A GEMA approach triangulates conventional ecological momentary assessment (EMA) methods, in which repeated assessments of participants’ cognitions, affect, behaviors, and biomarkers (eg, salivary cortisol [64]) are taken at multiple daily intervals, contextualized by GPS coordinates [49,65]. Through GEMA, recall bias is mitigated, and naturalistic spatiotemporal influences on individual health can be inferred [65]. It is particularly well suited for detecting context-specific environmental stressors and facilitators of individual resilience [66,67]. Using GEMA, aim 1 intended to describe activity spaces, the spatial configurations in which routine events occur with characteristic rhythms [68,69] associated with durable within-person patterns (both risk markers and risk buffers) of adherence and care among people living with HIV in New Orleans. Particular attention was paid to detecting potential mediated pathways via mood, mental health, substance use and misuse, trauma cognitions, coping skills, and domain-general self-efficacy variables, indicating the potential to operationalize these constructs as intervention touch points or proximal outcomes [70,71].

A small observational pilot of the GEMA method was reported by Theall et al [66]. Detailed methods and full (N=89) aim-1 GEMA findings are forthcoming.

As a nested, microlongitudinal study, aim 1 recruited and used the baseline and medical record data of a subset of participants already enrolled in the New Orleans area longitudinal New Orleans Alcohol Use in HIV cohort study [72,73]. This parent study recruited through the HIV outpatient clinic of a large university-affiliated medical center at which approximately 50% of New Orleans people living with HIV receive care.

Inclusion criteria for the parent study (N=365) included age of ≥18 years, documented HIV serostatus, and being in care at an HIV specialty clinic; any patient experiencing an acute illness within the previous 6 weeks, acute intoxication, or pregnancy was excluded, with the option to participate once the then-acute medical issue was resolved. Enrolling a longitudinal cohort, the New Orleans Alcohol Use in HIV study’s primary aims are to advance clinically actionable knowledge on the biological mechanisms and socioenvironmental drivers of HIV- and alcohol-related comorbidities [72,73]. Details of the parent study protocol are available in Appendix 1 of the study by Welsh et al [72]. For the sake of aim-2 intervention development, aim 1 recruited people living with HIV (n=49; a limited subsample because of scheduling delays incurred during the recovery from 2021 Hurricane Ida) on ART who either self-reported a personal history of violence or lived in a high-violence neighborhood in the parent study. These “either/or” multilevel eligibility criteria were formulated pragmatically to ensure sufficient statistical power. History of violence criteria included reporting childhood physical or sexual abuse or physical assault as an adult or endorsing moderate to high stress because of neighborhood crime and family or relational violence. Living in a high-violence neighborhood was defined as the respondent’s residential Census tract having violent crime rates at or above the 75th percentile for violent crime rates for tracts within Orleans Parish. According to these criteria, 96% (47/49) of the sample indicated an individual personal history of violence exposure, with 4% (2/49) endorsing tract-level violence exposure exclusively. Parent study participants who met the inclusion criteria were contacted initially by phone after randomizing the order in which outreach was conducted. If interested, they were asked to schedule a 1-hour dedicated baseline appointment at which informed consent procedures were undertaken, a baseline assessment was conducted using REDCap (Research Electronic Data Capture; Vanderbilt University) [74], the daily diary EMA measures were reviewed for comprehension, and a GPS app was installed on participants’ smartphones. Eligible people living with HIV who did not possess a smartphone were provided one by study staff. Once enrolled, daily diary and geospatial tracking data were collected for each individual over a continuous 14-day daily diary survey period.

The 49 aim-1 participants whose GEMA responses informed intervention development recounted a range of lifetime violence exposures at baseline: 24 (49%) recounted childhood domestic violence, 18 (37%) recounted childhood sexual assault, 26 (53%) recounted physical assault in adulthood, 20 (41%) recounted current crime and violence exposures, and 9 (18%) recounted current intrafamilial violence. Screener responses suggested a range of potential mental health challenges: of the 49 participants, 13 (27%) had borderline or clinical PTSD as defined by the DSM-5, a total of 19 (39%) had borderline or clinical anxiety, and 12 (24%) had borderline or clinical depression. ART nonadherence was reported by 28 (57%). The broader sociodemographic attributes of aim-1 participants are presented in Table 1.

