Hermes: A Digital Assistant for Coordinating Invisible Work in Family Elderly Caregiving Scenarios

Aging in place, i.e., encouraging people to live in their homes for as long as possible, is nowadays the preferred option of living for older adults and public services. However, this requires that the elderly become self-sufficient in a large number of daily activities, either independently or with the support of formal or informal caregivers. In most societies, informal care provision is supported partially or fully by members of the older adult’s family network, who usually struggle addressing their own professional and personal lives, as well as those of their own families. These informal caregivers contribute with time or resources to help mitigate the needs of their older adults.

In order to support an effective aging in place schema, a collective effort sustained by the involvement of a considerable part of the family network is required. Therefore, informal elderly caregiving is a process that becomes unpredictable in terms of response time, given that assistance and active involvement both depend on the cooperative, voluntary, and articulated action of family members [3]. In that respect, the informal care experience at home needs to address several factors, such as effectively managing the expertise of caregivers, dealing with feelings of social isolation among older adults, and caregiver burden [9].

Given the distributed and interdependent role of family members when fulfilling caregiving tasks, a major challenge that emerges is articulating the distributed collective effort within the network. In particular, this means understanding how to coordinate the actions of family members and allocate resources to improve the efficiency and effectiveness of the process. Building upon the discourse proposed by Star and Strauss on the ecology of visible and invisible work [10], there are confronting visions regarding the informal care provision to their older adults in family networks. In that respect, informal caregivers usually feel overwhelmed and burdened by fulfilling their tasks, which seem to be invisible to other members in the family network [4]. However, the coordination, resource allocation, and articulation of these tasks are all crucial endeavors to sustain the effectiveness of the family elderly caregiving process.

Previous research works have shown that technology-mediated collaboration can help address invisible work in family elderly caregiving scenarios. However, most initiatives have been focused on supporting the needs of the elderly or their caregivers but have not addressed the caregiving process as a holistic system that involves at least these two types of participants. Trying to address such a process in a holistic way, this article proposes a digital assistant named Hermes that helps coordinate the invisible work conducted in family elderly caregiving scenarios. This assistant is autonomous, but it uses as input the information of the caregiving activities that is added by the users (mainly family members) through a mobile application.

Next section presents and discusses the related work. Section 3 presents the architecture of the proposed system and introduces its main components. Sections 4 and 5 describe the system front-end and back-end respectively. Section 6 explains the usability and usefulness evaluations performed to the system and the obtained results. Finally, Sect. 7 presents the conclusions and provides perspectives on future work.

In HCI literature, the social nuances of caregiving work have not been deeply considered [1]. In particular, the work of family caregivers is typically invisible [2, 4] and lacks formal support [9].

Family caregiving is considered an unstructured (or unframed) process, given that it does not have a predefined workflow that can be specified at the process design time. Instead, only the participants and the potential interactions among them can be specified and supported with technology. In this scenario, the coordination activities are triggered on-demand by the participants, considering their own work context and needs. Therefore, counting on a centralized component, able to successfully coordinate the caregiving activities, is low feasible in practice given the heterogeneity and dynamism of the work context of the participants. In that case, coordination should be promoted through the use of awareness and persuasion mechanisms implemented by distributed as well as centralized components.

Currently, the cooperative foundations of care provision and assistance to older adults is sustained through the involvement of a considerable part of the family network. In that respect, due to the distributed and interdependent role of family members when fulfilling caregiving tasks, a major challenge that emerges is articulating the distributed collective effort within the network [7]. In other words, any socio-technical system aiming to support this application domain needs to sustain the coordination of family members and the allocation of resources to improve the efficiency of the process. According to Star and Strauss [10], family elderly caregiving corresponds to a particular example of invisible work, where duties—and consequent burden—are mainly assumed by one of the adult children in the family the work [6].

Ensuring the assumption of family caregiving roles requires articulating the collective distributed effort of family members. However, this can turn out to be demanding due to the complex and dynamic arrangement of the family network with regard to this process. Indeed, promoting articulation work is more an expectation rather than an established practice in family contexts, where improvisation and individualism are often sources of tension and conflict [4].

According to Schmidt and Simone [7], the articulation of the distributed activities that are part of a cooperative work requires appropriate mechanisms of communication and coordination among the involved participants. Likewise, classical approaches of CSCW design suggest that using common information spaces, such as shared views or board systems [7, 8] could contribute to make visible these aspects of the caregiving process.

Therefore, one plausible alternative to assist older adults into embracing caregiving technology is to design meaningful experiences that do not limit physical exchanges [5]. In addition, respecting the existing routines and expectations of the different involved stakeholders is required, not only for facilitating the technology adoption and promoting effective and meaningful exchanges, but also to try aligning the implicit and explicit asymmetries existing across generations [4].

