Evaluating digital intervention approaches for supporting immigrant women with intimate partner violence experiences: Findings from the It's weWomen plus sequential multiple assignment randomized trial (SMART)

Abstract

Background

Intimate partner violence (IPV) disproportionately affects immigrant women, who often face barriers to accessing in-person services. Digital interventions offer a promising alternative by providing tailored, remote support.

Methods

In this SMART trial, 1265 foreign-born immigrant women across the U.S. were randomized to a personalized online (n = 660) or standard online safety information (n = 605) intervention. At 3 months, low responders (n = 366) were re-randomized to receive text-only (n = 183) or text + phone support (n = 183). Outcomes were assessed at 6 and 12 months.

Results

All groups showed reduced physical and sexual IPV over time, with no significant differences between first-stage conditions. Low responders in the text + phone group demonstrated significantly greater reductions in physical and sexual IPV (d = −0.25, p < 0.01), depression (d = −0.22, p < 0.01), and increased empowerment (d = 0.22, p < 0.01), from 3 to 12 months, compared to responders. These between-group effects were supported by significant within-group improvements, with the text + phone group narrowing or closing the gap with responders in most outcomes by 12 months. Among low responders initially assigned to the personalized online intervention, those re-randomized to text + phone support outperformed those receiving text-only support—showing significantly greater reductions in IPV (d = −0.32, p < 0.05), depression (d = −0.33, p < 0.05), and greater gains in empowerment (d = 0.27, p < 0.05). The text-only group also improved, particularly in depression and PTSD, with outcomes approaching those of responders by 12 months. Across conditions, low responders also showed substantial improvements in safety behaviors (d = 0.24–0.25; p < 0.05).

Conclusion

These findings highlight the value of stepped-care, adaptive approaches in addressing persistent IPV-related needs. Integrating personalized phone support into digital interventions can enhance outcomes for survivors who do not respond to brief, initial support alone.

Highlights

• •Digital interventions can be effective for immigrant women facing IPV.
• •Tailored digital support is key for women unresponsive to standard interventions.
• •Combined text and phone support can be effective for enhanced intervention.

1. Background

Intimate partner violence (IPV) is a critical public health problem that particularly impacts immigrant women in the United States. Studies indicate a broad prevalence range for IPV among this group, with lifetime IPV victimization rates ranging from 13.9 % to 93 %, depending on sample characteristics such as gender composition, sampling methods, and definitions of IPV [1]. For instance, studies that focus exclusively on immigrant women, employ convenience sampling or use broader definitions of IPV that include psychological abuse tend to report higher prevalence rates. In contrast, lower rates are often found in studies with mixed-gender samples, randomized sampling, or narrower definitions of IPV [1]. Being in abusive relationships places immigrant women at a high risk of severe violence or even homicide by male partners [2,3]. Their vulnerability stems from barriers to accessing services, such as undocumented immigration status, a lack of knowledge about available U.S. services, and limited social support networks [4]. Additional barriers to accessing services include socio-cultural norms, institutional racism, issues of cultural competence, and a lack of diversity within frontline services [5]. These obstacles can intensify feelings of isolation, powerlessness, and fear [5], increasing their likelihood of ongoing abuse, severe IPV, and the potential for lethal outcomes. To address these barriers, there is a need for evidence-based interventions that immigrant women can access from any location and can support them in a remote format.

Digital interventions offer a promising solution to overcome barriers immigrant women often encounter in traditional IPV services. These digital interventions can be delivered through many forms, such as web-based platforms, mobile devices, and text messages. Such interventions may be a viable source of support for historically marginalized survivors of IPV, including immigrant women [6]. Safety planning is a critical component of digital interventions designed to help women mitigate IPV-related risks. This approach involves working with survivors to assess their unique situation, identify potential dangers, evaluate current safety strategies, and outline a personalized plan to prevent or respond to IPV [[7], [8], [9], [10]]. Digitally delivered safety planning interventions can empower immigrant women to make safety plans, build knowledge of available resources, and access care discreetly. These can also help women enact their safety and minimize harm from abuse [11,12]. Given the privacy offered by digital platforms, such interventions can reduce exposure to stigma and the discomfort associated with discussing sensitive topics in person [6,11,12].

Given that immigrant women may experience distinct forms of abuse (e.g., threat of deportation) and face specific risk factors [13], their safety planning needs differ from those of non-immigrant women. Studies have highlighted the necessity of culturally tailored safety planning strategies that address these unique needs, such as ensuring services are accessible without formal identification and providing information on legal rights in the U.S. [4,11,12]. However, research on digital interventions tailored specifically for immigrant women who experience IPV remains limited. Additionally, a single intervention approach may not be effective for this heterogeneous population, as variations in responses can be influenced by individual circumstances such as fear of partners, everyday discrimination, immigration-related stressors, or insufficient social support [11]. Consequently, our study utilized a Sequential Multiple Assignment Randomized Trial (SMART) design to evaluate the effectiveness of digital intervention approaches for immigrant women. The SMART design provides a framework for adapting intervention components based on participant response, enabling a more personalized approach to intervention delivery [14]. An adaptive intervention adapts the type or dosage of an intervention based on the responses or characteristics of participants [15]. By tailoring the type and intensity of intervention based on specific needs and response patterns, this design aims to enhance the intervention's effectiveness and applicability in real-world settings [11,16,17]. The adaptive intervention approach is essential for addressing the complex and varied needs of immigrant survivors of IPV.

This study used a SMART design to identify strategies for improving the safety and well-being of immigrant women in abusive relationships. In the first stage, women were randomized to receive either a personalized online intervention (a culturally adapted immigrant version of myPlan) [18] or standard online safety information alone. This initial comparison evaluated the effectiveness of the two approaches in improving the outcomes related to safety, mental health, and empowerment for survivors of IPV. Women who did not show significant improvement—classified as low responders —were re-randomized to second-stage interventions to receive either supportive text messaging alone or a combination of text messaging and phone support. This adaptive design enabled the evaluation of both initial intervention effects and optimal strategies for supporting low responders. The primary aims were to (1) assess the relative effectiveness of supportive text messaging versus combined text and phone support in improving outcomes among low responders and (2) assess whether outcomes among low responders who received second-stage interventions differed significantly from those of initial responders to first-stage interventions. We hypothesized that women who received the personalized online intervention would experience better outcomes than those who received standard online safety information. Among low responders, we expected both second-stage interventions to be beneficial, with the combination of text messaging and phone support yielding greater improvements than text messaging alone. Finally, we anticipated that low responders who received a second-stage intervention would demonstrate improvements in safety, mental health, and empowerment and that their outcomes would approach those of initial responders.

2. Method

2.1. Study design

The study employed a SMART design to evaluate the impact of distinct digital intervention components—personalized online intervention, standard online safety information, text messaging, and phone support (Table 1)—on safety, mental health, and empowerment outcomes for recent immigrant women survivors of IPV. This study is registered with ClinicalTrials.gov as NCT04098276. The study flow diagram is presented in Fig. 1.

Table 1.

WeWomen plus SMART trial: Intervention components and delivery timeframe.

Digital Interventions	Description	Timing/Duration
Personalized Online Intervention	Safety decision and planning support delivered via a secure web app, including: •Danger Assessment (DA) to assess the level of risk •Priority-setting activities •Tailored safety plan with links to resources	•Immediately after baseline assessment •One-time use with the option to return •Users could access the intervention at any time and complete it at their own pace, with multiple logins permitted.
Standard Online Safety Information	Online website with •Usual safety planning resources modeled on national/state domestic violence resources •No DA or personalized safety plan	•Immediately after baseline assessment •One-time use with the option to return •Users could access the intervention at any time and complete it at their own pace, with multiple logins permitted.
Text Messaging Intervention	Strengths-based empowerment text messages focusing on: •Week 1: Safety/Risk assessment and planning •Week 2: Self-care and support resources •Week 3: Risk/Danger assessment and service referrals •Week 4: Resources and long-term safety goals	•Once weekly for 4 weeks
Phone Call Intervention	Trauma-informed, empowerment-focused calls integrating: •Strengths-based perspective •Motivational enhancement •Solution-focused techniques •Safety assessment •Goal setting and reinforcement	•Delivered after completion of the text component • 2 phone sessions total: - First call: 1 week after text ends - Second call: 1 week after first call

Fig. 1.

The study flow diagram.

Note. The safety outcome included measures of physical and sexual abuse (primary outcome) and safety behaviors (secondary outcome).

2.2. Participants and procedures

Eligible participants were 18–64 years of age, who identified as a woman (assigned female at birth or had a gender identity as a woman), experienced IPV in the past 12 months, were foreign-born immigrant women from any race and background, had access to a private, secure computer or smartphone that was not monitored or shared by an abusive partner, and were comfortable using it to access the study site. Participants also needed to be willing and able to be contacted via phone or text for enhanced intervention evaluation. IPV eligibility was based on a brief self-report screening; participants were eligible if they reported any past-year experience of physical, emotional, or sexual abuse or fear of a current or former partner. Participants were excluded if they were U.S.-born, did not report IPV in the past year (i.e., answered “no” to all screening questions), or lacked access to a safe device.

Participants were recruited using multiple in-person and remote recruitment strategies from multiple states across the US. In-person recruitment took place in high-traffic areas frequented by immigrant women, including healthcare clinics, community facilities, community events and festivals, beauty parlors, markets, grocery stores, and religious centers. The study information was also disseminated through immigrant-specific general service providers, health clinics, and immigrant IPV support agencies throughout the US. Additional outreach strategies included listservs, social media platforms (e.g., Facebook), Research Match, and peer-to-peer referrals to inform the immigrant community about the opportunity. Flyers were posted at various locations within the city, including around the university. University student groups, including cultural or ethnic student associations, also distributed the study information within their networks.

