This study was a single-center, two-parallel-armed, assessor-blinded, 1:1-allocated RCT with a 32-week follow up. Pregnant women who scored at or above the threshold for positive depressive symptoms on the Edinburgh Postnatal Depression Scale (EPDS) or Patient Health Questionnaire-9 (PHQ-9) were randomly assigned to the mindfulness intervention group (receiving 8-week smartphone-based mindfulness training) or attention control group (receiving 8-week regular WeChat health consultations). This study was approved by the ethical review board of the School of Nursing, Shandong University (2018-R-015). The trial was registered in the Chinese Clinical Trial Registry (ChiCTR1900028521) in December 2019 and no significant changes were made between the start of the trial and the registration confirmation.

The study was conducted between March 2018 and January 2020, and participants were recruited in the obstetrics clinic of a tertiary hospital in Jinan, Shandong, a city located in the east of China. The hospital provides perinatal services for around 5000 pregnant women each year.

Inclusion criteria for women to participate in the study included: (1) aged 18 years and over, (2) in the 12th to 20th week of gestation, (3) singleton pregnancy, (4) no plan to terminate pregnancy, (5) planned to receive antenatal examination and deliver in the study hospital, (6) completed junior high school education or above, (7) positive in depressive symptoms screening with an EPDS score>9 or a PHQ-9 score>4, (8) able to use the app on a smartphone for the study, and (9) able to understand and respond to the questionnaire.

Exclusion criteria were: (1) at risk of suicide or self-harm, (2) currently receiving psychiatric treatment or using psychiatric medications, (3) history of substance abuse or addiction in the past 6 months, (4) prior experience with mindfulness meditation, and (5) declined to participate.

According to the national health policy in China, health status for all pregnant women is recorded in the 12th gestational week and they start prenatal visits regularly thereafter. Pregnant women were recruited from the records of the tertiary hospital where the study was based, during the time they are ordinarily required to report pregnancy-related information and receive preventive depression screening at their first regular visit in the obstetrics clinic. Informed consent for preventive psychological assessment was first obtained, and once consent was granted, printed questionnaires, including sociodemographics, pregnancy-related characteristics, and mental health indicators, were distributed for participants to complete and return. All participant records were screened according to the inclusion and exclusion criteria, and eligible participants were then contacted by a research assistant through telephone or WeChat (a popular instant social communicating software in China) within 2 weeks. Online informed consent for participation in the RCT was obtained through the online survey platform Wjx.cn. The RCT program was introduced to potential participants while discussing results of their preventive psychological assessment. This initial baseline evaluation of symptoms was classified as T1 for consented trial participants. The procedure and objectives of the trial were explained, and then online informed consent was obtained if they agreed to participate.

Recruited participants were randomly assigned to the mindfulness training during pregnancy or attention control group; the mindfulness group received 8-week mindfulness training and the control group received 8-week regular WeChat health consultation to control attention. Participants who were allocated to the mindfulness training group received the URL of the app through WeChat. They could use the app as long as they signed up. Data collection and assessment of outcomes took place over four time points in the follow-up period. T2 assessment took place at 4 weeks after allocation (intermediate period of intervention), T3 assessment took place at 8 weeks after allocation (endpoint of the intervention), T4 took place at 18 weeks after allocation (before childbirth), and T5 took place at 6 weeks after delivery. Follow-up assessments were collected by computer/smartphone-assisted self-administered surveys. All participants were awarded 2 yuan (US $0.30) when completing an assessment.

Depression symptoms as assessed by the EPDS was the primary outcome in this study. The EPDS is a self-report scale that assesses depressive symptoms experienced within the last week during both prenatal and postnatal periods [34]. The EPDS contains 10 items with responses on a 4-point Likert scale ranging from 0 to 3, where higher scores represent greater intensity of a depressive symptom. The EPDS was recommended as a valid depression screening tool across different cultures and different trimesters in a validation study review [35], and the cut-off score of 9/10 was used to identify positive depressive symptoms for screening purposes in this study. Additionally, another commonly used depression screening tool, the PHQ-9 developed on the basis of Diagnostic and Statistical Manual of Mental Disorders-IV criteria, was also used in the screening period with a cutoff of 4/5 indicative of positive depression criteria [36].

