Introduction

Postpartum depression (PPD) is a common mental disorder characterized by low mood, loss of pleasure, and sleep disturbance during the postpartum period []. The prevalence of PPD ranges from 3% to 38% in different nations and is higher in limited-income countries [,]. PPD leads to adverse consequences for the mother and family members, such as emotional strain and increased caregiving burden. Women with PPD may experience prolonged periods of distress and are more vulnerable to recurrent depressive episodes []. Previous studies revealed that PPD can impair a mother’s parenting ability, such as breastfeeding, potentially resulting in enduring adverse effects on the child’s development across emotional, cognitive, and physical domains [,]. Moreover, PPD can strain family relationships and impose economic burdens due to increased health care needs and reduced productivity [].

With such a profound impact, mothers should be routinely screened for PPD, and early interventions should be implemented. However, current screening for PPD is mainly based on existing depressive symptoms such as fatigue and sleep disturbance, which are believed to be overlooked due to overlap with normal physiological manifestations after delivery [,]. In addition, the diagnosis of PPD depends on patients’ subjective reporting of personal health conditions []. It is urgent to identify individuals with high risk for PPD before clinical symptoms appear, while no effective and validated screening tools are currently available [,].

Previous studies have identified several risk factors of PPD such as unplanned pregnancy, lack of social support, and family history of mental disorders [,]. However, limited variables in such studies led to a lack of integrity. Machine learning algorithms provide support for the development of predictive models to prevent and intervene adverse health outcomes, offering avenues for personalized prediction and intervention strategies [,]. Several studies have adapted machine learning into the prediction of PPD risk in the last few years and achieved impressive performance [-]. However, insufficient model explanations leave obstacles for actual implementation. Besides, mental disorders are strongly associated with cultural backgrounds and study populations. Thus, the challenge remains to develop more nuanced and culturally adaptable machine learning models for the early detection and effective management of PPD, bridging the gap in current research and practice.

Given the importance of early screening for PPD and the limitations mentioned earlier, we conducted a retrospective study at our institution. This study comprehensively collected variables from multiple aspects, adopted machine learning algorithms to identify risk factors, and aimed to achieve precise prediction of PPD.

MethodsParticipants

Pregnant women who underwent perinatal examinations and delivered at West China Second University Hospital, Sichuan University, from January 2017 to December 2020 were invited to participate in this study. The study cohort was divided into training and validation sets by random sampling. Participants were screened for eligibility. The inclusion criteria were as follows: (1) participants who completed regular examinations and delivered at our institution, (2) participants with a gestational age of ≥28 weeks, and (3) participants who gave consent to participation and be followed up. The exclusion criteria were (1) participants with a psychiatric history in the 6 months before conception and (2) participants with missing data.

Outcome

Participants were assessed for PPD 3 months post partum with the Edinburgh Postnatal Depression Scale []. The Edinburgh Postnatal Depression Scale has 10 items concerning depressive symptoms, and each item is evaluated using scores ranging from 0 to 3, constituting a total score of 30. Participants who scored 13 or more were regarded as having PPD []. The diagnosis of PPD was confirmed by 2 experienced senior psychiatrists using the Structured Clinical Interview for Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) [] and the Chinese Classification and Diagnostic Criteria of Mental Disorders, Third Edition (CCMD-3) [].

Variable Screening

Demographic variables were collected from the electronic medical record system of our institution. Clinical variables were assessed and documented by qualified clinicians. Relevant laboratory indicators were collected at 28 weeks of gestation from the medical laboratory system of the institution.

Participants were assessed for antepartum depression before delivery with the Zung Self-Rating Depression Scale []. The Self-Rating Depression Scale is a self-reported scale with 20 items concerning depressive symptoms, and each item is evaluated with scores ranging from 0 to 4, according to the severity of symptoms. All participants with more than 53 points were regarded as having antepartum depression [].

Social variables, including education, income, exposure to suspected adverse factors, and family and social relations, were collected using scales and self-administered questionnaires. Income level was assessed using the local minimum income standard. Suspected adverse factors included alcohol consumption and smoking. Family and social relations comprised spouses in good health, only child, planned pregnancy, social support, family satisfaction, adverse marital status, and family history of mental illness. The level of social support was measured using the Social Support Rating Scale, which is widely used to assess social support with great reliability []. Scores higher than 35 are considered normal; scores of 35 or lower indicate low levels of social support []. Family satisfaction was assessed using the Family Adaptation, Partnership, Growth, Affection, Resolve index []. The Family Adaptation, Partnership, Growth, Affection, Resolve index consists of 5 items, each with a score ranging from 0 to 2. It systematically evaluates the level of family care a pregnant woman receives. A total score of 0‐3 represents a low level of family satisfaction, and a score higher than 4 represents a normal level.