Study advertisements used accessible language to emphasize women's health, safety, and well-being, such as, “Do you feel unsafe in your relationship?” “Be part of a research study for immigrant women's safety, health, and well-being.” These messages aimed to engage immigrant women seeking support, information, or resources. All materials highlighted that participation was voluntary, confidential, and could be completed online. Interested women were directed to a secure website with detailed information about eligibility criteria, confidentiality protections, and the scope of the study. For in-person recruitment, staff approached potentially eligible women and helped them complete the screening on study tablets or laptops. Women who chose not to screen or who were not found to be eligible were encouraged to take study flyers and share the information with their networks.

The screening process was completed online, where women could provide their contact information—such as a secure phone number and email address—to receive study-related information. Study staff then contacted eligible participants to confirm their eligibility over the phone and prevent false or repeat registrations. Verbal consent was obtained from all valid, eligible, interested, and willing participants, who also completed an electronic, written consent when logging into the study website and were required to indicate consent before proceeding to the survey. Study staff implemented safety protocols to protect participants during survey completion and while engaging in online, text, and phone-based interventions. These protocols included strategies such as reminders to delete text messages after each session, the use of unique login credentials, and the use of code words during participant interactions. Data were collected via web-based questionnaires at baseline and 3-, 6-, and 12 months following the initial baseline assessment. Participants received a $40 incentive for each completed survey. All study materials, including questionnaires and interventions, were available in English, Spanish, and Arabic to enhance accessibility and inclusivity for participants from diverse language backgrounds. The study procedures were approved by the Johns Hopkins School of Medicine Institutional Review Board.

2.3. Intervention components

2.3.1. Personalized online intervention

The personalized online intervention included the culturally adapted evidence-based myPlan intervention [18,19], a free, secure web-based app designed as a safety decision-making tool based on Dutton's empowerment model [20]. It was adapted for immigrant populations, specifically drawing on qualitative data from Asian, African, and Latina women to enhance cultural relevance [4]. The intervention, which was self-administered without a separate pre-intervention tutorial, involved education about healthy relationships, users' assessment of the severity of danger in their relationship using the Danger Assessment (DA) [21], clarification of safety priorities, reduction of decisional conflict, and access to resources for healing from the impacts of IPV. Before using the intervention, users were guided through a safe technology tutorial and asked to create a 4-digit password. After onboarding, they received a general safety plan with strategies such as contacting a hotline, preparing an emergency bag, and safeguarding money and important documents. Users could further personalize their safety plan by completing three sections: identifying warning signs of unhealthy relationships, completing Danger Assessment (DA)—a validated measure that evaluates the risk of severe or lethal violence [21]—receiving immediate feedback based on their danger score, and setting personal priorities based on their situation. An evidence-based algorithm informed by the DA scoring protocol [21], integrated these responses to generate a tailored safety plan with links to relevant local and national services. The intervention was designed for flexible access, allowing participants to use their preferred digital devices (e.g., computer or smartphone) at a time and place that felt safe and convenient.

2.3.2. Standard online safety information

The standard online safety information website provided women with usual safety planning resources modeled on national and state domestic violence online resources. This component was also self-administered without a separate pre-intervention tutorial. Women in the standard safety information component did not complete the DA or receive safety planning informed by the DA. However, safety check-ins were provided to participants in all conditions.

2.3.3. Text-messaging

The text messaging component was grounded in a strengths-based empowerment framework [20,22] and the Psychosocial readiness model [23], which informed the theory of change underlying the intervention. The hypothesized mechanism of change was that increasing women's awareness, self-efficacy, and access to supportive resources would enhance their readiness to engage in safety-related behaviors—even in the presence of ambivalence, fear, or logistical barriers. The messages were designed to gently engage women along a continuum of change by meeting them where they were, supporting autonomy, and honoring their choices, priorities, and needs. Depending on each woman's level of readiness, the text messages aimed to increase awareness of safety planning, highlight available community services, provide safety check-ins, assess risk for severe violence, identify barriers to implementing safety steps, and build confidence in carrying them out. Messages were tailored and delivered once a week for four weeks, at times and on days selected by women for their safety and convenience. Recognizing that behavior change in the context of IPV is rarely linear, the messages emphasized incremental progress and supported self-directed action rather than offering prescriptive guidance. Sessions were primarily automated, with approximately 98 % of women receiving messages delivered through the Clinical Trial Management System following a structured dialogue flow based on participant input. A small subset (2.3 %) of women who preferred manual delivery received messages from research assistants using the same scripted dialogue. Regardless of the delivery method, the content used an empathetic tone, reinforced self-worth, emphasized self-care, and concluded with a list of emergency resources and safety reminders, such as instructions to delete messages for protection.

2.3.4. Phone call

The phone call component complemented the text messaging by offering a more personalized, interactive means of support grounded in a trauma-informed, strengths-based approach. The theory of change guiding this component drew from motivational enhancement [24,25] and solution-focused therapy [26], which suggest that empathic engagement, recognition of strengths, and collaborative goal-setting can build motivation and foster behavior change. The calls helped reduce ambivalence, increase engagement, and support incremental steps toward safety, health, and well-being by validating women's emotions, reframing challenges as opportunities, and reinforcing personal agency—while recognizing that meaningful change in the context of IPV often unfolds through non-linear and individualized processes. Trained facilitators used empathetic, strengths-based communication to support healing and recovery, emphasizing women's protective factors and helping them reframe challenges into accomplishments. Each phone session also included a safety assessment, allowing facilitators to tailor support to women's specific risk contexts. Two calls were delivered: the first occurred one week after the text message series concluded, and the second one week later. During the initial call, facilitators established rapport and trust, addressed women's primary concerns, emphasized their achievements, and reframed negative thoughts into strengths. They used motivational interviewing techniques to foster self-reflection and jointly set attainable goals aligned with the women's values and priorities. The second call focused on reviewing progress, identifying barriers, and reinforcing goals. Facilitators explored exceptions to problem patterns, highlighted cultural and personal strengths, and helped women assess available resources. The phone call component was offered in addition to the text messaging intervention for participants randomized to the text + phone group. Women in this group received a total of six weeks of support (four weeks of text messages plus two weeks of phone contact), compared to four weeks of support for participants randomized to the text-only group.

2.4. Measures

2.4.1. Primary outcome

2.4.1.1. Physical and sexual abuse

Physical and sexual abuse were measured using the adapted version of the Revised Conflict Tactics Scale (CTS2; 15 items) [27] (alpha = 0.93). The subscales of CTS2 include physical aggression, injury, psychological aggression, and sexual coercion. Response categories ranged from −0 = never to 4 = very frequently. Higher values on CTS-2 within the past 12 months indicated severe or more frequent experiences of violence. The items were scored using the severity-times-frequency weighted score for physical and sexual abuse, as recommended by Straus [27]. While the CTS-2 administered at baseline assessed experiences in the past 12 months, follow-up surveys at 3, 6, and 12 months asked participants to report on changes in their experiences since the previous survey.

2.4.2. Secondary outcomes

2.4.2.1. Depression

Depressive symptoms were measured using The Patient Health Questionnaire (PHQ-9) [28] (alpha = 0.90), a nine-item measure designed to assess depression symptoms over the past two weeks based on the diagnostic criteria for major depressive disorder in the Diagnostic and Statistical Manual (DSM-IV). Each of the nine items scored from 0 (not at all) to 3 (nearly every day). A total score was computed by summing the items to measure the severity of depression, with higher scores indicating more severe depression symptoms.

2.4.2.2. PTSD

Past week PTSD symptoms were measured using the Harvard Trauma Questionnaire (16 items) [29] (alpha = 0.94), with response options ranging from 1 (Not at all) to 5 (Extremely). The items represented intrusion/re-experiencing, avoidance/numbing, and hypervigilance/arousal symptom clusters. Higher scores indicated more severe PTSD symptoms.

2.4.2.3. Empowerment

The Personal Progress Scale-Revised (PPS-R) (23 items) [30] (alpha = 0.88) was used to assess multiple areas associated with overall empowerment, such as positive self-evaluation, self-esteem, ability to regulate emotional distress, gender-role and cultural identity awareness, self-efficacy, self-care, problem-solving, assertiveness skills, and access to resources. The responses were rated on a 7-point scale ranging from 1 (Almost Never) to 7 (Almost Always), with a total maximum composite score of 161. Higher scores indicated a higher level of empowerment.

2.4.2.4. Safety behaviors

Safety Behaviors were measured using 27 items from the safety strategies checklist [19,31,32]. The items assessed help-seeking strategies for IPV and how helpful they were to safety. The safety behavior score was calculated by summing the number of helpful safety behaviors and dividing by the total number tried [19].

2.4.3. Covariates

Socio-demographics. Age was recorded in years. Race was categorized as White, Black, Asian, or Other. Ethnicity was classified as Hispanic or non-Hispanic. Women's geographic region of origin —defined as the broad area of the world where they were born—was categorized as Asia, Latin America, Europe and the Middle East, Africa, U.S. nearby regions, or unknown. For length of time in the US, the time women lived in the US was divided into four categories: less than one year, 1-to 4 years, 5-to 10 years, and more than 10 years. Women's number of children was classified as 0, 1, 2, or 3 or more. Education was categorized into three levels: Associate degree or lower, Bachelor's degree, and Master's degree or higher. Relationship status was categorized as married, partnered, formerly married, or former non-marital. Sexual orientation status was classified as non-heterosexual and heterosexual. Financial stress was assessed using an item that asked women to indicate how often they ran out of money for basic necessities during each month: (1) never, (2) once or twice, (3) every month or monthly, (4) every week or weekly, or 5 (every day or daily). Employment status was categorized as full-time, part-time, seeking opportunities, or other. Living situation was categorized as living with a partner or not. Everyday discrimination experiences were measured using nine items from the Everyday Discrimination Scale [33] (alpha = 0.97). The items were rated as 1 (Almost every day) to 6 (Never). Higher scores indicated higher levels of discrimination.

