This study was nested within a 2 × 2 × 2 cluster-randomized full factorial experiment implemented across 16 low-income urban and periurban communities in Mwanza, Tanzania [26]. Each community (cluster) was randomly assigned to one of 8 experimental conditions, comprising all possible combinations of 3 components: guidance (guided vs self-guided), app design (structured vs unstructured), and digital support (enhanced vs basic). The trial aimed to optimize user engagement with ParentApp, which was developed in partnership with the United Nations Children’s Fund, the World Health Organization, and the Tanzanian national government. The trial was preregistered on the Pan-African Clinical Trial Registry (PACTR202210657553944) on October 11, 2022. Further information about the trial design and findings can be found in the study protocol and related trial publications [26-28].

Eligible participants were required to be at least 18 years old, be caring for an adolescent aged 10 to 17 years, reside in the same household as the adolescent for at least 4 nights per week during the previous month, and have regular access to an Android smartphone (OS 5.1.1 or higher). To address the underrepresentation of fathers in parenting interventions, the study aimed to enroll at least 20% men. Participants were recruited on a rolling basis between October 24, 2022 and December 1, 2022, through local community leaders, who were predominantly men. Community leaders introduced and endorsed the study during community meetings, helping to build trust and acceptability among families.

Cluster-level onboarding sessions were conducted in local schools and community centers by staff from Tanzania’s National Institute for Medical Research (NIMR) and trained facilitators from Investing in Children and Strengthening Their Societies, a local nongovernmental organization. Sessions included study information, eligibility screening, and informed consent, followed by app onboarding and baseline data collection. After obtaining informed consent, participants were divided into small groups (5‐10 caregivers) to facilitate hands-on support with downloading and installing ParentApp from the Google Play Store. Following installation, NIMR research assistants moved between participants to configure each app according to the cluster’s allocated intervention condition before guiding caregivers through the app’s core features, including navigation, activity completion, and available resources. Participants then completed the introductory module, which included the embedded baseline assessment. Clusters allocated to the enhanced digital support condition also participated in a brief group-based smartphone literacy session, while those allocated to the guided condition were added to facilitator-moderated WhatsApp groups.

Analyses were conducted in R (version 4.5.0; R Foundation for Statistical Computing). Sample characteristics were summarized overall and by caregiver gender. Gender differences in age, financial stress, food insecurity, positive parenting, harsh parenting, and depressive symptoms were assessed using 2-tailed Welch independent-samples t tests, with Cohen d calculated to estimate effect sizes. Engagement across the 12 modules was visualized using Kaplan-Meier survival curves stratified by caregiver gender.

Predictors of engagement were estimated using Poisson generalized linear mixed models with cluster-level random intercepts. Models were fit to multiply imputed datasets to address missing baseline data, and estimates were pooled across imputations [36]. The primary outcome (number of modules completed) was modeled using an offset for exposure, defined as the number of modules available to each participant. Secondary outcomes (first nonstart and first noncompletion) were modeled as event indicators (1=event, 0=none), with an offset for the number of modules at risk. For participants who experienced an event, the number of modules at risk was defined by the module at which the first disengagement event occurred; for those who did not experience an event, it was defined as the total number of modules available.

Each predictor was tested in a separate model, as the aim was to identify caregiver characteristics associated with engagement that could inform future intervention design and optimization. Predictors were group-mean centered within clusters to isolate within-cluster effects. All models adjusted for the experimental components (dummy-coded for guidance, app design, and digital support). Caregiver age and gender (grand-mean centered) were also included as covariates, except when either variable was the predictor of interest. The potential moderating effect of caregiver gender was examined by including an interaction term between the predictor of interest and gender. Gender was represented both as a group-mean centered variable and as a cluster-level mean to disaggregate individual- and cluster-level effects. Model estimates are reported as incidence rate ratios with 95% CIs and corresponding P values. The Benjamini-Hochberg procedure was applied to control the false discovery rate across models, with significance determined using the adjusted critical P value thresholds [37].

