Participants were women (aged ≥18 years) receiving prenatal care from 1 of 4 University of Pittsburgh Medical Center obstetric practices. Electronic health records in the participating practices were screened for study eligibility on a weekly basis between October 2024 and January 2025. Eligibility criteria included being in the first trimester of pregnancy and screening negative for active first-trimester depressive symptoms using the Patient Health Questionnaire–9 [15].

Eligible pregnant women were sent an invitation via their patient portal to complete the 6-question predictive screener and subsequent survey questions related to the screener. The invitation contained a unique survey link that expired at the end of the first trimester. Upon accepting the link and completing informed consent, a pregnant woman was considered a study participant. Participants were not provided with a risk score. This allowed for independent validation of the screener in this population through passive observation of depressive symptoms and diagnoses over the course of the pregnancy. However, mental health and other pregnancy resources were provided to all participants, regardless of risk, to reflect best clinical practice.

Feasibility of screening was assessed by measuring (1) the proportion of pregnant women eligible from the larger clinical population (eligibility rate), (2) the number of eligible individuals who engaged with the screener study invitation (reach), and (3) the number of completed surveys among those who accessed the survey (completion rate).

Acceptability was assessed with a series of Likert-scale questions (1=strongly disagree to 5=strongly agree), which included participant evaluations of screener question clarity (“The screener questions were clear”), ease of completion (“The screener was easy to complete”), and comfort with and honesty in disclosing responses (“I would be comfortable sharing my answers with my provider” and “I would have no concerns about answering the questions honestly,” respectively).

Participants were asked about the usefulness of the screener as part of routine prenatal care, including the benefits of knowing the results of the screener (“What would be the benefit(s) to you personally of knowing you are likely to develop depression at some point during your pregnancy?”), their preferred method of receiving screener results (“If you completed a screener like this as part of your prenatal care, how would you prefer to receive your results?”), perceived helpfulness of knowing depression risk (“How helpful would it be to know whether you are likely to develop depression at some point during your pregnancy?”), and any concerns they might have if they were to learn that they were likely to develop depression at some point during their pregnancy (“What concern(s) would you personally have about knowing you are likely to develop depression at some point during your pregnancy?”). Participants were also asked to endorse from a prespecified list and/or offer open-ended options for desired preventive care resources (“If you learned you were likely to develop depression at some point in your pregnancy, which of the following resources would you be most likely to use if available to you at low or no-cost?”) and offer any additional thoughts or comments they have about the survey or depression prevention (“We welcome any other thoughts you have about your participation in this survey or preventing depression in pregnancy.”). Deployment feasibility and acceptability of the screener were the primary outcomes of this study. Secondary analyses examined differences in attitudes between high- and low-risk pregnant women.

Descriptive analyses used means and SDs for continuous variables (eg, maternal age), medians and IQRs for ordinal variables (eg, Likert-scale responses), and distributions for categorical variables. The study sample was stratified by screener score into those with and without risk of developing moderate to severe depression symptoms. Differences in survey responses by risk category were assessed using chi-square tests for categorical and binary responses. The distributions of continuous and Likert-scale responses were assessed for normality using the D’Agostino-Pearson test. Differences between normally distributed responses were evaluated using 2-sample 2-tailed t tests, and nonnormally distributed responses were evaluated using Mann-Whitney U tests. Effect size was calculated using Cliff d, a robust nonparametric method appropriate for ordinal data such as Likert-scale data [27]. An α level of .05 was used for statistical significance. All respondents completed every survey item, resulting in no missing data. Analyses were performed using Python (version 3.12.4).

Rapid inductive thematic coding was applied to open-ended survey responses by 2 coders, each with qualitative coding experience and one with formal training in rapid qualitative analysis. After reading all open-ended survey responses, the coders cocreated a codebook. Responses were then independently coded. Once independent coding was complete, codes were compared, and discrepancies were discussed until consensus (ie, consistent classification) was reached.

