From Assessment to Intervention: Leveraging Ecological Momentary Assessment (EMA) to Develop a Personalized mobile-health (mHealth) Ecological Momentary Intervention (EMI) for Young Adults With ADHD and High-Risk Alcohol Use

Abstract

Ecological momentary assessment (EMA) can be a powerful and flexible tool for collecting data on alcohol use, particularly to understand proximal precursors and consequences. EMA can also be leveraged both to inform the development of and to deploy mobile-health (mHealth) interventions. This article describes the development of an mHealth ecological momentary intervention (EMI) for young adults with high-risk alcohol use and attention-deficit/hyperactivity disorder (ADHD). This novel intervention uses EMA as an intervention component to increase self-awareness via symptom monitoring. It also incorporates additional EMI components, including personalized feedback and behavioral strategy suggestions (“tips”), which operate synergistically with EMA questions and are tailored by EMA data. The theoretical underpinnings of this intervention are described, and its distinct relevance for young adults with ADHD who engage in high-risk alcohol use, are discussed. The process of developing this mHealth EMI is detailed, including examining EMA data to generate intervention content, considering participant feedback through iterative pilot testing, and applying human-centered design methods with end users and community partners. Finally, practical considerations of this intervention approach are discussed, including unique benefits, key challenges, and exciting future opportunities.

Ecological momentary assessment (EMA) can be a powerful and flexible tool for collecting data close in time to when a phenomenon unfolds or is experienced in real life (i.e., outside of the laboratory), particularly to understand fluctuations and within-person, temporal associations surrounding alcohol use (Shiffman, 2009). EMA is also a valuable tool for intervening on alcohol use – both to inform the development of real-time and mobile-health interventions (mHealth), and applied as an intervention component in an ecological momentary intervention (EMI) – particularly for individuals with attention-deficit/hyperactivity disorder (ADHD) who tend to experience high-risk alcohol use. In this article, we describe the ongoing process of developing an mHealth EMI for young adults that simultaneously addresses high-risk alcohol use and ADHD called TIPS (Training Inhibitory control using Personalized Strategies). We discuss its theoretical underpinnings, the EMA components of the intervention, how we have leveraged EMA data to design and modify the intervention through five stages of development, and practical considerations for EMIs and EMA-informed intervention development.

Background, Premise, & Theoretical Framework

High-Risk Alcohol Use Among Young Adults & Need for In-the-Moment mHealth Interventions

Alcohol use peaks in young adulthood (ages 18–25; Grant et al., 2017), particularly high-risk and problematic alcohol use such as heavy episodic or “binge drinking,” typically defined as 4 (female) to 5 (male) or more drinks in a single drinking session (Grant et al., 2017; SAMHSA, 2018) and high-intensity drinking (i.e., consuming 2–3 times these binge thresholds, Patrick & Azar, 2018). Young adults who engage in these patterns of heavy or high-risk alcohol use tend to experience a concerning number of alcohol problems, such as hangovers, blackouts, poor school and work performance, interpersonal problems, risky sex, disrupted sleep, and sleep disorders (Hasler & Pedersen, 2020; Lee et al., 2017; Park, 2004).

Up to 23% of drinkers aged 18–25 in the U.S. meet DSM criteria for past 12-month alcohol use disorder (AUD), which encapsulates both high levels of alcohol consumption and resulting consequences; this prevalence rate is higher than in any other age group (Grant et al., 2017). Despite this, few young adults with high-risk alcohol use perceive the need for treatment (SAMHSA, 2018; Venegas et al., 2021) nor actually receive treatment (i.e., only about 5% of those with AUD; Chen et al., 2016). Additional barriers include limited access like insufficient transportation, high cost, and the profound stigma surrounding alcohol use treatment (Schomerus et al., 2010; Schuler et al., 2015; Venegas et al., 2021). Removing this stigma, perhaps through mHealth interventions that offer greater access and privacy (Carpenter et al., 2020; Heron & Smyth, 2010), could make young adults more open to using supports like harm-reduction strategies for reducing their drinking (Charlet & Heinz, 2017). mHealth could also overcome stigma and other treatment barriers by targeting less stigmatized transdiagnostic risk factors upstream to alcohol use rather than solely targeting alcohol use directly, which could reach more young adults with high-risk alcohol use who would benefit from support but may not think they need help (Capron et al., 2018). Indeed, an emerging model of indirect interventions that target less stigmatizing, transdiagnostic risk factors upstream to the presenting condition has shown promise for both uptake and efficacy (Cuijpers, 2021; van Ballegooijen et al., 2024; van der Zweede et al., 2019).

High-Risk Alcohol Use & Unique Intervention Needs Among Young Adults with ADHD

ADHD, which is characterized by enduring and impairing symptoms of inattention (e.g., forgetfulness, distractibility, difficulty focusing), hyperactivity (e.g., fidgeting, restlessness), and impulsivity (e.g., interrupting others, impatience; APA, 2013) is a well-established risk factor for high-risk alcohol use and AUD (Charach et al., 2011; Groenman et al., 2017; Lee et al., 2011). For instance, despite similar rates of alcohol use, young adults with ADHD experience more alcohol-related problems than their peers without ADHD (Lee et al., 2011; Oddo et al., 2024; Rooney et al., 2012; Wang et al., 2021); some findings also indicate more frequent heavy episodic drinking (Garcia et al., 2020; Lundervold et al., 2020) and faster consumption (McKone et al., 2019), patterns of high-risk use that can have negative consequences. ADHD is prevalent, with 11% of youth aged 3–17 years in the U.S. having been given a diagnosis of ADHD by a doctor or other health care provider (Danielson et al., 2024), and it persists into adulthood, with fewer than 10% estimated to experience sustained recovery (Sibley et al., 2022). In conjunction, the unique challenges they face navigating the transition to adulthood (duPaul et al., 2021; Hechtman et al., 2016; LaCount et al., 2018; Wilens et al., 2018) make young adults with ADHD an especially crucial and prevalent target population for the prevention and reduction of high-risk alcohol use.

Tailoring treatment or support to the unique needs of individuals with ADHD is a key consideration, because traditional treatment models are not an ideal fit for this population. Stimulant medication is the first-line treatment for ADHD and is quite effective for ADHD symptom management (Wolraich et al., 2019), but it does not appear to reduce high-risk alcohol use (Molina et al., 2023; Molina et al., 2013; Thurstone et al., 2010), and most young adults stop using medication or use it only inconsistently (Biederman et al., 2019; Brinkman et al., 2020; McCarthy et al., 2009; Molina et al., 2009). Moreover, ADHD symptoms interfere with attending therapy (e.g., forgetting appointments, running late, trouble concentrating in session), and clients must independently remember to practice skills outside of therapy – a key challenge for those with ADHD. In contrast, individuals with ADHD tend to know what strategies to use but require real-time and intensive external support to remember when and where to use them (Barkley, 2015; LaCount et al., 2019). Individuals with ADHD also benefit from brief and frequent touches of support through their day-to-day lives to ensure consistent practice and help form generalizable habits across time and situations (Barkley, 2015; LaCount et al., 2018), which would extend to support to reduce drinking. mHealth tools offer an ideal solution to address these treatment needs, offering in-the-moment, in-context, and frequent support (Kennedy et al., 2024; Koch et al., 2021; Nahum-Shani et al., 2018) while conveniently leveraging the smartphones virtually all young adults use daily (Pew Research Center, 2024) including for behavioral health support (Wartella et al., 2016). Thus, we have endeavored to develop an mHealth intervention for young adults with ADHD.

