Alerts, prompts, and cues are all stimuli designed to prompt desired actions and reactions from users. We use the term “trigger” to refer to this broad category of digital stimuli. Triggers are especially important when developing interventions aimed at engaging people in desired therapeutic activities in a nonclinical setting characterized by an abundance of competing internal and external events. Designed to prompt an internal or external reaction, these stimuli give salience to an internal or external goal by focusing an individual on the desired goal at the appropriate time. For example, placing an item near the door that we will pass on our way out or a reminder from a loved one to take an umbrella as it is about to rain is a useful environmental engineering trigger that helps us to better prepare for a desired outcome (eg, to not get wet) [6]. Digital triggers include calendars, push notifications (both standard and in-app), texts, images, haptics, and other types of digital alerts, most often delivered through a mobile phone, watch, or computer.

There is growing empirical evidence supporting the use of triggers to enhance the effectiveness of health interventions or as stand-alone health interventions [7]. Evidence suggests that digital triggers improve individuals’ engagement in interventions with a specific target, such as appointment adherence, medication adherence, homework completion, and engagement in medical and psychological treatments [8-12], especially when compared with no treatment control conditions [8]. For example, Spohr and colleagues [13] found that individuals increased their physical activity immediately after receiving a push notification compared with individuals who solely completed a mobile self-report questionnaire. Despite the potential of triggers to foster engagement, however, reviews have suggested that the evidence has been mixed due to the heterogeneity in populations, delivery types (eg, email, phone), characteristics of the target itself, and different outcome measures used [14-17].

There is a robust body of literature on the use of alerts, primarily short message service (SMS), otherwise known as text messaging, and email as an intervention in itself. Meta-analyses and reviews have highlighted the effectiveness of digital triggers as interventions for smoking cessation, alcohol use reduction, and prenatal care [8,11,18]. Interestingly, a recent meta-analysis revealed that adding other components (eg, a website) does not significantly improve pure alert-based interventions for health behaviors [8]. However, there have been mixed results regarding the efficacy of pure digital trigger-based interventions for weight loss and physical activity. Results suggest that trigger-based interventions that are designed to increase education about a topic (eg, proper prenatal care) or those that are designed to reduce impulsive responses or increase psychological well-being may be most effective because they increase goal salience in one’s natural environment at the right time [19].

Despite the promising outcomes of alert-based interventions and the prevalence of these alerts and triggers in daily life, little attention has been paid to dismantling triggers into their component parts to deepen the understanding of how they function to improve physical and mental health outcomes. While not targeting health outcomes directly, the marketing community has embraced digital triggers as core components of engagement in products and services [20]; therefore, we will draw on findings from marketing research where applicable. Our goal is to present the core components of triggers so that those developing digital health interventions can understand how to tailor those components to best engage health intervention users in the context of their environment and daily lives.

Tailoring is the method of personalizing an intervention with respect to characteristics such as content and timing to ensure an intervention’s highest level of receptivity and engagement. Most reviews and meta-analyses of text message-based interventions have indicated that tailoring interventions to the individual, particularly when also targeting a specific condition, yields larger effect sizes than not tailoring [8,11,12,19,21]. Although the health intervention itself is typically tailored according to components of behavior change theories, with readiness or importance of change being the most frequently used [5,12,22], nearly every component of a digital trigger (ie, content, frequency, interactivity) can be tailored to increase engagement.

As mechanisms studies have revealed, tailoring can improve engagement by increasing self-referential and heuristic processing [23,24] and attention to a desired stimulus, and reducing effortful processing. Accordingly, the more that items are tailored, the more receptive an end user may become. For example, evidence from both the health and marketing literature has shown that tailoring images to end users’ demographic characteristics increases engagement [25,26] along with other strategies that match user characteristics with a persuasive strategy. Additional components, such as personalization by name, time of a trigger, end-point goal, frequency, current state, or location of receiver, can be tailored [27]. Nowhere is tailoring more evident than in the targeted advertisements on social media pages. For instance, Facebook uses 98 different data points, ranging from simple demographics (eg, gender) to life events (eg, just married) and consumer preferences (eg, preferred types of restaurants), to enable advertisers to tailor their message to specific audiences [28]. Therefore, we consider tailoring as an overall framework when developing trigger-based interventions. Below we describe each trigger component and, where possible, its relationship to other components. Figure 1 presents the concept of baseline and adaptive trigger tailoring within the context of product and engagement planning; Table 1 provides an explanatory overview of the digital trigger components described in the body of this work; and Table 2 offers a summary of our examination.