All study procedures were approved by the University of Pittsburgh Institutional Review Board (STUDY24020098) and are reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for reporting observational studies [28]. We conducted pretests of the survey with pregnant or recently pregnant colleagues, gathering feedback on the design and clarity of the instructions and survey questions. Participants provided digital consent for study participation and were informed that they could withdraw their consent at any time. They received a US $20 gift card for completing the screener and follow-up survey, which took approximately 10 minutes. All participant data were stored in a confidential and secure manner and access was restricted to authorized members of the research team.

During the study period, 1925 women received prenatal care at one of the participating obstetric practices. Of these 1925 women, 810 (42.1%) had an active patient portal, were at a gestational age of <14 weeks, and had scored <5 on the Patient Health Questionnaire–9 measure of active depression symptoms (eligibility rate). Of the 810 sent an invitation through their patient portal, only 2.8% (n=23) did not access it. In total, 36.7% (289/787) who accessed the invitation opened the study link (reach) and, of those, 88.2% (255/289) completed it (completion rate).

All participants were female with a mean age of 32 (SD 4.5) years. Most were White (217/255, 85.1%) and non-Hispanic (232/255, 91%), married (196/255, 76.9%), and had commercial insurance (196/255, 76.9%). Full demographic information is available in Table 1. Participants were demographically representative of the larger eligible population.

On the basis of established cutoff scores [17], 80% (204/255) of participants were identified by the predictive screener as having no or minimal depression risk and 20% (51/255) were identified as at risk of developing moderate to severe depression at a later stage of pregnancy.

The predictive screener demonstrated high acceptability among participants. Most agreed or strongly agreed (236/255, 92.5%) that the screener questions were clear (median 5, IQR 4-5), and 97.3% (248/255) reported that the questions were easy to complete (median 5, IQR 5-5). Most (236/255, 92.5%) agreed or strongly agreed that they would feel comfortable sharing the results of the screener with their obstetric care team providers (median 5, IQR 4-5). Similarly, the majority (247/255, 96.9%) agreed or strongly agreed that they would have no concerns about answering the questions honestly (median 5, IQR 5-5).

Table 2 presents acceptability metrics stratified by depression risk category, with those at risk of developing moderate-to-severe depression symptoms being less likely to endorse strong agreement with comfort and ease of honesty in responding to the screener.

Overall, 80% (204/255) reported that the screener was helpful or extremely helpful (median 4, IQR 4-5). Participants largely voiced a preference for receiving screener results through the patient portal (150/255, 58.8%), compared to receiving results in person (76/255, 29.8%) or by phone (13/255, 5.1%). Perspectives did not differ by depression risk.

When asked about potential benefits of receiving their depression risk screening score, participants most frequently cited seeking appropriate preventive resources and/or treatments (63/204, 30.9%) and the opportunity to receive education from their obstetric health provider (53/204, 26%). One survey participant wrote that the feedback could help them “reduce risk factors associated with postpartum depression and postpartum psychosis.”

Despite generally positive attitudes toward score disclosure, participants also expressed concerns. The most frequently reported concerns included increased anxiety or worry about developing depression (71/204, 34.8%) and worry about the inability to prevent depression (34/204, 16.7%). Several open-ended responses were offered, including concern about parenting ability and emotional connection to children if depression were to occur, premature medication use, and false positives.

Participants at risk of developing moderate to severe depression were significantly more likely than those with no or minimal depression risk to endorse the concern “fear that your health care provider will view you negatively” (5.9% vs 0.5%; P=.03). However, at-risk participants were also more likely to cite the ability to identify factors that might be affecting their mood as a benefit (17.6% vs 5.9%; P=.01). Table 3 presents the complete breakdown of perceived benefits and risks.

When asked about desired preventive health resources, participants endorsed multiple options. Preferred resources included counseling (197/255, 77.3%), with 61.8% (126/197) endorsing in-person therapy and 41.6% (82/197) endorsing app-based delivery. Mind- or body-based interventions (166/255, 65.1%) were frequently endorsed, with “yoga, exercise classes, or gym membership” (114/166, 68.7%) being a highly desired intervention, followed by prenatal classes or support groups (81/255, 31.8%).