Ethics approval was granted by the University of Oxford’s Departmental Research Ethics Committee (R69744/RE001) and the Tanzanian NIMR (NIMR/HQ/R.8s/Vol.IX/3856). Written informed consent was obtained from all study participants at the onboarding session. To support WhatsApp participation and app-server data synchronization, all participants received 1 GB of mobile data per month for the study duration. Participants received US $2 (US $1=TZS 2319.79 as of October 24, 2022) after completing the onboarding session.

Baseline characteristics for the full sample and by caregiver gender are presented in Table 1. The trial included 614 caregivers (409 women and 205 men), with ages ranging from 18 to 75 years (mean age 35.94, SD 11.84). Independent-samples 2-tailed t tests indicated no significant gender differences in age or positive parenting practices. However, women reported significantly higher levels of several socioeconomic and psychosocial stressors compared to men, including financial stress (mean 4.55, SD 2.73 vs mean 3.72, SD 2.73; P<.001; Cohen d=0.30), food insecurity (mean 7.88, SD 8.09 vs mean 6.17, SD 7.01; P=.009; Cohen d=0.22), harsh parenting (mean 7.40, SD 6.42 vs mean 5.07, SD 6.14; P<.001; Cohen d=0.37), and depressive symptoms (mean 8.53, SD 4.12 vs mean 7.80, SD 3.56; P=.02; Cohen d=0.19). Although effect sizes were small, the findings indicate consistent gender differences in exposure to adversity in the sample.

This study provides the first known analysis of caregiver-level predictors of engagement with a digital parenting program in an LMIC, with the notable strength of including a large proportion of men and fathers (205/614, 33.4%). While nearly all participants initiated the program, engagement dropped substantially after onboarding, with the steepest decline occurring after the first module. This mirrors broader trends in digital health interventions, where attrition is typically concentrated in the early stages of program use [38]. These findings highlight the importance of prioritizing retention strategies early in the intervention.

Of the baseline characteristics examined, caregiver gender was the only statistically significant predictor of engagement. Women completed approximately 15% more modules than men, a finding that contrasts with several digital parenting studies in HICs, where engagement is often similar across genders [17]. One possible explanation is that women’s higher engagement may reflect greater exposure to socioeconomic and psychosocial adversity, including financial stress, food insecurity, depressive symptoms, and harsh disciplinary practices. Women reported significantly higher levels of each of these stressors, which may have increased the perceived relevance of the intervention and, in turn, their motivation to engage.

This interpretation aligns with the Health Belief Model [39], which posits that individuals are more likely to adopt health behaviors when they perceive the risks of inaction as high and the benefits of action as tangible. From this perspective, socioeconomic and psychosocial adversity may have served as a cue to action, outweighing emotional or cognitive barriers that might otherwise prevent engagement [40]. That women sustained engagement despite multiple, overlapping stressors is a notable finding. It suggests that digital parenting interventions may offer a feasible and acceptable modality for supporting high-risk caregivers in LMICs.

In contrast, the lower engagement observed among men may reflect a complex interplay of identity, norms, and structural constraints. In Tanzania and many other LMICs, caregiving is typically viewed as a maternal responsibility, while fatherhood is closely tied to provision, protection, and authority [41-43]. When program content is perceived as incongruent with these gendered expectations or fails to affirm men’s aspirations as fathers, it may appear irrelevant or incompatible with their roles. This is echoed in global fatherhood research showing that interventions which acknowledge and support men’s caregiving identities and align with their aspirations tend to enhance perceived relevance and achieve higher engagement [13,44].

Lower engagement among men may also reflect deeper identity tensions under conditions of economic precarity. Qualitative research from Mwanza, Tanzania describes how many men aspire to be nurturing and involved fathers but face conflicting practical and social pressures from work demands, limited time, and community norms [43]. These experiences not only restrict men’s capacity to participate in caregiving but may also lead to internal dissonance when they are unable to fulfill culturally sanctioned roles of fatherhood, a phenomenon that has been described as moral injury in fatherhood research [45,46]. Similar findings from Uganda indicate that even when men express egalitarian views, structural constraints and social stigma around caregiving can limit their involvement in parenting programs [47]. These dynamics highlight the need for interventions that actively affirm men’s caregiving identities and address the structural barriers that limit engagement. While women reported greater socioeconomic adversity in the present study, integrating digital parenting programs with broader economic and social protection initiatives may help reduce the opportunity costs of participation for both men and women [48].

