The primary aim of this study is to evaluate the impact of the MBN intervention on changes in maternal depression, parenting strategies and knowledge, as well as infant social-emotional and social-communication behavior at postintervention and 6-month follow-up. We will also examine the relationship between maternal and infant changes. The study is being conducted in the urban core of a large southern US city, which is one of the fastest growing and most segregated and economically inequitable in the country.

The study is registered as a clinical trial at ClinicalTrials.gov (NCT03464630). Before human subjects’ activity, the full, detailed study protocol #H18217 was approved by the Georgia State University Institutional Review Board (IRB) on November 21, 2017 (see Multimedia Appendix 1 for peer review summary statements of the grant proposal).

The study uses a two-arm, randomized controlled intent-to-treat trial design, with random assignment in a 1:1 allocation to one of the two parallel mobile intervention conditions. IRB approval is obtained before the involvement of human subjects. Recruitment strategies include the distribution of study information to health and social service agencies serving low-income women. Print materials were provided in agency offices, web-based posts were placed on health and social service websites, and text blasts were sent by community service providers to women on their service lists. Referral is conducted through a project web-based referral system to support mother self-referral, provider referral, and research team referral. Referred mothers are screened by phone to determine their eligibility for inclusion.

The eligibility criteria include the following: biological mothers aged 18 years or older with an infant younger than 1 year, who are English speaking, and who meet the Patient Health Questionnaire (PHQ)-2 criteria for elevated depression symptoms [47]. Exclusion criteria at the time of screening include maternal homelessness or shelter residence, major physical or mental illness that would hinder meaningful participation, infant major physical illness, and not having custody of the infant. Inclusion and exclusion criteria are established to ensure that mothers were not burdened by severe stressors that might restrict their ability to participate in the study.

Sample size determination is based on anticipated effect sizes and the minimum sample size needed to have sufficient power to detect these effects. In this study, a moderate-to-large effect size (Cohen d=0.5-1.03) is anticipated for maternal outcomes and a small-to-moderate effect size (Cohen d=0.2-0.4) is anticipated for infant outcomes based on the PALS program evidence [18,19], our Baby-Net results [23,46], and evidence of Coping with Depression Course [48] and internet-based cognitive behavioral therapy treatment success [49]. For the smallest anticipated effect sizes (ie, those for infants), to detect these effects within a 2×2 analysis based on α=.05, a sample of n=75 per condition is needed (total n=150). Within this sample size estimation, effects were viewed relative to both high (r=0.68) and low (r=0.21) repeated measures correlation. We found, with power at 0.95, we could detect an effect as low as Cohen d=0.37 (with low repeated correlation) and Cohen d=0.23 (with high repeated correlation), and with a power of 0.80, we could detect an effect size as low as Cohen d=0.29 (with low repeated correlation) and Cohen d=0.18 (with high repeated correlation). To view latent growth curve model maintenance trajectories through follow-up, the number of cases per estimated parameter needs to be sufficient, with a rough guideline of 5:1 [50,51]. Using this guideline, with an estimated sample size of 150, we will have the sample size to estimate 30 parameters, which is sufficient for modeling three time points and a condition predictor.

On the basis of a sample size of 150 needed across the three study assessment points (pre, post, and follow-up), we used anticipated attrition rates to estimate the initial sample size needed to achieve this number. We expect a pre-post attrition rate of 10% and an overall 17% rate at follow-up assessments based on traditional PALS, across a series of randomized control studies, with attrition rates ranging from 9% to 24% [52,53]. In our recent Baby-Net R01 study [45], we observed a 7% attrition from pre to postassessment and 15% at 6-month follow-up working with low-income mothers, some of whom were experiencing elevated levels of depressive symptoms. As such, we estimate that we will need an initial sample of at least 180 mothers to initiate the study, consent, and complete preassessment.