Secondary outcomes consisted of multidimensional health issues for perinatal women, including anxiety symptoms, perceived stress, positive affect, negative affect, sleep-related problems, fatigue, prospective memory, retrospective memory, and fear of childbirth. Anxiety in the previous 2 weeks was evaluated by the 7-item Generalized Anxiety Disorder scale, a clinically useful assessment for detection of symptoms of anxiety in the perinatal period [37]. We assessed perceived stress of pregnant women using the 4-item validated Perceived Stress Scale to measure the degree in the past month that their situation was appraised as uncontrollable, unpredictable, and overwhelming [38]. The Positive and Negative Affect Schedule [39] was used to measure individuals’ agreement and endorsement on statements related to positive and negative affect. Sleep-related problems in the past month were evaluated by the self-administered Pittsburgh Sleep Quality Index questionnaire [40] and the degree of fatigue in the past week was evaluated by the 9-item Fatigue Severity Scale [41]. Subjective prospective and retrospective memory failures in daily life were self-rated by participants using the Prospective and Retrospective Memory Questionnaire [42]. In addition, the Wijma Delivery Expectancy Questionnaire [43] with items related to women’s cognitive appraisal regarding the delivery process was used to measure the level of fear of childbirth. For all measurements other than positive affect, higher scores represent worse mental health outcomes, and for positive affect, higher scores represent a higher positive affect. Time points for the outcome indicators assessed are shown in Multimedia Appendix 2.

Sociodemographic characteristics such as age, gestational age, BMI, education level, work status, marital status, family economic status, and pregnancy-related characteristics were self-reported by the participants at baseline (T1). Study researchers were allowed to collect additional clinical data on participants from medical records following birth.

Previous meta-analyses indicated that mindfulness practice during pregnancy reduced depression scores with an effect size of 0.59 [44] and self-help mindfulness resulted in small to medium effect sizes on anxiety/depression [45]. On the basis of a medium intervention effect (Cohen d=0.50), an estimated sample size of 128 was required to compare between-group differences on depression at postintervention with 80% statistical power, a 1:1 allocation rate, and a two-tailed significance level of .05. In the above-mentioned meta-analysis of self-help mindfulness [45], on average, 27% of the participants were lost to follow up in the postintervention assessment. Considering a 30% attrition rate, a final sample of 168 with 84 individuals in each group was required. The sample size was calculated using G*Power [46].

The trial used a simple randomization approach. The random number sequence was generated by a researcher who did not participate in this study using the Random Number Generators function in SPSS 23.0. The sequence was kept in sealed, opaque, numbered envelopes. When each participant was enrolled, a research assistant opened the envelope in sequence and assigned the participant to the group. Participants received allocation according to the order of enrollment. Throughout the RCT procedure, the assessor who collected follow-up data did not know the assignment. Considering the use of the smartphone app only in the mindfulness intervention, the participants were able to infer the assignment.

Logs of practice on formal mindfulness training were recorded and used to evaluate the fidelity of mindfulness training. At least 3 days of practice per week was considered a completed training week, and at least 4/8 completed training weeks were considered as completion of the mindfulness training program. The completion rate in this study was calculated as the percentage of participants who completed the mindfulness training divided by the number of participants who received the intervention.

Statistical analyses were performed with SPSS 23.0. The primary analysis used an intention-to-treat (ITT) approach. Normality of outcomes at baseline was visually examined, and negative affect, sleep, and positive affect were log-transformed due to non-normal data distributions. No more than 5 individuals for each measure at T1 had a single entry missing, which were considered as missing at random and were imputed with the median of the relevant item. Independent sample t tests and χ² tests were used to compare the baseline characteristics between the intervention and control groups, and between the study sample and dropout sample. To assess the population-averaged mindfulness training vs attention control intervention effect on outcomes, generalized estimating equations (GEEs) were formulated. This approach has been recommended because it is able to handle missing data appropriately and remains stable in different correlation matrices [47]. For GEEs, the participants and assessment time points were designated as subject variables and within-subject variables, respectively, with an exchangeable working correlation matrix and full maximum-likelihood estimation applied. The continuous outcome variables at five time points were the dependent variables. Main effects of group, time, and group × time interaction effects were examined. Testing of a simple main effect for group was also explored by examining differences between groups at each time point. To further assess the intervention effect on depression symptoms remission postintervention (T3), binary positive depressive symptoms at T3 and EPDS reductions from T1 to T3 were compared between mindfulness and attention control groups using logistic regression models, adjusting for the EPDS score at baseline and between-group imbalanced factor (intended pregnancy) after randomization. In addition, subgroup analysis was also conducted to explore the impact of parity (primipara/multipara) on the intervention effect.

Effect sizes are presented as Cohen d based on the ITT rule. Cohen d was calculated between groups and follow ups referring to baseline data. Effect sizes were considered to be small (d=0.2), medium (d=0.5), and large (d=0.8). A two-sided P value less than .05 in the primary analysis and less than .025 in the subgroup analysis after Bonferroni multiple-comparison correction were considered statistically significant.