To avoid the potential bias of multicollinearity and overfitting, least absolute shrinkage and selection operator (LASSO) regression was performed to select and filter the variables in the training set. LASSO is a regression-based methodology that can reduce model complexity; multicollinearity and overfitting are avoided by constructing a penalty function []. LASSO regression is applied to filter a large number of variables and remove those that are insignificant [,]. The 5-fold cross-validation method was used to calculate the optimal λ values, and variables with nonzero coefficients were selected as the final predictive factors. After LASSO regression, the variance inflation factor (VIF) was calculated among the included variables to assess multicollinearity. The VIF was introduced to understand the impact of collinearity in regression models and has since been widely applied in various fields, including medical research [-]. VIF helps ensure that machine learning models or statistical models are not adversely affected by collinear predictors []. Typically, a VIF value greater than 10 is considered indicative of high multicollinearity, which may necessitate removing or adjusting variables to improve model stability [,].

Model Development

We used the following 3 machine learning algorithms to develop the PPD prediction model: extreme gradient boosting (XGBoost), random forest (RF), and gradient boosting machine (GBM). XGBoost is a powerful and efficient machine learning algorithm known for its exceptional performance in regression, classification, and ranking problems. It is an extension of the traditional gradient boosting method that combines multiple weak classifiers to create a strong classifier that minimizes the loss function []. RF is an ensemble machine learning algorithm based on decision trees. It creates multiple decision trees, each based on a randomly sampled subset of the training data to create a more accurate and robust output []. GBM is a popular machine learning algorithm that combines the principles of boosting and gradient descent to create a powerful predictive model []. Additionally, logistic regression, a traditional method, was implemented to predict PPD as a control.

Machine learning models were developed in the training set. To mitigate overfitting and achieve ideal model performance, hyperparameters for each machine learning model were tuned by grid search. In each session of hyperparameter tuning, 3-fold cross-validation was implemented, and the area under the receiver operating characteristic curve (AUC) was the criterion to assess model performance []. The combination of hyperparameters with the largest AUC value was further evaluated in the validation set.

Model Evaluation

Predictive models were evaluated with the receiver operating characteristic curve (ROC) and decision curve analysis (DCA) in the validation set. ROC reflects the ability of a model to discriminate PPD []. DCA is used to evaluate and compare the clinical utility of different diagnostic or predictive models. It provides a framework for assessing the net benefit of a model by taking into account the potential harms and benefits associated with different decision thresholds []. Additionally, accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of each model were calculated for comprehensive evaluation. Based on ROC, the predictive model with the greatest AUC value was considered as the optimal model, which would be further explored for interpretation.

Model Interpretation

We performed variable importance, partial dependence plot (PDP), and Shapley Additive Explanation (SHAP) to interpret the optimal predictive model. Variable importance assesses the contribution of each input variable by calculating the decrease in error when split by a variable []. PDPs calculate the partial dependence of a variable by fixing the values of other variables and observing the variation in the outcome []. It helps to explain how the outcome changes with changes in input variables. SHAP measures the contribution of variables in each individual sample []. The SHAP values show how much each variable contributes, either positively or negatively, to the outcome.

Statistical Software

All statistical analyses were performed with R software (version 4.3.1; R Foundation for Statistical Computing). LASSO regression was performed using the R package glmnet. Logistic regression model was implemented using the R package glm. XGBoost, RF, and GBM models were developed and assessed with R package mlr3. Model interpretation was performed with R packages fastshap and pdp. Other involved packages include xgboost, randomForest, gbm, pROC, ggplot2, and their various dependencies.

Ethical Considerations

This study was approved by the ethics committee of West China Second University Hospital, Sichuan University (approval 2021-186). Informed consent was obtained from all individual participants involved in the study. The original informed consent covered the secondary use of the data without the need for additional consent. All participant data were anonymized to protect privacy and confidentiality. No compensation was provided for participation in this study.