An additional consideration in this study is our need to screen for maternal depression to arrive at an initial sample of 180. The estimated range of depression within our targeted recruitment population of low-income, diverse women ranges from 30% to 50% [11,12]. Hence, at least one out of three women we screen will likely be eligible. This estimate will require us to screen 540 mothers to yield an eligible sample of at least 180 mothers. In previous studies [37,44], 4 out of 5 women who self-select to be screened and are eligible go on to consent to preassessment and intervention. Hence, it is necessary to screen a total sample of 675 to obtain a sample of 180 mothers.

The study includes two parallel mobile internet remote coaching intervention programs that are identical in intervention structure. To reduce literacy demands and maximize accessibility, both programs are video- and narration-based. The structure of intervention delivery for both programs includes the following: (1) web-based administration of a 14-session intervention with video, narration, and activities to present session content and check-in questions to assess knowledge acquisition, recorded in the database for review by both parent and coach; (2) creation of a 5-minute app-collected video of mother-infant interactions for later review by coach and parent; (3) summary of topics; (4) daily activities (homework); (5) participant-rated satisfaction, ease of use, and effectiveness recorded in the database; and (6) weekly video coach calls to coview the mother-infant video. All mothers receive an iPhone with access to their assigned intervention program and unlimited mobile calls, data, and texting. Participants complete the study activities in their homes using these mobile devices.

The content of the MBN intervention sessions on mood improvement focuses on mood monitoring, behavioral activation by increasing mother pleasant activities, and cognitive coping strategies [37]. Parenting content focuses on recognition of infant signals, warm and contingent responding to infant signals, maintaining infant attention and interest, and early language literacy promotion strategies [23,45]. Within the app, mothers receive daily reminders to rate their mood based on their preferred schedule.

As a comparison condition, the Depression and Developmental Awareness (DDAS) program [46] serves as an attention control for the time spent in intervention and remote coaching contact. In contrast to targeting maternal mood and parent practices, the content focuses on awareness of maternal depression and infant developmental milestones.

Project protocols for consent, assessment, and intervention are used to train all project staff before study initiation. Our consent protocol consists of staff training on a checklist to ensure ethical informed consent. Supervisors observe staff conducting mock consent administration and view their performance relative to the checklist, with a requirement that staff reach 100% accuracy on checklist coverage. For conducting assessments, detailed project protocols are used to train assessors before conducting assessments with mothers. Assessors are trained on assessment protocols focused on providing appropriate assessment instructions, helping mothers understand questions in a manner that will not influence their responses, and administering assessments to mothers verbally if desired by the mother. For interventions, coaches in both conditions are trained to conduct weekly review calls with mothers, rate maternal progress, and complete implementation fidelity checks. A total of 20% of the recorded coach calls are randomly selected for independent completion of fidelity checklists to calculate interobserver agreement of fidelity. In addition, all staff have regular supervision meetings with the principal investigator (PI) to monitor adherence to the consent, assessment, and intervention protocols.

The primary study outcomes include the following: (1) maternal depressive symptoms, (2) parent-sensitive and responsive practices, (3) parent knowledge of infant social-emotional and social-communication behaviors and their promotion, and (4) infant social-emotional and social-communication engagement.

Following consent, preintervention assessment is completed face-to-face in home or via a mobile video meeting. This comprehensive, multimethod assessment includes interviews, questionnaires, and observational procedures to obtain demographic information and community service receipt and to assess the domains of maternal functioning, including depression symptoms, parenting attitudes and beliefs, parent practices, infant social-emotional and social-communication functioning, and parent-infant interaction. This comprehensive assessment protocol is repeated at postassessment. A 6-month follow-up assessment is administered electronically to assess maternal depression symptoms and parenting knowledge of infant social-emotional and social-communication behaviors and their promotion.

