Development and Internal Validation of a Psychosocial Nomogram for Antenatal Depressive Symptoms in Third-Trimester Pregnant Women at a Regional Hospital in Southern Vietnam

Abstract

Purpose

Antenatal depression is common and under-recognized, yet few individualized tools exist to support its identification in obstetric practice in low- and middle-income countries. We estimated the prevalence of a positive antenatal depression screen among third-trimester pregnant women in southern Vietnam and developed and internally validated a psychosocial nomogram for the concurrent probability of a positive screen.

Patients and Methods

Within a prospective study at Long Khanh Regional General Hospital (July 2025 to May 2026), 968 women at 28 weeks of gestation or later were screened with a Vietnamese-language version of the Edinburgh Postnatal Depression Scale (EPDS; positive screen defined as a score ≥13). Sociodemographic, obstetric, marital, family, and social variables were recorded at the same visit, so predictors and outcome were measured concurrently (a diagnostic rather than a prognostic model). Candidate predictors with p<0.25 in univariable logistic regression entered a multivariable model reduced by backward elimination (p<0.05).

Results

The prevalence of a positive screen was 16.5% (160/968; 95% CI 14.3–19.0). Six variables were retained in the final model: an adverse family event (adjusted odds ratio [aOR] 5.25), marital conflict (aOR 3.64), a family history of depression (aOR 2.76), neighbourhood or workplace conflict (aOR 2.48), work-related pressure (aOR 2.22), and a lack of psychosocial support (aOR 1.77). Apparent discrimination was good (area under the curve 0.807, 95% CI 0.769–0.847) and fell to an optimism-corrected 0.782 after bootstrapping the full procedure; the optimism-corrected calibration slope was 0.88 and calibration-in-the-large was near zero. On decision curve analysis the nomogram gave a higher net benefit than screening all women at threshold probabilities above about 10%.

Conclusion

About one in six third-trimester women screened positive for depressive symptoms. A nomogram based on six readily recorded psychosocial variables showed good apparent and promising internally validated performance and may support psychosocial risk stratification where universal formal screening is difficult.

Introduction

Antenatal depression is a common but frequently overlooked complication of pregnancy and a major public-health concern for maternal and child health. Reported prevalence varies with country income level, trimester, screening instrument, and threshold, and the burden is generally higher in low- and middle-income settings, where pooled estimates commonly exceed 15%.^1–3 The third trimester has been identified as a period when depressive symptoms are particularly frequent.¹ Antenatal depression has been associated with reduced adherence to antenatal care and, in pooled analyses, with a modest increase in adverse birth outcomes such as preterm birth and low birth weight,² and depressive symptoms in early pregnancy have been linked to altered fetal growth.⁴ Antenatal depressive symptoms also frequently persist into the postpartum period, with roughly two in five affected women remaining depressed after childbirth.⁵ Early identification of affected women is therefore a reasonable public-health priority.

In Vietnam, clinical and research attention has centred on postpartum rather than antenatal depression, although prospective population-based work has documented a substantial burden of common mental disorders during pregnancy.⁶ Screening is the usual entry point. The Edinburgh Postnatal Depression Scale (EPDS) is the most widely used self-report instrument for perinatal depression, and a culturally verified Vietnamese version has been validated against a structured clinical interview, although optimal thresholds vary with setting and the diagnostic target.^7,9 Most local studies remain descriptive, reporting prevalence and associated factors without converting them into a tool that clinicians can apply to an individual woman. A nomogram is a graphical calculator that turns a small set of variables from a regression model into an individual predicted probability. For antenatal depression such tools are scarce and have been developed almost entirely outside Vietnam. A recent Chinese nomogram achieved high discrimination, but its final model relied on neuroticism, negative-coping, and anxiety scores obtained from administered psychometric instruments, which are difficult to apply in busy or low-resource antenatal clinics.¹⁰ No tool built from brief interview-based psychosocial variables that can be elicited during routine history-taking has been reported for the Vietnamese obstetric population. Marital quality, family support, and adverse life events are recurrently reported correlates of perinatal depression and are plausible candidates for such a tool.^11–13 We therefore estimated the prevalence of a positive antenatal depression screen among third-trimester women at a regional hospital in southern Vietnam and developed and internally validated a psychosocial nomogram for the concurrent probability of a positive screen.