Conceptual Framework for the TIPS mHealth intervention: Executive Functioning Model of ADHD and Alcohol Use

Prominent models of ADHD emphasize neurobiologically driven executive functioning deficits as key contributors to ADHD symptoms and downstream impairments (Barkley, 2012, 2015; Nigg, 2001; Sonuga-Barke, 2003). Importantly, these executive functioning deficits are also transdiagnostic risk factors for AUD and high-risk alcohol use that precedes it, as outlined in several etiological models such as the Research Domain Criteria (RDoC; National Advisory Mental Health Council Workgroup on Changes to the Research Domain Criteria Matrix, 2018; Sanislow et al., 2010), Addictions Neuroclinical Assessment (ANA; Kwako et al., 2016), Alcohol and Addiction RDoC (AARDoC; Witkiewitz et al., 2019), and the Etiologic, Theory-Based, Ontogenetic Hierarchical framework (ETOH; Boness et al., 2021). Thus, to support individuals with ADHD and high-risk alcohol use, our TIPS intervention model targets these joint executive functioning risks to help reduce both in-the-moment ADHD symptoms and high-risk alcohol use.

Three Core Executive Functions

Clinically relevant expressions of executive dysfunction describe “top-down” (requiring intentional thought and effort) cognitive and behavioral processes that involve directing thought and/or action toward oneself to pursue future goals (Barkley, 2012; 2015; Diamond, 2013; Sonuga-Barke, 2003). Informed predominantly by Barkley’s executive functioning theory of ADHD (Barkley, 1997; 2012; 2015) and supported by addictions models (Boness et al., 2021; Kwako et al., 2016; Sanislow et al., 2010; Witkiewitz et al., 2019), three core executive functions are thought to underlie ADHD and high-risk alcohol use: self-awareness, inhibitory control, and working memory.¹ Self-awareness is defined as the direction of one’s attention onto one’s self, thereby creating recognition of one’s own internal states and behaviors (Barkley, 2012; 2015) and effective self-monitoring of behavior (Sonuga-Barke, 2002). Inhibitory control is conceptualized as the ability to stop, prevent, and/or modify a prepotent, dominant, or automatic response in favor of a more adaptive one (Antshel et al., 2014; Barkley, 2012; 2015; Diamond, 2013; Nigg, 2017). Inhibitory control encapsulates both the inhibition of behavior, typically measured by response inhibition and/or impulsivity, and the inhibition of thoughts and attention, typically captured by concepts like interference control, selective attention, and cognitive inhibition (Barkley, 2015; Diamond, 2013). Working memory is the mental maintenance and manipulation of information (both visual-spatial and verbal), including both past and future experiences and how they relate to one another, allowing self-directed actions and planning to unfold across time toward a goal – in other words, extending one’s experience beyond the “here and now” (Baddeley, 2012; Barkley, 2012; 2015; Diamond, 2013; Kane & Engle, 2002). Together, these three executive functions are thought to operate synergistically; a breakdown in one entails deficits in the others, resulting in downstream symptoms and impairments of ADHD and high-risk alcohol use (Barkley, 2012; 2015; Diamond, 2013). For instance, self-awareness is thought of as the precursor to other executive functions (Barkley, 2012; 2015); by necessity, one must first be aware of a behavior in which they are engaging or about to engage (e.g., interrupting) to inhibit it and consider an alternative behavior. Likewise, in order to hold information in mind about the past and/or the future and act on it (working memory), one must inhibit responses or thoughts that disrupt those mental representations (Diamond, 2013; Kane & Engle, 2002). One must also hold their future goal(s) in mind in order to guide their current behavior and inhibit certain behaviors in favor of others that will help achieve that goal (Diamond, 2013).

Executive Functioning in Relation to Alcohol Use.

Transdiagnostic models of AUD have implicated all three of these executive functions in various drinking behaviors and AUD more broadly (Boness et al., 2021; Kwako et al., 2016; Sanislow et al., 2010; Witkiewitz et al., 2019). For example, low self-awareness and insight are associated with AUD and are thought to impair an individual’s regulation over their drinking, leading to over-consumption (Boness et al., 2021). Poor inhibitory control has been consistently shown to predict a range of high-risk drinking behaviors (Lee et al., 2019), including difficulty stopping a drinking session (Rooney et al., 2015)), drinking a greater quantity (Weafer et al., 2011), and experiencing more alcohol-related consequences (Pedersen et al., 2016). Further, working memory impairments co-occur with high-risk alcohol use, in part as a neurobiological effect of drinking that can then worsen one’s overall ADHD symptom profile and in turn perpetuate high-risk drinking (Kwako et al., 2016). As illustrated in Figure 1, our conceptual model focuses on these three core transdiagnostic executive functions as candidate treatment targets for both ADHD and high-risk alcohol use.

Figure 1. Conceptual Model of Executive Functioning, ADHD, and Alcohol Use Informing the TIPS mHealth Ecological Momentary Intervention.

Notes. Red arrows indicate acute processes (i.e., operating across moments and hours), and blue arrows indicate cumulative processes (i.e., developing over weeks and years). Double-headed arrows demarcate hypothesized bidirectional effects, and dashed lines among the executive functions contained within the gray box suggest that they are highly interrelated with one another, being necessary but not sufficient for the others.

Momentary Executive Functioning Processes in ADHD and Alcohol Use

Figure 1 highlights the value of EMA for assessing and intervening on the momentary processes within our conceptual model. These processes play out cumulatively over time on the order of weeks, months, and years (blue arrows); critically, they also play out acutely, on the timescale of seconds, moments, and hours (red arrows). For instance, lapses in executive functioning can manifest as ADHD symptoms, like interrupting someone who is talking, and in drinking contexts, being quick to have another drink without thinking through the potential consequences. These momentary processes can then spur acute impairments, such as strained interpersonal relations (in the case of interrupting) or missing school or work the day after drinking to intoxication (in the case of being quick to consume another drink). As these acute processes repeat across time and situations, they accumulate to produce persistent patterns of ADHD symptoms, high-risk alcohol use, and more serious impairments (e.g., being fired or failing school).² It is the momentary dynamics among these constructs that an mHealth EMI is poised to disrupt by acutely targeting executive functioning in the moments and real-world situations where they are needed to regulate alcohol use, ADHD symptoms, and their downstream impairments. As noted above, individuals with ADHD especially need support embedded into these acute experiences “at the point of performance” (Barkley, 2012; 2015). In addition to buttressing in-the-moment decisions and actions, mHealth EMI can also increase access to intervention relative to standard psychosocial treatments, including to those who are not in college—the majority of young adults with ADHD (Fowler et al., 2016; Hechtman et al., 2016; Iribarren et al., 2017; Kalbag & Levin, 2005).

In summary, the executive functioning model of ADHD informing our mHealth intervention posits that deficits in three core executive functions (self-awareness, inhibitory control, and working memory) underlie both the symptoms of ADHD and high-risk alcohol use, both acutely in the moment and cumulatively over time, making these upstream transdiagnostic risk factors excellent and destigmatizing candidate mHealth EMI targets to disrupt the maladaptive dual course of ADHD and alcohol use.

TIPS Intervention Model

Figure 2 shows the components of the TIPS EMI developed to address the three core executive function domains and, in turn, both ADHD symptoms and alcohol use. Briefly, TIPS is a 4.5-week intervention (including 5 weekends to maximally capture drinking) delivered multiple times throughout the day via smartphone that incorporates EMA surveys with personalized feedback on how target ADHD symptoms are changing, along with personalized behavioral strategy suggestions, or “tips,” including both drinking behaviors and ADHD symptoms (Figure 3).