The daily diary measures, which operationalized the EMA aspect of aim 1, assessed a range of intrapsychic constructs, behaviors, and environmental stressors, combining brief prevalidated scales with novel study-specific measures. Participants were prompted by SMS text message and asked to complete the daily diary measures within a 3-hour window 3 times daily: 8 AM to 11 AM, 1 PM to 4 PM, and 7 PM to 10 PM Central Time Zone. Qualtrics (Qualtrics International, Inc) hosted the survey measures and administered the preprogrammed SMS text message prompts [76].

The daily diary measures for aim 1, with sample items, is shown in Multimedia Appendix 1.

At aim-1 baseline appointments, participants were guided through the installation of Actsoft Encore GPS-tracking app [77]. Throughout each individual’s 14-day daily diary survey period, Encore captured longitudinal and latitudinal coordinates at 1-minute intervals when moving and 1-hour intervals when stationary (for aim 3, the NOLA GEM app itself will deliver daily diaries and record geospatial trajectories). GPS data were stored on a password-protected cloud server. Participant GPS data were then transformed into activity paths at 50-, 100-, and 200-meter buffers using Python (Python Software Foundation) [78]. Environmental exposures were assessed through the creation of activity paths and interceptions with the 3 buffer sizes. Both static and dynamic environmental exposures were captured. Static data included, for example, annual rates per US Census tract level of exposure such as alcohol outlet density, community violent crime rate, poverty, and concentrated disadvantage. Dynamic data included time-stamped data linked to daily GPS points, including time-incident violent and nonviolent 9-1-1 calls to service data points and police stop-and-frisk calls. Activity path data were then aggregated to be merged with daily diary data for triangulation.

In the course of exploratory and interim analyses of baseline and daily diary data, a range of environmental stressors with salience to the everyday lives of New Orleans area people living with HIV emerged. The study’s original objective was to understand how local spatial exposure and daily experiences may serve as determinants of traumatic stress or retraumatization connected to downstream maladaptive coping behaviors such as alcohol use and consequent patterns of ART nonadherence [79,80]. Our preliminary analyses did not find associations between outcomes related to alcohol use and daily mood across activity path–based spatial alcohol and violence exposure. These null results necessitated a departure from the original aim-1 goal of building generalizable place-based exposures as just-in-time adaptive intervention (JITAI) inputs into the NOLA GEM app. As such, tailoring variables, and the daily diary items through which they were captured in aim 1, were selected based on their overall prevalence at baseline, capturing past and cumulative experiences and associations with key outcomes within individuals on a given day among aim-1 participants. In recognition of the irreducibly idiographic nature of spatial traumatic stress determinants evidenced in this study, an individually customizable geofencing solution was built into the eventual NOLA GEM app (detailed in the NOLA GEM App section).

These stressors, with relevant statistics illustrating the rationale for their inclusion as JITAI tailoring variables, are provided in Table 2.

Although GEMA can undergird rich, sophisticated statistical insights on the activity spaces of New Orleans people living with HIV, an overreliance on the “technocratic thrust” of decontextualized mathematical modeling risks overlooking the lived expertise of longtime New Orleanian survivors, with implications for external validity specifically [83,84] and epistemic justice, which is broadly defined as affording appropriate credibility to the knowledge- and sense-making practices of structurally marginalized groups such as people living with HIV [80,85]. As such, aim 2 recruited a subcohort of people living with HIV (n=12) enrolled in aim 1 to contribute their insights on stress, coping, and survivorship via in-depth interviews. This phase aimed to adapt LIFT in a manner that (1) activated the unique, uniquely powerful value propositions of mHealth [51,53,54,86] and (2) interlocked GEMA-derived intervention touch points with the core components of an extant evidence-based intervention (EBI) logic [33,34,36-39] while (3) recognizing the singular cultural and historical context of HIV survivorship in contemporary New Orleans. As such, aim 2 embraced principles of design justice, which call for a nonhierarchical, nonexploitative stance toward engaging the lived expertise of local communities [87] while leveraging the pragmatic, product-focused toolkit of Lean User Experience, which emphasizes an accelerated, iterative build-measure-learn approach to software development [88].