Materials and Methods

Study Design and Setting

Women in their third trimester were enrolled prospectively at the Department of Obstetrics and Gynaecology, Long Khanh Regional General Hospital, Dong Nai Province, Vietnam. Recruitment ran from July 2025 to May 2026. Each woman contributed a single assessment at her antenatal visit, at which the candidate variables and the EPDS were recorded together.

Participants

Eligible women were in the third trimester, defined by the conventional obstetric threshold of 28 weeks 0 days of gestation or later, attended and completed their pregnancy at the study hospital, could read Vietnamese, and gave informed consent. Women who could not complete the questionnaire because of an organic neurological condition or a communication-impairing disability, and those who declined, were excluded. Recruitment was consecutive.

Outcome

The outcome was a positive antenatal depression screen on a Vietnamese-language version of the EPDS, a ten-item scale scored 0 to 3 per item (total 0 to 30).⁷ A total score of 13 or greater defined a positive screen. We prespecified this threshold to identify women with a higher depressive symptom burden and to favour specificity; it was not intended to define a psychiatric diagnosis, and the individual-participant-data meta-analysis of the EPDS reports that lower thresholds maximise combined sensitivity and specificity, so our estimates should be read as screen positivity rather than diagnosed depression.⁹ The Vietnamese EPDS has been culturally verified and validated against a structured clinical interview in perinatal women.¹⁴

Candidate Predictors and Definitions

Candidate variables were specified in advance and collected by structured face-to-face interview using a pre-designed questionnaire, in five domains: sociodemographic, obstetric and reproductive, partner-related, family, and social. Each was coded as a binary contrast of a hypothesized higher-symptom category versus a reference category. The six variables retained in the final model were defined as follows: adverse family event, a serious event affecting a close relative during the pregnancy (for example serious illness, death, or loss of the husband’s employment) versus none; marital conflict, occasional or frequent conflict with the husband versus never; family history of depression, any first-degree relative with previously diagnosed depression versus none; neighbourhood or workplace conflict, present versus absent; work-related pressure, present versus absent; and lack of psychosocial support, the woman reporting that she did not have someone to turn to for emotional support versus having such support. Full items and coding for all candidate variables are given in Supplementary Table S1.

Sample Size

For the descriptive objective, the sample size was estimated using the single-proportion formula, assuming a prevalence of late-pregnancy common mental disorder symptoms of 28.1%,⁶ a 95% confidence level, and an absolute precision of 0.05, yielding a minimum of 311 women. Recruitment was consecutive throughout the prespecified study period, and all eligible participants were included; the final analytic sample comprised 968 women, including 160 with a positive EPDS screen.

Data Handling

Responses were entered into an electronic database. One blank record was excluded. Values outside prespecified plausible ranges could not be reliably reconciled and were set to missing; no other values were altered. The analytic set comprised 968 women.

Model Development and Internal Validation

Categorical variables were summarized as frequencies and percentages and continuous variables as mean and standard deviation. Each candidate variable was examined in univariable logistic regression, and those with p<0.25 entered a multivariable model reduced by backward elimination retaining p<0.05. Adjusted odds ratios (aOR) with 95% confidence intervals were reported, and a nomogram was built from the coefficients of the final model, with points proportional to each coefficient and the largest scaled to 100. Univariable pre-screening followed by stepwise selection can inflate apparent performance and destabilise the selected set. Because the coefficients were not penalised, the complete model-development procedure was repeated within each bootstrap sample to quantify the optimism arising from both variable selection and coefficient estimation.

Discrimination was quantified by the area under the receiver operating characteristic curve (AUC). Internal validation used 500 bootstrap resamples, and within each resample the complete model-building procedure, including univariable screening and backward elimination, was repeated, so that the optimism estimate reflects the whole selection process rather than a fixed six-variable model.¹⁵ We report the apparent and optimism-corrected AUC, the optimism-corrected calibration slope and calibration-in-the-large, the apparent and optimism-corrected Brier score, and a calibration plot showing both the apparent and the optimism-corrected calibration. The proportion of bootstrap resamples in which each candidate variable was retained is given in Supplementary Table S2. Clinical utility was assessed by decision curve analysis, in which the clinical action evaluated was prioritisation for EPDS administration; net benefit was compared across threshold probabilities with administering the EPDS to all or to no women.¹⁶ A threshold from the Youden index is reported descriptively, with the associated sensitivity, specificity, and predictive values presented as apparent (development-sample) values. Analyses used R version 4.3.3; logistic models were fitted with the stats package (glm), and bootstrapping and graphics used base R. A two-sided p-value below 0.05 was considered significant.