ResultsParticipant Characteristics

The overall procedure of this study is shown in . After eligibility screening, 2055 participants were included in the study cohort. A total of 78 variables were incorporated in our study including 16 psychosocial characteristics, 43 obstetric characteristics, and 19 laboratory indicators. The baseline characteristics were analyzed by χ² test and Wilcoxon test for category variables and continuous variables, respectively. The detailed characteristics are shown in -. Of these participants, 697 (33.9%) participants were diagnosed with PPD, 621 (30.2%) participants had antepartum depression, 101 (4.9%) were unemployed, 187 (9.1%) had an income below the local minimum income standard, 1947 (94.7%) pregnant women were the only child of the family, and 45 (2.2%) reported low family satisfaction; the median age of participants was 32 (IQR 29-35) years.

‎Figure 1. The flow chart of the study design. Table 1. Psychosocial characteristics of participants.Psychosocial characteristicsNon-PPD (n=1358)PPD (n=697)P valueMethodAge (years), median (IQR)31 (29-35)32 (29-35).03WilcoxonAntepartum depression, n (%)<.001χ² testNo1170 (86.2)264 (37.9)Yes188 (13.8)433 (62.1)Ethnics, n (%).15χ² testHan1326 (97.6)673 (96.6)Others32 (2.4)24 (3.4)Work status, n (%).002χ² testEmployed1277 (94)677 (97.1)Unemployed81 (6)20 (2.9)Season of delivery, n (%).51χ² testSpring343 (25.3)193 (27.7)Summer350 (25.8)165 (23.7)Autumn411 (30.3)217 (31.1)Winter254 (18.7)122 (17.5)Education, n (%).63χ² testBelow higher education459 (33.8)243 (34.9)Higher education899 (66.2)454 (65.1)Income, n (%).46χ² testBelow normal level119 (8.8)68 (9.8)At or above normal level1239 (91.2)629 (90.2)Smoking, n (%).69Yates’ correctionNo1355 (99.8)694 (99.6)Yes3 (0.2)3 (0.4)Drinking, n (%)>.99Yates’ correctionNo1355 (99.8)695 (99.7)Yes3 (0.2)2 (0.3)Spouse in good health, n (%).95χ² testYes1339 (98.6)687 (98.6)No19 (1.4)10 (1.4)Only child, n (%).17χ² testYes1280 (94.3)667 (95.7)No78 (5.7)30 (4.3)Planned pregnancy, n (%).31χ² testYes1310 (96.5)666 (95.6)No48 (3.5)31 (4.4)Social support, n (%).52χ² testNormal1329 (97.9)679 (97.4)Low29 (2.1)18 (2.6)Family satisfaction, n (%).13χ² testNormal1333 (98.2)677 (97.1)Low25 (1.8)20 (2.9)Adverse marital status, n (%).006χ² testNo1349 (99.3)683 (98)Yes9 (0.7)14 (2)Family history of mental illness, n (%).04Yates’ correctionNo1356 (99.9)691 (99.1)Yes2 (0.1)6 (0.9)

aPPD: postpartum depression.