Other Variable. Psychological Abuse was measured using the Women's Experience with Battering (WEB; 10 items) [34] (alpha = 0.93). The WEB was scored using the sum of items rated on a 6-point Likert scale ranging from 1 (disagree strongly) to 6 (strongly agree). The total scores were dichotomized, and a score of 20 points or higher on the WEB was considered positive for psychological abuse [34].

2.5. Sample size and power

A sample size of 1266 eligible women was calculated a priori, with 633 women randomized to the personalized online intervention and 633 to the standard online safety information conditions. The sample size was determined based on a power of 0.80 and an alpha of 0.05, requiring 633 participants per group to detect significant differences over time between the two groups, with an anticipated effect size of 0.25 or greater. We estimated that 30 % of participants in the personalized online intervention condition would be low responders. As a result, a second-stage randomization was applied to the low responders in the personalized online intervention arm. This process led to an expectation of 190 women being randomized to either the text-only or the text + phone intervention components (95 women per group). In contrast, we anticipated a higher low responder rate in the standard online safety information condition, estimating that 50 % of women (n = 316) would be randomized to the text-only or the text + phone intervention components, resulting in 158 women per group. Assuming an 80 % retention rate at 12 months, we calculated the sample size per group to detect an effect of 0.25 or greater. For the comparison between the text + phone group and the text-only group, we expected an effect size in the small to moderate range (d = 0.2 to 0.4). Ultimately, data collection resulted in 1265 participants, with 660 in the personalized online intervention condition and 605 in the standard online safety information condition. Of these, 181 low responders were in the personalized online condition, and 185 in the standard online safety information condition (86–95 per group). Our overall retention rate at 12 months was 88.4 %.

2.6. Randomization and masking

For first-stage randomization, women survivors of IPV were initially randomized to receive the personalized online intervention condition or standard online safety information condition using computer-generated block randomization stratified by length of time in the US, with study team members blinded to the assignment group. At the three-month follow-up, based on women's responses to the 3-month follow-up survey, women who had not shown significant improvements—specifically, a reduction of at least 0.25 standard deviation in IPV severity and frequency and an increase of at least 0.25 standard deviation in safety-related empowerment compared to baseline—were classified as low responders and re-randomized. These two outcomes were chosen to comprehensively assess the intervention's impact. They captured both the prevention of further harm (through reductions in IPV severity and frequency) and the promotion of empowerment (through increases in safety-related empowerment). This dual focus ensured the intervention addressed not only the immediate safety and well-being of survivors but also supported growth and resilience — such as enhancing independence, confidence, and self-determination — recognizing that the journey toward healing may begin even while survivors are still navigating abuse, or continue after leaving an unsafe relationship. The team members were not masked to the study condition for second-stage randomization. These low responders, from both the personalized online and the standard online safety information conditions, were re-randomized to receive either (1) text messaging alone or (2) a combination of text messaging and phone-based support to augment their original assignment. In contrast, participants classified as responders continued with their initial intervention assignment only and received no additional components.

2.7. Data analysis

Demographic and other baseline characteristics (Table 2) were summarized using descriptive statistics: means and standard deviations for continuous variables and frequency distributions for categorical variables. To assess differences across study conditions, chi-square tests were used for categorical variables, while two-sample t-tests were applied for continuous variables.

Table 2.

Baseline sample characteristics.

	Personalized Online Intervention (n = 660)	Standard Safety Information (n = 605)	Low responders			Responders (n = 825)	Total sample (n = 1265)
			Text-Only (n = 183)	Text + Phone (n = 183)	Total (n = 366)
Age [Mean (SD)]	36.70 (9.88)	37.39 (9.67)	36.10 (9.02)	35.98 (9.75)	36.04 (9.37)	37.33 (9.99)1	37.03 (9.79)
Race (N, %)
White	167 (25.3 %)	189 (31.3 %)	47 (25.8 %)	41 (22.4 %)	88 (24.0 %)	244 (29.6 %)	356 (28.2 %)
Black	85 (12.9 %)	86 (14.2 %)	31 (16.9 %)	24 (13.1 %)	55 (15.0 %)	104 (12.6 %)	171 (13.5 %)
Asian	222 (33.6 %)	164 (27.2 %)	57 (31.1 %)	69 (37.7 %)	126 (34.4 %)	246 (29.9 %)	386 (30.5 %)
Other	186 (28.2 %)1	165 (27.3 %)1	48 (26.2 %)	49 (26.8 %)	97 (26.5 %)	230 (27.9 %)	351 (27.8 %)
Ethnicity (N, %)
Hispanic	396 (60.8 %)	329 (55.0 %)	114 (62.6 %)	120 (66.7 %)	128 (35.4 %)	354 (43.5 %)	725 (58.0 %)
non-Hispanic	255 (39.2 %)1	269 (45.0 %)1	68 (37.4 %)2	60 (33.3 %)2	234 (64.6 %)	460 (56.5 %)2	524 (42.0 %)
Region of origin (N, %)
US nearby regions	46 (7.0 %)	41 (6.8 %)	18 (9.8 %)	7 (3.8 %)	25 (6.8 %)	57 (6.9 %)	87 (6.9 %)
Asia	180 (27.3 %)	142 (23.5 %)	45 (24.6 %)	62 (33.9 %)	107 (29.2 %)	204 (24.7 %)	322 (25.5 %)
Latin America	207 (31.4 %)	206 (34.0 %)	52 (28.4 %)	50 (27.3 %)	102 (27.9 %)	276 (33.5 %)	413 (32.6 %)
Europe and Middle East	47 (7.1 %)	40 (6.6 %)	12 (6.6 %)	16 (8.7 %)	28 (7.7 %)	56 (6.8 %)	87 (6.9 %)
Africa	46 (7.0 %)	44 (7.3 %)	22 (12.0 %)	12 (6.6 %)	34 (9.3 %)	48 (5.8 %)	90 (7.1 %)
Unknown	134 (20.3 %)	132 (21.8 %)	34 (18.6 %)2	36 (19.7 %)2	70 (19.1 %)	184 (22.3 %)2	266 (21.0 %)
Length of time in the US
Less than one year	48 (7.3 %)	43 (7.1 %)	12 (6.6 %)	15 (8.2 %)	27 (7.4 %)	62 (7.5 %)	91 (7.2)
1–4 years	117 (17.7 %)	106 (17.5 %)	33 (18.0 %)	35 (19.1 %)	68 (18.6 %)	142 (17.2 %)	223 (17.6)
5–10 years	135 (20.5 %)	128 (21.2 %)	44 (24.0 %)	39 (21.3 %)	83 (22.7 %)	165 (20.0 %)	263 (20.8)
More than 10 years	360 (54.5 %)	327 (54.2 %)	94 (51.4 %)	94 (51.4 %)	188 (51.4 %)	455 (55.2 %)	687 (54.4)
Number of children (N, %)
0	234 (35.5 %)	193 (32.0 %)	59 (32.2 %)	61 (33.3 %)	120 (32.8 %)	289 (35.1 %)	427 (33.8 %)
1	153 (23.2 %)	128 (21.2 %)	50 (27.3 %)	55 (30.1 %)	105 (28.7 %)	161 (19.5 %)	281 (22.2 %)
2	151 (22.9 %)	158 (26.2 %)	39 (21.3 %)	43 (23.5 %)	82 (22.4 %)	213 (25.8 %)	309 (24.4 %)
3 or more	122 (18.5 %)	125 (20.7 %)	35 (19.1 %)2	24 (13.1 %)2	59 (16.1 %)	161 (19.5 %)2	247 (19.5 %)
Education (N, %)
Associate degree or lower	267 (40.8 %)	228 (38.4 %)	77 (42.5 %)	81 (44.8 %)	158 (43.6 %)	301 (37.0 %)	495 (39.7 %)
Bachelor's degree	199 (30.4 %)	186 (31.4 %)	49 (27.1 %)	45 (24.9 %)	94 (26.0 %)	267 (32.8 %)	385 (30.9 %)
Master's degree or higher	188 (28.7 %)	179 (30.2 %)	55 (30.4 %)	55 (30.4 %)	110 (30.4 %)	245 (30.1 %)	367 (29.4 %)
Relationship status (N, %)
Married	331 (50.2 %)	300 (49.8 %)	89 (48.6 %)	88 (48.1 %)	177 (48.4 %)	418 (50.8 %)	631 (50.0 %)
Partnered/Coupled	117 (17.7 %)	111 (18.4 %)	35 (19.1 %)	36 (19.7 %)	71 (19.4 %)	145 (17.6 %)	228 (18.1 %)
Formerly Married	63 (9.5 %)	73 (12.1 %)	20 (10.9 %)	28 (15.3 %)	48 (13.1 %)	77 (9.4 %)	136 (10.8 %)
Single/Former	149 (22.6 %)	119 (19.7 %)	39 (21.4 %)	31 (16.9 %)	70 (19.1 %)	183 (22.2 %)	268 (21.2 %)
Sexual Orientation
Non-Heterosexual	103 (15.6 %)	97 (16.1 %)	28 (15.3 %)	31 (16.9 %)	59 (16.1 %)	127 (15.5 %)	200 (15.8 %)
Heterosexual	557 (84.4 %)	505 (83.9 %)	155 (84.7 %)	152 (83.1 %)	307 (83.9 %)	695 (84.5 %)	1062 (84.2 %)
Financial stress (N, %)
Seldom or no	266 (42.1 %)	269 (46.3 %)	65 (37.1 %)	77 (43.8 %)	142 (40.5 %)	370 (46.8 %)	535 (44.1 %)
Often	366 (57.9 %)	312 (53.7 %)	110 (62.9 %)	99 (56.3 %)	209 (59.5 %)	420 (53.2 %)	678 (55.9 %)
Employment status (N, %)
Employed full time	263 (41.1 %)	244 (42.0 %)	69 (39.4 %)	71 (39.9 %)	140 (39.7 %)	337 (42.3 %)	507 (41.5 %)
Employed part time	136 (21.3 %)	133 (22.9 %)	42 (24.0 %)	47 (26.4 %)	89 (25.2 %)	162 (20.3 %)	269 (22.0 %)
Seeking opportunities	138 (21.6 %)	121 (20.8 %)	40 (22.9 %)	40 (22.5 %)	80 (22.7 %)	163 (20.5 %)	259 (21.2 %)
Other	103 (16.1 %)	83 (14.3 %)	24 (13.7 %)	20 (11.2 %)	44 (12.5 %)	135 (16.9 %)	186 (15.2 %)
Living situation (N, %)
Not living with a partner	80 (12.1 %)	59 (9.8 %)	14 (7.7 %)	16 (8.7 %)	30 (8.2 %)	104 (12.6 %)	139 (11.0 %)
living with a partner	580 (87.9 %)	546 (90.2 %)	169 (92.3 %)	167 (91.3 %)	336 (91.8 %)	721 (87.4 %)	1126 (89.0 %)
Discrimination (Mean (SD))	13.72 (8.83)	14.65 (9.02)	15.62 (8.64)2	15.72 (9.83)2	15.67 (9.24)	13.45 (8.69)2	14.16 (8.93)
Physical IPV
No	181 (27.4 %)	181 (26.6 %)	26 (14.2 %)	19 (10.4 %)	45 (12.3 %)	281 (34.1 %)	342 (27.0 %)
Yes	479 (72.6 %)	444 (73.4 %)	157 (85.8 %)2	164 (89.6 %)2	321 (87.7 %)	544 (65.9 %)2	923 (73.0 %)
Sexual IPV
No	255 (38.6 %)	221 (36.5 %)	59 (32.2 %)	52 (28.4 %)	111 (30.3 %)	341 (41.3 %)	476 (37.6 %)
Yes	405 (61.4 %)	384 (63.5 %)	124 (67.8 %)2	131 (71.6 %)2	255 (69.7 %)	484 (58.7 %)2	789 (62.4 %)
Psychological IPV
No	117 (17.7 %)	108 (17.9 %)	28 (15.3 %)	23 (13.7 %)	53 (14.5 %)	156 (18.9 %)	225 (17.8 %)
Yes	543 (82.3 %)	496 (82.1 %)	155 (84.7 %)	158 (86.3 %)	313 (85.5 %)	668 (81.1 %)	1039 (82.2 %)
Multiple IPV
All three types of IPV	299 (47.6 %)	286 (50.8 %)	98 (55.7 %)	108 (59.3 %)	206 (57.5 %)	343 (44.9 %)	585 (49.1 %)
Two types of IPV	201 (32.0 %)	189 (33.6 %)	64 (36.3 %)	55 (30.2 %)	119 (33.2 %)	246 (32.2 %)	390 (32.7 %)
One type of IPV	128 (20.4 %)	88 (15.6 %)	14 (8.0 %)2	19 (10.4 %)2	33 (9.3 %)	175 (22.9 %)2	88 (18.2 %)
Outcomes
Physical and Sexual IPV	18.04 (25.52)	17.83 (25.08)	20.32 (23.27)2	22.12 (24.34)2	21.23 (23.80)	16.03 (25.51)2	17.94 (25.30)
Depression	9.99 (6.40)	10.39 (6.62)	10.05 (6.12)	10.31 (6.67)	10.18 (6.39)	10.13 (6.58)	10.18 (6.50)
PTSD	2.42 (0.94)	2.50 (1.00)	2.53 (0.97)	2.56 (1.02)	2.55 (0.99)	2.41 (0.95)	2.46 (0.97)
Safety Behaviors	66.93 (27.88)	64.90 (26.87)	62.19 (27.13)	64.70 (29.97)	63.44 (28.57)	67.31 (26.70)	65.96 (27.41)
Empowerment	119.61 (21.19)	118.86 (21.67)	119.65 (19.43)	121.33 (22.78)	120.49 (21.16)	118.67 (21.49)	119.25 (21.42)

