Search

BACKGROUND Hypoglycemia is a serious complication in diabetes, it impairs cognitive and psychomotor function, and is linked to driving mishaps. In-vehicle voice assistants (VAs) have been designed to proactively deliver a warning of hypoglycemia while driving. However, proactive VAs can cause driving impairments through startling. Hence, we complement the voice warning from the VA with ambient light-emitting diodes (LED) and investigate the effect of this addition on emotional reaction. OBJECTIVE Designing an in-vehicle voice warning for hypoglycemia and assessing its effect on the emotional reaction and technology acceptance. METHODS We present two studies investigating the emotional reactions of drivers with diabetes to different hypoglycemia warning modalities. The same procedure was replicated in two settings: simulated and real driving. A quasi-experimental design, with two independent variables (blood glucose phase and warning modality) and one main dependent variable (emotional reaction), was implemented. The material and apparatus included intravenous catheters to manipulate blood glucose and a tablet with an app to simulate hypoglycemia warnings. The warnings had three possible modalities: Standard, Voice, and Voice + LED. Objective emotional reaction (arousal) was measured physiologically via skin conductance response (SCR). Subjective emotional reaction was measured with the Affective Slider (valence and arousal). Both emotional reaction measures were tested with a mixed-effect linear model. Secondary outcomes included self-reported measures of technology acceptance. RESULTS Our results showed that in the simulated-driving setting, the Voice + LED warning modality was preferred over other modalities. However, this advantage decreased in real-world driving because the LED was less visible outdoors. The Voice modality was more effective than the Standard warning modality in both simulated and real-world driving studies. The mixed model on self-reported emotional reaction yielded significant results. Self-reported arousal was higher during Decreasing blood glucose and Hypoglycemia for Voice and Voice + LED in the real-world driving study. In contrast, self-reported valence was influenced by blood glucose manipulation rather than the warning modality. The mixed model on SCR did not yield significant results. Participants consistently ranked the Voice and Voice + LED warning modalities superior to the Standard modality. CONCLUSIONS This paper proposes using the in-vehicle VA and ambient lighting system installed in a car to deliver a warning of hypoglycemia. It investigated how the warning modality affects emotional response and acceptance in virtual and real-world environments. The study provides insight into the potential of implementing VA-based health warnings in cars and contributes to a better understanding of the challenges and opportunities of in-vehicle VA-based health interventions.

BACKGROUND Hypoglycemia is a frequent and acute complication in type-1 diabetes mellitus (T1DM) and is associated with a higher risk of car mishaps. Currently, hypoglycemia can be detected and signaled through flash glucose monitoring (FGM) or continuous glucose monitoring (CGM) devices. These devices, however, require manual and visual interaction, removing the focus of attention from the driving task. Hypoglycemia is known to cause a decrease in attention, challenging the safety of using such devices behind the wheel. Here, we present an investigation of hands- and distraction-free technology: an in–vehicle voice–assistant–based warning. OBJECTIVE Designing and assessing a voice–assistant–based health warning for hypoglycemia while driving and addressing the limitations of the current warning solutions. METHODS We developed and assessed the warning in three studies, where participants received the warning while driving. In all studies, we measured participants’ self-reported technology readiness, perception of the warning, and compliance behavior (whether they stopped the car and their reaction time), and assessed any room for improvement through participants’ feedback. In Study 0, 10 healthy participants drove in a simulator and assessed the feasibility of using a voice assistant to deliver a warning. In Study 1, 18 participants with T1DM drove in a simulator and assessed the revised version of the warning. In Study 2, 20 participants with T1DM undergoing hypoglycemia assessed a further revised version of the warning while driving in a real car on a test track. In all studies, we also measured self-reported technology readiness, and acceptance of the warning, and assessed compliance behavior and reaction time in response to the warning. RESULTS In all studies, 100% of participants complied with the warning. In Study 0, healthy participants perceived the warning as usable and useful, and their feedback suggested reducing speech rate and increasing driver-assistant interaction. In Study 1, participants with T1DM reported good perception and their feedback suggested the warning to be less instructive. In Study 2, we observed moderate perception (lower than in Study 1), and participants’ feedback revealed the warning was too overloading. CONCLUSIONS To the best of the authors’ knowledge, this is the first study investigating the feasibility of an in–vehicle voice–assistant–based warning for hypoglycemia while driving. Drivers find such an implementation useful and effective, although individuals with T1DM preferred a simple and direct voice warning, rather than a conversational one. This may reflect the utility and unfamiliarity of proactive behavior in voice assistants. We anticipate this research to be a starting point for the combination of driver-state warnings and for voice–assistant–based health support, and to be a guide for the design of such a combination. CLINICALTRIAL ClinicalTrials.gov NCT04035993, NCT04569630