In order to help family members coordinate their caregiving actions and be aware of the activities conducted by others, this section presents Hermes, an autonomous digital assistant that helps reduce the load and tension generated by the assignment and coordination of caregiving duties. This assistant acts as an intermediary that provides awareness to the participants and tries to persuade people to coordinate their activities and help address the needs of older adults.

The system considers the following four user profiles, which were identified based on the roles structure proposed in [4]:

• Older adult: This is an elderly person who is the target of the caregiving actions. The participation of the older adult as a user is not mandatory. However, if he/she does, this user can interact with the digital assistant in several ways as well as receiving feedback and notifications.

• Assistant (or caregiver): These are people that perform direct activities in favor of the elderly as part of the caregiving process. Typically, they are the children of the older adult that need to coordinate their activities with other caregivers. These users have access to all services provided by the system.

• Helper: Users playing this role perform occasional activities in favor of the older adults; for instance, when a caregiver or the older adult asks them for help. Helpers interact with the digital assistant through WhatsApp text messages. A caregiver can propose to assign a specific activity to a helper through the regular task assignment process. Examples of helpers are neighbors and friends of the elderly being cared for, as well as relatives that are not part of the close caregiving network.

• Health professional: These users correspond to physicians, nurses, or other medical personnel that are involved in monitoring the health condition of the older adults being cared for, and eventually treat them. If they agree, these users can be contacted by the assistants, in case of need; e.g., about health checkpoints of the older adults being monitored by them.

Figure 1 shows the participation of these user roles and illustrates the structure of the system including its main components. The system considers a front-end application that allows users to interact with the services, and a back-end that provides those services. Typically, the older adults and caregivers use, as front-end, a mobile application specifically designed to interact with the task manager and the chatbot, both providing most of the services offered by the system. The users playing the role of helper and health professional use WhatsApp to interact with other roles and also to receive notifications, invitations, and awareness information sent by the chatbot. The decision of using WhatsApp by these roles was made to ease their participation; i.e., to lower the entry barriers to participate in this caregiving ecosystem. Thus, they do not need to install any extra application in their mobile phones, except WhatsApp.

Structure of the Hermes system.

Full size image

The main end-users are the family caregivers, who usually add and update tasks into the system (e.g., to make a medical appointment, or buy groceries for the elderlies). They are also the target of the most interactions triggered by the chatbot, e.g., to coordinate tasks or try to get a commitment on those that are pending. The older adults can also use the application, mainly to be aware of the status of the caregiving tasks. As shown in Fig. 1, these users interact with the system through the mobile application that represents the front-end of Hermes. This application allows to create chat groups, without the presence of the virtual assistant.

As mentioned before, the main functionality provided by the back-end involves the task manager and a chatbot. The former manages three types of tasks (or activities) that are scheduled and updated by the users or the chatbot. The latter is mainly in charge to provide awareness about these tasks and reminders to the users in charge of performing them. Next sections explain in more detail the front-end and back-end of the system.

This application manages tasks, contacts, and conversations (chats). The services related to tasks are one of three types: recording medical appointments, where a caregiver has to accompany the older adult to some place; recording a general task that caregivers can perform whenever they need; and recording reminders that help caregivers address periodical activities in favor of the older adults.

Once a task is created, the system runs an “assignment cycle” splitting the caregivers in three priority groups. It then asks each group whether a family member can take the task, until someone does. Then, the task creator is notified about the assignment. On the day of the scheduled event, the virtual assistant sends a task reminder to the user in charge. Once the task is completed, the system will ask for feedback, record it for future events, and post a notification to the Notice panel.

The users can chat with the virtual assistant to ask about the state and history of different tasks, as well as browse the history records as a way to be informed without asking another person. This application also has the feature of adding medical support as another contact, e.g., to perform quick questions or notify emergency situations. In order to do that, the health professional just has to add Hermes (for a specific older adult) as a contact in their personal WhatsApp account. Next sections present a summary of the user interfaces provided by the mobile application to manage tasks and contacts.

4.1 Task Management Front-End

The mobile application has few users interfaces in order to make the user interaction simple and easy. After launching the application, and once the user is logged in, the system shows a main menu with a list of dialogs (or chat rooms) available for that user (Fig. 2 center). This interface contains a message panel, the dialog with the virtual assistant Hermes, and a list of private and group chat rooms.

Sample of the user interfaces of Hermes.

Full size image

Clicking on the ‘message panel’, the users can read messages about the coordination process and result of the tasks. The Hermes’s chat is the only service that can be utilized by the user to interact with the virtual assistant (i.e., the chatbot). Private chats are 1-to-1 dialogs with other members of the caregiving group and they are sorted by user types starting with older adults, assistants, helpers, and professionals who support the group.