Notes. Percentages were within columns. ∗p < 0.05.¹p value indicated whether sample characteristics significantly differed between personalized online intervention and standard safety information at baseline. ²p value indicated whether sample characteristics significantly differed between the three groups: low responder group (text only), low responder group (text + phone) and responders at baseline.

Following the guidelines of Nahum-Shani et al. (2020) for evaluating data from SMART studies with longitudinal outcomes [35], we assessed the overall differences between the "personalized online intervention” and "standard online safety information" over time using population-average regression models with a Generalized Estimating Equations (GEE) approach. The GEE analysis incorporated first-stage random assignment (personalized online vs. standard online safety information), time (baseline, 3, 6, and 12 months), and the interaction between treatment and time. The primary outcomes (physical and sexual IPV) and secondary outcomes (PTSD, depression, safety behaviors, and empowerment) were continuous variables. As such, the model utilized a normal distribution with an identity link function and an exchangeable covariance structure. This analysis was designed to test whether the changes over time differed significantly between personalized and standard online safety information conditions.

To assess the relative effectiveness of the two strategies for augmenting the personalized online and standard online safety information conditions with either follow-up text messaging only (text-only) or a combination of text and phone (text + phone) components on safety, mental health, and empowerment outcomes among low responders to the first-stage randomization, we used GEE model with exchangeable correlation matrix. The initial model used timepoints at baseline, 6 months, and 12 months, allowing us to assess overall changes over the full study period and account for any early response prior to re-randomization.

To more precisely isolate the effects of the second-stage interventions, we re-ran the analysis using the 3-month timepoint as the analytic baseline. This alternative model aligned with the study's adaptive design, in which low responders were re-randomized at 3 months, and allowed us to focus exclusively on change attributable to the second-stage intervention period. Both models included group (text-only vs. text + phone), time, and group-by-time interaction and were adjusted for intervention dose (Appendix). In both models, the text-only group was coded as the reference category.

To evaluate whether the second-stage augmentation strategies brought low responders to the level of responders on primary and secondary outcomes, we used the weighted GEE model with an exchangeable correlation matrix, adjusting for intervention dose. This analysis was performed using both the initial baseline and the 3-month re-randomization timepoint as alternative baselines, allowing us to compare trajectories both across the full study period and during the second-stage intervention phase. The model included time (baseline or 3 months, 6 months, and 12 months), final node group (responders, low responders assigned to text-only, and low responders assigned to text + phone), and the interaction between time and group. To account for the study design, we applied inverse probability weighting, assigning weights of 2 to responders and 4 to low responders, reflecting their respective randomization probabilities. Responders served as the reference group, enabling comparisons with both low responder groups that received augmentation. We reported p-values and 95 % confidence intervals from the GEE models without adjusting for multiple comparisons, as all primary and secondary outcomes were pre-specified and hypothesis-driven [11].

While the primary comparisons were conducted using the GEE model to assess differences in outcome trajectories across groups, we also performed within-group paired t-tests to evaluate whether each group demonstrated significant improvement over time—from 3 to 6 months and from 3 to 12 months. These analyses provided additional insight into the magnitude of change within each group. Because these involved 10 comparisons (two time intervals across five outcomes), we applied a Bonferroni correction to control for multiple testing, setting the significance threshold at p < 0.005. All analyses were conducted using SPSS 27.

3. Results

Table 2 presents participants’ characteristics at baseline by group assignment. The average age of the women in the total sample (n = 1265) was 37 years (SD = 9.79). Nearly 46.0 % (n = 577) had been in the US for ten years or less, with 24.8 % (n = 314) being recent immigrants who had been in the US for four years or less. Over 40.0 % of participants were Hispanic (42.0 %, n = 524). The racial/ethnic distribution included approximately 30.5 % Asian (n = 386), 28.2 % White (n = 356), 13.5 % Black (n = 171), and 27.8 % from other racial categories (n = 351).

Regarding region of origin, 32.6 % (n = 413) of the immigrant women were from Latin America, 25.5 % (n = 322) were from Asia, and the rest were from other regions. More than half of the participants were in a relationship (68.1 %, n = 859), with the remainder in former relationships. The majority of the women's partners were not born in the US (61.2 %, n = 773), while 38.8 % (n = 490) had US-born partners. About 84.2 % (n = 1062) of the women identified as heterosexual, with the rest reporting other sexual orientations. Regarding education, most women (60.3 %, n = 752) had completed a bachelor's degree or higher, yet only 41.5 % (n = 507) reported full-time employment. Almost half of the women (49.1 %, n = 585) had experienced all three types of IPV: physical, sexual, and psychological abuse. A substantial 73.0 % (n = 923) reported physical abuse, 62.4 % (n = 789) reported sexual abuse, and 82.2 % (n = 1039) reported psychological abuse.

3.1. Baseline equivalence

In the first-stage randomization, 1265 women were randomized to the personalized online intervention or standard online safety information website (Fig. 1). At 3 months follow-up, 366 (30.8 %) participants were found to be low responders and, therefore, were re-randomized to receive either the text messaging intervention only (text-only) (n = 183) or the combination of text messaging and phone call intervention (text + phone) (n = 183). No differences in sociodemographic or baseline measures were evident among women randomized to receive the personalized or standard online safety information in the first-stage randomization. However, women who were not re-randomized (responders) versus those who were re-randomized (total low responders randomized to text-only or text + phone) at 3 months' follow-up differed in the frequency of financial stress and discrimination. A significantly higher proportion of total low responders reported experiencing frequent financial stress (59.5 %, n = 209) than responders (53.2 %, n = 420; p = 0.04). Additionally, low responders reported significantly higher levels of everyday discrimination (M = 15.67, SD = 9.24) than responders (M = 13.45, SD = 8.69; p < 0.01). Further, the severity and frequency of physical and sexual IPV were significantly greater among low responders (M = 21.23, SD = 23.80) at baseline than among responders (M = 16.03, SD = 25.51; p < 0.01). Low-responders also exhibited more severe symptoms of PTSD (M = 2.55, SD = 0.99) compared to responders (M = 2.41, SD = 0.95; p = 0.03). Additionally, they reported lower engagement in safety behaviors (M = 63.44, SD = 28.57) than responders (M = 67.31, SD = 26.7, p = 0.03) at baseline.