Participants (n=61) shared several resources not otherwise listed in open-ended responses. From most to least frequently listed, these included mental health resources for their spouse or support person (eg, “Education for family/spouses so they can possibly help the individual battling depression”), self-care opportunities (eg, “massage”), practical support such as cleaning services or meal preparation (eg, “A service similar to task rabbit that is geared towards pregnancy needs. Perhaps people that could help identify inexpensive sources of second hand furniture/items, assembly help, specific errand support, meal prep/planning for once baby arrives, etc.”), parental leave, increased medical touchpoints during pregnancy, additional mental health education (eg, books and fact sheets), medication for depression prevention, structural changes (eg, improved health care access and transportation), access to a live mental health support person (“Mental health support chat forums with a licensed counselor”), a more holistic approach to health care, and additional faith or spiritual support.

Figure 1 shows preferred resources stratified by depression risk. Notably, less than 10% (20/204) of those with no or minimal depression risk expressed interest in childcare support compared to more than 25% (14/51) of those at risk of developing moderate to severe depression (χ²₁=9.52; n=255; P=.002). Although highly endorsed by both risk groups, participants with no or minimal depression risk expressed greater interest in exercise opportunities (χ²₁=3.94; n=255; P=.05).

After completing the survey, 55.3% (141/255) of participants shared optional thoughts about screener participation and about preventing depression in pregnancy. The most frequent open-ended comments expressed appreciation for the study or confirmed that participants would find a screener of this kind helpful. Additional responses ranged from suggestions for the screener or resources (eg, “A big area of concern for working moms is their jobs. I felt that the questions did not address the concern as much.”) to specific stressors (eg, “I think one of the biggest stressors that can contribute greatly to mood in the first trimester is the viability and health/wellness of the baby, i.e., risk of miscarriage or is there any genetic or chromosomal issue?”) and barriers to care (“I feel like cost is one of the biggest barriers to seeking help, just ahead of the stigma that still exists.”). One participant reported feeling that a discussion of depression may exacerbate depressive symptoms. Those at risk of developing moderate to severe depression more frequently expressed the need for help and social support during pregnancy, while those with no or minimal depression risk largely highlighted the need for depression- and anxiety-specific resources.

This study demonstrates the acceptability of a machine learning–based predictive screener designed to support perinatal depression prevention. Predictive screening is an emerging field, and, to the best of our knowledge, this is the first study to examine receptivity to a predictive mental health screener, to query pregnant women about their desired preventive care options, and to examine differences in desired resources by predicted risk level. While machine learning algorithms are being newly applied to mental health prediction, their ultimate success depends on adoption into clinical workflow, which requires patient engagement and a feasible implementation process [29]. Here, we highlight an approach to evaluating receptivity to such predictive screeners while testing the feasibility of sending the screener through our health care system’s digital patient portal.

In this study, pregnant women who completed the screener largely reported positive screener experiences, with over 90% agreeing that it was clear, easy, and comfortable to respond to and that they had no concerns about responding honestly. More than 80% thought it would be helpful to know their risk of developing depression symptoms during pregnancy, regardless of their risk score. Those at risk of developing moderate to severe depression were slightly more likely to report feeling uncomfortable sharing their honest responses with an obstetric health care provider, suggesting that the screener questions touch on risk factors that are challenging to disclose. Perceived benefits and risks of the screener were similar between the no- or minimal-risk and at-risk groups, with 2 exceptions: those at risk of developing depression reported greater value in the screener for helping them understand contributors to their mood. They also reported greater fear about being judged by their provider. These anticipated challenges may be grounded in stigma and prior experiences of discrimination [30,31].

Deploying the screener through the patient portal reached a little more than one-third of the eligible pregnant women receiving prenatal care in the participating obstetric practices. This suggests that in-person screening may need to be offered as part of a clinical workflow to deploy the screener effectively to a majority of the patient population. However, among those who initiated the screening survey, engagement was encouraging, with almost 90% of participants completing the entire screening process. Thus, if the eligible patient reach is optimized through multiple outreach channels, the screener could have a high completion rate.

