Designing an artefact to help users make intervention decisions about their wellness

Abstract

Healthcare systems have been evolving towards more decentralised, patient- empowered, and holistic approaches. This places a greater expectation on patients to monitor and report changes in their general wellness so they can make decisions as to when to seek clinical interventions. However, findings from this study suggest individuals find it challenging to detect deteriorations in wellness, due to the vast and multifaceted nature of the concept, the gradual onset of symptoms, and the difficulty in articulating change. Thus a mobile application is developed to help users with these issues. The design of this mobile application draws upon existing cognitive neuroscience research on change detection, both for external stimuli and internal ‘interoceptive’ sensations. This highlights several key factors to be considered, if wellness-related decision-making is to be supported. In particular, this identifies the role of patients’ top-down (attentional) and bottom-up (less-voluntary) processes for detecting wellness deteriorations.

Keywords:

• Wellness
• design science
• change detection
• neuroIS
• prototype development

Introduction and problem definition

Healthcare systems around the world are facing capacity-related challenges in the near future due to ageing populations (Anderson & Hussey, 2000; Schneider & Guralnik, 1990) and the increasing prevalence of issues such as obesity (Wang, McPherson, Marsh, Gortmaker, & Brown, 2011) and diabetes (Zhang et al., 2010). This has resulted in several key trends, including: (i) movement from a disease-centred model to a patient-centred model; (ii) an emphasis on quality rather than quantity across healthcare enterprises (iii) movement of patient care out of the clinic and into homes and communities; and (iv) a shift in focus from curbing illness to facilitating healthy living (Blatt, Crounse, & Wilson, 2012; Gianchandani, 2011; Saltman, Bankauskaite, & Vrangbaek, 2006). These trends place a decision-making burden on patients to participate more mindfully in their care and to report deteriorations before they become critical (Balaban, Weissman, Samuel, & Woolhandler, 2008). As part of this, there is significant demand on IT to better enable knowledge management and communication among actors (Wills, Sarnikar, El-Gayar, & Deokar, 2010) and so increase patient empowerment and self-care (Ballegaard, Hansen, & Kyng, 2008; Barry & Edgman-Levitan, 2012; Haux, Ammenwerth, Herzog, & Knaup, 2002).

Coinciding with this move towards decentralised IT-enabled healthcare is increasing interest among healthcare professionals in the idea of patients’ ‘wellness’, a concept defined by the World Health Organisation (WHO) as ‘the optimal state of health of individuals and groups…with two focal concerns: the realisation of the fullest potential of an individual, physically, psychologically, socially, spiritually, and economically; and the fulfilment of one’s role expectations in the family, community, place of worship, workplace and other settings’ (Smith, Tang, & Nutbeam, 2006). This requires factors to be considered across a range of dimensions, including physical, emotional, occupational, spiritual, intellectual, social (Hettler, 1984), psychological, environmental (Roscoe, 2009), and economic (Osberg & Sharpe, 2002). Such a holistic and multifaceted view of wellness is intended to capture the continuum of states lying above the threshold of that which requires medical intervention (c.f. Clark, 1996; Lorion, 2000; Sarason, 2000). Thus, detecting changes in patients’ wellness is seen as a powerful enabler of early intervention decision-making, so increasing patients’ quality of life and reducing the need for centralised care (Carnethon et al., 2009; Jacobson & Gostin, 2010; Miller, Roehrig, Hughes-Cromwick, & Lake, 2008; Naydeck, Pearson, Ozminkowski, Day, & Goetzel, 2008; Schmidt, Voigt, & Wikler, 2010). Yet the actual detection of changes in such holistic and multifaceted concepts is often difficult for various reasons, including: (i) the computational complexity involved lends itself to flawed biases and heuristics (Kahneman & Frederick, 2002); (ii) individuals are limited by their own understanding of what is relevant (Simon, 1982); and (iii) many environmental or emotional factors that influence behaviour go unnoticed (Damasio, 1994).

This study adopts a design science approach (Hevner, March, Park, & Ram,2004; March & Smith, 1995) in which a software artefact is developed to address the problem of detecting changes in users’ wellness. The focus of the study is at the level of an instantiation (Gregor & Hevner, 2013; Purao, 2002). However, development is informed by a neuroIS perspective (c.f. Dimoka et al., 2010; Loos et al., 2010; Riedl et al., 2010) that draws on existing cognitive neuroscience research on change detection as ‘justificatory knowledge’ (Gregor & Jones, 2007). For clarity, the following sections are organised primarily around the nominal design science research methodology presented by Peffers, Tuunanen, Rothenberger, and Chatterjee (2007). However, it should be noted that the scientific focus of this study is on the design, rather than evaluation. This is a predictable characteristic when engaging with radically novel domains, wherein rigorous theorising often needs to be prioritised over rigorous evaluation (which may be impossible until design has stabilised and matured) (c.f. Iivari, 2007).