Figure 2. TIPS mHealth Ecological Momentary Intervention Model.

Notes. Intervention components are listed in the yellow boxes. As in Figure 1, red arrows indicate acute processes (i.e., across moments/hours), blue arrow indicate cumulative processes (i.e., across weeks/years), double-headed arrows demarcate bidirectional processes, and dashed lines among executive functions suggest they are tightly interrelated and necessary for the others, but not necessarily sufficient. As such, we theorize these intervention components to operate synergistically. See Figure 3 for visual examples of each intervention component.

Figure 3. Overview of the TIPS mHealth Ecological Momentary Intervention (EMI) Components.

Notes. Pictures of the intervention components reflect simplified examples. The ecological momentary intervention (EMI) prompts are deployed 4x/day via smartphone at self-selected fixed times (wake time, afternoon, early evening, bedtime), in addition to self-initiated prompts during drinking episodes. The intervention length is 4.5 weeks (beginning on a Friday to capture 5 weekends and thus maximally capture drinking episodes). Before beginning the intervention, users select one target ADHD symptom on which to view graphical personalized feedback – a visual summary of their own self-ratings across time. Users may add a new target symptom each week. Behavioral strategy suggestions, or “tips,” are drawn from a menu of either general ADHD strategies that are applicable across situations or drinking-specific tips that are relevant surrounding drinking episodes. At the start of the intervention, users generate up to three of their own ADHD-general tips and three drinking-specific tips that will be shown to them over the course of the intervention. As displayed in the rightmost column, reminders of tips are also directly texted to users between each scheduled prompt and during drinking episodes so that users see a reminder of their tip between self-monitoring prompts, and also so that users receive a tip even when they have missed a prompt and thus may not be engaged in the self-awareness components. Prospective planning (when a new tip is given) and retrospective appraisal of tips use (after having the chance to apply a tip) are also facilitated by the types of EMA questions shown. In addition to the intervention components illustrated here, additional EMA questions capture information about current context, alcohol and other substance use, drinking intentions and motives, and affect, both for data collection and to tailor the content of a given tip to the user’s current circumstances. © 2025 by Traci Kennedy. All rights reserved.

Self-Awareness

TIPS employs two intervention elements to enhance self-awareness in individuals with ADHD. The first is symptom monitoring via EMA questions about ADHD symptoms throughout the day, which is meant to enhance awareness of the underlying executive functioning lapses expressed as ADHD symptoms. Users are prompted on their smartphones 4 times per day (and during drinking episodes) to rate the severity of their ADHD symptoms in the preceding few hours. They receive up to 3 additional reminder text messages within each prompt window. Once users indicate they have started drinking, they also receive automated reminders to complete monitoring surveys during drinking episodes. Intensively self-monitoring one’s behavior via repeated assessment can increase self-awareness acutely (in the moment) and cumulatively over time, and is an initial treatment component for many disorders to enhance symptom recognition and motivate behavior change (Bartholomew et al., 2016; McCarthy et al., 2015). EMA, although traditionally used as an assessment technique, can be used as an EMI to self-monitor in-the-moment symptoms and behaviors. In fact, symptom monitoring alone has been shown to help improve a range of behavioral health outcomes (e.g., smoking, McCarthy et al., 2015; alcohol use, Hufford et al., 2002; Schrimsher & Filtz, 2011; depressive symptoms, van Ballegooijen et al., 2016; anxiety/worry, Magnan et al., 2013; physical activity, van Sluijs et al., 2006). The presumed mechanism is that heightened self-awareness, facilitated by repeated self-reflection in EMA, triggers the use of strategies to mitigate symptoms (Barta et al., 2012; Darwin et al., 2013). In the EMA literature, this type of “reactivity,” where a respondent’s behavior changes due to repeatedly answering questions about the behavior, can interfere with the EMA process intended to purely observe and measure the behavior as it naturally occurs (e.g., Barta et al., 2012; Clifford et al., 2007; McCarthy et al., 2015; Rowan et al., 2007; Trull & Ebner-Priemer, 2013), but when viewed as an intervention, it is beneficial. Some findings suggest that such EMA reactivity only minimally interferes with the validity of its assessments (e.g., Cajita et al., 2023; Shiffman, 2009; Stone et al., 2003), though even small behavior changes can be a clinically meaningful component of a larger mHealth paradigm explicitly designed to increase self-awareness and modify behavior, particularly when conditions for behavior change are maximized in EMA (e.g., high frequency of monitoring, perceived desirability of the behavior, context-specific prompts at high-risk time points; Barta et al., 2012; Clifford et al., 2007; Shiffman, 2007).

The second intervention element that we deploy to enhance self-awareness is personalized symptom feedback. Consistent with social cognitive theory, feedback on behavior amplifies the impact of self-monitoring on behavior change (Barta et al., 2012; Kazdin, 1974; Suffoletto et al., 2015). Thus, to further boost the potential impact of TIPS, self-ratings of symptoms are provided back to the user in a visual line graph that charts their course across the intervention period. Feedback is likely crucial for young adults with ADHD who require external aids (e.g., visual feedback) to overcome difficulties with self-awareness deficits (Barkley, 2015; Shiels & Hawk, 2010). By viewing self-rated symptoms over time, the EMA data may be visually reinterpreted in relation to their responses over time. The goal is to increase awareness of their behavior in a given moment and to facilitate insight into temporal patterns and conditions under which their symptoms may tend to be better or worse.

Given the need for ADHD interventions to provide real-time support as described above, a goal of these two EMI elements is to help an individual realize when a different, more adaptive behavior is needed in the moment. This enhanced awareness of the need for inhibitory control should catalyze the use of the inhibitory control intervention elements described next (Knouse et al., 2015; LaCount et al., 2019). As such, raising self-awareness through symptom monitoring and feedback is viewed as a necessary initial component of the TIPS intervention model that sets the stage for the next components.³

Inhibitory Control

Next, to capitalize on increased self-awareness and directly improve in-the-moment inhibitory control, the intervention offers personalized behavioral strategy suggestions, or “tips.” EMA is again leveraged to tailor the content of these tips: responses on EMA questions about the user’s current context (e.g., location, time of day, whether drinking alcohol or not) and preferences drive the selection of a tip that is relevant to the individual’s circumstances from a standardized set of available strategies. The content of the tips is based on evidence-based CBT for ADHD grounded in executive functioning theory (Barkley, 2010; 2015; Ramsay & Rostain, 2015; Solanto, 2013), evidence-based behavioral treatment for alcohol use (Carroll & Kiluk, 2017; O’Donnell et al., 2019), and motivational interviewing principles (Miller & Rollnick, 2002), and has been iteratively refined based on user feedback as described below. Some tips focus broadly on inhibitory control in ADHD (e.g., put your tongue to the roof of your mouth to avoid interrupting), while others are specific to drinking situations (e.g., before grabbing your next drink, it may be helpful to think about what you plan to accomplish tomorrow and consider whether you want to stop for the night). This EMA-driven personalization is important to help overcome limitations of one-size-fits-all interventions (Boness & Witkiewitz, 2023), facilitates the real-time, in-context support that is needed for ADHD (Barkley, 2015), and reflects a major advantage afforded by mHealth EMI (e.g., Tong et al., 2021). The provision of tips throughout the day, including surrounding drinking episodes, is meant to enhance inhibitory control over drinking directly and support practice of inhibitory control across situations to build generalizable, context-cued habits (Wood et al., 2022).

