People engage in tracking activities for a wide range of purposes, from understanding finances to improving productivity to supporting artistic expression [10]. At the forefront of tracking technology and practice is the Quantified Self (QS) community [15], an enthusiastic group of trackers who describe themselves as an “international collaboration of users and makers of self-tracking tools.” QS members are active around the world with regional meetings and an annual conference. At meetings, trackers give Show and Tell presentations to describe their experiences.

Researchers study the practices of these and other trackers to understand their collection and sense-making strategies [10] and develop models of the tracking process [11,16] (discussed later in this section). People track using a variety of methods including automatic methods, such as bank logs, and manual methods, such as calendars [11]. Tools may be analog, such as paper journals, or digital, such as smartphone calendars [11]. After collecting data, people reflect on relationships among and patterns in data. Reflection can lead to insights about behavior or decisions to change behavior. However, barriers to successful tracking include difficulties with (1) deciding what data to collect and what tools to use, (2) using tools, (3) collating and formatting data, (4) understanding and reflecting on the data, and (5) formulating action plans based on the data [10,11].

Some QS members track to identify and solve health problems, although most health-related tracking is for general wellness. Similarly, much research focuses on tracking for general health and wellness (eg, [17-21]). Tracking has become an important and prevalent activity among the general public; the Pew Internet & American Life Project [6] recently reported that 69% of Americans track health data for themselves or a loved one.

However, many people managing a chronic illness must monitor their symptoms and their health through tracking. As of 2013, 19% of adults with no chronic illness tracked a health factor, whereas 40% of those with 1 chronic condition engaged in tracking, and 62% of adults with 2 or more chronic illnesses track one or more health factors [6]. This tracking can be an effective part of managing chronic illness, improving health outcomes [22], and communicating with health care providers [23]. Studies examining the lived experiences of tracking find that it is often coordinated or influenced by communication with health experts (eg, [24-26]), peers (eg, [14,20,21]), family members (eg, [27-29]), and colleagues and workplace programs (eg, [30]). Although much of the literature on tracking acknowledges that these practices happen in and are influenced by various social contexts, the models that guide design and study of tracking tools focus on individual tracking [11,16].

As more people managing chronic illness participate in health tracking, health care providers and researchers need to understand their work and social ecosystem. Moreover, although technology facilitates tracking, limited evidence supports the efficacy of specific tools to accomplish successful illness management [31]. Barriers to successful tracking for health include (1) insufficient support for collaboration with a provider [32], (2) difficulty making sense of data leading to lapses in tracking [16], and (3) difficulty remembering to track or deciding what to track [33]. Furthermore, many studies find that the apps available to support tracking for chronic illnesses are of poor quality [34-40], pointing to a lack of understanding of the needs of people managing chronic illness.

MacLeod et al [24] interviewed 12 people with a range of chronic illnesses who tracked some aspect of their health. They found that people wished to understand how their illnesses affected their lives within the context of information from health care providers. Mamykina et al [41] developed and tested MAHI—a mobile tracking and communication tool for people with diabetes. Their work revealed that, even with coaching from a diabetes educator, people newly diagnosed with diabetes struggled to develop self-efficacy and reflective thinking skills with regard to the data they captured. Our study sought to complement these studies by contributing an understanding of the processes by which trackers engage with a social dimension of tracking behavior.

A total of 4 informatics models have articulated the tracking process: Li et al [11]; Epstein et al [16]; Swan [42]; and Murnane et al [43]. Li et al [11] and Epstein et al [16] focus on general personal informatics, whereas Swan [42] and Murnane et al [43] discuss informatics in the context of health.

Li et al [11] conducted an interview study of 11 self-trackers to derive a stage-based personal informatics model. Their model describes the process of general self-tracking for any purpose and is composed of preparation, collection, integration, reflection, and action stages. This model focuses solely on the stages through which a single person progresses in self-tracking. The authors highlight the dependencies between each stage: a mistake in preparation can cause someone to collect the wrong data, and these problems then cascade to the integration, reflection, and action stages. Whooley et al [44] expand on the integration stage of Li et al’s personal informatics model with a discussion of why people track and how they integrate their data. More recently, Mamykina et al [26] studied the process of self-discovery in a structured diabetes education program, showing how tracking can scaffold learning and reflection for diabetes management.

The stage-based personal informatics model describes an ideal process for tracking, but that process can break down when it encounters the realities of everyday life [45]. To describe general tracking in everyday life, Epstein et al [16] propose the Lived Informatics model. This model is based on 184 surveys and 22 interviews on self-tracking behaviors for physical activity, location, and finances. The authors refined Li et al’s model by dividing its original preparation stage into 2 stages: deciding and selecting. Introducing a cycle named tracking and acting, they describe an iterative progression through collection, integration, and reflection. Finally, this model anticipates that people will lapse in their tracking practice either temporarily or permanently.

Swan [42] proposes a model for Patient-driven Health Care that includes self-tracking among the actions in which a patient might engage. The patient is the only one engaging in work, and people other than the patient are included only peripherally. Furthermore, the evidence base for developing the model is unclear.

Murnane et al [43] examined applying Ecological Systems Theory [46] to the work of tracking in long-term management of severe mental illness [43]. The resulting model describes the influences and resources available to people, including close personal ties, indirect institutional influences, and an individual’s sociocultural context. They discuss how personal informatics tools and data form an informatics layer that can mediate interactions with these other services, though the model does not articulate activities in the tracking process.

