device data to improve state-level real-time surveillance of influenza-like illness in the a population-based Digit 2020; Full Text Full Text PDF PubMed Scopus Google ScholarResting heart rate and of data significantly for and models of influenza-like AC Singh I Koehler E Polgreen PM A smartphone-driven thermometer application for real-time population- and individual-level influenza surveillance.Clin Infect Dis. 2018; 67: 388-397Crossref PubMed Scopus (19) Google Scholar, SD Singh I Ariza CA Daitch AL Philips PB Dalziel BD Real-time detection of COVID-19 epicenters within the United States using a network of smart thermometers.medRxiv. 2020; (published online April 10.) (preprint).https://doi.org/10.1101/2020.04.06.20039909Google smart anomalies are significantly with COVID-19 counts at the and state and with national influenza-like illness activity in the DatenspendeNews.https://corona-datenspende.de/science/en/Date: Dec 27, 2020Date accessed: December 27, 2020Google ScholarResting heart rate, physical of data might predict fever anomalies in G Li J Meng Z et al.Learning from large-scale wearable device data for predicting epidemics trend of COVID-19.Discrete Dyn Nat Soc. 2020; 20206152041Crossref Scopus (28) Google ScholarResting heart rate, rate with COVID-19 counts in the data not available specific of COVID-19 positive individuals were not to have a or supporting data on their As of in a new the data not available specific of COVID-19 positive individuals were not to have a or supporting data on their diagnosis using continuously collected data from wearables has including COVID-19 especially when are not available at continuously collected heart rate data has shown that an individual's heart rate to beats per min for in P I The between body temperature, heart rate and respiratory rate in Med J. 26: PubMed Scopus (59) Google Scholar, J M and cardiac Intern Med. PubMed Scopus Google Scholar and data from models that changes in heart rate can be detected days before L S T et the using physiological 2017; (published online Scholar have also shown early promise to improve models to symptomatic individuals who have COVID-19 versus those who have other infections. of self-reported symptom data has identified key symptoms of COVID-19 with other with loss of and as the C et tracking of self-reported symptoms to predict potential Med. 2020; 26: PubMed Scopus Google Scholar The of wearable data into these models has shown potential to improve the of these with an the from CI symptoms to 0·80 using also wearable G Radin JM Gadaleta M et al.Wearable sensor data and self-reported symptoms for COVID-19 detection.Nat Med. 2021; 27: 73-77Crossref PubMed Scopus (61) Google Scholar sensor data also has shown promise to identify X Dunn J D et tracking and activity using wearable useful 2017; PubMed Scopus Google Scholar including for individuals who positive for T Wang M Metwally AA et al.Pre-symptomatic detection of COVID-19 from smartwatch data.Nat Biomed Eng. 2020; 4: 1208-1220Crossref PubMed Scopus (59) Google Scholar which would be especially have also at identifying the need for at symptoms A Su HW Heneghan C Assessment of physiological signs associated with COVID-19 measured using wearable devices.NPJ Digit Med. 2020; 3: 156Crossref PubMed Scopus (36) Google Scholar how elevations in temperature correlate with self-reported BL Aschbacher K Fisher SM et al.Feasibility of continuous fever monitoring using wearable devices.Sci Rep. 2020; 1021640Crossref PubMed Scopus (13) Google Scholar and how symptoms and physiological changes are more for COVID-19 positive individuals than for influenza positive A Marinsek N Clay I et al.Characterizing COVID-19 and influenza illnesses in the real world via person-generated health data.Patterns. 2021; 2100188Summary Full Text Full Text PDF PubMed Scopus Google Scholar As new metrics are to substantially is to better understand wearable changes for different and tracking long-term such as with of infection. early signs of can be especially useful for early of care, and individual Additionally, as are monitoring might prove useful for identifying to and potential and improve to track infections and The value of the growing availability of continuous sensor technologies is on the of detection or prediction which should standards before in the B et new reporting to the in clinical Med. 2021; 27: PubMed Scopus Google Scholar to their AL M to the of models in health 2020; PubMed Scopus (59) Google Scholar, Wang S Marinsek N R M Foschini L in for 2019Date accessed: May 6, 2021Google Scholar with early and models in COVID-19 or individuals has several and they are at risk of due to bias on the participant when the participants might not be representative of the (2) bias in the when the are not available at the intended time and can be by the measurement bias in the when it is measured and bias in the due to sample and which can be only with a between and test (table L B et models for diagnosis and of systematic and 2020; PubMed Scopus Google Scholar There are often in reporting important parameters such as the of the or the of COVID-19 and other viral infections such as influenza in the considered population, different models report with different making it more to them for the prediction the studies in the work of and L B et models for diagnosis and of systematic and 2020; PubMed Scopus Google Scholar all to prediction but of the risk of bias due to both reporting and the results could be and not This bias is a which was in the of a prediction for COVID-19 patients that was based on several blood sample and found a accuracy of L J et prediction for COVID-19 2020; Scopus Google Scholar M S S validation clinical utility of the prediction for patients with 2020; (published online Scholar, C E M B S of a COVID-19 specific prediction to the 2020; 3: Scopus (5) Google Scholar, TC et of a prediction in patients with 2020; 3: Scopus Google Scholar, laboratory in patient prediction 2021; 3: 18Crossref Scopus Google Scholar, of an for patients with 2021; 3: Scopus (5) Google Scholar and

Topics & Concepts