Working Memory

Finally, four intervention elements, delivered within the EMA platform, support tip utilization by compensating for working memory deficits that can interfere with skill use. These working memory components are designed to operate both acutely and cumulatively across weeks (the red and blue arrows in Figure 2, respectively); intensively folding frequent reminders of and reflections on inhibitory control tips into daily routines can help users form lasting habits that will sustain improved executive functioning over time. This occurs by making inhibitory control more habitual, automatic, and triggered by environmental cues rather than relying as heavily on working memory (Barkley, 2015; Buabang et al., 2024; Diamond, 2013; Wood et al., 2022). First, repeated tips reminders are texted to users outside of their EMA prompt windows to help keep the strategy suggestions at the forefront of their minds. This should enable users to think of and use a given tip when it may be helpful acutely in the subsequent minutes and hours, thereby making these inhibitory control strategies acutely accessible. Second, users have exposure to recycled tips they previously used, with multiple chances to select and use the suggested strategy (i.e., users may swipe through tips and select one they want to use; at the next prompt, they indicate whether or not they want to be offered the tip again in the future, and if so, it is recycled back to the menu of tips). This repeated exposure is meant to build a user’s repository of strategies to help with inhibitory control across time and contexts. In this way, the goal is to scaffold working memory by filling in gaps across time when users may otherwise not remember to use a strategy. Multiple nudges to practice these strategies can help form long-lasting habits and establish contextual cues that can eventually activate the habitual, automatic use of these strategies (Buabang et al., 2024; Diamond, 2013; Wood et al., 2022). Third, the TIPS intervention further compensates for and scaffolds working memory by requiring users to prospectively plan their use of a tip in the moment it is suggested and, fourth, to retrospectively appraise their use of the tip and its effectiveness after receiving a tip and applying it. Specifically, EMA questions support this reflection in each prompt: one brief set after a strategy is suggested asks the user to rate how likely they are to use the tip and to consider how they will do so; another set at the start of the subsequent prompt, asks the user to rate the extent to which they used the tip, its effectiveness, and what made it helpful or unhelpful for future planning. This EMA-facilitated process is intended to tap the aspects of working memory that relate past experiences to future plans and goals, and should theoretically help transfer the inhibitory control tips from working memory into short-term memory. This process also supports the user in executing tips in the moments and situations when they are needed. The application of EMA to deliver these reminders and facilitate the prospective and retrospective queries represents another novel and practical function of this powerful tool beyond data collection.

TIPS mHealth Intervention Development

Below we describe the iterative process of the TIPS mHealth EMI development and refinement (see Figure 4). Three goals drove the intervention development process: (1) Create and refine the content of the intervention components (EMA questions, feedback, and strategy suggestions); (2) Refine and optimize the specifications of the intervention components, such as EMA frequency, tailoring decisions for each strategy suggestion, and duration of the entire intervention; and (3) Examine the feasibility and acceptability of the intervention as a whole and its components. We have pursued these goals simultaneously across five stages of intervention development thus far that have proceeded organically, with each stage adding a component and refining the previous ones based on insights gleaned. Throughout these five stages of intervention development, we have applied EMA in two ways: (1) we have utilized EMA as the intervention components themselves (symptom monitoring, personalized feedback, and tailored strategy suggestions); and (2) we have leveraged the rich quantitative and qualitative EMA data afforded by these intervention components, plus additional (non-intervention) EMA responses, to guide intervention development and refinement.

Figure 4. Overview of the TIPS mHealth EMI Development Process.

Notes. Each column demarcated by stage number summarizes a stage of the TIPS mHealth EMI development process. As indicated by the shaded box and downward arrows at the top of the figure, a community-engaged HCD approach permeates all stages of intervention development. The gray arrows between stages at the top of the figure represent the build-up of intervention components across stages, with each stage adding on components and refining the content and specifications (e.g., prompt frequency and timing; intervention length; format) for the prior components. Box 1 (blue) summarizes the primary EMI component begin developed in each stage (illustrated in more detail in Figure 3) and the executive function it targets (described in more detail in Figures 1 and 2). Box 2 (red) outlines how each stage addresses each of the three overarching intervention development goals. The double-headed arrows connecting the goals represent the iterative, cyclic nature of the intervention development process, with the outcomes of each goal further informing the others. Box 3 (green) summarizes the study design for each stage. Box 4 (purple) lists the components that constitute the intervention being tested at each stage to provide a sense of how the intervention grows over the course of the development process. Box 5 (black) highlights several of the key insights gleaned from the EMA data and other sources of data at each stage, which are incorporated into the subsequent iterations. ^© 2025 by Traci Kennedy. All rights reserved.

EMA: ecological momentary assessment; BL: baseline; FU: follow-up

In line with our conceptual model (Figure 1), the early stages of intervention development focused broadly on young adults with ADHD irrespective of drinking patterns with our ultimate goal of improving wide-ranging ADHD impairments (not solely alcohol use). Once the groundwork was laid for the general feasibility and acceptability of the intervention model in this population, we incorporated more of a focus on alcohol use beginning in Stage 3 of intervention development and therefore restricted the samples to high-risk drinkers with ADHD. We anticipate that future iterations of intervention development may include a similar focus on other downstream impairments of ADHD we hope to reduce (e.g., academic and occupational problems, interpersonal difficulties; see Figure 1, far right box) with additional content on these areas to amplify the intervention’s impact on them.

Our intervention development process is infused with a community engaged human-centered design (HCD) approach (Figure 4). This approach centers and empowers end users throughout all stages of development, employs methods of problem definition and solution generation that are intended to maximize efficiency and equity in tackling complex problems, and is increasingly applied to intervention development to make interventions optimally impactful, appealing, and useful from the outset (Vial et al., 2022). Our goals of iterative intervention development, refinement, and feasibility/acceptability testing align with HCD frameworks (e.g., Lyon et al., 2019). Further, HCD promotes the application of users’ EMA responses to inform intervention development, because they offer real-time indications of how users are experiencing the intervention as it unfolds, and these data provide quick, continuous insights throughout testing that enable the HCD principle of ongoing iterative refinement.

Stage 1 – Initial Pilot Work: Symptom Monitoring via EMA

In the first stage of intervention development, we piloted EMA questions and explored the preliminary feasibility and acceptability of EMA-based symptom monitoring as an intervention component. Fifteen young adults (age 18–21) with ADHD (not required to drink alcohol) completed: 1) a baseline visit including a diagnostic ADHD assessment, baseline questionnaires, and training in EMA; 2) one week of 4 daily EMA prompts assessing momentary ADHD symptoms; and 3) a follow-up visit that included questionnaire measures and a qualitative interview soliciting participants’ experiences with the EMA as well as needs and preferences for future intervention elements.. The timing of the four EMA prompts was: 15 minutes after self-selected typical wake time (unique to weekdays and weekends), two prompts sent at random times within fixed time windows during the afternoon (2:50 pm – 4:20 pm) and early evening (6:30 pm – 8:00 pm), and 15 minutes before self-selected typical bedtime (unique to weekdays and weekends). The question stem was, “Since the last assessment prompt…” and included 10 DSM ADHD symptom items adapted for EMA (4 inattention, e.g., “I forgot things;” 3 hyperactivity, e.g., “I fidgeted with my hands and feet or felt restless;” 3 impulsivity, e.g., “I interrupted people”), with a response scale ranging from not at all (0) to very much (3). Participants were compensated up to $100 for completing all portions of the study ($20 for the baseline visit, $30 for the follow-up visit, $5 per day of any completed EMA, and a $15 bonus for achieving at least an 80% EMA completion rate).