By invoking principles of epistemic and design justice, we aimed, in practice, to learn about and integrate community perspectives, needs, and especially strengths into the functionality of NOLA GEM [89]. This resulted in JITAI tailoring variables closely attuned to the needs of—and intervention content and proximal outcomes that reliably tap into the inductively derived strengths of—longtime violence-affected HIV survivors in New Orleans as “socially embedded, socially situated” community members [90]. In line with Jaworski et al [91] and Maestre et al [92], we synthesized mixed methods–derived insights in a manner calibrated to (1) appreciate local, culturally congruent nuances in our constructs of interest; (2) understand the precise conditions in which digital tools are used (and not used) among New Orleans people living with HIV; and (3) foreground coping focally and emotional well-being broadly as determinants of distal biomedical outcomes such as virologic control while calling on an extant EBI [33,34,36-39]. This approach is in line with core design justice principles, specifically design intentions committed to healing, sustainment, and empowerment; the catalyzation of people living with HIV end users’ lived expertise with formal design techniques; and, in particular, embracing “what is already working,” both via an adaptation of a demonstrated EBI and by incorporating aim-2 participants’ own drivers of survivorship and resilience while respecting users’ personal beliefs and local knowledge [87].

Overarchingly, LIFT activates the CTSC, relying on that theory’s distinction between problem- and emotion-focused coping [33-35]. Through this lens, a range of adjunctive coping skillsets is taught, practiced, and discussed by group members in the course of a LIFT implementation in the field. These include breathing awareness and retraining, mindfulness meditation, journaling and distinguishing problem- versus emotion-focused and adaptive versus maladaptive coping styles, setting attainable personal goals, and surmounting common trauma cognitions (eg, all-or-nothing thinking). Theoretically grounded details of their precise trauma-informed mechanisms of action are provided by Sikkema et al [9,34,36] and in the LIFT Manual (hereafter, the Manual) [33]. The essentials of integrating each skillset into a LIFT participant’s daily life are reinforced through a range of skillset-specific handouts (eg, “Handout 4: Guidelines for Journal Writing”). The content of the NOLA GEM app was adapted by combining the Manual’s facilitator notes, sample (facilitator) scripts, and handouts [33]. In resituating the provision of this skill-building content from an in-person sequential group modality to a smartphone-delivered just-in-time modality, we envisioned use cases in which, for instance, the stress of hearing gunshots in the morning could be ameliorated via real-time access to breathing awareness or journaling (both emotion-focused coping skills per the CTSC [35]), interrupting the possibility of resorting to maladaptive coping strategies such as alcohol misuse.

A JITAI provides individually tailored intervention content precisely when needed, typically by tapping into the key affordances of smartphone-based models of delivery. Key elements of a JITAI include decision points, the time points at which the precise provision of intervention content is decided; tailoring variables, the data inputs that determine the nature of the intervention content offered; and decision rules, which embed the logics of tailoring variables at each prespecified decision point [60,61]. The end points of a JITAI logic are proximal (or short-term) outcomes, typically mediators of an intervention’s ultimate distal outcomes [60]. JITAIs have been reliably demonstrated to drive significant improvements across behavioral health domains in pretest-posttest trials and against waitlist, inert, and active controls [93].

To attain a balanced age distribution, aim 2 planned, initially, to recruit 5 to 10 participants from two studies: (1) the parent New Orleans Alcohol Use in HIV cohort (age range 18-65 years [72,73]), whose participants’ ages were concentrated in middle adulthood (median 50, IQR 41-56 years), and (2) the adolescent and young adult–focused Adolescent Medicine Trials Network for HIV/AIDS Interventions cohort (age range 12-24 years [94]; aim 2 planned to recruit participants aged ≥18 years exclusively). Ultimately, we were only able to enroll participants (n=12; mean age 57.83, SD 10.12 years; 58.33% Black and 75% male) from the parent longitudinal study [72]. Interviews ranged from 34 minutes, 31 seconds to 1 hour, 32 minutes, 18 seconds in length. Participants were given the option of a Zoom (Zoom Video Communications)-conducted [95] versus an in-person interview; 2 preferred Zoom. Adherent to the Tulane University Health Sciences Center COVID-19 control protocols then in place, all participants were required to bring proof of vaccination. Interviewers, who conducted the interviews in pairs, and participants remained masked at all times. Data collection proceeded subsequent to a verbal informed consent script.

The sociodemographic attributes of the aim-2 participants are shown in Table 3.