Maternal depressive symptoms and severity are measured using the PHQ-9, a 9-item self-report instrument for screening, diagnosing, monitoring, and measuring the severity of depression [54]. Question 9 screens for suicidal ideation. The PHQ-9 has an internal reliability of 0.89 in a primary care setting and 0.86 in an obstetric setting. Maternal parenting behavior, attitudes, beliefs, knowledge, and stress are assessed as follows: the Landry Parent-Child Interaction Scale observational coding system [55], designed to assess naturalistic parent-child interaction during play at home, is used to code video-recorded mother-infant interaction behavior. The parent scales of interest in this study relative to maternal responsiveness include ratings of maternal positive affect, warmth, flexibility, and positive verbal content. Relative to maternal negative behavior, scales of interest include ratings of maternal physical intrusiveness as well as verbal and affective negativity. The Landry Parent-Child Interaction Scale has been used in a series of federally funded longitudinal and intervention studies over the past 15 years and has yielded adequate reliability and demonstrated the predictive validity of child social-emotional outcomes [55]. Behaviors will be coded across a semistructured play activity over a 5-minute period. Coders, blinded to intervention conditions and time points, conduct coding based on observing parent-infant interaction videos of participants in both intervention conditions. To assess reliability, 20% of all interactions are scored by 2 independent coders. The Indicator of Parent Child Interaction-2 [56] includes a brief semistructured book activity, which is video recorded, and will be used to code the following behaviors: (1) mother facilitative behaviors including conveyance of acceptance and warmth, descriptive language, following the child’s lead, and maintaining the child’s interest; (2) mother interrupting behaviors including intrusions, restrictions, and critical comments; (3) infant engagement behaviors, including positive social engagement, follow through, and sustained engagement; and (4) infant behaviors that interfere with engagement, such as fussiness, disruptive behaviors, and withdrawn behavior. The Indicator of Parent-Child Interaction has adequate psychometric features [57] and has been used to assess mother and infant behavior in multiple studies of high-risk infants and in population-based studies of universal interventions to promote early positive parent support behavior [58-60]. To assess reliability, 20% of all interactions are scored by 2 independent coders.

The Knowledge of Infant Social-Emotional Behavior and Promotion [45,61] has been used in previous intervention studies and is geared toward an understanding of the concepts of infant social-emotional behavior and its promotion by caregivers, assessing both definitional and applied concept knowledge, with items structured in multiple response formats, including open-ended, true or false, and multiple choice formats. The Concepts of Development Questionnaire [62] is a 20-item, Likert-type four-point scale (4=strongly agree and 1=strongly disagree) that assesses parenting beliefs. Specifically of interest in assessing our maternal responsiveness domain, the Concepts of Development Questionnaire focuses on constructs of flexibility and child centeredness, in contrast to parent centeredness. Parenting Sense of Competence [63] is a 17-item scale assessing parents’ satisfaction and self-perceived competency in the parenting role. Adequate internal consistency, factor structure, and construct validity have been reported [64]. The Parenting Stress Index-Short Form [65] is a 36-item self-report instrument that assesses stress directly associated with the parenting role using a five-point scale to indicate the degree to which that item has been stressful, with validity demonstrated for at-risk mothers [66]. The Brief Child Abuse Potential (BCAP) [67] is a 34-item self-report screening instrument that contains seven domains. The primary clinical scale (abuse) comprises six factor scales: distress, rigidity, unhappiness, persecution, loneliness, family conflict, and poverty. In addition, the BCAP contains three validity scales: lie, random response, and inconsistency. Overall, the 24-item BCAP abuse scale has high internal consistency (0.89); temporal stability estimates for the abuse scale are also adequate (ie, 0.91 and 0.75 for 1-day and 3-month intervals, respectively). The Automatic Thoughts Questionnaire [68] is a 30-item self-report instrument that measures the frequency of automatic negative thoughts related to depression. It contains four domains: personal maladjustment and desire for change, negative self-concepts and negative expectations, low self-esteem, and helplessness. The Automatic Thoughts Questionnaire has a high internal consistency (0.97). In our own work, we have demonstrated negative thoughts as a mechanistic explanation of Mom-Net intervention effects on maternal depression [69]. The Revised Dyadic Adjustment Scale [70] is a 14-item self-report instrument that assesses seven dimensions of relationships in three domains: consensus, satisfaction, and cohesion. Overall, the Revised Dyadic Adjustment Scale has a reliability (Cronbach α) of .90.

