Epilepsy, a chronic neurological disorder, is characterized by recurrent and unpredictable seizures, affecting individuals across all age demographics [1]. As a leading cause of chronic morbidity globally, epilepsy presents significant challenges, including a heightened risk of premature death, injuries, and diminished quality of life [2]. With a prevalence rate of 5-10 cases per 1000 people [1], the disorder imposes a considerable burden on individuals and health care systems alike. The financial impact is also notable, with the UK National Health Service spending approximately £1.5 billion (US $2 billion) annually on care that ranges from primary treatment and management to addressing seizure-related injuries and mental health issues [3,4]. The disability-adjusted life-years rates in the United Kingdom, estimated at 92,400 per 100,000 population [5], highlight the extensive health loss attributed to epilepsy. Sudden unexpected death in epilepsy remains a critical concern, with an incidence rate of approximately 1 in 1000 people with epilepsy annually, emphasizing the need for ongoing research and innovative management strategies [6,7]. Accurate forecasting of seizures could allow people with epilepsy to take additional safety precautions at times of increased risk.

Despite advancements in treatment—such as resective surgery, neurostimulation devices, new antiseizure medication, many with novel modes of action, not to mention the increasing availability of treatment across the world, with several antiseizure medications on the World Health Organization’s list of essential medicines—approximately one-third of the population (more than 50 million people) living with epilepsy globally still experience uncontrolled and apparently unpredictable seizures [8]. A large community survey (n=1056) [5] and the UK Epilepsy Priority Setting Partnership [9] ranked the unpredictability of seizures as one of the most impactful aspects of the condition and a research priority, underscoring an urgent need for solutions that can offer predictability and management of seizure risk. Seizure forecasting technologies emerge as a promising avenue in this regard, potentially enabling individuals to take pre-emptive measures during high-risk periods, therefore mitigating the risk of injury and premature death, and alleviating the psychological burden associated with the unpredictability of seizures [10].

One potential avenue for seizure forecasting is leveraging technologies that directly monitor brain activity, combining this with data analytics to explore seizure risk modeling. Video–electroencephalography (EEG) is the gold standard for monitoring brain biosignals to detect seizure activity. However, this is not well-suited to explore as an out-of-clinic seizure forecasting solution, as it is not feasible for long-term use in real-world scenarios [11]. Ultra–long-term electroencephalography monitoring (UNEEG) represents a significant advancement in the field [12]. UNEEG’s technology, involving surgical implantation, offers continuous EEG data capture in the user’s real-world context. People with epilepsy who have trialed the technology have found the continual monitoring valuable, both in terms of internalizing their locus of control and providing information to inform treatment changes. However, UNEEG is a minimally invasive method with intrinsic risks and costs of surgery, making it inaccessible to a wider range of users [12,13]. Noninvasive scalp EEG represents a more accessible solution that is designed for use outside of clinical settings. Yet, the adoption of this technology is hampered by limitations such as the stigma of visible devices, practicality, acceptance among users [13], and the technical and practical challenges of long-term outpatient monitoring [14].

An alternative seizure forecasting approach is to explore predictive analytics based on seizure precipitants that can be captured through noninvasive, nonmedical grade wearable devices (WDs), such as smartwatches. This presents a high-potential solution, as ubiquitous wearable technologies can be easily integrated into the end user’s everyday life, to continuously capture real-time data over the long term. Noninvasive WDs that can be used outside of clinical settings have been successfully used for epilepsy management in the form of the Empatica E4 wristband [11]. This device has shown promise in detecting epileptic seizures by monitoring physiological parameters such as heart rate and acceleration. However, the Empatica E4 focuses primarily on seizure detection rather than forecasting [11].

Exploration of seizure forecasting based on seizure precipitant data remains in its infancy. Emerging research has explored seizure cyclicity for forecasting models. However, these approaches on their own have limited accuracy for seizure forecasting [15-17]. There is potential to build upon and improve these models by including a greater range of seizure precipitants. This is a promising approach, given the existing evidence on the range of external and physiological variables that play a precipitating role in seizure propagation. Measurable biomarkers, such as emotional stress, infections, menstrual cycles, sleep deprivation, and alcohol use, have all been identified as seizure precipitants. The relationship between precipitants and seizure propensity could be quantified using mathematical models [13,17], with these models informing algorithms to analyze person-specific data and facilitating personalized seizure forecasts.