In this phase, we focused on refining the content and number of the EMA items used for ADHD symptom monitoring, along with their response scales and anchors. We examined descriptive statistics and obtained in-depth input from participants through cognitive interviewing to refine the EMA questions. We also redesigned the response scale to be a continuous slider scale from 0 to 100, partly in response to participants’ preferences for greater response range and flexibility, and partly to help maximize the opportunity for within-person variability. We felt this would be important in an intervention supporting self-awareness of in-the-moment fluctuations in ADHD symptom severity with reassurance from the measurement literature that we would not sacrifice reliability, validity, or usability (Funke & Reips, 2012; Imbault et al., 2018; Lewis & Erdinç, 2017).

The EMA data and follow-up quantitative and qualitative data supported the feasibility and acceptability of EMA-based self-monitoring for young adults with ADHD (e.g., 92% EMA completion rate; average completion time of 1 minute 20 seconds per EMA survey; 0 participants endorsed that completing EMA caused a problem). Moreover, even though EMA was not described to participants as an intervention, participants overwhelmingly found it helpful in enhancing their in-the-moment awareness of their ADHD symptoms. For instance, 13 of 15 participants (86.7%) agreed or strongly agreed that answering EMA questions about ADHD symptoms was helpful. Participants shared, “It made me more aware of my behavior and the effects that my ADHD has on my day-to-day life,” and, “Sometimes it would motivate me to improve on those behaviors.” Importantly, though, participants also voiced a desire for a feedback component that helped track their responses over time and situation, which reinforced our plans to incorporate feedback into the next iteration.

Stage 2 – Pilot Open Trial: Refining Symptom Monitoring via EMA and Adding Personalized Feedback

In Stage 2, we added a personalized feedback component to enhance self-awareness of ADHD symptoms and underlying executive functioning lapses, explored user acceptability of varying EMA prompt frequencies and protocol durations, and collected data to inform the addition of the inhibitory control and working memory elements of the intervention. Specifically, 72 young adults (ages 18–21) with ADHD (not required to drink alcohol) underwent similar procedures as those described above, with several key differences: (1) the EMA period was extended from one to three weeks; (2) participants were randomized to either a high-dose (5 EMA prompts per day) or a low-dose (1 EMA prompt per day at bedtime) group (double-masked); (3) feedback was provided at the end of each EMA prompt on whether a single ADHD “target” symptom, selected by the participant with clinician input at baseline, improved, worsened, or stayed the same since the prior completed prompt, and participants typed in what was either helpful (when symptoms improved) or what they could do to help with this target symptom (when symptoms worsened or remained stable); and (4) participants were explicitly told that a potential benefit of the EMA protocol was increased awareness of their ADHD symptoms (i.e., presented as an intervention versus solely assessment). Participants were compensated up to $235 for completing all portions of the study ($50 baseline visit, $50 follow-up visit, $2 per completed EMA prompt in the high-dose group or $10 in the low-dose groups, and a $25 bonus for achieving an EMA completion rate of at least 85%).

As in the prior pilot phase, most (52%–84%) participants shared qualitatively and/or quantitatively at follow-up that both the EMA questions and feedback were helpful for improving their self-regulation, and 90% indicated that it enhanced their symptom awareness (e.g., the feedback “made me reflect on how I was that day and if I improved or not, and it helped me think about what I could do better the next day”). In comparing the perceived acceptability of 5 times versus once-per-day EMA prompts, participants had mixed responses. This input helped us revert to a 4 prompts-per-day structure to balance user burden with the desire for ample self-reflection and feedback. Feasibility was high: the average completion rate was 82% and participants generally agreed that three weeks of EMA was not too burdensome (only 4.4% agreed or strongly agreed that it was too long); in fact, 43.5% reported they would be willing and able to participate for even longer and 44.6% independently suggested an ideal length of a month or more, leading us to lengthen the intervention period to 4.5 weeks. While we considered elongating the duration even more, we settled on 4.5 weeks based in part on the mHealth literature showing that interventions of similar intensity/prompt frequency last about a month on average, and in part on our need to balance participant burden and study resources with maximizing efficacy, particularly as interventions tend to have diminishing returns over time (e.g., Hayes et al., 2007), possibly due to intervention fatigue. Finally, we further refined the EMA response scale for the next iteration to include more helpful anchor words on the 0–100 slider scale as well as the format of the feedback from a verbal summary to a visual line graph synthesizing all the user’s responses over time, based on participant input (e.g., 83.3% indicated they would prefer graphical feedback).

Excitingly, we made use of the rich momentary data to refine the intervention. For instance, we culled participants’ real-time qualitative responses on their personalized feedback (“What do you think was helpful?”; “What could you do to improve?”). Although many helpful ideas were provided, the most common response (18% of all responses) was “I don’t know.” Reinforcing this in-the-moment uncertainty and working memory deficits, many participants shared in qualitative follow-up interviews that although symptom monitoring was helpful to enhance self-awareness, they needed direct suggestions in the moment to help manage their ADHD symptoms, even if they could identify helpful strategies retrospectively. This pattern resoundingly confirmed our plans to incorporate behavioral strategy suggestions into the intervention in the next iteration to directly help users exert inhibitory control once they were aware in the moment of the need to do so. Additionally, at each end-of-day assessment, participants indicated which of a list of executive functioning behavioral strategies they used during the day (Ramsay & Rostain, 2015; Solanto, 2013). Participants’ responses informed our initial thinking about the types of behavioral strategies we would develop to suggest to participants as part of the next phase of intervention development.

Stage 3 – Secondary Data Analysis: Behavioral Strategies (“Tips”) Content Development

In this next stage, we generated behavioral strategy suggestions (the “tips”). We began by analyzing participants’ quantitative and qualitative EMA data, as well as qualitative interview data, from Stage 2. For instance, we examined frequencies of participant endorsement of the executive functioning strategies in the end-of-day EMA survey to begin establishing what types of strategies participants might prefer. We also examined qualitative responses to the real-time open-ended feedback questions to map the types of strategies participants reported using in the moment onto our evolving list of theoretically and empirically based executive functioning strategies. We synthesized these with participants’ suggested strategies in follow-up questionnaires and qualitative interviews, including drinking-specific strategies that participants would find acceptable and feasible. We reviewed the literature on effective inhibitory control strategies for alcohol use (e.g., protective behavior strategies) that could feasibly be integrated into an mHealth intervention. We combined these data to generate an initial menu of tips on which to gain input for further refinement. To specify the timing and conditions under which each tip would be eligible for selection at a given EMA prompt (e.g., when self-reporting being with others vs. alone; morning vs. later in the day; before vs. after initiating drinking), we also incorporated users’ ideas and worked closely with computer programmers.