Although ParentApp was intentionally designed to support all caregivers rather than assuming mothers as the primary users, these findings suggest that additional strategies may be needed to improve engagement among men. Future iterations could benefit from further co-design with fathers to better understand how intervention content, tone, imagery, and delivery align with men’s caregiving identities and everyday constraints. Emerging work on digital fatherhood interventions illustrates how user-centered design can be used to develop app-based support tailored to fathers’ mental health, well-being, and parenting needs [49]. However, improving engagement among men may require more than tailoring content or delivery; it may also require gender-transformative approaches that actively challenge inequitable norms and support identity shifts in caregiving. As highlighted in a recent systematic review of fatherhood programs in LMICs, gender-transformative interventions foster critical reflection, promote shared decision-making, and engage with normative change at individual and community levels [48]. Promising examples include in-person parenting programs such as REAL Fathers in Uganda [50], Bandebereho in Rwanda [51], and Parenting for Respectability in Uganda [52], which have achieved shifts in father involvement, reductions in intimate partner violence (IPV), and improvements in household dynamics through facilitated dialogue, peer role-modeling, and local leadership.

Emerging evidence suggests that gender-transformative components can be successfully integrated into digital formats. One example is ParentText, a chatbot-based parenting intervention adapted from the PLH program suite and piloted in South Africa and Jamaica. ParentText incorporates content on gender-equitable behaviors and IPV prevention to support shared caregiving, challenge harmful gender norms, and reduce familial violence [53]. Qualitative findings suggest that men valued the anonymity, accessibility, and nonjudgmental tone of the chatbot, which facilitated engagement with sensitive topics such as relationships and caregiving [54]. A parallel pre-post pilot study reported modest reductions in women’s experience of IPV and improvements in men’s attitudes and gender-equitable behaviors. However, overall engagement with the program and with the IPV-related content in particular was limited, especially among men [55]. Together with the findings of the present study, this emerging evidence suggests that moving beyond one-size-fits-all approaches may be key to improving engagement with digital parenting interventions.

Several limitations should be acknowledged when interpreting these findings. First, engagement was operationalized using digital metrics (eg, module completion), which offer behavioral indicators of app use but do not capture content comprehension or real-world application. This reflects a broader challenge in digital health research, where engagement is often equated with usage alone, without consideration of learning or behavior change outcomes [56]. Moreover, these metrics do not capture users’ subjective experiences, such as interest, attention, or emotional resonance, which may be equally important in understanding meaningful engagement [9]. Future research should therefore incorporate multidimensional measures that capture both the quantity and quality of engagement, alongside users’ subjective experiences.

Second, the range of caregiver characteristics assessed at baseline was limited by the brief, app-embedded format of the survey. Key variables such as perceived digital skills and prior experience with technology were not captured, although these factors have been shown to influence engagement with digital interventions [57,58]. The study also did not measure adolescent-level characteristics, such as child mental health symptoms, which have been positively associated with parental engagement in previous research [56]. Accounting for these variables in future studies may improve the precision of engagement models and support more targeted intervention strategies.

Third, while the overall sample size was relatively large, the study may have been underpowered to detect gender interaction effects, especially given the smaller proportion of men in the sample. Future studies should consider adequately powered subgroup analyses to explore gender-specific patterns of engagement. Furthermore, although the sample was drawn from low-income urban and periurban settings, all participants were required to have regular access to a smartphone. As such, findings may not be generalizable to caregivers in rural areas or those with limited access to digital technology. Broader implementation research is needed to understand how structural barriers such as digital access, network reliability, and device sharing influence engagement, including in rural and more resource-constrained settings.

This study provides important evidence on caregiver engagement with a digital parenting program in an LMIC. Across demographic, socioeconomic, behavioral, and psychological factors, only caregiver gender significantly predicted engagement, with women completing more modules than men. The absence of significant associations with other baseline characteristics provides encouraging evidence that socioeconomic and psychosocial factors may not pose substantial barriers to engagement with ParentApp. However, high rates of attrition among both men and women immediately after onboarding highlight early disengagement as a critical challenge. Addressing this challenge will be essential if digital parenting programs are to achieve meaningful and lasting benefits for children and families in LMICs.