Survey participants identified specific perceived benefits and concerns associated with learning about being at risk of developing depression symptoms. The desire for therapy, whether in person or through an app, is reflective of the current best practices for preventive care, which include cognitive behavioral therapy and interpersonal therapy. Notably, participants also reported a desire for exercise opportunities and support groups, echoing research showing the protective effects of physical activity and social support on depression [18,21,32,33]. Child care was highly desired among the study population at risk of developing moderate to severe depression, suggesting that direct tangible support may impact the trajectory of depression risk. Financial stressors during pregnancy have previously been associated with poor outcomes for both mothers and infants. Findings from recent studies of guaranteed income programs for perinatal women show promising evidence of improvements in maternal mental health and infant developmental outcomes associated with alleviating financial stress [32,34-36].

This study contributes to a burgeoning literature on the importance of providing preventive mental health care during pregnancy and the postpartum period [37]. The perinatal period is uniquely vulnerable to depression due to a combination of hormonal changes, novel stressors (eg, pregnancy-related anxiety), and environmental factors [38]. However, up to half of the cases remain undiagnosed [39]. Although women without a psychiatric history are at lower baseline risk, they are more likely to be overlooked when they do experience depression, with lower rates of both depression screening and referral to care [40]. One solution to address the lack of diagnosis is more frequent, routine depression screening [13]. Care navigators dedicated to monitoring mental health symptoms throughout pregnancy, for example, have been shown to improve depressive symptoms among the patients they serve [41,42]. However, to date, there have been no screening tools available to support those professionals in identifying risk among individuals not actively reporting depressive symptoms [11]. If successfully implemented, predictive screeners could make the work of such care navigators more targeted and efficient.

This study has several limitations. Eligible participants were limited to those who could be contacted through an active patient portal. Although most of our prenatal patients have an active patient portal account, it is possible that these portals are not frequently accessed. Other studies suggest that individuals with limited health literacy are less likely to use patient portals to access health information [43-45]. Thus, the moderate response rate may introduce selection bias if individuals more familiar with portal-based screening were more likely to participate, whereas those with lower health literacy were less likely to engage. There were some differences in responses by participant risk level. Although these differences seem intuitive (eg, participants with greater depression risk were more likely to find the screener helpful in identifying influences on mood yet were more concerned about potential stigma from providers), they should be interpreted with caution given the modest sample size of the at-risk group and the number of comparisons conducted. The study sample was generally representative of the patient population of the academic medical system in which it was conducted, and the survey was developed using a patient-centered design approach with a diverse population. However, the results cannot be generalized to other health care systems in the United States or to clinics with more diverse patient populations. While the screener showed high accuracy within the sample on which it was built [17], further work is needed to establish independent external validation. Additional research must address how to best implement this screener within our health care system, including a testing workflow to fully close the loop on preventive care. A future goal would be to conduct validation and acceptability research of the screener in non-US populations.

This study is a first step toward understanding how a machine learning–based prevention-focused predictive depression screener can be integrated into a clinical workflow for preventive care. We found that only one-third of the eligible pregnant women in our clinics completed the screener via their digital patient portal. Ultimately, deploying preventive screening tools will require a hybrid digital and in-person screening approach to comprehensively screen all pregnant women who may be at risk of developing perinatal depression. Our risk screener was designed to identify individuals who may need additional evaluation and care from a trained provider. As such, it is intended as a supplement to current prenatal care, allowing for earlier identification of those at risk rather than serving as a replacement for professional mental health evaluation. Our findings suggest that the screener was acceptable to those who completed it. Moreover, participants voiced clear preferences for certain forms of preventive care in response to learning their risk. Participants’ preference for receiving predictive depression screening results through the patient portal suggests a promising approach to expanding access to mental health assessment among individuals already engaged with digital health platforms. Our findings also suggest some next steps on how to inform the clinical workflow to support a more personalized approach to preventive care in pregnancy, leveraging the most endorsed preventive care resources (eg, counseling, peer support groups, exercise, and childcare support) to address the individual risk factors identified by the predictive screener.