Infant social-emotional behavior and development change are assessed as follows: observed infant behavior in interaction with the mother was assessed using the Landry Parent-Child Interaction Scale, as described above [55]. Rating scales of interest in assessing infant behaviors include attention or arousal, warmth-seeking, and behavioral regulation. The Devereux Early Childhood Assessment for Infants [71] is a 33-item behavior rating scale that assesses child protective factors central to social and emotional health and resilience in infants aged 4 weeks up to 18 months, which displays adequate reliability and validity for use in this study [72].

The Family Profile Report Form [73] uses demographic and life course history data, including maternal relationship status and support, health status, psychiatric history, and other services received, to describe study participants and examine the potential moderating effects of the intervention. Socioeconomic stress will be assessed using Conger and Elder measure of economic hardship based on their family process model of economic hardship [74], which assesses different areas of financial stress and has been used in many studies to describe important aspects of societal disadvantage in samples [75-77]. Intervention dosage will be assessed by electronic recording of the mother: (1) number of sessions and homework completed, (2) time on the web, and (3) number of times the intervention was visited. System activity logs will provide descriptive statistics on (1) time of day at log on; (2) length of time in the intervention; (3) number of times logged on per week; and (4) length of time spent in the information, support, video, and assessment areas of the intervention. Program attrition will be documented, and for these subjects, we will conduct exit interviews to determine the reasons for discontinuation. Table 1 presents our project enrollment, intervention, and assessment schedule.

All data will be deidentified by using a project-specific identification number for each participant. Links between participant names and identification codes will exist in written form only on consent forms, and these forms will be stored in a locked room and file cabinet accessible only to the Georgia State University project staff. In addition, electronic files containing participant identifiers or data will be accessible only to the approved project staff. All nonidentifiable data (ie, those labeled with codes only) will be stored in the same manner in locked file cabinets or on password-protected, secure computer networks.

The data used for progress monitoring will be obtained electronically through the iOS app. State-of-the-art security protocols are used in all data collection and monitoring activities, as used for electronic commerce, using VeriSign SSL. The program app will be established within the Oregon Research Institute network with firewall to maintain security. The program app will be accessible via an iPhone with username and password protection.

Our postassessment n, to examine acute intervention effects, is expected to be 150 (75 per condition). We will initially view intervention effects on our mother or infant outcomes using a 2 (pre-post)×2 (intervention group) repeated measure analysis of variance (ANOVA). We will examine the intercorrelations among outcomes and, if significant, will use a MANOVA approach to examine intervention effects. The F test is robust to nonnormality if such nonnormality is caused by skewness rather than outliers. We will take appropriate measures to reduce outlier influences.

To examine the relationship between maternal change (parenting and maternal depression) and infant change, one approach will be to create individual β estimates for mothers and infants using the polynomial contrast function within MANOVA to produce individual trajectory scores reflecting parent and infant change from pre to post that can then be used in external between-condition covariate analysis. The trajectory scores will be subjected to an analysis of covariance, with infant change trajectories as the dependent variable and parent change as the covariate. We will determine whether the parent change covariate is significantly related to the dependent variable (demonstrating that changes in parenting behavior and child functioning covary). We will determine the statistical significance and effect sizes of the intervention group effects (ignoring the covariate). We will then determine if entering the parent functioning covariate modifies the intervention group effect size and statistical significance. If parent change is strongly linked to infant change, then entering the covariate should result in nonsignificant intervention effects. Estimates of covariance-adjusted effect sizes will provide estimates of the proportion of the intervention group effect size, which can be explained by the parenting change variable. We will test for between-group heterogeneity of covariance to determine whether the strength of the association between change in parenting and infant functioning differs by intervention group.

