The paper analyses the use of voice assistants in people’s homes through the lens of activity theory guided by concepts such as mental models, context, the relationship between subject and tool as well as contradictions. Activity theory sees conversational devices including voice assistants not as ends in themselves, but as tools to aid the performance of a particular activity or practice. After a brief overview of the use of activity theory in HCI research, the empirical data gathered in an interdisciplinary project on the use of voice-enabled technology are discussed with a focus on the contradictions that have emerged in the course of our study. Contradictions have emerged with respect to the traditional view of home as private, which clashes with the fact that voice assistants may transmit personal data to external bodies. Another contradiction relates to the gulf between people’s expectations of smartness and the current shortcomings of commercial voice assistants. Finally, the paper presents suggestions for how one might design conversational technology in a way that helps resolve those contradictions and cope with the emerging phenomena such as autonomous agents.

Voice assistants have become embedded in people's private spaces and domestic lives where they gather enormous amounts of personal information which is why they evoke serious privacy concerns. The paper reports the findings from a mixed-method study with 65 digital natives, their attitudes to privacy and actual and intended behaviour in privacy-sensitive situations and contexts. It also presents their recommendations to governments or organisations with regard to protecting their data. The results show that the majority are concerned about privacy but are willing to disclose personal data if the benefits outweigh the risks. The prevailing attitude is one characterised by uncertainty about what happens with their data, powerlessness about controlling their use, mistrust in big tech companies and uneasiness about the lack of transparency. Few take steps to self-manage their privacy, but rely on the government to take measures at the political and regulatory level. The respondents, however, show scant awareness of existing or planned legislation such as the GDPR and the Digital Services Act, respectively. A few participants are anxious to defend the analogue world and limit digitalization in general which in their opinion only opens the gate to surveillance and misuse.

Compared to parents of healthy children, parents of severely or chronically ill children have significantly worse physical and mental health and a lower quality of life, e.g. because of lack of sleep. The proposed solution aims at assisting caregivers by means of a remote monitoring service run by professional nursing staff which should allow parents to get a good night’s sleep. A smart algorithm has been developed to detect if a particular parameter (heart rate, respiration rate or oxygen saturation) has exceeded a pre-defined threshold and thus may imply an emergency. Parents are only alerted after a professional nurse in the monitoring centre has cross-checked vital parameter trends and carried out an audio-visual inspection. The quality and accuracy of the system has been validated through iterative testing including a test performed in a children’s hospital to ensure that the monitoring system is not inferior to a hospital set-up.

Digitalization is playing an increasingly significant role in practically all areas of society. The domain of healthcare has been moving into the digital world relatively late and only recently has started with digitalizing processes and services on a larger scale. Still, the potential for disruption in the healthcare industry is enormous. Although approaches to healthcare financing and regulatory schemes differ greatly between countries, it is generally recognized that current healthcare systems are characterized by lack of transparency and inefficiencies and that digitalization of healthcare services can lead to improvements in quality, efficiency and accessibility of care. We use the term “digital health” as an umbrella concept which subsumes eHealth, mobile or mHealth, telehealth or telemedicine, among others. Digital health can be defined as “an improvement in the way healthcare provision is conceived and delivered by healthcare providers through the use of information and communication technologies to monitor and improve the wellbeing and health of patients and to empower patients in the management of their health and that of their families” (Iyawa et al., 2016, p. 246). The COVID-19 pandemic, in particular, has shown how much can be gained if health data were stored and processed digitally. A smart information system could even lead to medical staff spending more time with patients and less on their PCs, for example, if information about patients were available seamlessly and the results of devices were transferred directly into their health records. Digitalization is actually a prerequisite of a patient-centered continuum of care model, which has been considered the gold standard by health experts for years. Further, digital health services are available independently of time and location, encourage the empowerment of patients and enable shared decision-making. But although the benefits of digital health services such as better accessibility of care are widely recognized, there has been no largescale integration into regular healthcare delivery. Rather, there is an abundance of successful pilot projects, which fail to be introduced into the regular healthcare delivery systems, i.e. the so-called ‘first healthcare market’, which includes reimbursable products, medicines and services. We assume that this discrepancy between expected benefits and actual implementation is primarily due to the lack of viable business models. Such business models play an important role in establishing sustainable service platforms and promoting the uptake of digital health services by health providers and patients. The healthcare market is characterized by a complex multi-stakeholder environment, fragmented decision-making processes as well as a lack of incentives for cost-effectiveness (see e.g. Botti & Monda, 2020). Besides, healthcare products and services are heavily regulated. The motto “Move fast and break things”, which tech companies use to describe the world as a playground for their creativity, does not fit hospitals. No advertising promise can be made without hard scientific evidence. Engineers in tech companies, on the other hand, are used to launching products early and then adjusting or stopping them. In medical technology, such an approach to product development would be unthinkable. Moreover, the development process has to be documented from the beginning according to official guidelines. There are two main routes into the healthcare market for tech firms: 1) doing business with hospitals and healthcare This is the Preface of the Topical Collection "Digital Healthcare Services".

The paper describes the design rationales and algorithms underlying SmartCoping, a mobile solution for recognizing stress based on the continuous monitoring of heart rate variability (HRV). A central contribution is the calibration of HRV values to each user of the SmartCoping app. Calibration is crucial since HRV varies greatly among people and must be interpreted individually. Calibration is implemented in two variants, one which requires user feedbacks on perceived stress levels during an initial learning phase and one where no feedback is needed, and a user’s maximal and minimal HRV values are determined during an initial usage phase. The variant with user feedbacks results in stress warnings which better reflect a user’s stress perception. The app can be used to trigger stress warnings, identify recurrent stress situations from the history data, and give relaxation support via a biofeedback component based on breathing exercises. SmartCoping helps people improve their self-perception and develop appropriate stress avoidance and coping strategies. The system has been evaluated in a field test with healthy volunteers.

