Digital clinical system for the identification, risk stratification, and longitudinal monitoring of women at risk of postpartum depression

Postpartum depression (PPD) is a clinically significant psychiatric disorder that develops in the first weeks or months after childbirth. It is characterized by a persistent depressed mood, anhedonia, marked anxiety, sleep disturbances, cognitive impairment, and, in some cases, suicidal ideation. [1]. According to the World Health Organization (https://www.who.int/ru), PPD affects 10–20% of women after childbirth, with a prevalence of 25–30% among socially vulnerable populations at particularly high risk [2]. In the Russian Federation, the estimated prevalence ranges from 11% to 17%, while some studies have reported that up to 45% of postpartum women achieve screening scores indicative of clinically significant depression [3]. A major challenge is the absence of systematic national programs for the early detection of PPD, resulting in extremely low rates of formal diagnosis, with no more than 3% of cases being identified.

The clinical importance of PPD extends beyond its high prevalence and profound consequences for the mother, infant, and family [4]. PPD is associated with impaired mother–infant attachment, delayed neurodevelopment in children, an increased risk of chronic maternal depression, and adverse effects on partner relationships [5]. Nevertheless, many affected women do not receive timely diagnosis or treatment because of social stigma, limited awareness, and the inherent limitations of the currently available screening strategies.

The Edinburgh Postnatal Depression Scale (EPDS), introduced in 1987, remains the standard instrument for the initial PPD screening. Although the EPDS demonstrates high specificity (83–95%), its sensitivity following a single assessment ranges from 59% to 80%, implying that up to half of PPD cases may remain undetected during routine screening [6]. Furthermore, in standard clinical practice, the EPDS is typically administered only once or twice during the entire postpartum period, despite the fact that clinical manifestations of PPD may emerge at varying time points, frequently more than six weeks after discharge from the maternity hospital.

Importantly, the EPDS is a screening rather than a diagnostic tool. An elevated score indicates the need for a comprehensive clinical assessment, whereas the definitive diagnosis of PPD must be established by a qualified clinician according to the diagnostic criteria of the International Classification of Diseases, 11th Revision (ICD-11).

Another major limitation is the lack of tools for the longitudinal monitoring of patients after childbirth. The clinical course of PPD is inherently dynamic, with symptoms that may worsen, remit, or recur over time. Early recognition of clinical deterioration is essential for the timely adjustment of therapeutic interventions [3]. Consequently, the episodic nature of existing screening approaches precludes the real-time assessment of symptom trajectories.

In recent years, digital health technologies, including mobile applications, telemedicine platforms, and messenger-based chatbots, have demonstrated considerable potential for monitoring mental health [7]. Several studies have shown that patients are willing to engage with digital platforms for psychological follow-up, particularly when barriers to face-to-face care exist [8]. Nevertheless, comprehensive systems that integrate screening, risk stratification, and longitudinal monitoring within a single clinically integrated platform remain scarce. Evidence regarding such integrated clinical systems is limited, and comparable solutions have been sparsely described in Russian clinical practice.

Accordingly, there is a compelling need to develop an integrated digital platform capable of supporting both early identification and continuous monitoring of women at risk of postpartum depression throughout the postpartum period.

The objective of the present study was to develop and conduct a pilot clinical evaluation of a digital system for the early screening and longitudinal monitoring of postpartum depression risk among postpartum women.

The specific objectives were as follows: (1) to design and implement an integrated digital platform comprising a physician web interface, a patient web portal, a messenger-based chatbot, a database, and a scheduling and notification system; (2) to implement early in-hospital screening on postpartum days 2–5 using the Edinburgh Postnatal Depression Scale (EPDS) and the Perinatal Anxiety Screening Scale–Revised (PASS-R); (3) to establish longitudinal follow-up through digital communication channels; and (4) to evaluate participant coverage, adherence, and the organizational feasibility of the system in a pilot cohort.