3.2. Attrition analysis

A total of 15,694 women were assessed for eligibility, with 1265 (8.0 %) meeting inclusion criteria and randomized to the personalized online intervention (52.1 %, n = 660) or standard safety information (47.9 %, n = 605). Most exclusions (90.7 %) were due to not meeting eligibility requirements. Retention was high across follow-ups, with 94.1 % completing 3 months, 90.1 % completing 6 months, and 88.4 % completing 12 months. At 3 months, participants were categorized as responders or low responders; low responders were re-randomized to either a text + phone or text-only intervention. Among low responders, 84.0 %–93.0 % completed subsequent follow-ups depending on the re-randomization group. Responders maintained high retention, with over 90 % completing follow-ups. A small proportion of women initially lost at 3 months returned for 6- and 12-month assessments.

Attrition analyses showed no significant baseline differences between completers and non-completers in age, duration of residence in the US, financial stress, relationship status, or primary and secondary outcomes. However, women lost to follow-up were more likely to be from Latin America (38.4 %, n = 76), have two or more children (51.0 %, n = 100), have relatively lower educational attainment (associate degree or lower, 50.3 %, n = 98), higher financial stress (65.4 %, n = 125), and experience multiple types of IPV exposure (56.1 %, n = 105). Details of participant flow are shown in Fig. 2 (CONSORT diagram).

Fig. 2.

The study CONSORT diagram.

3.3. First stage randomization

3.3.1. Comparative efficacy between groups

There were no statistically significant differences in the rate of change in outcomes between the personalized online and standard online safety information arms over time. That is, the primary outcome of physical and sexual IPV, as well as secondary outcomes—depression, PTSD, empowerment, and safety behaviors—did not significantly differ between the two first-stage intervention groups across the follow-up time points.

3.4. Second-stage randomization

3.4.1. Comparison of text-only and combination of text + phone intervention components on improvement in outcomes among the low-responders

3.4.1.1. Total low-responders

Overall, in the total sample, no significant differences were found between the text-only and text + phone groups among low responders across any time points—baseline to 12 months, 3–6 months, or 3 to 12 months—for the primary outcome of physical and sexual IPV. Similarly, no significant group differences were observed in trends for secondary outcomes, including depression., PTSD, empowerment, and safety behaviors (Table 3).

Table 3.

GEE Analysis of Outcome Variables from Baseline to Follow-up: Text + Phone Group vs Text-Only Group.

Outcome Variables	Total Low Responders			Personalized Online Intervention			Standard Safety Information
	Text Only (n = 183)	Text + Phone (n = 183)		Text Only (n = 86)	Text + Phone (n = 95)		Text Only (n = 97)	Text + Phone (n = 88)
	Mean (SD)	Mean (SD)	d1	Mean (SD)	Mean (SD)	d1	Mean (SD)	Mean (SD)	d1
Physical and Sexual IPV
Baseline	20.56 (23.72)	21.80 (24.13)		19.03 (35.92)	24.25 (35.37)		21.81 (30.30)	18.87 (32.22)
6 months	15.87 (23.64)	13.90 (21.94)	−0.13	13.78 (35.40)	13.88 (31.02)	−0.21	17.65 (31.27)	13.75 (29.89)	−0.14
12 months	16.48 (24.03)	14.88 (23.98)	−0.12	15.93 (34.18)	13.34 (29.60)	−.32∗	16.86 (33.67)	16.19 (36.97)	0.06
Depression
Baseline	10.06 (6.17)	10.29 (6.72)		9.53 (8.23)	10.60 (9.32)		10.51 (8.91)	9.87 (10.01)
6 months	8.81 (5.81)	8.80 (6.05)	−0.06	8.75 (8.14)	9.16 (9.21)	−0.15	8.87 (8.27)	8.37 (8.07)	−0.02
12 months	9.10 (6.30)	8.41 (6.37)	−0.17	8.81 (8.48)	7.97 (9.24)	−.33∗	9.37 (9.12)	8.85 (8.49)	−0.05
PTSD
Baseline	2.53 (0.96)	2.56 (1.10)		2.44 (1.37)	2.57 (1.37)		2.61 (1.37)	2.53 (1.65)
6 months	2.29 (0.90)	2.35 (0.88)	0.03	2.22 (1.29)	2.40 (1.39)	−0.11	2.36 (1.29)	2.28 (1.26)	−0.03
12 months	2.37 (1.03)	2.29 (1.00)	−0.10	2.30 (1.41)	2.20 (1.37)	−0.282	2.44 (1.29)	2.36 (1.25)	−0.06
Empowerment
Baseline	119.63 (19.61)	121.41 (23.17)		120.43 (28.24)	122.45 (34.00)		118.97 (27.15)	120.67 (31.26)
6 months	119.40 (19.38)	121.71 (21.94)	0.08	119.03 (26.62)	122.59 (32.91)	0.24	119.69 (27.65)	121.14 (28.88)	0.03
12 months	118.80 (20.56)	120.51 (23.23)	0.02	119.39 (28.40)	125.42 (34.60)	0.31	118.22 (29.30)	115.57 (28.10)	−0.16
Safety Behavior
Baseline	62.53 (27.42)	64.35 (30.44)		63.45 (40.44)	67.20 (41.82)		61.68 (37.43)	61.31 44.28)
6 months	94.73 (10.72)	94.97 (8.45)	−0.06	94.36 (16.80)	95.03 (11.22)	−0.10	95.04 (13.18)	94.97 (12.48)	0.03
12 months	72.33 (25.44)	74.52 (25.10)	0.01	73.23 (36.37)	76.04 (33.10)	−0.09	71.48 (35.47)	72.93 (37.72)	0.08

Notes: Each GEE model was controlled for the dose. The models were run using the text-only group as the reference group. The means presented were marginal means. Standard deviation was calculated using marginal mean and standard error. ∗p < 0.05. ¹Cohen's d. ²p = 0.06.

3.4.1.2. Comparison of text + phone and text-only participants in the personalized online intervention condition

A stratified GEE analysis comparing text + phone and text-only groups among low responders initially assigned to personalized online intervention revealed significant differences. Low responders re-randomized to the text + phone group experienced a significantly greater reduction in physical and sexual IPV from baseline to 12 months compared to the text-only group (p < 0.05; d = −0.32; Estimate=-7.80; 95 % CI = −14.83, 0.76). Similarly, the text + phone group showed a significantly greater reduction in depression symptoms over the same period (p < 0.05; Estimate=-1.91; d = −0.33; 95 % CI = −3.53, 0.28) (Table 3).

From 3 to 12 months, participants in the text + phone group also reported significantly higher empowerment scores than those in the text-only group (p < 0.05; Estimate=5.98; d = 0.27; 95 % CI = 0.83, 11.14). Within-group changes aligned with these findings: the text + phone group showed a significant mean increase in empowerment over the same period (+8.49; p < 0.005), while the text-only group showed a smaller, non-significant change (+2.51; p > 0.005; Table 5).

Table 5.

Within-group changes from 3 Months to follow-up in responders and low responders (text only and text + phone).

Outcome Variables	Total Sample			Personalized Online Intervention			Standard Safety Information
	Responders (n = 825)	Text Only (n = 183)	Text + Phone (n = 183)	Responders (n = 437)	Text Only (n = 86)	Text + Phone (n = 95)	Responders (n = 388)	Text Only (n = 97)	Text + Phone (n = 88)
	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)
Physical and Sexual IPV
3 months	8.91 (21.25)	18.25 (24.21)	20.19 (26.78)	8.44 (20.70)	17.47 (21.24)	19.30 (25.24)	8.37 (24.40)	21.43 (29.75)	23.20 (31.2)
6 months	9.60 (21.54)	16.05 (24.49)	14.26 (24.08)∗	8.68 (20.90)	13.83 (25.50)	14.18 (24.56)∗	9.52 (25.62)	20.51 (26.30)	16.44 (3.11)∗
12 months	9.91 (22.69)	16.93 (25.16)	15.50 (26.65)∗	9.32 (22.58)	15.92 (24.48)	13.90 (24.27)	9.43 (25.41)	20.39 (29.54)	19.18 (3.46)
Depression
3 months	9.39 (6.61)	10.91 (6.36)	10.34 (6.90)	9.20 (6.69)	10.27 (6.49)	10.49 (7.51)	9.59 (7.29)	11.49 (7.19)	10.19 (0.69)
6 months	8.91 (6.61)	8.84 (6.22)∗	8.78 (6.49)∗	8.87 (6.48)	8.57 (5.94)∗	9.04 (7.21)	8.95 (7.48)	9.12 (7.38)∗	8.49 (5.91)∗
12 months	8.91 (6.89)	9.12 (6.63)∗	8.42 (6.90)∗	8.81 (6.69)	8.62 (6.31)	7.92 (7.41)∗	9.01 (7.48)	9.61 (7.78)∗	8.95 (6.66)
PTSD
3 months	2.28 (0.86)	2.55 (0.95)	2.47 (0.95)	2.26 (0.84)	2.47 (1.02)	2.45 (1.07)	2.27 (0.98)	2.68 (1.08)	2.54 (1.03)
6 months	2.24 (0.86)	2.32 (.95)∗	2.36 (0.95)	2.20 (0.84)	2.21 (.93)∗	2.39 (1.07)	2.26 (1.18)	2.49 (1.18)	2.37 (1.03)
12 months	2.25 (0.86)	2.39 (1.08)∗	2.31 (1.08)	2.22 (1.05)	2.28 (1.02)	2.23 (1.17)	2.26 (1.18)	2.54 (1.08)	2.45 (1.03)
Empowerment
3 months	121.30 (21.5)	116.43 (19.4)	116.24 (22.4)	121.07 (21.9)	117.52 (19.0)	117.26 (25.93)	121.84 (24.0)	114.71 (23.0)	114.6 (21.6)
6 months	121.20 (21.8)	119.39 (20.4)	121.75 (23.4)∗	121.23 (22.3)	119.28 (19.9)	122.96 (26.51)∗	121.45 (23.4)	118.68 (24.3)	119.0 (21.8)∗
12 months	120.71 (22.4)	118.92 (16.3)	120.42 (24.9)∗	121.02 (23.4)	120.03 (21.0)	125.75 (27.88)∗	120.69 (24.4)	117.09 (25.5)	114.2 (21.9)
aSafety Behavior
Baseline	67.29 (26.7)	62.33 (27.2)	64.62 (30.0)	68.15 (37.6)	63.25 (40.8)	67.06 (41.1)	65.62 (39.0)	63.12 (38.7)	63.1 (43.6)
6 months	95.10 (9.8)∗	94.55 (10.3)∗	95.24 (8.2)∗	96.15 (12.1)∗	94.15 (16.9)∗	94.92 (12.3)∗	93.26 (19.7)∗	96.50 (17.1)∗	96.8 (14.9)∗
12 months	70.57 (28.0)∗	72.15 (25.8)∗	74.79 (25.2)∗	72.87 (37.5)∗	73.02 (37.2)∗	75.93 (33.1)∗	67.32 (43.6)	72.96 (37.5)∗	74.8 (37.7)∗