Ethics Approval and Informed Consent

The protocol was approved by the Biomedical Research Ethics Council of Can Tho University of Medicine and Pharmacy (No. 25. 389. HV/PCT-HDDD, dated 30 June 2025), and recruitment began only after this date. The study followed the Declaration of Helsinki, and all participants gave written informed consent.

Results

Participants and Prevalence

Of 968 third-trimester women, the mean maternal age was 28.2 ± 5.6 years and the mean gestational age at enrolment was 37.2 ± 2.4 weeks. Most women were of Kinh ethnicity (89.2%) and lived in rural areas (71.1%); 8.0% were from poor or near-poor households and 1.9% were not living with their husband (Table 1). The prevalence of a positive antenatal depression screen (EPDS ≥ 13) was 16.5% (160/968; 95% CI 14.3–19.0), and the mean EPDS score was 6.6 ± 4.8.

Table 1.

Baseline Characteristics of the Study Participants

Characteristic	n (%) or Mean ± SD
Maternal age (years), mean ± SD	28.2 ± 5.6
<20	52 (5.4)
20–34	793 (81.9)
≥35	123 (12.7)
Gestational age at enrollment (weeks), mean ± SD	37.2 ± 2.4
Ethnicity: Kinh/other	863 (89.2) / 105 (10.8)
Religion: yes	544 (56.2)
Education ≤ secondary school	244 (25.2)
Residence: rural	688 (71.1)
Poor/near-poor household	77 (8.0)
Not living with husband	18 (1.9)
Irregular menstruation history	214 (22.1)
History of miscarriage	160 (16.5)
History of induced abortion	38 (3.9)
Unplanned pregnancy	457 (47.2)
Primiparity	407 (42.0)
History of infertility treatment	27 (2.8)
Exposure to violence during pregnancy	14 (1.4)
Marital conflict (ever)	309 (31.9)
Family history of depression	36 (3.7)
Family conflict	33 (3.4)
Adverse family event	72 (7.4)
Neighbourhood/workplace conflict	29 (3.0)
Work-related pressure	134 (13.8)
Lack of psychosocial support	308 (31.8)
EPDS score, mean ± SD	6.6 ± 4.8
Positive screen (EPDS ≥ 13)	160 (16.5)

Variables Associated with a Positive Screen

In univariable analysis the strongest associations with a positive screen were an adverse family event, family conflict, a family history of depression, marital conflict, work-related pressure, and neighbourhood or workplace conflict; poor or near-poor status, irregular menstruation, and previous miscarriage were also associated (Table 2). Nineteen variables with p<0.25 entered the multivariable model.

Table 2.

Univariable Logistic Regression of Candidate Variables for a Positive Antenatal Depression Screen

Variable	EPDS ≥13/ Exposed, n/N	Crude OR	95% CI	p
Maternal age ≥ 35 years	21/123	1.05	0.63–1.73	0.862
Ethnicity other than Kinh	24/105	1.58	0.97–2.59	0.066
Religion (yes)	99/544	1.32	0.93–1.87	0.114
Education ≤ secondary	37/244	0.87	0.59–1.30	0.507
Rural residence	115/688	1.05	0.72–1.53	0.807
Poor/near-poor household	23/77	2.34	1.39–3.95	0.001
Not living with husband	5/18	1.97	0.69–5.61	0.203
Irregular menstruation	48/214	1.66	1.13–2.42	0.009
History of miscarriage	36/160	1.60	1.05–2.43	0.027
History of induced abortion	5/38	0.76	0.29–1.97	0.569
Unplanned pregnancy	83/457	1.25	0.89–1.76	0.196
Primiparity	61/407	0.82	0.58–1.17	0.272
Infertility treatment	8/27	2.19	0.94–5.08	0.069
Fetal-sex preference by husband	73/421	1.11	0.79–1.56	0.551
Violence during pregnancy	3/14	1.38	0.38–5.02	0.621
Husband not attentive to health	13/29	4.38	2.06–9.29	<0.001
Marital conflict	106/309	5.85	4.06–8.42	<0.001
Marital dissatisfaction	14/30	4.75	2.27–9.94	<0.001
Does not turn to husband when in difficulty	23/57	3.82	2.18–6.69	<0.001
Family history of depression	20/36	7.07	3.58–13.98	<0.001
Family conflict	20/33	8.74	4.25–17.97	<0.001
Adverse family event	48/72	14.00	8.25–23.75	<0.001
Perceived family unhappiness	12/29	3.77	1.77–8.06	<0.001
Neighbourhood/workplace conflict	14/29	5.07	2.40–10.73	<0.001
Work-related pressure	57/134	5.25	3.52–7.84	<0.001
Lack of psychosocial support	83/308	2.79	1.98–3.95	<0.001