Materials and methods

A pilot phase for developing a digital clinical system was conducted as part of a prospective observational study. This study enrolled all postpartum women who met the eligibility criteria, regardless of their baseline depression status. The project was supported by a state assignment to develop a predictive test system for identifying individuals at high risk of postpartum depression (PPD). The study was reviewed and approved by the Research Ethics Committee of V.I. Kulakov National Medical Research Center for Obstetrics, Gynecology, and Perinatology (protocol no. 2, February 19, 2026).

Postpartum women receiving care in the Second Physiological Obstetric Department of the Center were consecutively enrolled after providing written informed consent. All consecutive postpartum women were included, irrespective of the presence of PPD symptoms at the time of enrollment.

The inclusion criteria were as follows: age ≥ 18 years and delivery at V.I. Kulakov National Medical Research Center for Obstetrics, Gynecology, and Perinatology of the Ministry of Health of the Russian Federation; provision of written informed consent; Russian-language proficiency; and access to a smartphone or computer for the digital pathway or willingness to participate using a paper-based and telephone format.

The exclusion criteria were mental disorders requiring continuous psychotropic medication before pregnancy and refusal to participate in the study.

The planned sample size was 2,000 participants (pregnant and postpartum women), calculated based on an expected PPD prevalence of 15%, a 95% confidence interval, a precision of 2%, and an anticipated dropout rate of 20%. The pilot version of the project included 237 postpartum women.

A consecutive sampling approach was used, including all postpartum women who met the eligibility criteria without randomization or selection based on additional characteristics.

Digital clinical system: architecture and implementation

The system was developed for automated screening, longitudinal monitoring, PPD risk stratification, clinical decision support, and remote patient interactions. System requirements were formulated based on an analysis of clinical protocols, including the Edinburgh Postnatal Depression Scale (EPDS) and the Perinatal Anxiety Screening Scale–Revised (PASS-R), and expert interviews with obstetrician-gynecologists, clinical psychologists, and a psychotherapist at the Center.

The system was implemented using a distributed, three-tier architecture (Table 1). The client tier comprises a clinician web interface, patient web portal, and messenger bot. Communication with patients occurs through a secure HTTPS channel with authentication. The data tier consists of a database. User access is managed through authentication via email and/or messaging.

The system's clinical assessment toolkit includes four questionnaires: the Edinburgh Postnatal Depression Scale (EPDS; 10 items; threshold ≥10), the Perinatal Anxiety Screening Scale–Revised (PASS-R; 31 items), a baseline risk-factor questionnaire (22 questions covering medical history and sociodemographic characteristics), and a week-6 questionnaire (15 questions). The EPDS is the primary screening instrument for PPD, consisting of 10 items rated on a four-point response scale. The total score ranges from 0 to 30 and correlates with the likelihood of depressive disorder; a score of 10 or higher is conventionally used as a threshold for further clinical assessment [6]. Within the system, the EPDS was administered during both initial screening and scheduled follow-up assessments.

The PASS-R is used to identify perinatal anxiety disorders, distinguish anxiety from depression, and assess comorbidity. Its inclusion was justified by the high comorbidity of anxiety and depression during the postpartum period, which was reported to reach 50% in some studies.

The baseline maternal questionnaire and the repeat questionnaire at week 6 were structured questionnaires developed specifically for this project. They cover medical history, sociodemographic characteristics, and clinical risk factors. An investigator-developed questionnaire was designed to collect standard clinical and demographic data. Its psychometric properties were not assessed because of the pilot nature of the study; however, all items were consistent with generally accepted recommendations for history-taking in perinatal care. Each questionnaire includes questions on parity, mode of delivery, support from the partner and immediate social environment, psychiatric history, sleep patterns, and infant feeding. Questionnaire data were used by the stratification algorithm as adjustment factors to refine the risk estimates obtained from the EPDS.

The clinician and system administrator functions allow the addition of new questionnaire versions and the selection of an arbitrary scheduling regimen. They also enable manual entry of questionnaire data, barcodes, and other patient information when necessary.

The system supports the management of patients across two clinical profiles: postpartum and prenatal. This enables the coverage of the entire perinatal period, from pregnancy through the completion of postpartum follow-up, within a single platform.