To evaluate the moderating influences on parent and infant behavior, our first focus is on how maternal depression before intervention (ie, chronicity, severity, and receipt of psychiatric treatment or medication) affects intervention dosage; second, we are interested in the moderating effect of dosage on mother and infant change. For depression experience before intervention, we will examine a 2 (high or low dosage)×2 (intervention condition) ANOVA using maternal depression indicators (eg, chronicity) as the dependent measure. A significant main effect for dosage would indicate that higher levels of maternal depression are found at different dosage levels; it is anticipated that higher depression will be evidenced in the low dosage group. A significant interaction would indicate that a higher level of maternal depression is found within a dosage by condition cell. Although we would not anticipate a significant interaction, we will examine if higher levels of depression are associated with low dosage only within the MBN intervention, which could indicate that the skills focus of MBN learning may have been too intense for highly depressed mothers.

To examine dosage as a moderating influence on mother and infant change, we will use the individual β slope estimates reflecting parent and infant change as the dependent measures in separate 2 (intervention condition)×2 (high or low dosage) ANOVA designs. We hypothesize that a significant interaction term indicating the highest positive change trajectories will be for MBN mothers with high dosage, when compared with low dosage MBN mothers and DDAS mothers, regardless of dosage level. If a significant dose-effect relationship exists within the MBN condition, we will determine if an effective dosage level can be identified that is less than the maximum number of intervention sessions offered.

For contextual risks, analysis will examine Pearson correlations between contextual risk and level of maternal depression and a biserial correlation for the relationship between contextual risk and dosage (high or low) to determine if contextual risks are related to both initial levels of depression and subsequent engagement in intervention. To further examine the contextual risk of mother and infant change, we will form a high or low risk categorical variable based on median split and use the same 2 (intervention condition)×2 (high or low risk). If the main intervention condition effect is significant, we would expect MBN mothers and infants, regardless of risk level, to show the greatest improvements when compared with DDAS mothers. If significant interaction effects occur, we would expect mothers and infants within the MBN condition with lower levels of risk to show the greatest improvement in functioning and, though not statistically significant, that MBN mothers and infants, even in the presence of high risk, would show higher positive change trajectories than those of DDAS mothers and infants.

Maintenance affects will be viewed by a single indicator for maternal depression and parenting knowledge administered at follow-up. To this end, we will examine maternal change trajectories using structural equation modeling methodology, perform latent growth curve model analyses, and include an intervention condition that predicts intercept and slope estimates. This analysis will supplement our aim 2 analyses, and if these follow-up variables generally reflect the acute intervention trajectories, this will provide support for our view of maternal change across time. Given the restricted nature of follow-up assessments, balancing participant burden, and the desire to maximize assessment completion across time, we will view these maintenance trajectories with caution.

Before the study activities, all staff complete human subjects training and participate in safety monitoring and safety responding training under the supervision of the PI and licensed psychologist. Training includes discussion and written responses on study safety monitoring forms, followed by completion of safety monitoring forms to fidelity based on vignette practice. Safety monitoring forms are completed at all assessment time points and upon reporting of any concern about potential harm. Mothers in both conditions receive biweekly automated texts to complete the PHQ-9 [54]. PHQ-9 results are monitored by project coaches for trends in increasing depression severity and harmful thoughts, which trigger an immediate safety response from coaches. All safety monitoring forms that include endorsement of harmful thoughts and required safety response documentation are submitted to the project PI for review. Adverse events and serious adverse events are logged and reported to the IRB and National Institutes of Health (NIH).

The study was funded by the NIH to begin September 1, 2016. Recruitment efforts were initiated immediately following IRB approval on November 17, 2017. A sample of 184 mothers and their infants were recruited into the randomized controlled trial study between February 15, 2018, and March 11, 2021. Study intervention is underway, and we anticipate that follow-up assessments, which mark the end of data collection, will be completed in March 2022. Following the onset of COVID-19, we published a formative descriptive report on recruitment strategies [46] and a descriptive report of progression from referral to intervention initiation for mothers with study experience before the pandemic and mothers with study experience during the pandemic [83].

Deidentified study data will be made available publicly through a Georgia State University website and through a digital object identifier–linked public repository OpenTrials [84] following a 1-year embargo from the date of publication of primary study outcomes to allow for any commercialization.