Although researchers have used these models as a lens through which to study tracking related to health [14,47], they found that they lack elements important to the health context. Costa Figueirido et al [14] studied women making sense of infertility and found that the stage-based model did not adequately represent the fluidity of work or the collaboration in which women engaged. Mishra et al [47] studied people tracking while hospitalized and also observed that the stage-based model did not characterize the collaboration occurring around data. Costa Figueirido et al [14], Mishra et al [47], and Valdez and Brennan [13] all explicitly indicate a need for a model that more closely aligns to the unique needs of the health context.

This study addresses key gaps by constructing a model representing the work and people involved in tracking to support chronic illness management.

For the QS dataset, we selected videos posted publicly on the QS blog [15] between January 2012 (blog inception) and December 2018 (end of data collection) that focused on managing a chronic illness, as defined in the Introduction. QS speakers are enthusiastic technology and tracking hobbyists who often characterize their tracking practices as innovative. They share their experiences at local and worldwide meetings to disseminate information about their routines and insights. In these videos, people described their work processes: information they tracked, how they analyzed and learned from that information, and when and how they shared information with others.

These presentations are meant to instruct other QSs, therefore giving us access to their expertise similar to an interview elicitation. Given their proficiency, we expected QS presentations to give us an overview of the cutting edge of how people build knowledge about chronic illness through tracking. As the videos are publicly accessible, we were not required to obtain consent for their use. Although presenter names are included on the QS blog, we chose to use anonymous identifiers for analysis and reporting.

The interview participants were patients of primary care providers in Group Health owned and operated clinics (now Kaiser Permanente Washington), a large integrated health care delivery system in Washington State providing care to over 300,000 people. We recruited adults managing type 2 diabetes and mothers managing asthma for at least one child aged 12 years or younger. These 2 diagnoses were chosen because each requires daily health-related tasks and frequent contact with health care providers and health care systems (eg, scheduling appointments, filling prescriptions, scheduling lab tests, and asking questions outside of formal appointments). We used purposeful sampling to identify participants representative of the general population in the Northwest United States based on gender, ethnicity, technology use (with recorded use of a patient portal as a proxy), and education. All interview participants completed an informed consent process. We conducted semistructured interviews in each participant’s home inquiring about health goals, priorities for completing health tasks, and workflow in attaining those goals and tasks. The workflows articulated by participants included information on tracking and communication in support of health management. Group Health Research Institute contracted a Health Insurance Portability and Accountability Act (HIPAA)–approved outside agency to transcribe and redact audio recordings of interviews.

From among the videos on the QS blog, we identified 30 publicly available videos meeting inclusion criteria, with a total running time of over 6 hours and 57 min. Videos were from 24 people; 4 speakers made 2 presentations each, and 1 speaker made 3 presentations. A total of 16 speakers were male (67%). One speaker acted as an informal carer (4%) for her child (ie, an unpaid provider of health-related care and support).

All speakers appeared to be of non-Hispanic white race and ethnicity and therefore do not represent the demographics of the general population. However, we judged that the videos still provide valuable insights, and the homogeneity of this sample is somewhat balanced by the diversity of the interview samples discussed in the next section. Table 1 describes the speakers’ gender, race, tracking role, employment, and diagnosed illnesses. The mean tracking interval that the speakers referenced was 3 years. The average video length was 13 min 50 seconds. QS videos are denoted with Q# identifiers in the quotes highlighted in the results.

We enrolled 20 adults with type 2 diabetes and 20 mothers of children aged 12 years and older with asthma from among the patients of Group Health clinics. Table 2 describes participant demographics. Interviews ranged in length from 45 to 90 min with an average of about 60 min. For diabetes cohort participants with an informal carer, we invited them to participate in the interviews if possible. Diabetes cohort participants have D# identifiers in the quotes highlighted in the results. Asthma cohort participants are denoted by A# identifiers.

The primary themes emerging from our analysis consist of 2 parts—actors and work. The types of actors and work are summarized in Textbox 1.

We have discussed each actor and type of work, supported with examples from our analysis.

Actors are the person with chronic illness, informal carers, health care providers, and community members. These actors interact with each other and can all perform aspects of work, as described below. This definition extends beyond people included in the models of personal and health informatics from Li et al [11], Epstein et al [16], and Swan [42].

The types of tracking work include communication, information, collection, integration, reflection, and action. These types of work are similar to the stage-based model [11] but add communication work (to incorporate interactions between actors) and redefine preparation to information. Perhaps most important for the mechanics of work processes, we observed that unconstrained transitions described the structure of tracking for chronic illness management better than discrete stages. This reflects both the continuous and social natures of work revealed in our datasets. As also described by Epstein et al [16] and Costa Figueirido et al [14], our analysis showed that different types of work can occur simultaneously. Furthermore, any actor can engage in any work, and actors often collaborate or hand off work. We have discussed each type of work, dependencies, and workflow.

To bridge the gap between current informatics models and important characteristics of tracking for chronic illness management, we proposed the CoMSHI (pronounced com-she; Figure 1). The CoMSHI is based on insights from previous research (eg, [11,14,20,21,24,42,47]) and new data analysis about tracking behavior and social interactions. It portrays the actors and work, described above, that drive successful tracking in support of chronic illness management. Actors perform work in no particular order, and work can be ongoing and overlapping. All actors may engage in work and communicate around that work.