All questionnaires were version-controlled, and responses could not be modified after submission. Questionnaire administration was automated based on protocol templates for postpartum and prenatal profiles.

Algorithms

The system automatically generates notifications according to predefined threshold rules (EPDS >10, "attention"; EPDS >12 on two consecutive occasions, "alert"; PASS-R >41). At the current stage, final risk stratification, classifying patients as being at low, moderate, or high risk, is performed by a physician based on these notifications and aggregated EPDS, PASS-R, and questionnaire data. Fully automated risk stratification will be performed in the next stage.

Security

Data transmitted through the system were encrypted using Transport Layer Security (TLS) 1.3. Personal data, including full name and contact information, and clinical information were stored separately, with clinical data maintained in a separate collection using a pseudonymized key. An audit trail recorded the IP address, time, and action. Access to personal data within the system is restricted to users assigned to the physician or system administrator roles. The system was designed with provisions for role-based access control, data pseudonymization, and user activity auditing.

The described mechanisms, including TLS 1.3, separate storage of personal and clinical data, pseudonymization, and access auditing, have been implemented at the prototype level in the system. The formal assessment of compliance with Federal Law No. 152-FZ, medical confidentiality requirements, and the institution's local regulations were not undertaken during the pilot phase and are planned for the production implementation stage.

Development and testing stages

Closed beta testing was conducted with five healthcare professionals and 20 patients in the study. Implementation involved deployment on a secure cloud infrastructure, connection of the domain www.stop-prd.ru, and staff training.

Statistical analysis

Statistical analyses and reporting were conducted in accordance with the Statistical Analyses and Methods in the Published Literature (SAMPL) guidelines. Categorical variables are presented as absolute and relative frequencies (n/N (%)). Continuous variables were assessed for normality using the Shapiro–Wilk test. Normally distributed variables are presented as the mean (M) and standard deviation (SD), in the format M (SD), whereas non-normally distributed variables are presented as the median (Me) and quartiles, in the format Me (Q1; Q3).

Adherence to monitoring was compared between messenger-bot and web-portal users as a binary categorical outcome using the χ² test or Fisher's exact test when expected cell frequencies were small. The absolute difference in proportions with a 95% confidence interval (CI) was calculated as the effect measure. Parametric methods requiring assumptions of normality and homogeneity of variance between comparison groups, including Student's t-test and analysis of variance, were not used.

The operational (diagnostic) characteristics of the notification algorithm – sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) – were calculated to assess the ability of the threshold rule of two EPDS scores >12 to identify clinically significant PPD. These measures were calculated in the entire cohort of 237 women who completed the questionnaires by constructing a 2×2 contingency table in which algorithm activation was compared with the reference clinical diagnosis of PPD. The reference standard was a psychotherapist's assessment according to the International Classification of Diseases, 11th Revision (ICD-11) criteria. The reference diagnosis was established not only among women for whom the algorithm was triggered but also through a review of the medical records of the entire cohort to identify potentially false-negative cases. Nevertheless, given the pilot observational design and the absence of universal clinical interviews for all participants, irrespective of EPDS scores, the estimated sensitivity and specificity should be regarded as preliminary and require confirmation in subsequent prospective studies with complete reference-standard verification.

Results

The system has been deployed for pilot operation and functions in real time on the platform www.stop-prd.ru. Clinical testing is being conducted within the framework of a state assignment, which provides institutional support and methodological oversight of the implementation.

Coverage and sample formation

A total of 280 postpartum women were initially invited to participate in the pilot study. Of these, 38/280 (13.6%) declined participation at the first stage. The remaining 242/280 (86.4%) women provided informed consent. Among the consenting participants, 54/242 (22.3%) preferred paper-based questionnaires with subsequent telephone follow-up (due to the absence of a smartphone or unwillingness to use digital channels), whereas 188/242 (77.7%) opted for the digital platform (messenger bot or web-based personal account). All 188/188 (100%) digital platform users successfully registered. Reasons for declining digital interaction among the 54 patients who chose the paper-based format were: absence of a smartphone in 1/54 (1.9%), technical difficulties in 6/54 (11.1%), and refusal of digital interaction in 47/54 (87.0%).