Notes. Bold means and SD indicate a significant within-group change from 3 months to the respective follow-up timepoint (p < 0.005), after Bonferroni correction for multiple comparisons.

^aSafety behaviors were not assessed at the 3-month follow-up.

3.4.1.3. Comparison of text + phone and text-only participants in the standard online safety information condition

There were no differences between the text + phone and text-only groups in the standard online safety information condition for the primary outcome of physical and sexual IPV from baseline to 12 months. Similarly, no significant group differences were observed for PTSD, depression, empowerment, or safety behaviors (Table 3).

3.4.2. Comparison of responders and low responders to assess whether text and Text + phone support improved outcomes for low responders

3.4.2.1. Total responders and low responders

From baseline to 12 months, low responders showed significantly greater improvements in safety behaviors than responders in both the text-only (p < 0.05; d = 0.24; Estimate =6.41; 95 % CI = 1.81, 11.00) and text + phone groups (p < 0.01; d = 0.25; Estimate=7.03; 95 % CI = 2.04, 12.03). However, no significant differences were found between low responders and responders for physical and sexual IPV, depression, PTSD, or empowerment over the same period (Table 4).

Table 4.

GEE analysis of outcome variables from baseline to Follow-up: Low responders (text only and text + phone) versus responders.

Outcome Variables	Total Sample					Personalized Online Intervention					Standard Safety Information
	Responders (n = 825)	Text Only (n = 183)		Text + Phone (n = 183)		Responders (n = 437)	Text Only (n = 86)		Text + Phone (n = 95)		Responders (n = 388)	Text Only (n = 97)		Text + Phone (n = 88)
	Mean (SD)	Mean (SD)	d1	Mean (SD)	d1	Mean (SD)	Mean (SD)	d1	Mean (SD)	d1	Mean (SD)	Mean (SD)	d1	Mean (SD)	d1
Physical and Sexual IPV
Baseline	16.11 (25.9)	20.67 (23.4)		22.12 (24.1)		15.94 (34.8)	18.75 (35.9)		24.07 (35.5)		15.28 (41.1)	24.8 (35.9)		22.03 (38.0)
6 months	9.49 (21.4)	15.96 (23.7)	0.07	14.18 (22.4)	−0.05	8.54 (28.1)	13.50 (35.7)	0.06	13.71 (31.4)	−0.12	9.49 (36.9)	20.62 (35.3)	0.07	16.86 (39.2)	0.02
12 months	9.84 (21.7)	16.58 (24.0)	0.07	15.16 (24.1)	−0.04	9.34 (29.1)	15.65 (34.4)	0.12	13.17 (30.1)	−0.21	9.31 (35.8)	19.83 (38.3)	0.01	19.29 (42.2)	0.13
Depression
Baseline	10.13 (6.6)	10.09 (6.1)		10.31 (6.6)		9.95 (8.9)	9.42 (8.7)		10.47 (9.5)		10.26 (10.3)	10.85 (9.6)		10.26 (9.6)
6 months	8.90 (6.5)	8.84 (5.9)	0.02	8.82 (6.0)	−0.08	8.86 (8.7)	8.70 (8.6)	0.04	9.00 (9.3)	−0.15	8.88 (10.7)	9.20 (9.4)	−0.07	8.74 (8.1)	−0.01
12 months	8.89 (6.7)	9.13 (6.3)	0.04	8.43 (6.4)	−0.13	8.81 (8.9)	8.70 (8.7)	0.06	7.81 (9.4)	−.28∗	8.91 (10.6)	9.70 (10.0)	0.02	9.23 (8.7)	0.04
PTSD
Baseline	2.41 (0.9)	2.54 (0.9)		2.56 (1.1)		2.39 (1.4)	2.43 (1.5)		2.56 (1.4)		2.41 (1.4)	2.70 (1.5)		2.62 (1.6)
6 months	2.24 (0.8)	2.30 (0.9)	−0.07	2.35 (0.9)	−0.07	2.20 (1.1)	2.21 (1.3)	−0.02	2.38 (1.4)	−0.06	2.26 (1.7)	2.45 (1.6)	−0.11	2.37 (1.4)	−0.09
12 months	2.25 (0.8)	2.38 (1.0)	−0.01	2.29 (1.0)	−0.14	2.22 (1.4)	2.29 (1.4)	0.03	2.19 (1.4)	−.26∗	2.26 (1.7)	2.52 (1.4)	−0.03	2.45 (1.4)	−0.04
Empowerment
Baseline	118.70 (21.5)	119.56 (19.3)		121.35 (22.7)		118.52 (29.5)	121.00 (29.3)		123.07 (35.5)		119.07 (35.3)	117.77 (30.4)		119.3 (32.1)
6 months	121.10 (21.4)	119.34 (19.3)	−0.10	121.72 (21.8)	−0.02	121.00 (30.1)	119.59 (27.6)	−.18∗	123.31 (34.2)	0.02	121.42 (34.9)	118.54 (31.5)	−0.04	119.8 (29.9)	−0.07
12 months	120.60 (21.9)	118.75 (20.5)	−0.11	120.49 (23.1)	−0.09	120.71 (31.4)	119.96 (29.4)	−0.14	126.11 (35.9)	0.11	120.68 (34.7)	117.08 (33.0)	−0.08	114.2 (29.7)	−.31∗
Safety Behavior
Baseline	67.29 (26.7)	62.33 (27.2)		64.62 (30.0)		68.15 (37.6)	63.25 (40.8)		67.06 (41.1)		65.62 (39.0)	63.12 (38.7)		63.17 (43.6)
6 months	95.10 (9.8)	94.55 (10.3)	0.16	95.24 (8.2)	0.10	96.15 (12.1)	94.15 (16.9)	0.11	94.92 (12.3)	−0.01	93.26 (19.7)	96.50 (17.1)	0.21	96.82 (14.9)	0.21
12 months	70.57 (28.0)	72.15 (25.8)	.24∗	74.79 (25.2)	.25∗	72.87 (37.5)	73.02 (37.2)	0.18	75.93 (33.1)	0.14	67.32 (43.6)	72.96 (37.5)	.30∗	74.80 (37.7)	.36∗

Notes: Each GEE model was controlled for the dose. The means presented were marginal means after controlling for dose. Standard deviations were calculated using marginal means and standard errors. Responders were the reference group in all GEE models. ∗p < 0.05.¹Cohen's d.

Following re-randomization at 3 months, low responders in the text + phone group exhibited significantly greater reductions in physical and sexual IPV compared to responders from both 3–6 months (p < 0.01; d = −0.30; Estimate=-6.62; 95 % CI = −9.84, −3.40) and from 3 to 12 months (p < 0.01; d = −0.25; Estimate=-5.69; 95 % CI = −9.73, −1.66). Within-group analyses further supported these findings: the text + phone group showed significant mean reductions in physical and sexual IPV from 3 to 6 months (−5.93, p < 0.005) and from 3 to 12 months (−4.69, p < 0.005), while responders exhibited non-significant increases (+0.69 at 6 months, +1.00 at 12 months, p > 0.005). Despite improvements, IPV levels in the text + phone group remained higher than those in responders at both 6-month (Mean (M) = 14.26 vs. M = 9.60 in responders) and 12-month (M = 15.50 vs. M = 9.91 in responders) follow-ups, though the gap was substantially reduced.

Depression outcomes showed a similar pattern. The text + phone group demonstrated significantly greater reductions in depression symptoms than responders from 3 to 6 months (p < 0.01; d = −0.16; Estimate=-1.09; 95 % CI = −1.99, −0.18) and from 3 to 12 months (p < 0.01; d = −0.22; Estimate=-1.44; 95 % CI = −2.41, −0.48). Within-group reductions in depression were also significant for the text + phone group from 3 to 6 months (−1.56, p < 0.005) and from 3 to 12 months (−1.92, p < 0.005), while responders showed modest and non-significant reductions (−0.48 at both timepoints, p > 0.005). These improvements resulted in the text + phone group achieving comparable depression levels to responders at both 6-month (M = 8.78 vs. M = 8.91 in responders) and 12-month (M = 8.42 vs. M = 8.91 in responders) follow-ups.

