Relevant peer-reviewed papers published in English were identified by comprehensively searching key electronic databases from the earliest available evidence to October 27, 2021. These were the Web of Science, Cochrane Central Register of Controlled Trials, CINAHL, MEDLINE, Embase, PsycINFO, Scopus, ClinicalTrials.gov, and EBSCO Open Dissertations databases. The backward snowballing technique was also used to identify potentially relevant papers [37]. This involved looking at the reference lists of the selected papers to see if any more papers could be identified. The search strategy was created and refined based on consultation with a library information specialist (in Multimedia Appendix 1).

The inclusion and exclusion criteria used to screen the articles were based on PICOS (population, intervention, comparator, outcome, and study design) [38].

Figure 1 displays the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram for the screening and study selection processes. The search and screening processes were independently conducted by 2 reviewers (WMAWMY and OW) based on the titles and abstracts and after any duplicates were removed. Any disagreements were discussed with a senior researcher (HMM). Next, the 2 reviewers independently conducted full-text assessments based on the predefined inclusion and exclusion criteria. Both reviewers cross-checked the papers that were suggested for inclusion, and any disagreements were discussed and resolved with the senior researcher (HMM) and professor (AS). The list of excluded studies with their reasons for exclusion are available in Multimedia Appendix 2.

The methodological quality of the included studies was assessed using the Revised Cochrane risk-of-bias tool for randomized trials (ROB2) [43]. This tool addresses specific domains that can influence the risk of bias in an RCT. It asks a series of questions to enable users to evaluate the potential risk of bias in RCTs based on 5 domains: the randomization process, deviations from the intended intervention, missing outcome data, outcome measurements, and how the reported results were selected. The process evaluates each domain and enables the user to reach an overall judgement, namely a low or high risk of bias or some concerns. According to the tool guidelines, identifying a high risk of bias for any individual domain will produce an overall high risk of bias. Some concerns about any individual domain will lead to the paper being categorized as some concerns or a high overall risk [43]. The risk of bias assessments were independently conducted by 2 reviewers (WMAWMY and OW). Any discrepancies were discussed and agreed with the senior researcher (HMM) and professor (AS). The specific reasons for the 3 categorizations, namely low, some concerns, or high, were recorded in a Microsoft Excel spreadsheet template provided with the ROB2 tool.

A customized data extraction Excel spreadsheet was used by the reviewers. Data that were extracted from each of the final 7 papers included the author, year of publication, country, recruitment, sample size, and randomization. We also extracted the name, type, and mode of the intervention, screening method, primary depression scale, other mother-related outcomes, infant-related outcomes, number of sessions and duration, intervention start and end points, funding sources, brief synopsis, CBT elements, type of therapeutic guidance, and other support, such as technical or peer support. The research team also extracted and calculated the between-group effect sizes and 95% CIs after the intervention, user feedback, and the postintervention attrition rates. The Hedges g effect sizes were either extracted, if provided, or calculated using the 2020 Effect Size Calculator from the Center for Evaluation and Monitoring [44]. We extracted the means, standard deviations, and sample sizes of the intervention and control groups at the postintervention assessment point, during pregnancy. The effect sizes were interpreted as small (≤0.32), moderate (0.33 to 0.55), or large (≥0.56) [45]. The acceptability of the interventions was assessed according to the Theoretical Framework of Acceptability [46]. This suggests that intervention acceptability can be assessed at 3 time points: before, during, and after an intervention has been delivered [46]. In addition, the attrition rate percentage was calculated using the total number of subjects dropped out in the intervention group post-intervention over the total number of subjects in the intervention groups at baseline, after randomization. The data extraction was conducted by the first author (WMAWMY) and then independently checked by the second reviewer (OW) for accuracy and completeness. Disagreements were resolved through consensus discussion with a senior researcher (HMM). WMAWMY also contacted authors of included studies with a request for missing or additional information (in Multimedia Appendix 3).