Aim: To develop a non-invasive machine learning (ML) approach to detect hypoglycemia during real car driving based on driving (CAN) , and eye and head motion (EHM) data. Methods: We logged CAN and EHM data in 21 subjects with type 1 diabetes (18 male, 41 ± yrs, A1c 6.8 ± 0.7 % [51 ± 7 mmol/mol]) during driving in eu- (EU) and hypoglycemia (< 3.0 mmol/L, HYPO) . Participants drove in a car (Volkswagen Touran) supervised by a driving instructor on a closed test-track. Using CAN and EHM data, we built ML models to predict the probability of the driver being in HYPO. To make our approach applicable to different generations of cars, we present 3 ML models: first, a model combining CAN+EHM, representing the modern car with integrated camera. Second, a CAN model using driving data only, since modern cars are not generally equipped with EHM tracking. Third, anticipating that autonomous driving will limit the role of CAN data in the future, we tested a model solely based on EHM. Results: Mean BG in EU and HYPO was 6.3 ± 0.8 mmol/L and 2.5 ± 0.5 mmol/L (p< 0.001) , respectively. The model CAN+EHM achieved an area under the receiver operating characteristic curve of 0.88 ± 0.05, sensitivity of 0.70 ± 0.30, and specificity of 0.83 ± 0.in detecting HYPO. Further results are in Fig. 1. Conclusion: We propose ML-based approaches to non-invasively detect HYPO from driver behavior, applicable to contemporary cars and anticipating developments in automotive technology. Disclosure V.Lehmann: None. E.Fleisch: None. T.Kowatsch: Advisory Panel; Pathmate Technologies AG, Research Support; CSS Insurance, Stock/Shareholder; Pathmate Technologies AG. S.N.Lagger: None. M.Laimer: None. F.Wortmann: None. C.Stettler: None. T.Zueger: None. M.Maritsch: None. M.Notter: None. S.Schallmoser: None. C.Bérubé: None. C.Albrecht: None. M.Kraus: None. S.Feuerriegel: None. Funding Swiss National Science Foundation (SNF CRSII5_183569) , Swiss Diabetes Foundation, Diabetes Center Berne, Automobile Club Switzerland (ACS) , Federal Department of Defence, Civil Protection and Sport (DDPS) and Department of Research of the University Hospital Berne