Empowerment scores also improved significantly among low responders in the text + phone group relative to responders between 3 and 6 months (p < 0.01; d = 0.26; Estimate=5.61; 95 % CI = 2.83, 8.39) and from 3 to 12 months (p < 0.01; d = 0.22; Estimate=4.76; 95 % CI = 1.87, 7.65). Within-group analysis showed significant mean empowerment gains in the text + phone group from 3 to 6 months (+5.51, p < 0.005) and from 3 to 12 months (+4.18, p < 0.005), while responders showed non-significant decreases (−0.10 at 6 months, −0.59 at 12 months, p > 0.005). Notably, these improvements brought mean empowerment scores in the text + phone group to levels comparable to or slightly exceeding those of responders at both 6-month (M = 121.75 vs. M = 121.20 in responders) and 12-month (M = 120.42 vs. M = 120.71 in responders) follow-up timepoints.

Similar but somewhat attenuated effects were observed in the text-only group. Here, low responders experienced significantly greater reductions in depression symptoms from 3 to 6 months (p < 0.01; d = −0.24; Estimate=-1.59; 95 % CI = −2.49, −0.69) and 3–12 months (p < 0.01; d = −0.20; Estimate=-1.30; 95 % CI = −2.27, −0.33). Within-group analyses demonstrated significant improvements in depression for the text-only group from 3 to 6 months (−2.07, p < 0.005) and from 3 to 12 months (−1.79, p < 0.005), substantially greater than the modest and non-significant reductions observed in responders (−0.48 at both timepoints, p > 0.005). These improvements resulted in the mean depression scores in the text-only group close to those of responders at both 6-month (M = 8.84 vs. M = 8.91 in responders) and 12-month (M = 9.12 vs. M = 8.91 in responders) timepoints.

The text-only group also showed significant reductions in PTSD symptoms relative to responders from 3 to 6 months (p < 0.05; d = −0.22; Estimate=-0.22; 95 % CI = −0.33, −0.07) and 3–12 months (p < 0.05; d = −0.15; Estimate=-0.14; 95 % CI = −0.28, −0.01). Within-group analysis confirmed these changes, with significant reduction in PTSD symptoms in the text-only group from 3 to 6 months (−0.23, p < 0.005) and from 3 to 12 months (−0.16, p < 0.005), while responders exhibited minimal, non-significant decreases (−0.04 at 6 months, −0.03 at 12 months, p > 0.005). These improvements resulted in comparable mean PTSD scores between the text-only group and responders at 6-month (M = 2.32 vs. M = 2.24 in responders) and 12-month (M = 2.39 vs. M = 2.25 in responders) timepoints.

Empowerment gains were evident in the text-only group, with significant between-group differences from 3 to 6 months (p < 0.05; d = 0.14; Estimate=3.06; 95 % CI = 0.39, 5.73) and maintained through 12 months (p < 0.05; d = 0.15; Estimate=3.07; 95 % CI = 0.43, 5.71). Within-group improvements were non-significant but positive (+2.96 at 6 months, +2.49 at 12 months, p < 0.005) and from 3 to 12 months (+2.49, p < 0.005), contrasting with non-significant decreases in empowerment scores observed in responders (−0.10 to −0.59, p > 0.005). Ultimately, these changes led to mean empowerment scores in the text-only group that closely matched those of responders at both 6-month (M = 119.39 vs. M = 121.20 in responders) and 12-month (M = 118.92 vs. M = 120.71 in responders) follow-up timepoints (Table 5).

3.4.2.2. Comparison of responders and low-responders in the personalized online intervention condition

From baseline to 12 months, among participants in the personalized online intervention, low responders assigned to the text + phone support group reported significantly greater reductions in depression (p < 0.05; d = −0.28; Estimate = −1.52; 95 % CI = −2.78, 0.25) and PTSD symptoms (p < 0.05; d = −0.26; Estimate = -0.20; 95 % CI = −0.39, 0.00) from baseline to 12 months than responders. While empowerment did not significantly differ between groups from baseline to 6 months, by 12 months, low responders reached comparable levels of empowerment, with no significant differences observed between responders and low responders (p=0.26–0.71), as shown in Table 4.

From 3 to 12 months post-re-randomization, low responders in the text + phone group showed significantly greater reductions in physical and sexual IPV compared to responders, both from 3 to 6 months (p < 0.01; d = −0.25; Estimate=-5.36; 95 % CI = −9.32, −1.40) and from 3 to 12 months (p < 0.01; d = −0.29; Estimate=-6.29; 95 % CI = −11.63, −0.96). Within-group analyses supported these findings, showing significant IPV reductions in the text + phone group (−5.12 at 6 months; p < 0.005), while responders exhibited a non-significant increase (+0.24; p > 0.005). Despite these improvements, IPV levels remained somewhat elevated in the text + phone group compared to responders at 6 months (M = 14.18 vs. M = 8.68 in responders) and 12 months (M = 13.90 vs. M = 9.32 in responders), though the initial disparity was substantially reduced.

Depression outcomes showed a similar pattern. The text + phone group reported significantly greater reductions in depressive symptoms from 3 to 12 months (p < 0.01; d = −0.32; Estimate=-2.18; 95 % CI = −3.53, −0.84). Within-group analysis showed a significant reduction in depression in the text + phone group (−2.57; p < 0.005), compared to a smaller but still significant reduction among responders (−0.39; p < 0.005). These changes brought the text + phone group's mean depression scores in line with responders at 6 months (M = 9.04 vs. M = 8.87 in responders) and lower at 12 months (M = 7.92 vs. M = 8.81).

Improvements in empowerment were also significantly greater in the text + phone group across both time intervals: 3–6 months (p < 0.01; d = 0.24; Estimate=5.53; 95 % CI = 1.39, 9.68) and 3–12 months (p < 0.01; d = 0.38; Estimate=8.53; 95 % CI = 4.31,12.76).Within-group results confirmed substantial increases in empowerment scores from 3 to 6 months (+5.70; p < 0.005) and from 3 to 12 months (+8.49; p < 0.005), while responders exhibited minimal and non-significant changes (+0.16 at 6 months, −0.05 at 12 months; p > 0.005). Notably, by 6 and 12 months, mean empowerment scores were higher in the text + phone group than among responders (6-month: M = 122.96 vs. M = 121.23 in responders; 12-month: M = 125.75 vs. M = 121.02 in responders) (Table 5).

3.4.2.3. Comparison of responders and low responders in the standard safety information condition

From baseline to 12 months, in the standard online safety information condition, both responders and low responders showed improvement in IPV, depression, PTSD, and empowerment at 6 and 12 months, with responders typically showing greater overall gains. However, low responders assigned to text-only or text + phone support reported higher rates of attempted helpful safety behaviors at 12 months than responders (p < 0.05; d = 0.30 to 0.36; Table 4).

Following re-randomization, low responders in the text + phone group showed significantly greater reductions in physical and sexual IPV from 3 to 6 months compared to responders (p < 0.01; d = −0.31; Estimate=-7.91; 95 % CI = −13.15, −2.67). Within-group analysis supported these findings, revealing a significant reduction in IPV for the text + phone group (−6.76; p < 0.005), whereas responders showed a non-significant increase (+1.15; p > 0.005). These improvements narrowed the initial disparity, though mean IPV levels remained higher for the text + phone group at 6 months (M = 16.44 vs. M = 9.52 in responders).

The text + phone group also showed significantly greater gains in empowerment from 3 to 6 months (p < 0.05; d = 0.24; Estimate=5.69; 95 % CI = 2.04, 9.33). Within-group changes indicated a significant increase in empowerment (+4.40; p < 0.005), while responders experienced a non-significant decline (−0.39; p > 0.005). As a result, mean empowerment scores in the text + phone group approached those of responders at 6 months (M = 119.0 vs. M = 121.45 in responders).

Similarly, low responders in the text-only group experienced significantly greater reductions in depression symptoms compared to responders from 3 to 6 months (p < 0.05; d = −0.24; Estimate=-1.74; 95 % CI = −3.06, −0.42) and from 3 to 12 months (p < 0.05; d = −0.18; Estimate=-1.30; 95 % CI = −2.56, −0.05). Within-group analyses showed significant reductions in depression for the text-only group (−2.37 at 6 months, −1.88 at 12 months; p < 0.005), while responders showed smaller improvements (−0.64 at 6 months, −0.58 at 12 months; p < 0.005). These changes brought the text-only group's mean depression scores in line with responders at 6 months (M = 9.12 vs. M = 8.95 in responders) and 12 months (M = 9.61 vs. M = 9.01 in responders).

In addition, the text-only group showed a significant reduction in PTSD symptoms from 3 to 6 months (p < 0.05; d = −0.18; Estimate=-0.19; 95 % CI = −0.37, −0.01). However, within-group analysis showed this reduction was not statistically significant (−0.19; p > 0.005). In contrast, responders exhibited minimal change (−0.01; p > 0.005).

The text-only group also showed significantly greater improvements in empowerment from 3 to 6 months (p < 0.05; d = 0.18; Estimate=4.37; 95 % CI = 0.59, 8.14), supported by within-group gains (+3.97; p < 0.005), while responders showed no significant change (−0.39; p > 0.005). These improvements resulted in the text-only group achieving mean levels of PTSD (M = 2.49 vs. M = 2.26 in responders) and empowerment (M = 118.68 vs. M = 121.45 in responders) that were comparable to those of responders at the 6-month follow-up (Table 5).

4. Discussion

Our study was innovative in its comparative evaluation of four types of support for immigrant survivors of IPV: (1) personalized online safety planning, (2) standard online safety information, (3) text-only support, and (4) a combination of text and phone support, provided specifically to low responders to the online interventions. Consistent with prior research demonstrating the benefits of safety planning interventions in reducing IPV [9,36,37], women across all groups reported reductions in the severity and frequency of physical and sexual IPV over time, regardless of the intervention received. Additionally, women in all groups experienced improvements in mental health and increased engagement in safety behaviors from baseline to the 12-month follow-up.