Table 1 summarizes the characteristics of the 7 studies: 2 were from the United States, 2 were from China, and there was 1 each from Australia, Norway, and Sweden. The papers covered a total of 2830 participants, and the sample sizes ranged from 25 to 1342 at the time of randomization. The 7 studies assessed 7 different interventions. The Swedish Internet Cognitive Behavior Therapy for Antenatal Depression program had 42 participants, and the intervention, which was delivered via a secure online platform, consisted of reading material, assessments, homework, and worksheets [49]. Videos, graphics, and an online diary book were used by the Chinese Internet-based Mindful Self-Compassion Program with 314 participants [50]. The Mothers and Babies Internet Course had 852 participants, making it the second largest study. Coordinated from the United States, but with a worldwide reach, it provided fully automated online lessons that included information pages, short audio and video clips, images, and worksheets [47]. The American-based Sunnyside Group-based Internet Intervention was an online website with various interactive tools and games that facilitated learning and peer interaction. It was the smallest study, with 25 participants [48]. Mamma Mia was by far the biggest program, with 1342 Norwegian participants. It was a fully web-based intervention that combined text, pictures, pre-recorded audio files, and user input [51]. The Australian MUMentum Pregnancy Program had 87 participants. This featured a virtual online clinic and used an illustrated story centered around fictional characters [52]. The final study was the Chinese smartphone-based Mindfulness Behavioral Cognitive Therapy, with 168 participants, delivered through a smartphone app. It consisted of thematic curriculum provided through text, audio, and visual materials [53].

Two of the studies, from the United States [47] and Norway [51], employed a universal approach, and the other 5 studies used an indicated approach. Three studies recruited participants directly from hospitals: the Chinese study from a reproductive mental health multidisciplinary clinic [50], another Chinese study from the obstetrics clinic of a tertiary hospital [53], and the largest Norwegian study via midwives during regular maternal check-ups [51]. The other 4 studies combined diverse recruitment strategies, using advertisements on social media and online forums and in newspapers, mass emails, Google Ads, and flyers distributed at maternity clinics and hospitals. None of the studies screened participants using population-based approaches. Four studies screened participants during recruitment using different measures and cut-off scores. The Swedish study used a Montgomery-Åsberg Depression Rating Scale (MADRS) [54] score of 5 to 35 and no or a low risk of suicide [49]. The Chinese study used an Edinburgh Postnatal Depression Scale (EPDS) [55] score of ≥9 [50], while the other Chinese study used an EPDS score of ≥9 or Patient Health Questionnaire (PHQ)-9 [56] of ≥4 [53]. The small American study used PHQ-8 scores of 5 to 14, which indicated no diagnosis of major depression [48]. When it came to measuring depression, 6 studies used the EPDS, as either the only measure or one of the outcome measures. However, the EPDS outcome cut-off scores used for probable depression differed: 4 studies used >9 points [47,50,51,53], and 2 studies used >12 points [49,52]. Other studies used the PHQ-9, Center for Epidemiological Studies-Depression [57], MADRS-Self Report, and Beck Depression Inventory II [58] to measure depression outcomes. The studies also assessed various other mother-related outcomes, but only 2 studies reported infant-related or mother-infant attachment outcomes. These were the Chinese study [50], which used the Infant Behavior Questionnaire Revised-Very Short Form [59] and the Australian study [52], which used the Maternal Antenatal Attachment Scale [60]. Five studies [48-50,52,53] also assessed anxiety as secondary outcomes, using the Generalized Anxiety Disorder 7 scale [61], the State-Trait Anxiety Inventory [62], or the Inventory of Depression and Anxiety Symptoms [63].

Table 2 summarizes the description of each intervention, while Figure 2 compares the earliest starting point to the latest end point for each intervention. All the digitalized study interventions were delivered on the internet but varied in terms of their session structure and duration, the type of support provided, and the duration. Six interventions were delivered via a website, and only 1 [53], which was recently published, was delivered via a smartphone app. The shortest intervention, from Australia, lasted for 4 weeks and was comprised of 3 brief unguided internet-based CBT (iCBT) sessions [52], while the largest Norwegian intervention, Mamma Mia, had the longest duration, of 11.5 months, and provided sessions during pregnancy and postnatal periods [51]. Two of the 7 interventions were guided, as they incorporated an element of therapeutic guidance. The Swedish intervention was provided by CBT-trained therapists, who were not mental health professionals, on a fully digital platform [49], but no telephone guidance was provided. The level of therapeutic guidance provided by this guided iCBT varied, and messages sent via the internet platform were responded to within 48 hours during weekdays. The Chinese intervention did not detail the type of guided instructions that were provided [50]. In addition to therapeutic guidance, several of the guided and unguided interventions employed other forms of support, such as peer support discussion groups or platforms, as well as technical support by the research team members. Table 3 shows a brief synopsis and the CBT elements of the 7 interventions.