In this first interface the users can make phone calls, if enabled, or directly log into the chatroom. Users can go to the Menu view (Fig. 2 left) or tap a shortcut (Fig. 2 right) for adding elements. These options allow the users to add new contacts, chat groups, or tasks. By tapping onto either of ‘Add task’, ‘Add reminder’ or ‘Add medical appointment’ options, the virtual assistant launches an interactive dialog with the user about creating the selected task type. The ‘Add chat group’ option shows a list with other family assistants, where the user has the option to create new many-to-many dialogs.

4.2 Contact Management Front-End

Adding a new contact prompts users to the interfaces shown in Fig. 3, where they need to choose the role to be played by the new user. These roles can be older adult (who will receive care), assistant (main responsible for the caregiving activities), helper (who occasionally perform activities in favor of the older adults), or medical support (health professionals). If the new user is an older adult or a helper, such a user needs to fill a data form with their name, phone number, and email. An assistant, helper, or professional contact needs to have an older adult to care for and will receive an email invitation to participate in the caregiving group. The helpers and health professional need to add a contact to their WhatsApp application to accept the invitation. Conversely, an assistant contact has to install the mobile application of Hermes and register him/herself with a family code sent in a personalized invitation.

User interfaces of Hermes to add contacts.

Full size image

From the main menu (Fig. 2a), the user can reach “My profile” and “Log Out”. In the first option, users can manage their profile data and preferences, quit the caregiving group, deactivate location sharing, or set their absences, i.e., the days when they cannot take any task, except if it is urgent. Figure 4 summarizes the navigation map of the mobile application.

Navigation schema of Hermes.

Full size image

As mentioned before, the main components of the back-end are the task manager and the chatbot. Next we explain each of them.

5.1 Task Manager

The task lifecycle involves four activities: creation, assignment, reminding, and reporting of results. Once finished the cycle, the information is stored and used by the assistant to make decisions in the future.

The system supports three types of tasks (or activities): medical appointments, delivery of a reminder, and general tasks. The first ones require that a caregiver go with the elderly to the medical appointment. The delivery of reminders is related to activities that have short-time periods to be performed or are recurrent. Finally, the general tasks allow addressing the rest of the activities involved in the caregiving process and that require some coordination of the participants in the caregiving network.

Once the information about the activity and its periodicity are recorded, the creator can suggest a participant to perform such an activity. Otherwise, the creator finds a responsible for such a task. The system creates three groups considering users playing the roles of assistants and helpers, as well as taking into account the activities assigned in the past, their absence agenda, and proximity to the place where to perform the activity (if any). As shown in Fig. 5, the groups are ranged from those with less assigned activities (higher priority) to those with more assignments (lower priority).

Algorithm to prioritize groups for assigning an event.

Full size image

In this case, the system sends a notification to all users of the first group, asking for a responsible for the task. The process ends if a user accepts the invitation and the virtual assistant informs the assignment to the task creator. If a user explicitly indicates that is not able to address the request, then he is removed from the list for future notifications related to such a request. Otherwise, the users are added to the second group and the process is performed again until getting a responsible to perform the task or complete the notification to the three groups.

When a user accepts to perform a task, the assistant asks if he/she wants to record a reminder for such an activity. Figure 6 shows the steps that the virtual assistant follows to set these reminders. Finally, if the activity is a medical appointment or a general task, after a while the virtual assistant will ask the user committed with the task, what the results of such an assignment were; e.g., it can be completed, pending, or failed. The answer of the user is recorded, and he/she can inform the rest of the group about this result through an automatic message that will be available in the message panel.

Reminders and feedback flow.

Full size image

5.2 Chatbot

The chatbot provides three major services: (1) to assist family members in the process of adding a caregiving task that requires the support of a family member; (2) to ask family members for help to address particular needs of older adults; and (3) to deliver reminders to the people committed with caregiving tasks and provide awareness about the status of these tasks, e.g., if they are pending, assigned, or addressed.

This chatbot runs in the background of a mobile application and assists their users (usually, the family caregivers, and eventually the older adults) in the three major activities explained before. Figure 7 summarizes the behavior of the chatbot.

Caregiving support process performed by the chatbot.

Full size image

First of all, when a family member adds a new caregiving task using the mobile application (e.g., to carry the older adult to the physician), the chatbot helps that user classify and characterize the task. Then, the chatbot identifies a set of candidate members within the family network for completing such an assignment, considering the type of activity to be performed, the usual availability of each person, and the caregiving activities already performed by such a person in favor of the older adult.

Taking into account these variables, the chatbot finds a voluntary family member that agrees to perform the task. Depending on the results of this process, the chatbot changes its strategy to find new voluntary people if required. Finally, it provides on-time automatic reminders to the involved people trying to ensure the assignment be successfully addressed. The chatbot also provides awareness about the caregiving tasks that are pending, assigned, and successfully/unsuccessfully closed.