Interviews were semistructured in format and mixed methods interpretivist in orientation [83]. Initially, participants were prompted to designate how helpful they would find the skillsets offered by LIFT and a range of potential mHealth modalities (eg, SMS text messages or games) on a 4-point forced-choice Likert-type scale. The aim-2 sample was LIFT naive; the particular coping skillsets were introduced via brief interviewer-administered synopses (eg, “When managing daily stress related to life with HIV, how helpful would you find breathing retraining or finding different ways to breathe?”), with scripted in-depth definitions abstracted from the Manual available as needed (eg, “The way we breathe can affect the way we feel. Daily stress and anxiety can change the way we breathe and keep us feeling stressed. Breathing retraining describes ways we can adjust our breathing to help us relax when managing everyday stress”). Unstructured probes reflecting interviewees’ own language wherever possible elicited further detail on within-person patterns of successful coping grounded in the lived experiences of violence-affected people living with HIV in New Orleans. In addition, participants were asked to describe stress-inducing and reliably calming places and personal histories of violence, spirituality, relationships, drugs, and alcohol. A series of items grounded in design justice principles [87] prompted participants to reflect on the idiographic influences of their gender, ethnoracial and cultural heritage, and ability or disability on (1) their preferred stress management techniques and (2) the ways in which they interact with and selectively disengage from digital technologies. Each interview culminated in an adapted think-aloud exercise calling upon tenets of human-centered design [96,97] in which participants were reminded of the LIFT skillsets and mHealth modalities that they had previously considered “helpful” or “very helpful” and of particular stressors that they had characterized as “able to throw [their] whole day’s plans off track,” incite drug and alcohol misuse, or disrupt ART adherence. Aiming to encourage creative leaps [96] while eliciting context-specific JITAI dynamics rooted within the life worlds of New Orleans people living with HIV, participants were asked to narrativize precisely how their preferred LIFT skillsets would aid their coping through their most acute stressors, with impromptu probes (whenever possible) articulating details of their ideal mode of smartphone-based delivery. All interviews were audio recorded using a Sony ICD-PX470 digital voice recorder and transcribed verbatim.

The qualitative analyses were subdivided into two phases, each with a dedicated intervention adaptation purpose:

The app development timeline necessitated this subdivision. Programming operationalized contender proximal outcomes as “SkillCheck” quizzes (described later in this paragraph) was an early, urgent development milestone, whereas in-app content adaptation could occur on a more generous, investigator-initiated schedule. The analytic procedure of the first phase of aim 2 relied, first, on a variant of rigorous and accelerated data reduction, a rapid matrix analysis technique [98]. Coders (n=7) condensed each interview into its richest, most salient excerpts using matrices in Microsoft Excel (Microsoft Corp) [99]. Independently, each coder inductively developed first-cycle (“open” in grounded theory terminology) descriptive, process, and in vivo codes [100]. Concurrently, each coder, consistent with a causation coding approach [100], identified contender proximal JITAI outcomes—inductively derived potential mediators of the NOLA GEM intervention’s predetermined distal outcomes [60,61]. The first author (SJS), in collaboration with LVH, synthesized these initial codes and contender proximal outcomes into second-cycle (“axial”) codes. These contender proximal outcomes were then preliminarily operationalized as “SkillCheck” quizzes, which would be provided to end users upon completion of prespecified units of intervention content (a process that is detailed in the Intervention Design section).

Subsequently, in the analysis in the second phase of aim 2, the second-cycle coding schema and complete interview data set were migrated to Dedoose (SocioCultural Research Consultants) [101]. In Dedoose, 3 coders, accompanied by the first author (SJS), pilot coded randomly selected full transcripts masked to each other’s coding applications, resolved coding discrepancies, and refined the final codebook in conference until a pooled intercoder reliability of κ=0.78 was achieved [102]. After obtaining acceptable intercoder reliability, the balance of transcripts was coded independently (KS, LVH, and MBS).

A total of 10 contender proximal outcomes were detected through the rapid matrix analyses in the first phase of aim 2. These constructs and the SkillCheck items used to operationalize them as potential mediators are provided in Textbox 1.

Building on the insights from the first phase of aim 2, refined survivorship and resilience themes, with 2 nested subthemes, were detected across the inductive analyses in the second phase of aim 2. Methodologically, we generated second-cycle pattern codes, formatted as process codes, which rely on gerunds to (consistent with dynamic JITAI logics [60]) “imply actions intertwined with the dynamics of time” [100]. The complete set of themes, subthemes, definitions, and illustrative excerpts is provided in Table 4.