Ubiquitous technologies with multimodal sensors, combined with ecological momentary assessment (EMA; repeated sampling of an individual’s cognitions, emotions, symptoms, and behaviors) could enable the real-time capture of the broad spectrum of known seizure precipitants. These real-time data, combined with predictive analytics, offer a feasible and accessible solution for seizure forecasting. To our knowledge, there is a pressing need to address the research gap in developing and evaluating technology for real-time seizure precipitant data collection, essential for supporting seizure forecasting interventions.

A technology to collect real-time seizure precipitant data relies on people with epilepsy actively engaging with the technology. As such, it is essential that a user-centered design approach is taken. The person-based approach [18] integrates methods from user-centered design and qualitative research to design and evaluate complex digital health technologies. This approach can lead to solutions that are more acceptable, feasible, meaningful, and optimally engaging to the people who will use them. The person-based approach advocates undertaking iterative design cycles with users through both formal qualitative research and patient and public involvement. Previous research [16,19] underscores the necessity of using person-based approach for the future design of epilepsy-related technologies, emphasizing iterative design cycles and user involvement to create solutions that are not only technologically effective but also deeply aligned with the needs and experiences of those living with epilepsy. Therefore, this study used the person-based approach, recognizing its critical role in developing effective, user-centered epilepsy management technologies.

The ATMOSPHERE study, funded by the Engineering and Physical Sciences Research Council (EPSRC) and in collaboration with the N-CODE network [20], represents an effort to leverage technology for the benefit of individuals with epilepsy. The ATMOSPHERE project is an early-stage, proof-of-concept research initiative seeking to develop and evaluate a noninvasive, wearable, epileptic seizure forecasting technological solution. This innovative approach aims to tackle the apparently unpredictable nature of seizures by collecting real-time data on various seizure precipitants, such as sleep quality, heart rate, menstrual cycle, medication adherence, and stress levels, from adults living with epilepsy.

The entire project is organized into several work packages, each targeting a specific aspect of seizure forecasting technology development, such as (1) work package 1—synthesizing the evidence based on seizure precipitants or triggers, (2) work package 2—designing and testing the usability of a mid-fidelity functional prototype to capture real-time precipitant data from individuals living with epilepsy, (3) work package 3—developing predictive models for seizure risk forecasting, and (4) work package 4—identifying user requirements for the end user app interface including presenting seizure forecasting output. Developing a high-fidelity functional prototype is based on the outcome of the usability study in work package 2.

This protocol paper specifically addresses workstream 2 of the ATMOSPHERE study, reporting the results from the completed co-design study and outlining the protocol for the upcoming usability study.

This multiphase research workstream aims to design, develop, and test prototypes for collecting real-time data on seizure and seizure precipitants.

The first objective is to collaboratively design a mid-fidelity functional prototype capable of collecting real-time seizure and seizure precipitant data from people with epilepsy. The second objective is to conduct “in-the-wild” usability testing of this prototype to evaluate its feasibility and user acceptance and engagement. This step is critical to assessing the data capture prototype’s real-world application and identifying areas for improvement based on user feedback. The third objective is to use insights gained from usability testing to refine and enhance the prototype, and focus on improving its functionality, ease of use, and overall effectiveness in collecting high-quality data for seizure forecasting.

Drawing on the person-based approach [21], we undertook iterative patient and public involvement to co-design the data collection prototype (phase 1). We are now preparing to conduct a mixed methods “in-the-wild” usability study in phase 2 to further investigate the prototype’s usability and implement refinements (Figure 1).

Phase 1 involved a co-design with patient and public involvement contributors, which does not necessitate formal ethical approval [25]. However, we adhered to the General Data Protection Regulation (GDPR) guidelines [26]. To protect patient and public involvement contributor privacy, we used pseudonymization for sampling forms, and we secured consent to report aggregated data. During co-design sessions, no direct quotes were captured; instead, we created summaries in the form of guiding principles and tables of changes. We offered patient and public involvement contributors the equivalent of £20 per hour (US $26) in shopping vouchers.

For phase 2, ethical approval was obtained from the University of Bristol, Faculty of Engineering Research Ethics Committee on March 10, 2022 (reference 10152). In alignment with our commitment to data privacy and protection, we carried out a data protection impact assessment [27]. This assessment (ensures compliance with the GDPR within the European Union [26]), aims to evaluate and minimize potential risks to the rights and freedoms associated with personal data. The information compliance manager and data protection officer from the University of Bristol approved the data protection impact assessment on January 29, 2024.

This phase focuses on usability and the refinement of the prototype. Data collection for phase 2 started in May 2024, and the findings are expected to be available in August 2024.