The total number of completed questionnaires (digital and paper-based) was 1186, with a mean of 4.6 per active patient (range 1–8). The mean time to complete the Edinburgh Postnatal Depression Scale (EPDS) was 2.8 (0.9) minutes, and the Perinatal Anxiety Screening Scale–Russian version (PASS-R) required 3.5 (1.2) minutes (for the digital format).

Of the 242 consenting women, 237/242 (98.0%) completed the first assessment (2–5 days postpartum); 5/242 (2.0%) did not complete the questionnaires due to technical or personal reasons. At the follow-up assessment (week 6), 42/237 (17.7%) of the 237 women who had completed the first assessment declined further participation. Thus, follow-up data were obtained for 195/237 (82.3%) women. Notably, all 54/54 (100%) patients who chose the paper-based telephone format remained adherent owing to active telephone support, whereas among digital platform users, the attrition rate was 42/188 (22.3%).

EPDS and PASS-R trajectories

When calculating the distribution of anxiety and depression levels at the follow-up assessment, the prevalence of clinically significant postpartum depression (PPD) (EPDS ≥13) at 6 weeks was 22/195 (11.3%), and the prevalence of severe anxiety (PASS-R ≥42) was 15/195 (7.7%). The results from both assessments are presented in Table 2.

The proportion of women with high depression levels increased from 5.9% (early postpartum days) to 11.3% (week 6), reflecting the typical delayed manifestation of PPD. The proportion of severe anxiety also increased, from 2.5% to 7.7%.

Monitoring adherence

Adherence to monitoring (completion of the assigned questionnaire within 24 hours after a reminder) was assessed among digital platform users (n=188). Among messenger bot users, adherence was 140/169 (82.8%), and among web-based personal account users, 13/19 (68.4%); the difference in proportions was 14% (95% CI -3 to 37%). Fisher’s exact test indicated that the difference was not statistically significant (p=0.13). Given the small size of the web-based personal account subsample (n=19), this estimate has low precision (wide CI for the difference in proportions); therefore, the observed difference should be interpreted with caution – a limitation of this pilot phase. Patients who chose the paper-based telephone format (n=54) demonstrated 100% adherence. However, this figure cannot be directly compared with the digital channels, because the paper-based telephone group received active, individualized telephone support, which constitutes an independent intervention.

Clinical and demographic characteristics of the cohort

The pilot study included 237 postpartum women who completed the initial assessment. The median age of participants was 31 years (27; 35), which corresponds to the typical age of primiparous women in the Russian Federation. Primiparous women accounted for 51.9% (123/237) and multiparous women for 48.1% (114/237). Preterm birth (<37 weeks of gestation) was recorded in 22/237 (9.3%) patients, which does not exceed general population rates. Delivery by cesarean section was performed in 73/237 (30.8%) women, of which 45/237 (19.0%) were elective and 28/237 (11.8%) were emergency procedures. A higher education degree was held by 161/237 (67.9%) participants, and 180/237 (75.9%) women were in a registered marriage. These characteristics reflect the socio-demographic and clinical structure of the examined cohort and describe the sample of this pilot study. Among the 237 women who completed the first assessment, the most prevalent risk factors were: postpartum sleep disturbance in 137/237 (57.8%), low social support (questionnaire scale score ≤3 out of 5) in 81/237 (34.2%), breastfeeding difficulties in 64/237 (27.0%), a history of a depressive episode (outside pregnancy) in 52/237 (21.9%), and complicated pregnancy (preeclampsia, threatened miscarriage, gestational diabetes mellitus) in 73/237 (30.8%). The presence of two or more risk factors was noted in 97/237 (40.9%) postpartum women.

Algorithm triggering and operational characteristics

During the operational period (March 10 to May 10, 2026), the system (incorporating data from digital and paper-based telephone monitoring) generated 15 alerts for a sustained elevation in EPDS score (>12 points on two consecutive occasions, 7 days apart) (Table 3). In all cases, a clinical psychologist conducted an unscheduled consultation, followed by follow-up with a psychotherapist at the Center. A clinical diagnosis of PPD (according to ICD-11 criteria) was confirmed in 12/15 patients (80%). Of the remaining three, two were diagnosed with an anxiety disorder (PASS-R >41) and one with an adjustment disorder.