After backward elimination six variables were retained in the final model: an adverse family event (aOR 5.25), marital conflict (aOR 3.64), a family history of depression (aOR 2.76), neighbourhood or workplace conflict (aOR 2.48), work-related pressure (aOR 2.22), and a lack of psychosocial support (aOR 1.77) (Table 3). The corresponding model, which underlies both the nomogram and Table 3, is logit(P) = −2.872 + 1.659 (adverse family event) + 1.293 (marital conflict) + 1.015 (family history of depression) + 0.909 (neighbourhood or workplace conflict) + 0.796 (work-related pressure) + 0.572 (lack of psychosocial support), with each variable coded 1 if present and 0 otherwise.

Table 3.

Multivariable Logistic Regression Model and Nomogram Points

Predictor	aOR	95% CI	p	Points
Adverse family event	5.25	2.84–9.70	<0.001	100
Marital conflict	3.64	2.45–5.43	<0.001	78
Family history of depression	2.76	1.20–6.35	0.017	61
Neighbourhood/workplace conflict	2.48	1.04–5.90	0.040	55
Work-related pressure	2.22	1.35–3.63	0.002	48
Lack of psychosocial support	1.77	1.19–2.65	0.005	35

Nomogram and Performance

The nomogram converts the six variables into an individual probability of a positive screen (Figure 1); points when a factor is present were adverse family event 100, marital conflict 78, family history 61, neighbourhood or workplace conflict 55, work-related pressure 48, and lack of support 35, spanning a predicted probability of about 5% to 97%. Apparent discrimination was good (AUC 0.807, 95% CI 0.769–0.847; Figure 2A). When the entire model-building procedure was repeated within 500 bootstrap resamples, optimism was 0.025 and the optimism-corrected AUC was 0.782; across resamples the number of retained variables ranged from 4 to 14 (median 8), and several candidates entered the model in only a minority of resamples (Supplementary Table S2), illustrating selection instability. Calibration was reasonable, with an optimism-corrected calibration slope of 0.88, an optimism-corrected calibration-in-the-large near zero, an apparent Brier score of 0.105, and an optimism-corrected Brier score of 0.113 (Figure 2B). At a Youden-derived threshold (predicted probability 0.11), the apparent sensitivity was 77.5%, specificity 69.6%, positive predictive value 33.5%, and negative predictive value 94.0%; these are development-sample values and are expected to be optimistic. For illustration, grouping women by predicted probability gave observed positivity of 5.7% below 10%, 21.3% between 10% and 30%, and 56.0% above 30%. On decision curve analysis the nomogram produced a higher net benefit than screening all women at threshold probabilities above roughly 10% (Figure 2C). Performance is summarized in Table S3.

Figure 1.

Nomogram for the probability of a positive antenatal depression screen.

Figure 2.

Performance of the model. (A) Receiver operating characteristic curve (apparent AUC 0.807, 95% CI 0.769–0.847). (B) Calibration plot of observed versus predicted probability, with grouped observations (points, with 95% confidence intervals), a lowess curve, and the ideal line. (C) Decision curve analysis comparing the nomogram with screening all and screening no women.

Discussion

Among 968 third-trimester women in southern Vietnam, 16.5% screened positive for depressive symptoms, and six routinely recorded psychosocial variables were independently associated with a positive screen. A nomogram built from them had good apparent discrimination (AUC 0.807) that fell to 0.782 once the whole selection process was bootstrapped, with an optimism-corrected calibration slope of 0.88 indicating modest over-fitting.