There is substantial literature discussing the impact of message source credibility on increasing engagement, receptivity, and actual behavior change [29]. The source of a digital trigger can be a provider, coach, peer, friend, individual, group, blog, online publication, mobile app, sensor, or anything that can be imagined by the receiver. Subtleties embedded in the message source can alter the impact and persuasiveness of an intervention, which highlights the need to carefully consider issues of framing. For example, one study revealed that visitors to a health information webpage were more likely to take action toward change when the information was on an official webpage rather than on a blog or personal page, regardless of the content [30].

Throughout the literature, there is evidence that (1) source credibility affects recipients’ perception of delivered content [31]; (2) anonymous reviews are less credible than personally sourced reviews [32]; (3) messages from sources that are demographically closer to the receiver are more salient, and messages from people most important to the receiver are more likely to be viewed [33]; (4) messages are more persuasive if the sender is perceived as an expert before the message is sent [34,35]; and (5) digital interventions are enhanced when they are coupled with human support [36], possibly because of the combination of a human touch and the expectation of the individual’s accountability [19].

It is worth noting that the receiver’s attribution of the source might often be as powerful as the actual source of the message. Data show that people apply social rules such as politeness and reciprocity to computers [37], attribute humanistic characteristics to robots based on the latter’s responsiveness [38], and develop a therapeutic alliance with self-help programs [39]. In a study conducted in a methadone clinic, participants responded to automated text messages as if it were a social interaction; they thanked the system for sending the message, despite having been informed that the messages were automated [40]. In this case, the participants’ perception of the sender was more salient than the actual source. Perception can also be influenced by unknown source triggers where there is social reciprocity or perceived social acceptance, such as with friend request alerts or other social media notifications. Taken together, it is clear that understanding and optimizing both the user’s perception and the actual message source can have a substantial impact on trigger receptivity beyond the content. This is particularly true when the goal is to engage an end user in an actionable step that requires effort or burden.

The “how” relates to the means by which the trigger is presented and sent. Here, we divide the “how” into 2 different concepts: stimulus type, which refers to how the trigger is presented, and delivery medium, which refers to how the trigger is sent.

The “when” refers to a certain point in time when the recipient is able to pay attention to a push notification and exert the desired amount of effort required to attain his or her goal. It derives from the concept “just-in-time,” which focuses on the individual’s receptivity within the context of ongoing tasks and daily routines [52]. People make quick decisions about how they attend to notifications as a combination of “Why am I receiving this?” and timing [53]. Although triggers can be disruptive, the recipient accepts them because they contain useful information (see “What” below). Nevertheless, adequate timing enables us to more effectively engage and support people when there is an opportunity for positive change [54,55]. Studies have repeatedly shown notifications to be effective in increasing people’s immediate engagement across a broad range of health issues such as medication uptake (eg, [56-58]), attending scheduled appointments (eg, [59,60], and brief exercise [19]; these timely reminders help people to remember to engage in beneficially perceived actions during their daily routine [61]. Most trigger timing is based on self-selected alert times and general pattern awareness (eg, medication in the morning).

People’s capacity to pay the desired amount of attention to a notification is influenced by the ongoing task they are engaged in when they receive the trigger. Mehrotra et al showed that perceived disruption increases as the complexity of an ongoing task increases [62]. In this study, mobile notifications were perceived as the most disruptive if they arrived when the user was in the middle of a task or just finishing it, and least disruptive if the user was idle or starting a new task. After work, when traveling, and at leisure were the tasks during which the users perceived notifications to be the most disruptive. However, knowing the target audience enabled the intervention developers to identify the right time windows for sending out triggers.

Similarly, MailChimp, an email marketing company that was sending billions of emails per month by 2014, analyzed the results of their time optimization system in terms of recipients’ subscription following email distributions [63]. They came to the broad conclusion that the optimal time to send emails peaked at 10:00 AM in the recipients’ own time zone between Monday and Friday (eg, work days). This finding is in line with another analysis presented by Localytics, a marketing company [64]. When categorizing the response rates for college students and bartenders, however, the optimal send time peak shifted to 1:00 PM; for neonatal nurses, who work in shifts, the results were not consistent. In effect, people tended to be more receptive to email notifications several hours after waking, but this time varied based on people’s occupation. Baumel and Schueller [65] presented similar results showing that women with perinatal mood disorders preferred to receive services late at night, when clinicians were not available, due to the fact that they were awake and had pockets of availability at that time.