Lean user experience (UX) recognizes the challenge of user-centered outcomes losing focus over the course of complex, multistakeholder technology development cycles—a challenge that is intensified when faced with extant, theoretically grounded EBI logics [33,34]. As such, we called on Lean UX techniques to integrate the full gamut of formative insights cultivated across aims 1 and 2, counteracting the (interrelated) possibilities of stakeholder misalignment or theoretical fragmentation [103] (as an added strength of this approach, Lean UX exists in tandem with the Agile workflows adopted by most contemporary software developers [88]).

Specifically, using the Miro whiteboarding platform [104], we assembled a feature hypothesis chart and conducted a design workshop in which the formative research undertaken to date was organized according to a JITAI-consistent intervention logic [60,61,91]. Hypothesis in this context refers to an easily communicable single-sentence association between a user outcome; a software feature; and (in terms by Mohr et al [105]) the usage aims and elements of a behavioral intervention technology logic, whereby each element is positioned as a column header in the feature hypothesis chart. The workshop, facilitated by the first author (SJS) in June 2022, formulated feature hypotheses accordingly: “We will achieve [distal outcome] if user can overcome [environmental stressor (derived from Aim I descriptives)], operationalized via [daily diary input (selected from among Aim I measures)], to attain [contender proximal outcome (derived from aim-2 rapid matrix analyses)], by mastering [LIFT coping skillset (abstracted from the Manual)], in a manner informed by [survivorship and resilience themes (derived from Aim IIb thematic analysis)], delivered through [mHealth multimedia modality].” All faculty investigators and interviewers (serving as end-user advocates) were present. A mocked-up feature hypothesis chart is shown in Figure 1 [88,104].

Subsequently, these JITAI logics were refined into user story maps, a user experience research technique that uses simple but summative flow diagrams to portray complex in-app interactions and contingencies [88]. An illustrative excerpt of the user story maps that we provided to our developers, Crowdbotics, is shown in Figure 2 (hosted in full resolution [106]).

The complete user story maps, encompassing all possible daily diary–coping skills recommendation–SkillCheck sequences, are provided in Multimedia Appendix 2 (which can be downloaded in full resolution [107]).

Fundamental to a JITAI adaptation is the determination of (1) tailoring variables; (2) the timely provision of appropriate intervention content, mapped to a plausible contender proximal outcome, at each predefined threshold of a tailoring variable; and (3) clear, succinct, programmable logics defining the mechanisms of (1) and (2) [60,61]. NOLA GEM incorporates a subset of aim-1 daily diary inputs as tailoring variables. On the basis of aim-1 findings, the constructs and environmental stressors that will initialize the NOLA GEM app’s JITAI logics are mental health challenges; financial problems; direct violence experienced firsthand, such as an assault (with a built-in emergency response management protocol, if warranted); indirect violence, witnessed or overheard; past violence, resurfaced in the form of intrusive thoughts, memories, or trauma cognitions [82]; interpersonal discord or lack of social support; HIV-related stigma; alcohol-related craving and coping; and a general stress measure, the triggering of which will offer the end user a selection of all LIFT skills.

These stressors are operationalized via the daily diary measures piloted initially in aim 1. If the end user indicates an acute stressor during a daily diary beyond a prespecified cutoff value, they receive an in-app notification after completion of the daily diary inviting them to open a recommended LIFT skill. Should multiple postthreshold stressors be endorsed, the user is given the option to select their preferred LIFT skill. The microcopy of these messages is uniformly strengths-based, encouraging, and nonpathologizing (eg, “It looks like you might be dealing with some stress today. Based on your daily diary responses, NOLA GEM recommends reviewing these skills”). Embedded hyperlinks will automatically launch the recommended LIFT session if a user chooses to. Aiming to minimize “interaction cost,” or the cognitive load associated with navigating a user interface [108,109], multifaceted interactivity was largely eschewed in favor of brief, “snackable” content [110] in which practical skills are foregrounded, available in text and audio formats. Each session encapsulates the CTSC-based strategies taught by the manualized LIFT intervention [33,35]. NOLA GEM users are not required to complete each skill within a single session. Upon completion at a user-directed pace, end users are presented with single-item “SkillChecks,” which assess the degree to which each LIFT skillset was successfully imparted by the app and operationalize the proximal outcome to which the LIFT session and tailoring variable were mapped on the feature hypothesis chart using dedicated items in a Likert-type format. Structuring SkillChecks in this fashion further reduces interaction costs and permits both knowledge delivery and proximal outcomes to be examined as potentially distinct mediators of distal outcomes in aim 3.