This paper reports (1) the co-design process of a novel technology designed to collect real-time data needed for seizure forecasting and (2) the protocol for the usability testing of the prototype. Our co-design identified guiding principles for the prototype, including:

This project emphasized the co-design of the prototype, ensuring that the technology aligns with the preferences and requirements of its target community. We engaged with clinicians from a range of relevant clinical specialties. We also engaged with a diverse group of individuals living with epilepsy, varying in aspects such as gender, age, seizure type, and socioeconomic background (determined by indices of deprivation). Individuals living with epilepsy are a diverse group, and we acknowledge that it will be important to continue to gather more views on the technology beyond the 15 individuals involved in the phase-1 co-design. A notable limitation of phase 1 is the lack of ethnic diversity among contributors and absence of individuals from the very lowest indices of deprivation. To find a remedy, in phase 2, we are working with a specialized organization to actively engage and involve ethnic minority groups and those from more deprived areas. This approach aims to make the development more inclusive and the technology more representative of the community it serves.

The second phase of our study is designed to evaluate the usability of our technology in a naturalistic setting (in the wild). The prototype undergoing testing is currently focused exclusively on the capabilities related to data collection. Consequently, it does not yet incorporate the advanced features, such as predictive analytics, for determining seizure risk scores and the provision of seizure forecasting outputs to end users. These more sophisticated functionalities are currently under development within separate workstreams. It is hypothesized that the availability of seizure forecasting output would serve as the primary motivation for individuals to continuously use the technology over an extended period. Therefore, the current limitation of the prototype, in its inability to replicate this aspect of real-world application and motivation, is acknowledged. To offset this limitation, we have endeavored to introduce an alternative form of motivation for participants by offering summaries of their data (data visualizations of seizure and precipitant data) at the conclusion of the study period.

The data collection protocol, including the type and frequency of prompts, was co-designed with individuals with lived experience. Subsequently, the prototype was tested with 3 patient and public involvement contributors who did not raise concerns about the frequency of prompts. Instead, users welcomed the reminders, especially those experiencing the well-documented memory impairments associated with epilepsy [30]. However, there is a possibility that the frequency of prompts could be burdensome when the technology is used in the wild over an extended period. During qualitative interviews, we will explicitly explore participants’ views on the data collection protocol, and we will refine the protocol, where necessary, based on this feedback.

The prototype developed for this phase is characterized as a mid-fidelity and functional prototype, which permits the testing of its core functionalities, such as passive data collection and EMA. Despite its capabilities, this prototype is not without its limitations, including a user experience interface that may not meet high standards and the necessity for some manual operations. However, the adoption of an agile methodology for this phase offers significant advantages. This approach allows for rapid learning and adjustments based on real-world, in-the-wild use, aligning with the principles of technological probing. While acknowledging its current constraints, this phase is pivotal in facilitating immediate, invaluable insights into the usability and potential impact of the technology in everyday contexts, guiding subsequent developments and enhancements.

This phase of the study did not include individuals with intellectual disabilities, an important population given their increased risk of epilepsy [31]. Recognizing the necessity of designing interventions that cater specifically to their needs, it is evident that this group requires a thorough investigation into their specialized design requirements (beyond the scope of this study). Such an exploration should be a priority for future research to ensure the development of more inclusive and effective solutions. Furthermore, the study focuses on adults, and a separate investigation for pediatric epilepsies is needed.

Overall, this work will inform the development of a high-fidelity functional prototype. The assessments of that prototype will be conducted through user experience walkthroughs and qualitative interviews to identify usability and accessibility challenges. The results from these upcoming assessments will be reported in a separate paper for work package 4.

The co-design presented here identified user requirements for technology to capture the real-time data needed for seizure forecasting. Requirements include the ability to collect data passively where possible (with options for manual data input); collect custom, personal seizure precipitant data (in addition to evidence-based precipitant data); enable cotracking, where family or carers can input data; and consider inclusive design (due to variation in user characteristics and needs). The planned research in phase 2 will test the usability test of a mid-fidelity and functional prototype developed to meet these requirements. This co-design and usability testing contributes to a broader research aim; to develop seizure forecasting technology, leveraging real-time data on seizure precipitants, and using predictive analytics.

This technology could significantly improve the quality of life for individuals with epilepsy. By providing insights into the likelihood of seizure occurrence, such technology could offer a new level of independence and safety, reduce the economic and health care burdens associated with epilepsy, and facilitate a more personalized and efficient use of health care resources for epilepsy management. This technology development will be explored through proof-of-concept research of developing a noninvasive, WD specifically designed for seizure forecasting, integrating machine learning models for personalized predictive analytics, and using smartphone apps for real-time feedback to end users.