Stage 4 – Pilot Testing: Refining Tips⁴

In Stage 4, we aimed to refine the tips content, optimize their tailoring criteria (decision rules for conditions under which each tip should be delivered), and add the intervention components intended to support working memory (tips reminders, recycled exposure to tips across time, and prospective planning and retrospective appraisal of tips use). We piloted the now 4.5-week intervention in a pre-posttest open (uncontrolled) trial with 21 young adults (ages 18–25) who drank alcohol (at least weekly “binge” drinking). We included EMA-based symptom monitoring, personalized graphical feedback, and semi-tailored tips 4 times per day. This prompt frequency was based on feedback from Stage 3 that 5 prompts per day felt too frequent (e.g., 77% completion rate in the 5x/day group compared to 85% in the 1x/day group; 44.5% of participants in the 5x/day group agreed or strongly agreed that there were too many prompts per day) and that half of participants felt that the ideal prompt frequency would be 3–5 prompts per day. Although the other half felt that 1–2 per day would suffice, we balanced this preference with our need to offer sufficient opportunities for monitoring, feedback, and strategy suggestions in diverse contexts throughout the day, in line with our conceptual model. Finally, the average overall response rate of about 80% would ensure that at least 3 of the 4 prompts would be completed per day, leaving room for multiple intervention touches even with imperfect adherence. We conducted follow-up focus groups with participants, and we obtained in-depth feedback on all aspects of the intervention, with a particular focus on the tips (content, format, timing, etc.). Participants were compensated up to $400 for completing all portions of the study ($50 baseline visit, $75 follow-up visit and focus group, $2 per completed EMA prompt, and a $25 bonus for achieving an EMA completion rate of at least 85%). All participants (100%) agreed or strongly agreed that they were satisfied with the amount of compensation.

We interspersed focus groups with virtual meetings with a community advisory board comprising young adults with ADHD, clinicians, and researchers. Using HCD methods, we iteratively shared participant feedback with the advisory board, edited the tips in real time together as a group, brought the recommended changes to the next focus group for further feedback, and so on. To complement this cyclic input, we again leveraged the EMA data to inform intervention development. For instance, from the EMA questions that facilitate prospective planning and retrospective appraisal of tips use, we will examine real-time quantitative responses on frequency and perceived helpfulness of tips use as well as the qualitative responses indicating momentary barriers and facilitators to their use. Additionally, participants were asked throughout this stage to rate the overall intervention periodically in EMA using a 5-star rating system and to share reactions, questions, and feedback with the research team as they arose, including screen captures to help the team identify problematic aspects of the intervention to be revised. Together, these efforts will result in a finalized menu of tips and a further refined intervention overall for testing in Stage 5. For example, in Stage 4 and prior stages, participants shared a desire to view how other behaviors track with their ADHD. Thus, we have enhanced the graphical feedback to demarcate drinking days and low sleep nights (less than 6 hours, which may cause dips in inhibitory control; Lowe et al., 2017) so users can deepen their awareness of potential associations between these factors and their tracked ADHD symptoms (see Figure 3, Personalized Feedback).

Notably, Stage 4 feasibility and acceptability metrics were high: average completion rate was 82%, and remained stable across the 4.5 weeks, which has encouraged us to retain this length for the next iteration. In support of the increased duration to 4.5 weeks, 55% of participants disagreed or strongly disagreed that this duration was too long; only one participant (5%) strongly agreed. Additionally, 65% and 64% of participants reported finding the ADHD-general and drinking-specific tips helpful, respectively – a clue that even users who are not explicitly seeking support for drinking find the intervention approach acceptable. In fact, supporting our indirect approach to drinking support in this intervention, most participants in this stage (85%) agreed or strongly agreed that they would prefer “an intervention that helps with my ADHD symptoms and may help with safer drinking than an intervention that focuses on drinking directly.” In contrast, only 25% agreed or strongly agreed that they would be willing to use a digital intervention directly focused on drinking (55% disagreed or strongly disagreed).

Stage 5 – Putting Everything Together: Pilot RCT of TIPS Intervention Feasibility

Finally, in a pilot RCT of the refined TIPS intervention, we will randomize 70 young adults (age 18–25) with ADHD with at least weekly heavy (“binge”) drinking to either the TIPS intervention or to an alcohol monitoring-only active control group for equal duration (4.5 weeks) with compensation equal to that in Stage 4. This design holds constant the alcohol monitoring portions of TIPS, which has itself been shown to reduce drinking (McCambridge & Kypri, 2011; Michie et al., 2012), so that we can isolate the combined effect of the unique components central to our intervention model (symptom monitoring, personalized feedback, and strategy suggestions) that are theorized to operate synergistically. Quantitative and qualitative data from follow-up questionnaires along with qualitative interviews will enable an appraisal of intervention feasibility and highlight areas for future additional intervention refinement. Moreover, real-time EMA responses (e.g., helpfulness of tips) will aid further curation of the tips in future iterations. Beyond testing overall preliminary intervention efficacy, we will leverage the quantitative EMA data to examine momentary associations implied by the conceptual model, as well as cumulative changes in these associations across the intervention period. The EMA data generated as part of this intervention will therefore be utilized for substantive and novel theoretical advances in our understanding of momentary dynamics among executive functioning, ADHD symptoms, alcohol use, and resulting impairment.

Practical Considerations and Lessons Learned: Benefits, Challenges, & Future Opportunities for Incorporating EMA into mHealth Interventions

We believe the TIPS intervention is illustrative of the many benefits, unique challenges, and future opportunities of EMA leveraged for mHealth interventions for alcohol use. We describe several of these practical considerations below with the goal of elucidating promising avenues for alcohol intervention researchers.

Benefits

mHealth EMI programs are an appealing and scalable method to provide young adults support for high-risk drinking, given that most young adults use their phones frequently throughout the day (Pew Research Center, 2024) and increasingly so for behavioral health support (Wartella et al., 2016). Additionally, EMI is a fairly simple approach that holds potential to maximize feasibility and acceptability among users, in turn supporting uptake and engagement. Keeping interventions simple is also key for individuals with ADHD who can struggle to organize many different features and requirements of a more complex system. Indeed, data increasingly and convincingly demonstrate the feasibility of EMA with young people with ADHD, making it a valuable option with this population (e.g., Karalunas et al., 2024; Kennedy et al., 2022; Koch et al., 2021; Miguelez-Fernandez et al., 2018).

A strength of EMA and EMI is their ecological validity, which makes it ideal for identifying the real-world contexts where support is needed and directing support to those moments and situations. This feature is crucial for alcohol interventions, where the focus is on drinking episodes and the circumstances leading up to them, and especially so for those with ADHD, for whom here-and-now support is needed to overcome executive functioning challenges of independently applying inhibitory control strategies when they are needed (e.g., Barkley, 2015). The repetitive nature of EMA may also be considered advantageous insofar as repeated exposure to and practice with self-reflection and implementing inhibitory control strategies is vital to building habits over time. Indeed, this repeated practice embedded into everyday life is considered indispensable to making executive functioning skills less effortful and more automatic, which can in turn enable weaning the active support of mHealth over time as these skills become internalized (Buabang et al., 2024; Diamond, 2013; Wood et al., 2022). For instance, a future goal is to gradually reduce the intensity of TIPS (e.g., fewer prompts per day, less active support for self-reflection) as users transition to using the skills independently. If supported by acceptability data in Stage 5, this may also facilitate elongating the intervention duration for maximum benefit, since user burden would decrease in the latter stages.