The next section defines the objectives of a solution using a series of interviews and focus groups designed to elucidate requirements. In particular, these interviews point to the need to address the subjectivity of wellness measurement, the difficulty in detecting deteriorations that emerge gradually, and users’ lack of patience for cumbersome measurement. Design and development then builds on kernel theory from cognitive neuroscience that identifies cognitive load and focus as key moderators of top-town change detection and punctuation and noise as key moderators of bottom-up detection. A mobile application is developed that targets these key moderators by integrating validated wellness questionnaires in a novel design process that uses initial recordings as a baseline for future measurements. We conclude with a discussion of findings from the demonstration and evaluation of the artefact.

Gathering requirements for IT-enabled wellness change detection

Description of informants

Requirements gathering took place over an 11 month period from January 2014 to November 2014. Sampling sought to target individuals for whom the issue of detecting changes in wellness was likely to be most pronounced. This meant engaging with clinical healthcare professionals (to whom deteriorating individuals would be referred) and non- clinical individuals likely to be responsible for detecting changes in wellness (e.g. family members caring for loved ones). Requirements were collected through: (i) 8 initial in-depth personal interviews with key informants (some were interviewed twice); (ii) 5 focus groups with experienced family and clinical carers; and (iii) 4 supplementary interviews with longer-term family carers and senior healthcare professionals (see Table 1). All interviews and focus groups were digitally recorded and memos were taken during and after requirements gathering.

Table 1. Requirements gathering through interviews and focus groups.

Technique	Description of Participants	Num. of Participants	Contact Hours
Interview 1,2	Executive US National Council on Ageing, expert in family caregiving & evidence based practices in health	1	1.5
Interview 3	Senior Officer, e-Health and Innovation, WHO	1	1
Interview 4, 5	Head of Innovation, Carer Organisation, Ireland	1	3
Interview 6	Senior Manager, Carer Association, Ireland	1	1
Interview 7, 8	Carer and Training Officer, Carer Organisation Ireland	1	2
Focus Group 1	Experienced carers in Midwest of Ireland	4	2.5
Focus Group 2	Clinic Nurses in Carer Organisation, Midwest Ireland	2	2
Focus Group 3	Experienced Carers, Midwest of Ireland	6	3
Focus Group 4	Clinic Nurses in Carer Organisation, Southwest Ireland	2	1.5
Focus Group 5	Experienced Carers, Southwest of Ireland	4	2.5
Interview 9	Experienced long-term Carer A	1	1
Interview 10	Experienced long-term Carer B	1	1
Interview 11	General Practitioner	1	1.5
Interview 12	Senior Geriatrician, South of Ireland	1	1
		27	24.5

Key requirements identified

The first major finding from requirements gathering was widespread consensus that some form of IT solution is needed to assist the detection of changes in individuals’ wellness. One senior clinical professional noted that ‘what we need are tools to help people to manage their own wellness and to prioritise their needs’. Interestingly, this was not just a question of assisting in the detection of the change on the carer-side but also in appropriately communicating that change to clinical professionals. Clinical professionals readily admitted ‘carers have a lot of information that could be shared and leveraged, but their information is not considered valuable’ and ‘a big part of wellness management is managing interactions with healthcare professionals…it would be great to bridge the non- clinical and the clinical’. Carers agreed with this sentiment. For example one family carer argued that ‘you will be asked if you are a nurse and if you are not, you are not supposed to make a comment’ while another complained ‘what has grown into the healthcare system is that doctors only want to deal with the patient’. More clinical-style data from carers was proposed as a solution to this several by clinical healthcare professionals, e.g. one expert explained that some way of measuring and tracking wellness would ‘formalise the wellness information that carers hold and they would then be better positioned to share it’.