In addition, supplanting the originally envisioned generalized geospatial JITAI triggers [79], NOLA GEM users, during onboarding to the app, can leverage a geofencing feature to prespecify areas or locations that reliably evince stress. With such geofencing established, entering the prespecified area will activate an SMS text message prompt (“It looks like you may be in a stressful place”) recommending each user’s favored LIFT coping skills. In addition, foregrounding resilience, an option to prespecify a place that end users associate with personal restorativeness is included.

Illustrative screenshots of the NOLA GEM user interface are provided in Figure 3 (hosted in full resolution [111]).

The pilot trial will be a single-blind parallel superiority trial assessing NOLA GEM (treatment arm) against GEMA alone (control arm) for a period of 21 days, allocated at a 1:1 ratio. In contrast to aim 1, the NOLA GEM app is capable of delivering SMS text message–linked daily diaries and recording GPS coordinates without recourse to third-party software. The GEMA-alone control was chosen to control for any intervention effects of recording daily mood, experiences, and other psychosocial variables as some research has demonstrated improvements in outcomes such as PTSD, depressive symptoms, and emotional awareness in participants recording their experiences using EMA methods [112-115] (aim-2 interviewees, unprompted, disclosed similar enhancements in self-awareness during their experience as aim-1 participants). This design establishes a high-resolution comparator to NOLA GEM, capturing GEMA data across both trial arms while potentially isolating the effects of JITAI-based momentary provision of LIFT content as a hypothesized mechanism of action.

The aim-3 inclusion criteria are equivalent to those of aims 1 and 2, although both HIV outpatient clinic–based as well as community-based recruitment strategies will be used, inviting people living with HIV across the New Orleans metropolitan area to learn about the study through flyers and screening in real time via a QR code. Once the code is scanned, users will be brought to a Qualtrics survey that will ask basic demographic questions to facilitate quota sampling, history of living with HIV, current ART medications, and past violence exposure, which will be assessed via the Adverse Childhood Experiences Questionnaire and US Department of Veterans Affairs National Center for PTSD Life Events Checklist for DSM-5 [116].

An option to screen by phone, with a research assistant manually inputting each potential participant’s screener responses, will also be available. Once individuals are determined to meet the eligibility criteria, they will be called to set up an intake appointment.

The target sample size is approximately 100, which will accommodate an anticipated ≥10% attrition rate based on previous experience enrolling local violence-affected in-care people living with HIV, for an optimal analytic sample of 60 (30 per arm).

Participants will be recruited using flyers from several clinics and community locations in New Orleans that offer services to people living with HIV. Many of the community sites provide linkage to care by guiding people through insurance eligibility, appointments, and medication access. Two of these community sites were founded to provide services specifically to Black and African American women living with HIV. The clinic sites are federally qualified health centers that provide a range of care; however, we are working specifically with their HIV prevention and care teams.

By and large, aim 3 will comprise a digital trial subsequent to NOLA GEM onboarding. Screeners will be completed either on the participants’ phones or by calling research staff. Once participants are screened, intake appointments will take place either over Zoom or at private clinical facilities within the Tulane University School of Public Health and Tropical Medicine. All further aspects of the study will take place on the NOLA GEM app or over the phone.

With eligibility determined, participants will be randomized, and a baseline appointment will be scheduled. Quota sampling will ensure a ≥50% female, 50% aged >40 years and 50% aged <40 years, and ≥75% Black sample, reflecting the unique drivers and PTSD symptom constellations endured by women and Black Americans while also reflecting New Orleans people living with HIV demographics.

At baseline, informed consent will be obtained, and an in-person or Zoom assessment will be administered by the research staff. The NOLA GEM app (through which participants can be assigned to either treatment or control user types via an integrated back end) will be installed on participants’ smartphones from the Crowdbotics-hosted Apple App Store or Google Play Store, and a detailed onboarding will be conducted by research staff to ensure uniform comprehension of the app’s features, the nature of GEMA, and the wording of daily diary items. Although participants will be instructed on the expectations, functional affordances, and user expectations specific to their trial arm, they will be blinded to their explicit treatment versus control assignment.

An iPhone SE smartphone will be provided to participants without access to a personal smartphone. Devices will be provided to participants with a scheduled return date. Once returned, iPhones will be sanitized and recharged, and all device histories (calls, browser, and SMS text message) will be cleared. If participants do not return the study-provided iPhone and become unreachable, they will be considered dropped from the study.