As our pilot data suggest, an EMI approach may appeal to the majority of young adult drinkers with problematic alcohol use who do not seek help, primarily because of perceived lack of need. Consistent with a motivational interviewing framework (Miller & Rollnick, 2002), increasing self-awareness via EMA preserves the young person’s autonomy in recognizing and changing their behavior rather than being explicitly instructed to do so. In fact, one of the driving themes that emerged from participant feedback in the process of refining the tips was the need to optimize users’ autonomy, such as by altering the wording of tips (e.g., “Do you want to keep track of how many drinks you have tonight?” instead of “Keep track of how many drinks you have tonight”) and offering greater choice (e.g., allowing the user to select a tip from sequentially presented options rather than offering just one tip at a time). An added benefit of integrating EMA with behavioral nudges is that EMA responses permit personalization of the tips, including tailoring criteria that can only be gleaned from self-report, such as whether the user wants to receive a given tip again in the future. As we have described in the context of the TIPS intervention, perhaps the most unique aspect of mHealth EMI is that the data it provides, particularly on within-person changes across time, aid our understanding of the momentary dynamics surrounding alcohol use. In this way, EMA can be considered a dual-purpose intervention, as it simultaneously changes behavior and generates scientifically useful data to improve outcomes. EMIs align well with measurement-based care that is considered best practice for psychology and psychiatry treatment, which ensures effective care through consistent assessment of patient outcomes (Fortney et al., 2015; Resnick et al., 2020). In fact, EMA could enhance standard measurement-based practice by providing clinicians with more real-time and fine-grained information on patients’ symptoms between appointments to efficiently gauge and adjust treatment.

EMA data even allow examinations of whether proximal behaviors and symptoms (not just distal outcomes) are improving across days and weeks, which is an important goal of an mHealth intervention in and of itself, yet can also be tested as mediators of intervention effects on more distal outcomes (Nahum-Shani et al., 2018). For example, in a previous sample of adolescents with ADHD, we were able to track improvements in self-reported ADHD symptom severity over the course of 17 days of EMA; even more exciting, we leveraged missing data to discover that completing versus missing an EMA prompt was associated with within-person, acute improvements in ADHD symptom severity at the next prompt (Kennedy et al., 2022). Given that other constructs can simultaneously be assessed in EMA alongside the intervention targets, other momentary associations can be examined within person to test acute components of an EMI, such as momentary relations between impulsivity and drinking behavior, as well as how these associations may evolve over time. In TIPS, we also briefly measure constructs such as momentary mood and positive and negative urgency in EMA to permit examinations of their relations with our target constructs. In short, embedding an intervention within an EMA protocol offers considerable flexibility and unique opportunities for data collection and analysis, provided that participant burden is minimized.

Finally, we have found the data we collect in EMA to be extremely helpful in the process of developing the intervention. It elucidates users’ real-time experiences and the momentary dynamics an mHealth EMI seeks to disrupt, along with feedback on the intervention components precisely when they are being delivered. These raw, in-the-moment reactions can be challenging to obtain in post-intervention questionnaires and interviews when recall bias is substantial. The rich details contained in open-ended EMA responses, in users’ own words, have especially clarified how certain intervention elements are helpful or unhelpful as they are delivered. In summary, many of the same advantages of EMA that make it a useful research assessment method can be embraced both in applying it as part of mHealth EMI and to develop such interventions.

Challenges

mHealth EMI also presents important challenges. Perhaps most obvious is the burden of frequent self-report across several weeks, highlighting the need to balance intervention intensity with feasibility. Not only can EMA multiple times per day induce response fatigue and impede sustainability, especially during daily activities, but it can quickly feel repetitive. Both concerns are especially salient for young adults with ADHD given the executive functioning skills needed to persist at ongoing tasks perceived as boring (Barkley, 2015). Although research findings demonstrate that young adult samples can achieve high average EMA completion rates of 85% or more using smartphones (Trull & Ebner-Priemer, 2013), including in samples with ADHD (Miguelez-Fernandez et al., 2018), this does not guarantee that users will sustain their engagement absent the compensation structures used in EMA/EMI studies to incentivize compliance. Indeed, the issue of engagement in mHealth more broadly is a major challenge facing the field that must be addressed for mHealth to take hold as a viable intervention option (Nahum-Shani et al., 2022). On the other hand, the goal for TIPS users is not to persist with EMA prompts indefinitely, but rather to ultimately wean from high frequency intervention, and eventually build more autonomous, habitual self-reflection and use of inhibitory control strategies through repeated practice. Users may internalize these practices over time to facilitate symptom monitoring and inhibitory control beyond the intervention period, increasing their independence and control in carrying out these skills (e.g., setting digital reminders at key times throughout their day). The extent to which individuals would engage in such an intervention in the first place outside of a research study offering compensation, however, is an open question.

Our feasibility and acceptability data from several stages of development offer insights into this issue for TIPS. Qualitative interview data in Stage 2 indicated that 11% of participants would continue beyond the 3-week study period without being compensated. In quantitative questionnaires in Stages 2 (3 weeks) and 4 (4.5 weeks), participants were asked how much they agreed that they would continue the intervention after the study ends if given the opportunity (though we did not specify “without compensation”). Using a 5-point scale from strongly disagree to strongly agree, 51% of participants across the two studies agreed or strongly agreed, suggesting potential for real-world use alongside opportunity for improvement. In all stages, participants suggested ways to minimize user burden that would facilitate their use of the EMI for longer or without monetary incentives (e.g., reduce prompt frequency and length, enhance flexibility and personalization of prompt timing, make feedback more informative and visually engaging), which we have iteratively incorporated with an eye toward optimizing the EMI for future implementation and dissemination. To empirically determine the minimally effective incentive, EMI users could be randomized to varying incentives (monetary and otherwise, such as gamified rewards like points and badges) as in prior work (Abdelazeem et al., 2022), or a microrandomized trial could be conducted to test the within-person effects of varying incentives on a user’s engagement throughout the intervention (Rabbi et al., 2018). These efforts could occur within an effectiveness-implementation hybrid design to efficiently maximize ultimate public health impact while substantiating clinical effectiveness in preparation for real-world dissemination (Curran et al., 2012).

Another concern is whether increased self-awareness and continual receipt of inhibitory control strategies may actually worsen outcomes for some users. For populations with ADHD, an intervention requiring users to frequently check their phone may garner criticism, given that distracted smartphone use is exactly one of the dysregulated behaviors in ADHD such an intervention is designed to help curb and that could exacerbate symptoms (Thorell et al., 2024). More generally, despite the ample evidence that EMA improves behaviors on average, heterogeneity in individuals’ responses certainly exists. For instance, spotlighting one’s ADHD symptoms throughout the day may harm self-confidence and/or incite negative affect. Although this possibility has not been widely considered in the EMA literature, such iatrogenic effects, should they exist, must be monitored and examined, as in any intervention. Related to this challenge is the question of whether to not only suggest inhibitory control strategies, but also more intensively teach users how to implement them. Although the strategies in the TIPS intervention were designed to be straightforward and easy to understand, some individuals in some moments may struggle with implementation. This issue was discussed by our advisory board during TIPS intervention development, but at present, we have elected to prioritize simplicity over complexity while continuing to gather EMA data on users’ ability to apply the tips. More broadly, one of the challenges we have faced in developing TIPS is balancing simplicity and feasibility of the design with users’ interest in enhanced functionality afforded by mobile technology, all while preserving the core theoretical roots thought to drive the ultimate effectiveness of the intervention. For example, end users and advisory board members have suggested adding more advanced app features to TIPS, such as integrating calendar and organizational systems, building a social support network among users, and making tips more interactive. While we strive to be responsive to these preferences of end users and have incorporated much of their input, we have also heard from end users with ADHD that they appreciate the simplicity of TIPS, as systems with many “bells and whistles” can be overwhelming. Importantly, the relative benefits of simplicity and “push” (vs. “pull”) interventions in mHealth have also been supported empirically (e.g., Suffoletto, 2024). Thus, we have not adopted every recommendation in an effort to meet users’ needs around usability and simplicity, and we have prioritized changes that align with our conceptual model. This tension is not unique in intervention development, though it may be magnified in mHealth development because of the seemingly limitless options mobile technology offers.