BACKGROUND Noncommunicable diseases (NCDs) constitute a burden on public health. These are best controlled through self-management practices, such as self-information. Fostering patients’ access to health-related information through efficient and accessible channels, such as commercial voice assistants (VAs), may support the patients’ ability to make health-related decisions and manage their chronic conditions. OBJECTIVE This study aims to evaluate the reliability of the most common VAs (ie, Amazon Alexa, Apple Siri, and Google Assistant) in responding to questions about management of the main NCD. METHODS We generated health-related questions based on frequently asked questions from health organization, government, medical nonprofit, and other recognized health-related websites about conditions associated with Alzheimer’s disease (AD), lung cancer (LCA), chronic obstructive pulmonary disease, diabetes mellitus (DM), cardiovascular disease, chronic kidney disease (CKD), and cerebrovascular accident (CVA). We then validated them with practicing medical specialists, selecting the 10 most frequent ones. Given the low average frequency of the AD-related questions, we excluded such questions. This resulted in a pool of 60 questions. We submitted the selected questions to VAs in a 3×3×6 fractional factorial design experiment with 3 developers (ie, Amazon, Apple, and Google), 3 modalities (ie, voice only, voice and display, display only), and 6 diseases. We assessed the rate of error-free voice responses and classified the web sources based on previous research (ie, expert, commercial, crowdsourced, or not stated). RESULTS Google showed the highest total response rate, followed by Amazon and Apple. Moreover, although Amazon and Apple showed a comparable response rate in both voice-and-display and voice-only modalities, Google showed a slightly higher response rate in voice only. The same pattern was observed for the rate of expert sources. When considering the response and expert source rate across diseases, we observed that although Google remained comparable, with a slight advantage for LCA and CKD, both Amazon and Apple showed the highest response rate for LCA. However, both Google and Apple showed most often expert sources for CVA, while Amazon did so for DM. CONCLUSIONS Google showed the highest response rate and the highest rate of expert sources, leading to the conclusion that Google Assistant would be the most reliable tool in responding to questions about NCD management. However, the rate of expert sources differed across diseases. We urge health organizations to collaborate with Google, Amazon, and Apple to allow their VAs to consistently provide reliable answers to health-related questions on NCD management across the different diseases.

Aim: To non-invasively detect hypoglycemia in individuals with type 1 diabetes (T1D) based on gaze behavior while driving. Methods: Controlled hypoglycemia was induced in 19 individuals (12 males, age 32 ± 7.1 yrs) with T1D (HbA1c 7.1 ± 0.6% [54 ± 6 mmol/mol]) using an adapted hypoglycemic clamp protocol. Gaze and blood glucose (BG) data were gathered while driving in a simulator during three 18 min sessions: session 1 (BG 90-144 mg/dL), session 2 (BG declining from 72 to 45 mg/dL), and session 3 (BG 36-45 mg/dL). A gradient-boosting machine learning (ML) model was built for hypoglycemia (BG Results: Mean venous BG was 105.4 ± 11.4 mg/dL during session 1, declined from 61.4 ± 6.1 mg/dL to 47.2 ± 8.5 mg/dL during session 2, and was 42.7 ± 4.1 mg/dL during session 3, respectively. Gaze analysis provided 29,968 data samples (1,577.5 ± 52 per subject, 10,041 euglycemia, 19,927 hypoglycemia). Overall, ML achieved an area under the receiver-operating-characteristics curve of 0.83 ± 0.09 for hypoglycemia detection with leave-one-subject-out cross-validation. Conclusion: ML-based gaze analysis shows high accuracy in non-invasive hypoglycemia detection while driving. Our approach offers promising potential in various settings where cameras are available. Disclosure V. Lehmann: None. E. Fleisch: Research Support; Self; CSS Health Insurance, Switzerland, Stock/Shareholder; Self; Pathmate-Technologies AG, Switzerland. F. Wortmann: None. C. Stettler: None. M. Maritsch: None. T. Zueger: None. A. Marxer: None. C. Berube: None. M. Kraus: None. C. Albrecht: None. S. Feuerriegel: None. T. Kowatsch: Advisory Panel; Self; Pathmate Technologies AG, Switzerland and Germany, Stock/Shareholder; Self; Pathmate Technologies AG, Switzerland and Germany. Funding Swiss National Science Foundation (SNF CRSII5_183569); Insel Group; Diabetes Center Berne