The study app is Health Insurance Portability and Accountability Act (HIPAA) compliant and will transmit daily diary and GPS data to a Crowdbotics-hosted administrator panel, also HIPAA compliant, eschewing sensitive, identifiable data storage on participants’ personal devices. GPS data collection will occur continuously, recording latitudinal and longitudinal points at 1-minute intervals. In both arms, participants will be prompted twice each day by SMS text message to complete their daily diary items within a 3-hour window (9 AM-noon and 6 PM-9 PM) whenever ready and able to find privacy.

The aim-3 daily diary items are shown in Multimedia Appendix 3.

Throughout the conduct of the aim-3 trial, data will be securely stored on REDCap, and data sets that are downloaded for analysis will be stored on secure Tulane-hosted Box cloud servers. NOLA GEM user-level paradata, including daily diary responses, will be downloaded from the HIPAA-compliant Crowdbotics-hosted administrator panel, manually cleaned, and transformed into a standard “wide” data format in which each row corresponds to a unique participant using dedicated R scripts (R Foundation for Statistical Computing).

All data collected will be monitored weekly by the research staff to ensure the privacy and confidentiality of the participants. General quality assurance checks will also be run on the data during and after collection. Any adverse events or unanticipated problems will be reported immediately to the institutional review board. To ensure the confidentiality of the data, each participant is assigned a unique identifier that will not be linked to any identifying information. Any technology used (Zoom, Qualtrics, NOLA GEM app, or REDCap) is HIPAA secure. All SMS text message–delivered reminders sent to participants during the research period will be generic and will have no specific content in case participants wish to keep their participation private.

Intervention arm participants will be provided access to the SMS text message prompts; daily diary interface; a resource list that covers local health care, housing, domestic violence, family, and mental health services; a “My Progress” screen where they can track elements of their interaction with the app; and the full suite of LIFT sessions, skills, and SkillChecks. Following the submission of each daily diary, the study app will recommend specific LIFT content based on participants’ particular patterns of momentary stressors. In addition, participants can access all content in their own time.

Control arm participants will have access to the SMS text message and daily diary functions exclusively plus the resource list provided to treatment arm participants.

JITAI-specific proximal outcomes have been summarized previously. The aim-3 distal outcomes and their precise operationalization are outlined in this section.

Feasibility and acceptability will be assessed after participation in the 21-day trial period. Feasibility will be operationalized via screening and enrollment rates; completion rates of daily diaries; and paradata describing within-person patterns of use, including JITAI-triggered prompts [61,117]. Acceptability will be operationalized as follows. Intervention arm participants will complete the User Experience Questionnaire (short version), with ratings of the NOLA GEM app’s pragmatic and hedonic quality User Experience Questionnaire domains assessed against available benchmarks [118]. The single-item Net Promoter Score will ask about participants’ likelihood to recommend NOLA GEM on a 0 to 10 scale [119]. Both measures will be administered at the immediate postintervention assessment.

Attrition rates will also be recorded, and those who drop out of the study will receive a set of questions exploring motivations for discontinuation wherever possible.

Primary outcomes, representing recognized HIV CoC milestones [1,21,23], will be operationalized as follows via a postassessment survey.

Diverse domains of psychosocial and emotional well-being will be assessed among all aim-3 participants.

Consistent with aim 1, aim-3 data will be cleaned and saved in SAS data files (SAS Institute), with extensive consistency checks conducted at all levels of data [66]. At the completion of each check, the data will be added to the existing database for analysis. GPS data will be stored on a password-protected server and constructed into activity paths, replicating the aim-1 approach [66].

Preliminarily, descriptive and crude bivariate associations between variables of interest will be generated. Daily diary completion rates and agreement with aggregate retrospective survey data for comparable measures will be calculated. Data reduction techniques, mitigating the high volume and dimensionality of GEMA-derived data, will include the creation of composite indexes of cumulative risk, with exploratory factor analysis, assessment of internal consistency, and Bonferroni correction applied as warranted.

Aim-3 analyses of preliminary efficacy will include analysis of each day-to-day distal outcome as a continuous variable as well as average exposures across the total intervention period. We will explore potential transformations to approach normality on key response variables [133]. Linear mixed-effects models will relate day-level outcomes to day-level indexes of cumulative risk and protective exposure derived originally from aim 1, adjusted for baseline values [134]. Random effects will be included to account for within-subject series correlations. Control variables will be modeled as fixed effects.