It can also be challenging to strike a balance between designing EMA for assessment versus intervention when we aim to do both in this program of research. Several of these decision points for TIPS include: whether to standardize time windows for EMA prompts to permit comparability of between-person data versus maximize flexibility and personalization in the timing of prompts for the greatest possible impact on a given individual; whether to prompt participants randomly to enhance the representativeness of their data or at fixed, predictable intervals to encourage anticipatory behavior change; whether to keep EMA items constant across individuals and time or to individualize items thought to be most relevant for a given individual; and how to promote in-depth self-reflection while making EMA quick and easy to complete. In our view, there may be good reason to maximize assessment or intervention benefits for any given feature, and as described above, one of the exciting benefits of this dual-purpose intervention approach is the resulting high-frequency data to analyze.

Nonrandom missing data is a crucial consideration and can represent a major hindrance when using EMA for data collection. While missing data can be thoughtfully leveraged to examine acute effects of EMI (described above), it may also present unique problems. Specifically, in TIPS, completing an EMA prompt triggers a new tip that is tailored based on the user’s responses; however, it is possible that missing an EMA prompt is associated with moments of particularly poor executive functioning when self-awareness and inhibitory control are especially needed, in which moments the user would not receive a new tailored tip (though would be reminded of their prior tip). This may be especially true for drinking episodes, when it may be harder for users to keep up with EMA prompts while drinking or intoxicated, thereby missing the opportunity for in-the-moment drinking-specific support. This highlights a final challenge in EMI: the limitations of self-report. Objective metrics of ADHD symptoms, alcohol use behaviors, and underlying executive dysfunction obtained through passive mobile sensing hold promise to complement self-report EMA (e.g., Harari et al., 2016; Insel, 2017). Passive sensing could overcome limitations of self-report by being collected continuously and effortlessly to automatically trigger and tailor support in vulnerable moments even without user input. Passively detecting drinking episodes could be especially helpful (Bae et al., 2023), so users still receive a tip and reminders to self-monitor even when they are less likely to self-initiate intervention nudges while drinking. More objective indicators can complement self-report to enhance reliability in assessment, and they could be even more useful for enhancing self-awareness than subjective self-report when a user overlooks symptoms. On the other hand, relying too heavily on passively collected data would minimize the cognitive and behavioral engagement in the intervention content that active assessment promotes through intentional reflection and behavioral planning, which the TIPS model posits as key active ingredients. Therefore, while incorporating passive sensing into this intervention model is an exciting next step for alcohol mHealth interventions, representing both substantive and technical advancements, a balanced approach between active and passive data will likely prove crucial.

Future Opportunities

Beyond integrating passive sensing to trigger just-in-time support, the mHealth EMI described here offers abundant opportunities for further research and development, of which we mention just a few. First, varying EMI designs could be compared to test the optimal frequency, timing, and content of assessment items and tips. For example, a single prompt and tip per day may suffice and minimize user burden. The prompt may be best delivered at the end of the day, or anchored to a key event in the individual’s day (e.g., upon arriving to work, or when starting a drinking episode) rather than a specific time. A microrandomized trial is an essential design for isolating within-person causal effects of various mHealth intervention specifications, such as frequency, timing, and tip content (Klasnja et al., 2015; Qian et al., 2022). Comparative efficacy of the various components of an EMI like TIPS (e.g., symptom monitoring vs. feedback vs. strategy suggestions) can be tested via microrandomized trials or via factorial designs using the multiphase optimization strategy (MOST) framework (Collins et al., 2013). These approaches can efficiently dismantle the intervention package and potentially inform trimming of less effective components.

Second, an EMI like TIPS holds vast potential for greater personalization of any number of features – prompt timing, tips context, and greater tailoring to temporal context, to name a few. Advanced machine learning could support making EMA items and tips adaptive over the course of the intervention, such as prioritizing the presentation of high-need strategies and titrating these as improvements are sustained over time. In addition to personalizing the intervention based on user responses, users’ real-time preferences regarding the various intervention features could be incorporated into an adaptive algorithm to further customize the intervention, potentially enhancing effectiveness (Cornet & Holden, 2018). For instance, in TIPS, users could “favorite” especially helpful tips and/or ADHD symptom monitoring questions to upweight them for enhanced personalization.

Third, advanced engagement strategies could be pursued. Despite our commitment to intervention simplicity, gamification may enhance both user engagement and effectiveness (Cheng, 2020). This may be especially so for individuals with ADHD for whom immediate positive reinforcement especially promotes skill use (Barkley, 2015). For instance, users could earn points, badges, or elements in virtual environments (e.g., aquarium, garden) for completing an EMA prompt or reporting the use of a new tip, which have been shown to promote mHealth engagement (Cheng, 2020; Rabbi et al., 2018). A rich emerging literature on mHealth engagement strategies points to many options that can be incorporated into EMIs. Many of these strategies have been echoed by our TIPS participants throughout all stages of intervention development, including: varying prompt content over time to reduce repetitiveness and prevent boredom; increased personalization to optimize relevance; socialization and virtual support among users; using salient aesthetics and sensory design features (e.g., visual, audio, tactile) that capture the user’s attention; methods of making engagement intrinsically motivating, such as infusing fun and humor into EMI prompts and emphasizing the user’s autonomy in participating; and enhanced, visually appealing, comprehensive individualized feedback (Nahum-Shani et al., 2022; Nwosu et al., 2022; Rabbi et al., 2018).⁵ These types of design elements may be especially useful when engagement is expected to be lower, such as during alcohol intoxication. EMI prompts during drinking episodes should also be quick, easy, and motivating to complete to support engagement (Hufford, 2007). Human-centered or user-centered design frameworks are invaluable tools to prioritize these design elements alongside the theory-driven content of EMIs, as effective design is necessary to ensure acceptability and sustained engagement among end users (Lazard et al., 2021; Lyon & Koerner, 2016).

Finally, extensions of this approach could be considered to reduce the problematic use of other substances that are common among young adults with ADHD (e.g., cannabis, nicotine), as could extensions to other age groups. Moreover, although we focus on one conceptual model targeting executive functions underlying ADHD and problematic alcohol use, the principles may be useful to apply more broadly – including using EMI to enhance self-awareness and trigger behavioral nudges, integrating EMA as a data collection tool with mHealth, and soliciting real-time user feedback on mHealth elements via EMA to inform intervention development.

Conclusions

EMA prevails as a crucial tool for understanding the momentary, real-world dynamics of human behavior and related constructs pertinent to ADHD and alcohol use. Beyond its advantages for collecting data, EMA holds tremendous promise in mHealth EMIs to help self-monitor behavior and manage alcohol use, and to help shape such interventions as they are being developed. Continuous, iterative co-development with end users through approaches like HCD aligns well with an EMI approach and can yield invaluable insights to improve interventions. We have illustrated how we have harnessed some of the features of EMA in developing the TIPS mHealth EMI for young adults with ADHD and high-risk alcohol use, and we anticipate that other EMIs employing similar principles are currently under development across alcohol-using populations. We hope these examples can be readily generalized by researchers studying the use of alcohol and other substances. Despite the challenges associated with this method, it offers exciting opportunities for maximizing intervention personalization, appeal, and effectiveness.