The second major finding was the sense that changes in patients’ wellness were routinely going unnoticed. In part this was due to their subjectivity, e.g. a clinical healthcare professional commented

sometimes carers don’t have information or rules to guide them on how to care for someone…if a patient is going downhill, the carer needs some information/direction regarding what they should do…and it’s complex because wellness is a subjective thing which can be influenced by your perspective

It was also attributed to the gradual onset of deteriorations over time, meaning carers struggled to pinpoint their origin, e.g. one carer lamented that ‘your memory can play tricks on you and when the doctor says “how long is this going on?”…you don’t’ think it’s so long, but it might be longer’. This often meant that serious conditions snuck up on carers without manifesting seemingly out of the ordinary symptoms. This is typified by the following anecdote:

I left my husband one day to go to the chemist…only down as far as the square and back. When I came back the person who was looking after him was holding his hand and he was after having a mini TIA [Transient Ischemic Attack]. And while I was out, I met a woman who asked me how is [her husband] and I said ‘sure he’s grand’…. she must have thought that I’m a lying b**** because 15 min later, there was an ambulance outside my door

The third major finding was the importance of ease of use for any proposed solution. Carers and clinical professionals noted that they didn’t have large amounts of time and effort to invest, meaning there was some passive resistance (c.f. Lapointe & Rivard, 2005) towards any additional processes. Some carers took this further by suggesting they might prefer not to know about specific deteriorations, as there is often nothing they can do to address them. One carer whose husband was terminally ill commented ‘we all know what is going to happen so why bother recording wellness information’.

Design and development

Kernel theory: the neurology of change detection

The topic of change detection has received significant attention in cognitive neuroscience, both from the point of view of detecting external sensory changes as well as detecting internal or ‘interoceptive’ changes. Each of these is important for detecting changes in wellness and enabling decision-making; the former for onlookers and carers seeking to identify deteriorations in others, the latter for individuals seeking to self-identify deteriorations.

In terms of detecting external sensory changes, the dominant stream of literature engages with a phenomenon referred to as ‘change blindness’ (c.f. Rensink, 2002; Simons, 2000; Simons & Rensink, 2005), whereby novel stimuli are introduced without entering individuals’ conscious awareness. This lack of awareness can occur for surprisingly large changes, particularly where the introduction of novel stimuli coincides with other visual disruptions (e.g. Beck, Rees, Frith, & Lavie, 2001; Pessoa & Ungerleider, 2004). Examples of experimental paradigms that explore change blindness include ‘mudsplashes’, where small high-contrast shapes are introduced to draw attention away from other changes (e.g. O’Regan, Rensink, & Clark, 1999), as well as ‘flicker’ tasks, where scenes are modified between brief displays of blank screens (Rensink, O’Regan, & Clark, 1997).

The neurology responsible for sensory change detection involves both subcortical components involved in early stimulus processing, e.g. the superior colliculus in visual tasks (Cavanaugh & Wurtz, 2004), as well as a distributed multimodal cortical network responsible for attention, involving areas such as the temporoparietal junction, the insula, the cingulate, and areas of the pre-frontal cortex (Beck et al., 2001; Downar, Crawley, Mikulis, & Davis, 2000). The neurological process of sensory change detection is typically measured via electroencephalography (EEG) through the ‘P300’, a positive event related potential occurring parieto-centrally 300 ms or more after stimulus presentation (c.f. Linden, 2005). This signal illustrates correlated activity in perceptual and conceptual regions (Koivisto & Revonsuo, 2003) and contributes towards a consensus that sensory change perception involves both top-down attention processes and bottom-up less- voluntary processes (Corbetta & Shulman, 2002; Jiang, Summerfield, & Egner, 2013; Sarter, Givens, & Bruno, 2001).

In terms of detecting internal interoceptive changes, existing research identifies a similar relationship between perceptual and conceptual systems. Conceptually, a multimodal network for interoception is also thought to involve the insula, the cingulate, and areas of the pre-frontal cortex (Craig, 2002; Critchley, Wiens, Rotshtein, Öhman, & Dolan, 2004; Singer, Critchley, & Preuschoff, 2009). Perceptually, interoception is thought to take input from a system of somatosensory afferents distributed around the body and skin (Khalsa, Rudrauf, Feinstein, & Tranel, 2009; Olausson, Wessberg, McGlone, & Vallbo, 2010). The effect of this interoceptive system is more than simply detecting pain. Rather, input from this collective system acts to create awareness of body states, which consequently manifest as conscious feelings (Craig, 2002; Damasio & Carvalho, 2013) and contribute to a general sense of ‘self’ (Northoff et al., 2006). However, similar to external sensory change detection, this is not solely a bottom-up process, rather top-down attentional and predictive processes also play a role in consciously appraising changes in personal conditions, such as health and neuropsychiatric illness (c.f. Seth, 2013).

All this supports a similar interplay between top-down and bottom-up process for both external and internal perceptions of deteriorations in wellness, i.e. individuals need to afford some attention to particular deteriorations, and those deteriorations need to become readily noticeable according to their sensory salience. This also suggests that specific factors are likely to influence the efficiency of common top-down cortex-based elements of change detection and bottom-up subcortical/somatosensory elements (see Figure 1).