The prevalence is consistent with regional evidence, lying close to the 18.9% reported in urban Thailand and within the range described for South Asian and other low- and middle-income populations, and in keeping with the burden of antenatal common mental disorders documented prospectively in Vietnam.^3,6,12,17 Differences in screening instruments and thresholds, including the use of other scales in some of these studies, limit direct comparison. The variables that remained in the model were psychosocial rather than obstetric. An adverse family event was the strongest correlate, consistent with the role of acute stressors, and marital conflict and a lack of support were also prominent, echoing perinatal studies that link relationship quality and low support to antenatal mental disorders.^11–13 A family history of depression may reflect familial vulnerability arising from shared genetic and environmental factors. Because exposures and the screen were measured together, these associations cannot be interpreted causally: marital conflict, perceived pressure, and a sense of having no one to turn to may equally be consequences of current low mood. The model should therefore be understood as describing who currently screens positive, which is sufficient for stratification but not for identifying targets for intervention.

The most relevant comparison is the recent Chinese antenatal-depression nomogram, which reached higher discrimination (area under the curve about 0.90) in a larger sample, but its final model used neuroticism, negative-coping, and anxiety scores from administered psychometric instruments that are impractical for routine antenatal care.¹⁰ Our model had lower discrimination but used a simpler set of interview-based variables; direct comparison of the two figures is limited because the populations and validation strategies differ. Decision curve analysis suggested added value over administering the EPDS to all women at thresholds above about 10%.

Professional guidelines recommend screening every pregnant woman for depression and anxiety with a standardized validated instrument at the initial prenatal visit, later in pregnancy, and postpartum.¹⁸ The nomogram is not intended to replace this universal screening or to justify withholding it where universal screening is feasible. Its potential role is as a pre-screening prioritisation tool in settings where universal EPDS administration cannot be consistently implemented, drawing attention to women in higher-probability groups for earlier EPDS administration or referral, while accepting that repeated assessment is needed because symptoms change across pregnancy. The model has undergone internal validation only; its threshold-based classification metrics are apparent, and the proposed risk groups are illustrative rather than validated clinical categories.

Strengths include prospective enrolment, a reasonably large sample, a Vietnamese-language EPDS validated against a clinical interview, and internal validation that repeated the entire model-building procedure and so did not understate optimism. Limitations are important. First, the concurrent measurement of exposures and outcome makes reverse causation likely for several predictors, so the tool is diagnostic, not aetiological. Second, recruitment at a single hospital, with completion of pregnancy at the site required, invites selection bias and limits generalisability to community populations. Third, variable selection by univariable screening and backward elimination is unstable and optimistic; penalised estimation and, above all, external validation are needed, and the modest ratio of events to candidate parameters reinforces this.^15,19,20 Fourth, the EPDS identifies probable symptoms rather than a diagnosis, and the reported threshold metrics are apparent; the relatively high threshold of 13 may also have reduced sensitivity, given evidence that lower thresholds can perform better for perinatal common mental disorders in Vietnamese women.^8,9 Fifth, several predictors were infrequent, widening their confidence intervals, and all were self-reported, so recall and social-desirability bias are possible. Future work should pursue multicentre external validation, penalised model updating, and studies of whether nomogram-assisted triage changes maternal outcomes.

Conclusion

About one in six third-trimester women in this southern Vietnamese cohort screened positive for depressive symptoms, and a nomogram based on six brief interview-based psychosocial variables showed good apparent and promising internally validated performance for identifying who currently screens positive. It may help prioritise women for EPDS administration or referral where universal formal screening is difficult, but external, multicentre validation is required before any clinical use.

Funding Statement

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Abbreviations

aOR, adjusted odds ratio; AUC, area under the receiver operating characteristic curve; CI, confidence interval; EPDS, Edinburgh Postnatal Depression Scale; OR, odds ratio; SD, standard deviation.

Data Sharing Statement

The data sets generated and analysed during the current study are available from the corresponding author upon reasonable request.

Ethics Approval and Consent to Participate

Approved by the Biomedical Research Ethics Council of Can Tho University of Medicine and Pharmacy (No. 25. 389. HV/PCT-HDDD, 30 June 2025); all participants gave written informed consent.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors report there are no conflicts of interest in this work.