These empirical findings suggest that understanding the daily routines and availability of the target audience is key to planning the trigger time, because it enables the intervention developer to make sure that the trigger is received within the right context. Research also indicates that (1) relying on reminders supports repetition, but hinders habit development; (2) evidence-based cues (eg, sending data after lunch) increase automaticity; and (3) users prefer to customize and schedule alerts to support individual aspects of their daily routine [61,66] and their relevance. For example, Muench and colleagues [67] found that individuals wanted different behavior change techniques at different moments in the change process (eg, social support messages during relapse vs cognitive reframing during craving or prelapse states).

In the future, the rise of passive data collection systems via wearable, phone-based, and external sensors will provide significantly more opportunity to deliver just-in-time adaptive contextual digital triggers to end users when they need them most. While these systems will eventually allow for greater precision in triggering users at the highest levels of receptivity based on context, timing, and data from ongoing monitoring [68,69], they require significant learning. Thus, to benefit users in the short term, intervention developers should offer end users more flexibility in personally customizing their alerts. However, customization should not sacrifice best practices of when a message might be most effective. Consequently, guiding users to apply these alerts in certain time windows based on current knowledge is recommended.

“How much” refers to the frequency of triggers during the intervention phase. The question is whether sending out triggers more frequently increases users’ compliance with the trigger’s goal or results in trigger habituation, which eventually diminishes the reason for triggering. While “when” relates to the context of timing to maximize the trigger’s impact, “how much” relates to the impact of the trigger’s frequency on user behavior.

It is worth noting that the use of triggers in health interventions differs from the active utilization of a program. For example, there is extensive literature suggesting that, in Internet interventions, user utilization of the program results in better outcomes, implying a dose-response relationship [14,70-73]. However, this relationship can be mediated by the user’s level of motivation and persistence [74]. Triggers are different, in that sending out more triggers to the user does not equate to the user making a greater effort to engage with the intervention. Therefore, it is important for the trigger designer to determine the number and frequency of triggers that will best encourage the user’s efforts.

Determining this amount is a challenge, as data on the effect of a trigger’s frequency on health outcomes are sparse. A recent review of messaging interventions for preventive health revealed that sending more versus fewer messages resulted in better outcomes; however, this review was based on few studies, making it hard to draw conclusions about the effects of trigger intensity [11]. In a study on different interventions to increase fruit and vegetable consumption, Heimendinger and colleagues did not find any effect of tailoring from 1 mailing, but found that tailoring was effective in increasing engagement when participants received 4 mailings [75]. Another study considered variation in frequency by examining whether the frequency of text messaging feedback (1 vs 3 weekly messages) could affect smoking cessation [76]. While the timing of the first weekly text message feedback (sent to both groups) was event based—sent out to individuals when they were most receptive (ie, after sending out an assessment)—the timing of the other 2 messages, sent only to the 3-weekly messages group, did not consider users’ receptiveness; the messages were generally sent out on the 2 days following the initial message. No differences were identified between these 2 conditions. It is possible that the lack of “when” considerations for the subsequent 2 messages affected the outcomes, highlighting how the interdependency of components may affect a trigger’s effectiveness.

While research has not directly examined the habituation to triggers in patient-centered health interventions, data from related fields suggest that an uncontrolled increase in the frequency of triggers results in a decrease in the triggers’ effectiveness. Studies examining responsiveness to alerts among health providers have identified an alert fatigue: a decrease in the desired response to the alert due to its excessive appearance [77]. For example, in a literature review on physician response to drug safety alerts, van der Sijs et al found that these alerts were overridden by clinicians in 49% to 96% of cases because they receive so many irrelevant alerts [78]. When this occurs, the impact of all triggers becomes compromised. In a randomized study, Baseman et al [79] enrolled health care providers to receive public health messages and examined message content recall rates. The authors found that, for every increase of 1 local public health message per week, there was a statistically significant 41.2% decrease in the odds of the health care provider recalling the content of the study message.

Although direct empirical evidence on the impact of trigger frequency can be useful, looking more broadly into the research literature suggests that the frequency of triggers should take into account the user’s experience of importance and readiness for change. For example, people who reported that it was more important to change their drinking behavior preferred to receive more messages than those with lower importance scores [80]. Since readiness for change affects the perceived importance of the health intervention (eg, [81-84]), it results in higher adherence to treatment, engagement with treatment, and response to suggestions made by clinicians across behavioral health issues (eg, [85-87]). Accordingly, the frequency of triggers should be modified based on the perceived importance of the intervention in the user’s life along with other factors. Certainly, it is also possible to simply ask end users about their frequency preferences.