Despite the promises of artificial intelligence (AI) technology, its adoption in medicine has met with formidable obstacles due to the inherent opaqueness of the internal decision processes that are based on models which are difficult or even impossible to understand. In particular, the increasing usage of (deep) neural nets and the resulting black-box algorithms has led to wide-spread demands for explainability. Apart from discussing how explainability might be achieved, the paper also looks at other approaches at building trust such as certification or controlling bias. Trust is a crucial prerequisite for the use and acceptance of AI systems in the medical domain.

This paper introduces an alternative approach to conventional pedometer apps which measure the wide-spread goal of 10,000 steps a day. Instead we focus on the intensity of physical activity, which is in line with recent recommendations of renowned health institutions such as the WHO. These promote a minimum of moderate to vigorous physically active time per week to achieve the desired health benefits. The paper discusses how the guidelines have been implemented. It also outlines how we help maintain user motivation over time (e.g. by integrating and personalising "nudges") and how we intend to solve the challenges posed by different fitness levels and personal lifestyles.

The paper presents results from the SmartSleep project which aims at developing a smartphone app that gives users individual advice on how to change their behaviour to improve their sleep. The advice is generated by identifying correlations between behaviour during the day and sleep architecture. To this end, the project addressed two sub-tasks: detecting a user’s daytime behaviour and recognising sleep stages in an everyday setting. In the case of daytime activity detection the best results were achieved using an accelerometer at the wrist and another one at the ankle (87%). A subsequent smoothing step increased the accuracy to over 90%. For recognising sleep architecture we experimented with various consumer wearables that we used in addition to the usual PSG sensors in a sleep lab. Several sleep stage classifiers were learned from the resulting sensor data streams segmented into labelled sleep stages of 30 s each. Apart from handcrafted features we experimented with unsupervised feature learning based on the deep learning paradigm. Our best results for correctly classified sleep stages are between 86 and 90% for Wake, REM, N2 and N3, while the best recognition rate for N1 is 37%. Finally, we discuss a preliminary design of the algorithm for determining correlations between daytime behaviour and sleep architecture.

The advent of digitalization exposes enterprises to an ongoing transformation with the challenge to quickly capture relevant aspects of changes. This brings the demand to create or adapt domain-specific modeling languages (DSMLs) efficiently and in a timely manner, which, on the contrary, is a complex and time-consuming engineering task. This is not just due to the required high expertise in both knowledge engineering and targeted domain. It is also due to the sequential approach that still characterizes the accommodation of new requirements in modeling language engineering. In this paper we present a DSML adaptation approach where agility is fostered by merging engineering phases in a single modeling environment. This is supported by ontology concepts, which are tightly coupled with DSML constructs. Hence, a modeling environment is being developed that enables a modeling language to be adapted on-the-fly. An initial set of operators is presented for the rapid and efficient adaptation of both syntax and semantics of modeling languages. The approach allows modeling languages to be quickly released for usage.

The paper describes a mobile solution for the early recognition and management of stress based on continuous monitoring of heart rate variability (HRV) and contextual data (activity, location, etc.). A central contribution is the automatic calibration of measured HRV values to perceived stress levels during an initial learning phase where the user provides feedback when prompted by the system. This is crucial as HRV varies greatly among people. A data mining component identifies recurrent stress situations so that people can develop appropriate stress avoidance and coping strategies. A biofeedback component based on breathing exercises helps users relax. The solution is being tested by healthy volunteers before conducting a clinical study with patients after alcohol detoxification.

This paper presents a domain-specific modelling language for patient transferal management (DSML4PTM). To foster reusability within the modelling community, existing modelling languages were taken into account as far as possible and then extended as was needed by the application domain. The language was developed through iteration following the design science research methodology. For requirements elicitation purposes domain expertise and healthcare standards were taken into account. The new modelling language was evaluated first with respect to the elicited requirements and then through the creation of two models reflecting a reference process and an application scenario. Next, an evaluation on the perceived usefulness and cognitive effort of the language was performed using a focus group with modelling and domain experts.

The paper presents results from the SmartSleep project which aims at developing a smartphone app that gives users individual advice on how to change their behaviour to improve their sleep. The advice is generated by identifying correlations between behaviour during the day and sleep architecture. To this end, the project addresses two sub-tasks: detecting a user’s daytime behaviour and recognising sleep stages in an everyday setting. The focus of the paper is on the second task. Various sensor devices from the consumer market were used in addition to the usual PSG sensors in a sleep lab. An expert assigned a sleep stage for every 30 seconds. Subsequently, a sleep stage classifier was learned from the resulting sensor data streams segmented into labelled sleep stages of 30 seconds each. Apart from handcrafted features we also experimented with unsupervised feature learning based on the deep learning paradigm. Our best results for correctly classified sleep stages are in the range of 90 to 91% for Wake, REM and N3, while the best recognition rate for N2 is 83%. The classification results for N1 turned out to be much worse, N1 being mostly confused with N2.

The paper analyzes current weaknesses of behavioral change support systems such as the lack of adequately taking into account the heterogeneity of target users. Based on this analysis the paper presents an application framework that comprises various components to accommodate user preferences and to adapt system interventions to individual users: a goal hierarchy which users can tailor to their needs, dividing nudges into different types that correspond to speech acts, rules for context-specific triggering of nudges. User adaptation is realized with approaches from user modeling and collaborative filtering. The result is a self-learning application that changes in line with a user's progress, which is expected to enhance user acceptance and increase and sustain people's motivation for behavioral change. The application framework will be evaluated by comparing a mobile health app using the framework with a simplified version of the app that does not support user tailoring and adaptation.