Figure 3 and Figure 4 display the risk of bias assessment results for each study. Six studies had a high overall risk of bias, as a high risk of bias was recorded in at least one domain, while 1 study recorded some concern for overall risk of bias. Self-reported measures are commonly used to evaluate psychological outcomes, but there were at least some concerns about the risk of bias that these provided in 7 studies, based on the ROB2 guideline [43]. The missing outcome data domain was of some concern, given nearly half (3/7, 43%) of the studies did not provide sufficient information about the reasons for dropout between groups or analyses to address these missing data. In addition, the way that the reported results were selected was of high concern. These were particularly related to the fact that 5 of the study protocols were not registered or there were inconsistencies, without justification, between the registered protocols and the published papers. These inconsistencies included the absence of follow-up results in the published paper or differences between these and the intended analysis plan in the published protocol.

In view of the overall high risk of bias in all except one of the studies, the efficacy of the interventions needs to be interpreted with caution. Five studies provided sufficient data for effect size calculations. Figure 5 summarizes the Hedges g effect size for the 5 studies [48,49,51-53]. In general, the more positive results for the intervention than for the control groups showed promising evidence for the efficacy of digitalized CBT interventions for pregnant women. It was notable that there were differences in efficacy between the guided and unguided interventions. The highest postintervention effect size was recorded for the Swedish Internet Cognitive Behavior Therapy for Antenatal Depression program, which focused on pregnant women who were judged to have no or a low risk of suicide when screened for symptoms of depression [49]. The remaining 4 interventions were all unguided and recorded large [48], moderate [53], or small [51,52] effect sizes.

Table 4 compares the acceptability assessments conducted by individual studies based on 3 time points: before, during, and after the intervention was delivered. Overall, most of the participants were satisfied with the interventions they received, but high attrition rates were recorded for the 4 unguided interventions (22.1%-46.5%). The way that intervention acceptability was measured varied significantly between the studies, especially the scales that were chosen. All the studies obtained feedback after the interventions had finished. Five studies assessed acceptability during the interventions [48,49,51-53], while only 2 studies assessed acceptability before they were used [48,51]. The highest attrition rates for the unguided interventions were for the unguided MUMentum Pregnancy transdiagnostic intervention [52]. In contrast, the guided intervention recorded a very low attrition rate of 4.5% for the Swedish Internet Cognitive Behavior Therapy for Antenatal Depression program [49].

Our review identified 7 RCTs of digitalized CBT interventions for depression symptoms during pregnancy that were promisingly efficacious and had good acceptability. RCTs using digital CBT for depression during pregnancy have been scarce in contrast to the number of studies on digitalized CBT interventions for depression overall. We judged 6 of the 7 studies to have a high risk of bias. The high risk of bias was mainly due to missing outcome data and selection of reported results domains. These point to the need to provide accessible trial registration information with sufficient information prior to commencement of controlled trials. Although missing data in psychological intervention studies are common, a detailed description of how these missing data were addressed and reasons for dropouts in the intervention and control group need to be clarified and compared in each study. Therefore, the findings on efficacy should be considered with caution, and more high-quality studies are needed. Despite these limitations, this review produced 3 main findings. First, technology can be used to deliver CBT programs that target depression during pregnancy. Second, digitalized CBT during pregnancy showed promising evidence of efficacy and positive outcomes for depression (Hedges g for guided interventions=1.21; Hedges g for unguided interventions=0.14-0.99). Third, the digitalized CBT interventions were well-received by pregnant women, especially interventions that were guided, which had lower attrition rates than the unguided interventions (4.5% versus 22.1%-46.5%).

CBT interventions have traditionally been delivered face-to-face, but technological advances have revolutionized the way they are being adapted to various platforms. The COVID-19 pandemic has also led to an urgent increase in the need for digital care delivery. For the past 20 years, the progress in digitalized interventions has been rapid, and various evidence-based digitalized interventions have been developed and disseminated [66], especially for depression [67]. However, little is known about digitalized CBT interventions for depression during pregnancy. All 7 interventions identified in this review used the internet and website or smartphone app platforms to deliver core elements of CBT, such as psychoeducation, cognitive restructuring, problem-solving, and behavioral activation. The central themes of these were managing challenges and expectations during pregnancy by using CBT skills. These platforms contained a combination of text, images, and videos to make the interventions attractive and engaging. Other additional interactive features were unique to some interventions. For example, the American-based Sunnyside Group-based Internet Intervention included an activity feed that displayed each participant’s activity on the site. This meant that other participants could like, comment, or provide feedback on other women’s public posts. It also had an individual garden plot and community garden that were linked to individual profiles. The intervention provided individual and group games, where garden gnomes or flowers could be collected by completing various tasks [48]. Likewise, the Swedish Internet Cognitive Behavior Therapy for Antenatal Depression program provided an online discussion group, in which participants could interact with each other anonymously [49]. These unique interactive features allowed participants to share their lived experiences during pregnancy, as well as provided a platform for emotional peer support. These are both considered to be key functions of traditional online pregnancy forums [68].