Background: Chronic and mental health conditions are increasingly prevalent worldwide. As devices in our everyday lives offer more and more voice-based self-service, voice-based conversational agents (VCAs) have the potential to support the prevention and management of these conditions in a scalable manner. However, evidence on VCAs dedicated to the prevention and management of chronic and mental health conditions is unclear. Objective: This study provides a better understanding of the current methods used in the evaluation of health interventions for the prevention and management of chronic and mental health conditions delivered through VCAs. Methods: We conducted a systematic literature review using PubMed MEDLINE, Embase, PsycINFO, Scopus, and Web of Science databases. We included primary research involving the prevention or management of chronic or mental health conditions through a VCA and reporting an empirical evaluation of the system either in terms of system accuracy, technology acceptance, or both. A total of 2 independent reviewers conducted the screening and data extraction, and agreement between them was measured using Cohen kappa. A narrative approach was used to synthesize the selected records. Results: Of 7170 prescreened papers, 12 met the inclusion criteria. All studies were nonexperimental. The VCAs provided behavioral support (n=5), health monitoring services (n=3), or both (n=4). The interventions were delivered via smartphones (n=5), tablets (n=2), or smart speakers (n=3). In 2 cases, no device was specified. A total of 3 VCAs targeted cancer, whereas 2 VCAs targeted diabetes and heart failure. The other VCAs targeted hearing impairment, asthma, Parkinson disease, dementia, autism, intellectual disability, and depression. The majority of the studies (n=7) assessed technology acceptance, but only few studies (n=3) used validated instruments. Half of the studies (n=6) reported either performance measures on speech recognition or on the ability of VCAs to respond to health-related queries. Only a minority of the studies (n=2) reported behavioral measures or a measure of attitudes toward intervention-targeted health behavior. Moreover, only a minority of studies (n=4) reported controlling for participants’ previous experience with technology. Finally, risk bias varied markedly. Conclusions: The heterogeneity in the methods, the limited number of studies identified, and the high risk of bias show that research on VCAs for chronic and mental health conditions is still in its infancy. Although the results of system accuracy and technology acceptance are encouraging, there is still a need to establish more conclusive evidence on the efficacy of VCAs for the prevention and management of chronic and mental health conditions, both in absolute terms and in comparison with standard health care., Journal of Medical Internet Research, 23 (3), ISSN:1438-8871

Background: Noncommunicable diseases (NCDs) constitute a burden on public health. These are best controlled through self-management practices, such as self-information. Fostering patients’ access to health-related information through efficient and accessible channels, such as commercial voice assistants (VAs), may support the patients’ ability to make health-related decisions and manage their chronic conditions. Objective: This study aims to evaluate the reliability of the most common VAs (ie, Amazon Alexa, Apple Siri, and Google Assistant) in responding to questions about management of the main NCD. Methods: We generated health-related questions based on frequently asked questions from health organization, government, medical nonprofit, and other recognized health-related websites about conditions associated with Alzheimer’s disease (AD), lung cancer (LCA), chronic obstructive pulmonary disease, diabetes mellitus (DM), cardiovascular disease, chronic kidney disease (CKD), and cerebrovascular accident (CVA). We then validated them with practicing medical specialists, selecting the 10 most frequent ones. Given the low average frequency of the AD-related questions, we excluded such questions. This resulted in a pool of 60 questions. We submitted the selected questions to VAs in a 3×3×6 fractional factorial design experiment with 3 developers (ie, Amazon, Apple, and Google), 3 modalities (ie, voice only, voice and display, display only), and 6 diseases. We assessed the rate of error-free voice responses and classified the web sources based on previous research (ie, expert, commercial, crowdsourced, or not stated). Results: Google showed the highest total response rate, followed by Amazon and Apple. Moreover, although Amazon and Apple showed a comparable response rate in both voice-and-display and voice-only modalities, Google showed a slightly higher response rate in voice only. The same pattern was observed for the rate of expert sources. When considering the response and expert source rate across diseases, we observed that although Google remained comparable, with a slight advantage for LCA and CKD, both Amazon and Apple showed the highest response rate for LCA. However, both Google and Apple showed most often expert sources for CVA, while Amazon did so for DM. Conclusions: Google showed the highest response rate and the highest rate of expert sources, leading to the conclusion that Google Assistant would be the most reliable tool in responding to questions about NCD management. However, the rate of expert sources differed across diseases. We urge health organizations to collaborate with Google, Amazon, and Apple to allow their VAs to consistently provide reliable answers to health-related questions on NCD management across the different diseases., Journal of Medical Internet Research, 23 (12), ISSN:1438-8871