Subclinical EPDS elevations (10–12 points, single occasion) were recorded in six patients. The system sent a notification to the physician flagged as "attention," after which a preventive conversation was conducted and recommendations on sleep and routine correction were provided. None of these six women developed clinically significant PPD during the 6-week follow-up period. In an observational pilot design without a control group, this can be regarded as a potential preventive role of timely routing but not as a proven causal relationship.

A retrospective analysis of data from the 237 women who completed the first assessment showed the following: the sensitivity of the algorithm (EPDS threshold >12 on two occasions) for detecting clinically significant PPD was 80% (12 of 15 true positives), specificity reached 98.6% (219 of 222 true negatives), the positive predictive value was 80%, and the negative predictive value was 98.6% (219/222). These metrics were calculated for the entire cohort using the reference diagnosis established by a psychotherapist; however, given the pilot design, these indicators require confirmation in a prospective study with universal clinical verification.

According to an anonymous survey (n=180 among digital platform users), 94% of messenger bot users rated the system’s convenience as "good" or "excellent." The main advantages cited were: no need for an in-person visit (86%), clear reminders (79%), and confidentiality (68%). Among the 19 web-based personal account users, 73% gave a "good/excellent" rating. All patients who used the paper-based telephone format (n=54) noted the high quality of telephone support, but 42% expressed a desire to switch to a digital channel in the future if technically feasible.

During the observation period, the system identified four patients with suicidal ideation (a response of "3" on item 10 of the EPDS). An urgent consultation with a psychotherapist was arranged for all four patients with a positive response to EPDS item 10 within 2 hours. Given the small size and short duration of the pilot cohort, the absence of suicidal events cannot be directly causally attributed to the system’s operation.

Discussion

Existing digital tools for maternal mental health can be divided into several categories. Self-report applications, such as the Whooley Questions and several others, provide one-time screening but do not support monitoring and are not integrated into the clinical workflow [9]. Telemedicine platforms provide remote access to specialists but do not facilitate automated data collection between visits [10]. Messenger-based digital tools have been used in several research projects for reminders and data collection but generally do not incorporate risk stratification or full clinical integration [11].

The developed system differs from these existing tools in that it combines screening, risk stratification, and dynamic monitoring within a single clinically integrated platform, thereby providing continuous communication between the patient and healthcare professional.

In recent years, several systematic reviews have evaluated the effectiveness of digital interventions in perinatal psychiatry. A meta-analysis including 48 studies demonstrated a small but statistically significant positive effect of digital tools on reducing depressive symptoms in perinatal women (standardized mean difference, 0.29; p=0.003); however, none of the studies used a theoretical model to guide the development of the digital intervention [7]. Another meta-analysis of 31 randomized studies involving 11 024 participants demonstrated a more pronounced effect: the standardized mean difference for PPD symptoms was -0.64 and that for anxiety was -0.49. However, the authors noted that most studies were of limited quality and required replication in larger samples [12].

The limited evidence base is also relevant to mobile applications. A narrative review evaluating the US market for perinatal mental health applications identified only 38 of nearly 600 applications that offered a meaningful mental health intervention. Of these, only three had published patient-based studies, and only one product had moderate-quality evidence (USPSTF level B) [13]. The authors emphasized that obstetrician-gynecologists should consider such applications as adjunctive rather than stand-alone therapeutic tools [13]. In our digital system, the transition from episodic screening to a dynamic monitoring model enables the identification of cases with delayed PPD onset and real-time tracking of changes in clinical status. The system also incorporates behavioral indicators (patterns of interaction with the chatbot and web-based patient portal) as an additional source of clinical information, representing a promising area that remains insufficiently studied in the existing literature.