The limited evidence and high risk of bias in 6 out of the 7 papers we reviewed prevented us from coming to firm conclusions on the efficacy of digitalized CBT for depression during pregnancy. However, our review does provide promising evidence on the efficacy and positive impact of digitalized CBT for pregnant women. Previous research has reported that the prevalence of postnatal depression symptoms can be predicted by depression symptoms during pregnancy [3]. Interventions during pregnancy allow participants to acquire and apply coping skills at an early stage, potentially before their depressive symptoms worsen. It should also be noted that 5 of the included studies used standard care, an active control, or another intervention as control groups, which may partly explain the modest effect sizes in most of the studies. The Swedish intervention [49] had an exceptionally high effect size. Unsurprisingly, that intervention used a guided and indicated approach. The intervention was delivered via an online internet platform comprised of about 75,000 words of reading material, assessments, homework, case stories, and worksheets. The therapists provided regular individual feedback, encouragement, and support in written messages, with replies within 48 hours on weekdays. Interestingly, the therapeutic guidance was provided by CBT-trained therapists who were not mental health professionals. Compelling evidence from systematic reviews and meta-analyses of iCBT research for depression has highlighted the importance of providing support from a therapist or human guidance for effective interventions [69-72].

Six out of 7 studies we reviewed also assessed other self-reported maternal outcomes: anxiety, psychological distress, alcohol and drug use, mindfulness attention awareness, general health, stress, self-compassion, well-being, quality of life and work and social adjustment, positive and negative affect, sleep-related problems, fatigue, prospective memory, retrospective memory, and fear of childbirth. Only one study assessed infant outcomes, using the Infant Behavior Questionnaire, and another study assessed mother-child attachment or bonding using the Maternal Antenatal Attachment Scale. This was somewhat surprising, as there is compelling evidence that prenatal depression has been associated with various infant-related issues, including temperament, bonding, emotional and behavioral problems, cognitive impairment, and psychopathology [73-79]. The fact that there was so little research on infant-related or mother-infant attachment outcomes by the studies we reviewed limits any insightful discussion and highlights the need for further evaluation.

All the studies we reviewed reported diverse acceptability outcomes for at least 1 of the 3 time points: before, during, and after the intervention was delivered. All the interventions recorded good acceptability, especially after the intervention had been completed, as the participants were satisfied with the interventions and felt that they were useful. Acceptability data before and during intervention delivery are still lacking, and more acceptability components can be explored further. The Theoretical Framework of Acceptability outlines 7 components of acceptability that could be explored in intervention studies at these 3 time points: affective attitude, burden, perceived effectiveness, ethicality, intervention coherence, opportunity costs, and self-efficacy [46]. Including all the acceptability data in a single paper may be challenging due to word count limits or information overload. To deal with this issue, authors could provide acceptability evaluations in separate papers. For example, the Mamma Mia intervention included in our review was the focus of 3 separate papers on intervention development [64], acceptability and feasibility [65], and efficacy [51].

The attrition rates for the unguided interventions were high (22.1%-46.5%), and this finding was consistent with previous meta-analyses on digitalized CBT interventions for depression [70,72]. The terms “universal” and “indicated” were based on the Institute of Medicine’s spectrum of mental health interventions based on the risk for the target population: prevention (universal, selective, indicated), treatment (case identification, standard treatment for known disorders), and maintenance (compliance with long-term treatment, aftercare) [22]. Our review identified 2 “universal” interventions (developed for pregnant women with any risk level for diagnosable clinical depression) and 5 “indicated” interventions (developed for pregnant women already at risk for diagnosable clinical depression [eg, those that exceed a cut-off screening score]). These classifications (ie, universal, indicated) have also been used in other perinatal distress, depression, and anxiety research [80].