Figure 1. Model of change detection.

The’ mudsplash’ experiments demonstrate that top-down elements of change perception require some degree of focus. However, such demand for focus in top-down cognitive processes also creates a vulnerability to cognitive load (Kahneman, 2011; Lieberman, 2007). This was illustrated in the ‘invisible gorilla’ experiments by Simons and Chabris (1999), in which subjects were asked to count the number of times a ball was passed during a basketball training session. While this training session was underway, an individual in a gorilla suit walked into the centre of the scene. Approximately half of the subjects failed to notice the gorilla, even when the individual in the gorilla suit beat their chest with both hands and looked directly at the camera. More recently, this was replicated in an expert domain when radiographers were asked to inspect a series of medical scans, the last of which included a large image of a gorilla 48 times the size of an average nodule (Drew, Vo, & Wolfe, 2013). Eye-tracking data revealed most radiologists looked directly at said image, yet 83% did not report seeing it. This suggests that enabling top-down wellness change detection requires attention for each aspect of wellness at some manageable level of cognitive load.

Bottom-up processes are susceptible to different moderators, due to their sensory nature (Lieberman, 2007). The ability of sensory information to create an involuntary response in the cortical areas responsible for conscious awareness requires a stimulus has sufficient punctuation to spike activity in sensory subcortical/ somatosensory areas above some typical threshold of activity (e.g. Itti, Koch, & Niebur, 1998; Buschman & Miller, 2007; Knudsen, 2007). This means that gradual exposure to internal or external stimuli can decrease bottom-up sensitivity, a neurological phenomenon often leveraged in exposure therapy (Craske et al., 2008; McNally, 2007). The threshold-based mechanism for bottom- up change detection also makes it vulnerable to noise when the salience of other stimuli is high (Kelly & O’Connell, 2013; Smith & Ratcliff, 2004). This is the same limitation exploited by pickpockets and magicians, who seek to overload onlookers with distractor stimuli to decrease the salience of concealed activities (Macknik et al., 2008; Martinez-Conde & Macknik, 2008). This suggests that enabling bottom-up wellness change detection requires interoceptive comparisons are made periodically between isolated aspects of wellness.

Building an artefact

A mobile application was developed to assist wellness change detection in outpatients. This application targeted each of the four mediating factors identified using the process illustrated in Figure 2.

Figure 2. Process model for measuring wellness.

Focus was addressed by integrating an existing collection of questionnaires for measuring different aspects of wellness developed by the Promis® organisation. These items have been validated in existing medical research (Cella et al., 2007) and applied across a range of fields including pain management (Cook, Buckenmaier, & Gershon, 2014), oncology (Mesa et al., 2013), and neuromuscular disorders (Meilleur et al., 2015), to name but a few (see http://www.nihpromis.org/science/PublicationsYears for an extensive list).

Questionnaires target each aspect of wellness with a specific set of reflective items (typically 8–10). Twelve of these sets were digitalized for use in the mobile application. This included one set for global health, three for mental health (applied cognitive abilities, anxiety, and depression), four for physical health (physical function, fatigue, pain interference, and sleep disturbance), and four for social wellbeing (satisfaction with social roles and responsibilities, informational support, instrumental support, and emotional support).

Cognitive load for the total set of 98 items was addressed by creating a baseline measurement from users’ first recording, so that successive measurements could minimise the number of responses required. These successive recordings substituted the full set of reflective items for a single high-level question, e.g. ‘Have you noticed any change in your level of physical function since you last recorded?’ Users answering ‘yes’ were presented with the full set of items for that aspect of wellness to identify the nature of the change. Users answering ‘no’ were presented with an alternative follow-up question, designed to invoke bottom-up change detection and ensure users were not simply overlooking deteriorations.

Punctuation for bottom-up change detection was introduced in the alternative follow-up question by presenting a single item from the category to users, along with their previous answer for that item. The purpose of this step was to encourage users to relate their current and previous state by drawing on their visceral and emotional response to the comparison. Users who chose to repeat their previous answer skipped the detailed set of questions and progressed to the next aspect of wellness, while their previous set of detailed answers for that aspect of wellness was repeated automatically for their wellness recordings. Users opting to change their answer were presented with the full set of items for that aspect of wellness, again to identify the nature of the change.

Noise was minimised for bottom-up change detection by presenting these punctuated comparisons only as a safety check when a user indicated no change for some aspect of wellness, and only for one question. This was done to prevent the sensory recalibration that may occur if excessive comparisons were made. Figure 3 illustrates a screen shot of a comparison.