Artificial intelligence–based chatbots are of considerable interest. A randomized controlled trial (RCT) of a chatbot delivering cognitive behavioral therapy included 168 participants within 6 months postpartum who had screened positive on the Edinburgh Postnatal Depression Scale (EPDS). By week 6, the mean EPDS score had decreased by more than 5 points in the intervention group compared with 1 point in the control group, and clinically significant improvement was achieved in 70% versus 30% of participants, respectively [14]. However, in another RCT involving 192 postpartum women who were not prescreened according to symptom severity, no significant between-group differences in EPDS scores were observed between the chatbot and control groups, although chatbot users demonstrated a greater reduction in Patient Health Questionnaire-9 (PHQ-9) scores [15]. These discrepancies underscore the importance of the target population and eligibility thresholds.

The advances in telemedicine also warrant attention. The multicenter pragmatic SUMMIT randomized controlled trial (RCT) (n=1230) demonstrated the noninferiority of psychotherapy delivered by non-specialists (nurses and midwives) compared with specialist psychotherapists, as well as of telemedicine compared with in-person visits [8]. This finding has direct practical implications for health care systems with a shortage of qualified psychiatrists.

With regard to screening tools, the Edinburgh Postnatal Depression Scale (EPDS) remains the gold standard. The two-item Whooley Questions, endorsed by NICE, demonstrate high sensitivity (95%) but low specificity (65%) [16], which limits their use for dynamic monitoring.

In Russia, efforts are also underway to develop new questionnaires for women of reproductive age [17] and to advance the digitalization of perinatal psychiatry. For example, researchers at Northern State Medical University developed an online screening tool for assessing the psychoemotional status of pregnant and postpartum women [18]. Unlike our system, however, this tool does not provide dynamic monitoring, integration with a messenger bot, or administrative management of questionnaires.

The novelty of our proposed approach is defined by several characteristics: the transition from episodic screening to a dynamic monitoring model enables the identification of cases with delayed PPD onset and real-time tracking of changes in clinical status; support for multiple clinical profiles (postpartum and prenatal) with independently configurable questionnaire protocols enables coverage of the entire perinatal period within a single platform; and the built-in mechanism for managing arbitrary questionnaires through an administrative interface without code modification extends the potential application of the system beyond a specific condition, creating opportunities for use of the platform across a broad range of observational studies in perinatal psychiatry. Finally, the architectural principle of full data auditability ensures compliance with medical documentation requirements.

Limitations: a single-center pilot design without randomization; patient self-selection according to the mode of interaction; a small web-based patient portal subsample (n=19); a short follow-up period and lack of assessment of long-term outcomes; absence of automated stratification, with risk assessment performed manually by a physician, introducing a subjective component; absence of comprehensive clinical verification of all participants irrespective of EPDS scores, precluding definitive assessment of the algorithm's sensitivity and specificity (the reported estimates are preliminary); and limited patient engagement, with missed questionnaires reducing the quality of monitoring and system use potentially being challenging among groups with low digital literacy.

Conclusion

We developed a digital clinical system for the detection, risk stratification, and dynamic monitoring of PPD risk that integrates a messenger bot, a web-based patient portal, and a physician interface. Key distinguishing features from existing solutions include continuous rather than episodic monitoring; support for postpartum and prenatal profiles with configurable protocols; combined use of the validated EPDS and Perinatal Anxiety Screening Scale–Revised (PASS-R), as well as structured investigator-developed questionnaires for collecting clinical and demographic data; a role-based access model; and the ability to create arbitrary questionnaires without code modification.

The data obtained in the pilot study demonstrate the technical feasibility of the system, acceptable user adherence, and preliminary operational effectiveness of the notification algorithm. However, given the pilot observational design, limited sample size, absence of randomization, and incomplete reference-standard verification, all conclusions regarding the clinical effectiveness, sensitivity, and specificity of the algorithm should be considered preliminary.

The platform is technically ready for prospective clinical validation in an expanded cohort with a larger sample size, antenatal enrollment, automated risk stratification, and comprehensive clinical verification. Routine implementation in perinatal centers should be considered only after the effectiveness, reliability, and safety of the system have been confirmed in subsequent prospective studies. Collectively, these findings provide a basis for further development and validation of the system as a potential tool for improving the early detection of PPD in the perinatal population.