BACKGROUND Chronic and mental health conditions are increasingly prevalent worldwide. As devices in our everyday lives offer more and more voice-based self-service, voice-based conversational agents (VCAs) have the potential to support the prevention and management of these conditions in a scalable manner. However, evidence on VCAs dedicated to the prevention and management of chronic and mental health conditions is unclear. OBJECTIVE This study provides a better understanding of the current methods used in the evaluation of health interventions for the prevention and management of chronic and mental health conditions delivered through VCAs. METHODS We conducted a systematic literature review using PubMed MEDLINE, Embase, PsycINFO, Scopus, and Web of Science databases. We included primary research involving the prevention or management of chronic or mental health conditions through a VCA and reporting an empirical evaluation of the system either in terms of system accuracy, technology acceptance, or both. A total of 2 independent reviewers conducted the screening and data extraction, and agreement between them was measured using Cohen kappa. A narrative approach was used to synthesize the selected records. RESULTS Of 7170 prescreened papers, 12 met the inclusion criteria. All studies were nonexperimental. The VCAs provided behavioral support (n=5), health monitoring services (n=3), or both (n=4). The interventions were delivered via smartphones (n=5), tablets (n=2), or smart speakers (n=3). In 2 cases, no device was specified. A total of 3 VCAs targeted cancer, whereas 2 VCAs targeted diabetes and heart failure. The other VCAs targeted hearing impairment, asthma, Parkinson disease, dementia, autism, intellectual disability, and depression. The majority of the studies (n=7) assessed technology acceptance, but only few studies (n=3) used validated instruments. Half of the studies (n=6) reported either performance measures on speech recognition or on the ability of VCAs to respond to health-related queries. Only a minority of the studies (n=2) reported behavioral measures or a measure of attitudes toward intervention-targeted health behavior. Moreover, only a minority of studies (n=4) reported controlling for participants’ previous experience with technology. Finally, risk bias varied markedly. CONCLUSIONS The heterogeneity in the methods, the limited number of studies identified, and the high risk of bias show that research on VCAs for chronic and mental health conditions is still in its infancy. Although the results of system accuracy and technology acceptance are encouraging, there is still a need to establish more conclusive evidence on the efficacy of VCAs for the prevention and management of chronic and mental health conditions, both in absolute terms and in comparison with standard health care. CLINICALTRIAL

CHI EA '20: Extended Abstracts of the 2020 CHI Conference on Human Factors in Computing Systems, ISBN:978-1-4503-6819-3

This paper proposes an approach to detect information relevance during decision-making from eye movements in order to enable user interface adaptation. This is a challenging task because gaze behavior varies greatly across individual users and tasks and groundtruth data is difficult to obtain. Thus, prior work has mostly focused on simpler target-search tasks or on establishing general interest, where gaze behavior is less complex. From the literature, we identify six metrics that capture different aspects of the gaze behavior during decision-making and combine them in a voting scheme. We empirically show, that this accounts for the large variations in gaze behavior and out-performs standalone metrics. Importantly, it offers an intuitive way to control the amount of detected information, which is crucial for different UI adaptation schemes to succeed. We show the applicability of our approach by developing a room-search application that changes the visual saliency of content detected as relevant. In an empirical study, we show that it detects up to 97% of relevant elements with respect to user self-reporting, which allows us to meaningfully adapt the interface, as confirmed by participants. Our approach is fast, does not need any explicit user input and can be applied independent of task and user., Comment: The first two authors contributed equally to this work