As suggested by Thabane [48], several sensitivity analyses were performed. First, the mindfulness training vs attention control intervention effect was analyzed according to per-protocol group and as-treated group comparisons in addition to the ITT group. Second, adjusted GEE models with a baseline imbalanced factor (intended pregnancy) were also established. Third, intervention effects were also evaluated in participants who completed different numbers of follow-up assessments. Details of the statistical analysis are provided in Multimedia Appendix 3.

Recruitment began in March 2018 and ended in June 2019, and follow-up assessments ended in January 2020. Figure 2 illustrates the full study flow diagram. A total of 1140 pregnant women were contacted and a final sample of 168 was allocated. During the whole follow-up period for the 168 participants, 24/84 (29%) participants dropped out in the mindfulness training group and 34/84 (41%) dropped out in the attention control group with the overall dropout rate reaching 34.5%. More than half of the participants completed at least 3 follow ups (94/168, 55.6%) and 56/168 (33.3%) of participants completed all follow ups. A logistic regression model with binary dropout status as the dependent variable showed that participants with an advanced gestational age at baseline tended to drop out more frequently during follow up (odds ratio [OR] 1.033, 95% CI 1.004-1.063; see Multimedia Appendix 4).

Table 2 shows the details of participant characteristics. Overall, all of the pregnant women were married, with an average age of around 30 years, and were at 99 gestational days (around 14 gestational weeks) at baseline. A total of 41.1% (69/168) of the participants screened positive in the EPDS and 92.3% (155/168) screened positive for depression symptoms in the PHQ-9. The mean gestational week was 16.8 (SD 1.068) at randomization. There was no significant difference between the mindfulness training and attention control groups at baseline except for intended pregnancy. More pregnant women planned this pregnancy in the mindfulness training group than in the attention control group (χ²₁=8.4, P=.004).

Of the 84 participants allocated to the mindfulness training group, 10 of them did not activate the app, which was considered refusing the allocation. All of the remaining participants received the mindfulness training program. On the basis of the ITT sample, the mean number of completed training weeks was 3 weeks (SD 2.701). As a whole, 44/84 participants completed at least 4 weeks of training and the total completion rate was 52.4%. In all, 7/84 (8%) participants completed the entire 8-week training program.

For the binary EPDS variable at T3, fewer participants reported positive depressive symptoms (EPDS>9) in the mindfulness training group than in the attention control group (15/63, 24% vs 24/54, 44%; χ²₁=5.6, P=.02). Results of the logistic regression model indicated that mindfulness training led to a 0.609-times reduction on the risk of positive antenatal depressive symptoms compared to the attention control (Table 5) at postintervention. For EPDS reduction from T1 to T3, stepwise EPDS decreased scores ranging from 1 to 9 points were compared between groups; OR values remained statistically significant with EPDS score reduction>2. Mindfulness training showed a significant effect on EPDS score reduction over the attention control with the OR ranging from 3.471 to 27.986 (Table 5), indicating that mindfulness training had more potential to relieve depression symptoms than attention control.

The mindfulness training vs attention control intervention effects on EPDS scores differed between a primipara and multipara state. In nulliparous women, the group × time interaction effect was statistically significant and the between-group mean difference reached 3.72 points on the EPDS at T3 (Table 6). However, no significant intervention effect was found in multiparous women (Table 6). Specifically, the EPDS score decreased at T3 in the mindfulness training group but increased in the attention control group in primipara, whereas it increased in both groups at T3 in multipara (Table 6 and Figure 4). However, the three-way interaction test (group × time × parity) did not reach statistical significance in the GEE model (χ²₄=5.6, P=.24; see Multimedia Appendix 6). Results on subgroup analysis of anxiety and positive affect were consistent with those obtained for EPDS (see Multimedia Appendix 7).

Sensitivity analysis in adjusted GEE models supported the primary results (Multimedia Appendix 8). In sensitivity analysis related to the number of follow ups (Multimedia Appendix 9), results from those who completed at least 2 or 3 follow ups showed a significant group × time interaction effect on the EPDS as well as in the primary analysis. However, mindfulness training vs attention control intervention effect trends were only found among participants who completed all follow ups. For anxiety, between-group mean differences at T3 reached statistical significance in participants who completed different numbers of follow ups in addition to the significant mindfulness training vs attention control intervention effect. Results from participants who completed at least 2 follow ups and all follow ups also supported the primary results on positive affect.