Figure 3. Screenshot from measuring wellness.

This process reduced the number of items users were required to answer from 98 down to a potential minimum of 24 (provided no changes were observed). This was intended to allow users to measure wellness more often, with fewer time demands and less cognitive burnout. Ongoing readings could be viewed at any time; either in spreadsheet output, in graph format, or via alerts provided if a user’s score was five or more standard deviations outside the baseline score established by Promis® (see Figure 4).

Figure 4. Screenshot from viewing wellness trends.

Demonstration and evaluation

The mobile application was demonstrated to local carer groups, of whom 12 signed up for participation in alpha testing over several months. While this evaluation was qualitative and informal, three key findings emerged to inform future design iterations. The first finding came from reports of significant utility from a handful of users. Examples of feedback from different users include

I thought it was fantastic…it makes you think about things…when I did the sleep

questionnaire I realized that I never sleep

It’s definitely good…it tells you to do something and I am working towards changing a few things to get more support in order to alleviate the alerts that I am getting

I’ve got the same alerts twice, but I’ve made a few changes and next time I measure

wellness, I want to see a change

it’s great to have something that says ‘be careful here’

Several participants also reported making the decision to initiate interventions based on measurements, such as reviewing pain medication and organising social outings. Several carers also used the application to monitor their own wellness. One such individual, a full time carer for her ill husband, noticed deteriorations in her personal depression and anxiety and this motivated her to arrange counselling. She intended to show the wellness alerts to her husband’s specialists because she wanted them to understand ‘this is the state I’m in and I’m trying to be his carer…I’m the one they are releasing him to’.

The second finding was a reluctance to engage with the application among the remaining users. Of the 12 users to date, 6 have not entered more than the initial recording made during participatory demonstration. There are three possible explanations for this. Explanation one is simply that the application is inhibited by ease-of-use. This explanation would be more compelling if users had fallen away after two or three recordings, rather than simply not returning to use the measurement system at all. Explanation two is that the application is only suited to certain types of users. No obvious discriminatory factors have been identified to differentiate between adopters and non-adopters; however this remains open to future investigation. Explanation three is that all users must push through a seeming initial lack of usefulness before discovering the actual value of the system. This makes sense, given that the application is designed to assist users in identifying unknown problems (meaning a priori they may believe no problems exist, therefore have no motivation to use the application). This was confirmed by several users who explained their subsequently non-adoption because they ‘didn’t expect to see change’.

The third finding of this evaluation exercise was the requirement not only to detect deteriorations but also to prescribe interventions. Such a requirement is not surprising in itself; however the level of readiness (and confidence) to initiate autonomous care interventions was higher than anticipated and became salient immediately once users saw the possibility of independently identifying and characterising changes.

Conclusions

Clearly, the artefact is in need of ongoing development and more systematic evaluation. Nonetheless, the current work presents several contributions. First, observations from the requirements-gathering focus groups and interviews demonstrate the need for a decision support system(s) in the area of wellness-detection.

Second, a level 3 design theory (i.e. an embedded causal model linking different outcomes and behaviours, c.f. Gregor & Hevner, 2013; Purao, 2002) is presented that models individuals’ change detection, based on the synthesis and integration of established streams of research in cognitive neuroscience. This model enables decision support systems to be developed for personal wellness management by presenting actionable constructs, causally linked to wellness-related change detection. Thus, designers have a set of ‘general components’ (Baskerville & Pries-Heje, 2010) for a range of designs in this area.

Third, a process model is presented that builds on this actionable causal model, thus providing a reusable level 2 design theory (abstract principles that describe the artefact). These can be drawn upon to inform design across varying domains and instances, e.g. wellness tracking for the elderly, people with disabilities, and people recovering from major trauma.

Fourth, a level 1 software artefact is presented that instantiates abstract concepts into a working prototype, thereby demonstrating both ‘how’ this instantiation may take place, as well as ‘that’ it may take place (Gregor & Jones, 2007).

Finally, the evaluation of the software artefact supports the utility of the approach, as well as two subsequent emerging requirements. The first requirement is the need to facilitate and inform carer-related decision-making around wellness interventions, not simply identifying deteriorations. The most promising approach for this appears to be the integration of an online community that allows individuals to discuss issues and share experiences in approaching different wellness problems. The addition of such a community may also act as a form of ‘persuasive technology’ (c.f. Fogg, 2002) by functioning as a social actor that encourages users to measure wellness. This may also help with the second emerging requirement, which is to overcome initial resistance of some users to engage with the system due to their lack of awareness of wellness problems.

Disclosure statement

No potential conflict of interest was reported by the authors.