As described in the previous section, the chatbot flow depicted in Fig. 7 was developed as a background service of Hermes. The goal of such a service is to coordinate and articulate the activities performed by the family caregivers in a family. Given the complexity of the tightly-coupled cooperative scenario addressed by the system, the design and evaluation of the mobile application followed an iterative and incremental process. This section presents the evaluation of the last version of the prototype, which involved a sample of end-users in a simulated setting.

6.1 Evaluation Setting

Eleven families participated in this process, each composed by an assistant and an older adult or a helper. All families were taking care of an older adult and had an important past experience in this activity. Some of these families have older adults suffering from mental illness as Alzheimer and dementia, while others have an external person as formal caregiver. The participants indicated the role they play in the caregiving process. Next table summarizes the characteristics of the participants (Table 1).

As recognized by Gutierrez et al. [4], the roles assumed by the family members tend to fit their ages and relationship with the older adult. Typically, the older adults are care recipients, middle aged adults (usually child) play the role of assistant, and young adults (usually grandchildren) act as helpers.

In this evaluation process we used the Wizard-of-Oz (WoO) method to present the Hermes system to the users, representing all interfaces needed to reflect the navigation, functions, and processes as a mid-fidelity paper prototypes. We interpreted the chat dialogs between virtual assistant and the user as verbal conversations.

Before the WoO session starts, the users were informed about the objectives and goals of the system as well as about the characterization of three roles supported by the system. During the session, the users assumed the role they interpret in their current caregiving scenario and we asked them to follow the Thinking Aloud method as a way to gather user comments. In the evaluation session, users were asked to make different tasks aimed to let them learn about the system functionalities and to determine whether they were useful. The tasks involved comprised: creating and interacting with contacts, creating and assigning caregiving duties, creating reminders and notifications, and asking for feedback about the status of a task.

Once the session was completed, the participants were asked to fill out an exit questionnaire comprising 20 items graded in a 5-point Likert scale. The first ten items were focused on determining the usability of the system, and the other ten evaluated its perceived usefulness. Next we indicated the obtained results.

6.2 System Usability Evaluation

Table 2 shows the items used to evaluate the usability of the system and their results grouped by role (older adult, assistant and helper), indicating their mean (M) and standard deviation (SD). The instrument that we used for conducting the evaluation was the System Usability Scale (SUS).

Assistants and helpers rated the system with high scores, showing the application is usable for them (>68 pts.). However, it is still not easy-to-use for older adults. The elderly people participating in this evaluation did not use mobile applications to interact with others, and therefore they felt the need of requiring more training for using the system comfortably.

For assistants, the reactions over the things they need to know before using the system vary. It can be caused by their level of exposure or use of mobile technology. In the case of helpers, scores were good. However, these users were uncertain about the capability of their older adults and assistants to quickly learn to use the system.

Assistants and helpers have a similar perception of the system usability, perceiving the system as comfortable, easy-to-use and learn, where the functionalities were properly integrated. However, the willingness for using the application is different. Particularly, assistants would use it since they felt the system helped them manage their caregiving agenda and coordinate with other family members. However the helpers, who felt less responsible for the caregiving process, would be ‘annoyed’ by using the assistant.

6.3 System Usefulness Evaluation

Table 3 shows the items used to evaluate the perceived usefulness of the system and its functionalities. It includes the evaluation results, the average and standard deviation, grouped by role. Concerning the system information handling, support to the lifecycle, and assignment activities, all participants assigned a high score to these services, and recommend to other family members the use of Hermes to coordinate the caregiving activities.

Participating users had mixed feelings about the feasibility of using this application in practice. Particularly, the older adults and assistants would use it since they perceive a benefit on it. However, the helpers are not that sure since they are usually not strongly committed with the caregiving process of the older adults in their families.

Moreover, assistants and helpers think the system would be not easy-to-use for the older adults, mainly due to their usual cognitive impairments and reluctance to use digital technologies. However, the participating older adults wondered where they could have a dedicated interface to interact with friends and family, as they showed themselves to be active and curious about digital devices.

Finally, the obtained results show that most family members were keen on the idea of interacting with the virtual assistant, praising the system as a valuable asset on providing help to manage more smoothly the caregiving process.

In this paper we presented Hermes, a digital assistant designed to help family caregiving stakeholders deal with the underlying challenges of such a process. As main design principles, Hermes addresses the intrinsic challenges derived from the cooperative practices and articulation work of family caregiving work. We show that this challenge can effectively be addressed using technology, as well as generating a potential positive impact on informal family caregivers and older adults.

In the next step of our research we will conduct a longitudinal in-home deployment study with a representative sample of family networks, aiming to better understand the involvement of family members when caring for their older adults with the assistance of Hermes.