Among participants identified as low responders at three months, those in the personalized online safety planning condition who received both text and phone support demonstrated significantly greater improvements from baseline to 12 months. These improvements included reductions in depression and PTSD symptoms, as well as increased empowerment and greater utilization of safety strategies. Notably, this group showed greater mental health improvement than both the text-only group and the initial responders, highlighting the potential of enhanced support to not only close but surpass early response gaps.

Between 3 and 12-months, the broader sample of low responders who received combined text and phone support also showed notable declines in physical and sexual IPV, along with continued improvements in psychological well-being and empowerment—bringing their outcomes closer to, and in some cases comparable with, those of responders. This pattern suggests that additional support may facilitate later-stage gains among survivors of IPV who do not respond to early intervention, aligning with the objectives of adaptive intervention frameworks.

Combined text and phone support was more effective than text messaging alone, particularly among those initially assigned to the personalized online intervention. This subgroup demonstrated the largest reductions in physical and sexual IPV across follow-up timepoints. While some of these gains may reflect higher baseline symptom levels, the consistent within-group improvements and convergence with responder outcomes highlight the effectiveness of layering digital tools with human support.

At baseline, low responders reported significantly higher levels of physical and sexual IPV and PTSD symptoms than the responders. They also reported more everyday experiences of discrimination and financial stress. These contextual factors likely contributed to their limited response to the initial online interventions, underscoring the need for more tailored approaches. When provided with supplemental support, low responders demonstrated significant improvements in multiple outcomes. These findings reinforce the need for flexible, adaptive approaches that respond to survivors’ individual needs and evolving circumstances.

Interestingly, low responders also engaged in more helpful safety behaviors than responders. This suggests that even under conditions of elevated stress, some survivors may take protective action when appropriately supported. This underscores the importance of offering a range of intervention strategies that recognize varied pathways to safety, healing, and empowerment for survivors of IPV. Finally, even participants from the standard online information group benefited significantly when given additional support, suggesting that the human interaction itself—rather than the format of the initial intervention—was a key factor in driving positive change.

The phone intervention in our study included psychotherapeutic elements such as empathetic communication, emotional validation, and cognitive restructuring (reframing their challenges into accomplishment) with supportive or motivational strategies. Our findings align with prior research recommending that effective interventions for survivors of IPV with mental health concerns should incorporate psychotherapeutic approaches alongside safety planning [9,39]. On average, low responders who received enhanced support achieved outcomes comparable to or even better than initial responders. This demonstrates the potential of flexible, remote, multi-modal support systems, particularly for immigrant women who often face barriers to accessing traditional services, including social isolation, immigration-related stress, and mistrust of providers [38].

A systematic review identified that effective interventions for reducing IPV included an empowerment-focused approach. Key strategies involved educating women about IPV, helping them identify threats to safety, tailoring safety plans to their priorities, facilitating access to resources, and conducting periodic safety check-ins. These strategies have proven effective in improving outcomes in studies focusing on non-immigrant survivors of IPV [9]. Our digital interventions, which integrated these same strategies, extended their benefits to immigrant survivors as well. Overall, these findings underscore the importance of interventions that can adapt to survivors’ evolving needs. Factors such as trauma history, discrimination, and financial or immigration-related stress can shape how women engage with and benefit from support. Rather than relying on a one-size-fits-all model, interventions should be personalized and flexible—combining technology with human connection—to effectively address both safety and mental health, especially for those facing complex, intersecting challenges.

4.1. Strengths and limitations

A key strength of our research is that it is the first SMART trial to evaluate digital interventions specifically designed for immigrant survivors of IPV. It provides compelling evidence for using digital intervention approaches to support immigrant IPV survivors who face barriers to accessing in-person services. Our high retention rate reflects the positive experiences women had with the interventions. Additionally, the trial was conducted with a diverse sample of immigrant IPV survivors from multiple regions across the U.S., enhancing the generalizability of our findings to immigrant women facing common challenges related to IPV and associated stressors. In light of these strengths, comparing our findings with the existing literature is essential. In a systematic review and meta-analysis [6] of studies conducted between 2007 and 2021, digital interventions for IPV were found to significantly reduce physical and psychological IPV (baseline to 6 months) and depression in the short term (baseline to 3 months). However, these studies did not demonstrate a significant reduction in sexual violence at any time point [6]. In contrast, our findings are notable because our research showed that the digital intervention approach effectively reduced both physical and sexual IPV as well as depression and PTSD symptoms among IPV survivors over the long term at the 12-month follow-up.

Our trial has limitations, including reliance on self-reported data from women, which may introduce reporting bias. However, we anticipated this bias to be minimal due to the anonymity and privacy afforded by the web-based data collection format. Another limitation is the potential for underreporting of IPV in follow-up surveys, as repeated questioning may lead some participants to reduce disclosure due to anxiety or concerns about the consequences, even with assurances of confidentiality. However, we anticipated this bias to be minimal, given the anonymity and privacy afforded by the web-based data collection format. Additionally, some women may have still felt ambivalent or unready to take action against IPV, which could have impacted their engagement with the interventions or prevented them from fully utilizing them.

The surveys and interventions were translated into Spanish and Arabic, which reflected the primary non-English languages spoken within the target population, as identified by our community partners. Although we recognize that this choice may have limited accessibility for speakers of other languages, practical constraints made broader translation efforts infeasible. Notably, non-English speakers comprised approximately 20 % of the enrolled sample, suggesting that the translations addressed the needs of a substantial portion of our participants. Nevertheless, it remains possible that language barriers influenced participants’ decisions to enroll, potentially introducing selection bias.

The study was fully powered with 80 % power to detect small to moderate effect sizes. However, despite careful planning, the distribution of participants for low-responder comparisons deviated slightly from our original plan, resulting in a smaller sample size for stratified analyses. While our observed effect sizes were generally consistent with, and in some cases, higher than our initial assumptions, this was not true for all outcomes. This limitation suggests that we may not have reliably detected significant effects for certain outcomes. Additionally, although our goal was to assess whether the outcomes of low responders approached those of initial responders, our study was not formally designed as a non-inferiority or equivalence trial. This limits the extent to which we can make formal claims about similarity between groups. Given the reduced sample size in these groups and the focused set of pre-specified hypotheses aligned with our primary aims, we chose not to adjust the Type I error rate for multiple comparisons. This decision was made to preserve statistical power and minimize the risk of Type II errors, ensuring we could still detect meaningful changes in our outcomes that were both statistically and practically significant.

Other limitations include the finding that, while reductions in physical and sexual IPV were observed among low responders, they reported relatively higher levels than responders at follow-up, suggesting ongoing elevated risk and a need for continued support. Additionally, baseline differences between groups may have influenced outcomes—low responders in the text + phone group reported higher levels of physical and sexual IPV at baseline compared to those in the text-only group. The greater improvements observed in this group may also have been affected by the longer intervention duration (six weeks vs. four weeks), making it difficult to isolate the effect of the intervention content from the impact of increased time and attention. To address these concerns, we plan to conduct a larger trial comparing the text-only and text + phone interventions to more rigorously evaluate their effects and better isolate the impact of these interventions from differences in baseline risk and contact time.

4.2. Implications

This study underscores the importance of tailoring interventions to meet the diverse and evolving needs of IPV survivors, particularly those who do not respond to initial support strategies. The effectiveness of both personalized online and standard online safety information interventions suggests that a range of digital support strategies could be viable as first-line support options in real-world settings. Given the positive outcomes observed, these online interventions could be broadly disseminated for use with survivors through existing IPV networks, philanthropic initiatives, or, where possible, government-supported platforms.

Our findings on the benefits of the second-stage text + phone intervention for low responders indicate that multimodal, personalized strategies—combining text messaging with direct phone support—can significantly improve engagement and outcomes for survivors with complex needs who require more intensive support. Organizations serving survivors of IPV should consider integrating structured phone-based support with complementary digital support methods. Providing practitioners with training and guidance on text and phone intervention protocols could further strengthen service delivery across IPV networks.

To maximize the impact of these interventions, it is also crucial to implement effective strategies for identifying low responders through regular assessment and to adjust support mechanisms based on individual survivor needs. In addition to fostering empowerment, interventions should emphasize actionable safety behaviors, as these may offer critical pathways to improving survivors' immediate safety and long-term outcomes. While further research is needed to refine digital intervention strategies and understand mechanisms of response among diverse populations, the strength of the current evidence supports pilot implementation to begin translating findings into practice. Delaying this process risks missing critical opportunities to provide evidence-based support to vulnerable IPV survivors with complex needs.

5. Conclusions

This study demonstrates the effectiveness of both personalized online and standard online safety information interventions as viable first-line digital support strategies for immigrant IPV survivors. Among those who did not respond initially, the addition of personalized text and phone support resulted in the greatest reductions in IPV, depression, and PTSD symptoms and improvement in empowerment and safety behaviors. These findings highlight the need for adaptive, individualized approaches that respond to survivors’ varying levels of need. Tailoring both the intensity and mode of support is essential to ensure that survivors requiring more comprehensive assistance receive appropriate, personalized care.

CRediT authorship contribution statement

Bushra Sabri: Writing – original draft, Supervision, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Jian Li: Formal analysis. Subhash Aryal: Formal analysis. Theresa Mata: Project administration, Data curation. Sarah M. Murray: Writing – review & editing, Methodology. Nancy Glass: Methodology. Jacquelyn C. Campbell: Methodology.

Funding

This research was supported by the National Institute on Minority Health and Health Disparities grant numbers (R01MD013863 and R01MD018503). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Declaration of competing interest

No potential conflict of interest was reported by the authors.

Appendix A. Supplementary data

The following is the Supplementary data to this article:

Data availability

De-identified data from this study are not available in a public archive. The authors confirm that the data supporting the findings of this study are available within the paper.