Guided versus unguided and universal versus indicated approaches may have their own merits when it comes to impact. For instance, the study on the unguided and universal Mamma Mia intervention recruited the highest number of participants, 1342, with 678 in the intervention group and 664 receiving standard care. Although this study showed small effect sizes, partly due to its universal approach, more than one-half (345/678, 50.9%) of the total participants in the intervention group completed 80% or more of the year-long program, with the first phase during pregnancy and the second and third phases during the postnatal period. Any approach requires consideration based on the readiness of pregnant women to change their health behavior, particularly with reference to the Capability, Opportunity, Motivation, Behavior (COM-B) framework [81]. In general, pregnancy has been called a “teachable moment,” which is described by Phelan [82] as naturally occurring life transitions or health events that are thought to motivate individuals to spontaneously adopt risk-reducing health behaviors. Although pregnancy is a time when women may benefit from increased motivation for behavioral change, based on the COM-B framework, such changes can be difficult to realize if their capabilities and opportunities are neglected [83,84]. As discussed, earlier digitalized CBT may provide the opportunity for pregnant women to access digitalized interventions. However, their lack of capacity, namely reduced energy, may prevent the women from continuously engaging with the program. That is why it is crucial for those who develop interventions to consider the 3 elements of capacity, opportunity, and motivation together, in order to minimize attrition.

Pregnancy is a relatively short period, and shorter interventions could attract pregnant women to engage. CBT is problem focused and based on the way that people think and behave, and technology can make its delivery much more cost-effective. This was proven by the Australian MUMentum Pregnancy Program included in our review [52]. The transdiagnostic intervention targeted both depression and anxiety, by using a story that centered around 2 fictional characters experiencing depression and anxiety during pregnancy. The characters learned to use CBT skills to manage their symptoms, based on their experiences and challenges during pregnancy. The program received highly positive feedback from participants, in terms of satisfaction, quality, and how they became immersed in the storyline. The small effect size recorded by this unguided intervention should not mask its potential benefits. It should be emphasized that, during the study period, participants in the treatment-as-usual control group continued to receive existing health care services. This might have had a confounding impact on their symptom levels, compared with the MUMentum Pregnancy Program intervention group [52]. Virtually all pregnant women have regular contact with health care professionals, regardless of their psychosocial well-being, and that may also have a positive effect on any depressive symptoms. This is consistent with findings of a recent individual network meta-analysis that the effectiveness of guided iCBT interventions were higher for individuals with moderate to severe depression, but unguided iCBT showed similar effectiveness for individuals with mild or subthreshold depression [72].

Although this review was conducted with high scientific rigor and we carried out a full quality appraisal of the included studies, there are several limitations that need to be taken into account when interpreting the findings. We only included papers published in English, and this may have meant that some potential studies in other languages or grey literature from other databases were not included. All except one of the included studies had a high risk of bias. They also used participant-reported measures for most of the outcomes, which can be influenced by knowledge of the intervention being received. Using the ROB2, this resulted in at least some concern of a risk of bias for all the studies. As only 1 study reported an infant-related outcome, namely infant temperament, and only 1 study assessed mother-infant attachment, meaningful discussions about infant outcomes and maternal attachment are not possible. In addition, our review only focused on CBT and did not include interpersonal psychotherapy, although both psychotherapies were included in the US Preventive Services Task Force treatment guideline for perinatal depression that was issued in 2019 [11]. A recent review found that interpersonal psychotherapy interventions for psychological distress in perinatal populations were limited to those delivered face-to-face or by telephone [85]. The nonstandardized description of interventions in our review also made it difficult to compare them. This calls for more standardized descriptions of interventions, such as using the Template for Intervention Description and Replication checklist [86]. Future studies could also explore various acceptability outcomes before, during, and after interventions are delivered. This could be done by using the 7 components of the Theoretical Framework of Acceptability: affective attitude, burden, perceived effectiveness, ethicality, intervention coherence, opportunity costs, and self-efficacy [46].

This systematic review suggests promising evidence for the potential efficacy, acceptability, and integration of digitalized CBT interventions that start during pregnancy. However, all except one of the studies included in this review recorded a high overall risk of bias. In future, we need high-quality studies with larger population-based samples to comprehensively analyze the efficacy of the interventions and explore the mechanisms of change. Digital interventions may have significant global implications when planning effective, nonstigmatizing, and cost-effective mental health treatment to prevent the long-term consequences of psychosocial problems in pregnancy. These could also have other wide-ranging clinical applications. Furthermore, there is an urgent need to study digital and remote interventions, as mental health services struggle to deal with the considerable increase in demand due to the COVID-19 pandemic.