Table 2. Obstetric characteristics of participants.Obstetric characteristicsNon-PPD (n=1358)PPD (n=697)P valueMethodWeight gain during pregnancy, median (IQR)12.5 (9.425‐15)12.5 (9.7‐16).39WilcoxonBMI, median (IQR)20.83 (19.43‐22.68)20.83 (19.36‐23.01).63WilcoxonAge of menarche (years), median (IQR)13 (12-13)13 (12-14).30WilcoxonGestational days, median (IQR)274 (268-280)274 (267-278).05WilcoxonBleeding volume, median (IQR)400 (300-400)400 (300-400).07WilcoxonFetal weight, median (IQR)3.28 (2.94‐3.57)3.23 (2.8‐3.53).001WilcoxonFetal height, median (IQR)50 (48-51)49 (48-51).03WilcoxonApgar 1 minute, median (IQR)10 (10-10)10 (10-10)<.001WilcoxonApgar 5 minutes, median (IQR)10 (10-10)10 (10-10)<.001WilcoxonApgar 10 minutes, median (IQR)10 (10-10)10 (10-10)<.001WilcoxonLength of stay, median (IQR)4 (4-6)4 (4-6).79WilcoxonGravidity.19χ² test1, n (%)471 (22.9)217 (10.6)2, n (%)370 (18)207 (10.1)3, n (%)265 (12.9)136 (6.6)4, n (%)160 (7.8)79 (3.8)≥5, n (%)92 (4.5)58 (2.8)Median (IQR)2 (1-3)2 (1-3)Abortions, n (%).38χ² test0624 (45.9)301 (43.2)1409 (30.1)207 (29.7)2207 (15.2)115 (16.5)≥3118 (8.7)74 (10.6)Parity, n (%).89Yates’ correction0831 (40.4)434 (21.1)1497 (24.2)246 (12)229 (1.4)16 (0.8)≥31 (0)1 (0)Conception method, n (%).93χ² testNormal1169 (86.1)599 (85.9)Assisted reproduction189 (13.9)98 (14.1)Fetal malformation, n (%).02χ² testNo1308 (96.3)656 (94.1)Yes50 (3.7)41 (5.9)Amniotic fluid volume disorder, n (%).35χ² testNo1284 (94.6)652 (93.5)Yes74 (5.4)45 (6.5)Renal disease, n (%).58χ² testNo1334 (98.2)687 (98.6)Yes24 (1.8)10 (1.4)Systemic lupus erythematosus, n (%).69Yates’ correctionNo1351 (99.5)695 (99.7)Yes7 (0.5)2 (0.3)Gestational diabetes mellitus, n (%).92χ² testNo1030 (75.8)530 (76)Yes328 (24.2)167 (24)Gestational hypertension, n (%).25χ² testNo1290 (95)670 (96.1)Yes68 (5)27 (3.9)Threatened premature labor, n (%).002χ² testNo1183 (87.1)571 (81.9)Yes175 (12.9)126 (18.1)Hepatitis B, n (%).49χ² testNo72 (5.3)42 (6)Yes1286 (94.7)655 (94)Twin pregnancy, n (%).07χ² testNo1247 (91.8)623 (89.4)Yes111 (8.2)74 (10.6)Placenta previa, n (%).45χ² testNo1281 (94.3)663 (95.1)Yes77 (5.7)34 (4.9)Heart disease, n (%).29χ² testNo1344 (99)693 (99.4)Yes14 (1)4 (0.6)Scarred uterus, n (%).81χ² testNo342 (25.2)179 (25.7)Yes1016 (74.8)518 (74.3)Rh blood type, n (%).01χ² testPositive1350 (99.4)685 (98.3)Negative8 (0.6)12 (1.7)ABO blood type, n (%).63χ² testO491 (36.2)241 (34.6)B335 (24.7)179 (25.7)A423 (31.1)211 (30.3)AB109 (8)66 (9.5)Abnormal fetal position, n (%).92χ² testNo1206 (88.8)618 (88.7)Yes152 (11.2)79 (11.3)Uterine myoma, n (%).94χ² testNo1227 (90.4)629 (90.2)Yes131 (9.6)68 (9.8)Ovarian cyst, n (%)>.99Yates’ correctionNo1349 (99.3)693 (99.4)Yes9 (0.7)4 (0.6)Umbilical cord encirclements, n (%).65χ² testNo869 (64)453 (65)Yes489 (36)244 (35)Hypothyroidism, n (%).40χ² testNo1130 (83.2)590 (84.6)Yes228 (16.8)107 (15.4)Pelvic anomaly, n (%).56χ² testNo1346 (99.1)689 (98.9)Yes12 (0.9)8 (1.1)Intrauterine death, n (%)<.001Yates’ correctionNo1358 (100)684 (98.1)Yes0 (0)13 (1.9)Macrosomia, n (%).84χ² testNo1295 (95.4)666 (95.6)Yes63 (4.6)31 (4.4)Fetal growth restriction, n (%).34χ² testNo1335 (98.3)681 (97.7)Yes23 (1.7)16 (2.3)Premature labor, n (%)<.001χ² testNo1207 (88.9)579 (83.1)Yes151 (11.1)118 (16.9)Mode of delivery, n (%).45Yates’ correctionVaginal delivery842 (62)433 (62.1)Cesarean section506 (37.3)262 (37.6)Assisted delivery10 (0.7)2 (0.3)Fetal sex, n (%).96χ² testMale668 (49.2)342 (49.1)Female690 (50.8)355 (50.9)Fetal distress, n (%).001χ² testNo1333 (98.2)667 (95.7)Yes25 (1.8)30 (4.3)Breastfeeding, n (%).61χ² testNo32 (2.4)19 (2.7)Yes1326 (97.6)678 (97.3)

aPPD: postpartum depression.

bApgar: appearance, pulse, grimace, activity, and respiration.

Table 3. Laboratory indicators.Laboratory indicatorsNon-PPD (n=1358), median (IQR)PPD (n=697), median (IQR)P valueMethodHemoglobin (g/L)111 (104‐117.75)111 (104-118).893WilcoxonSerum ferroprotein (ng/nL)18.15 (12.4‐25.9)18.9 (11.9‐27.2).400WilcoxonInternational normalized ratio0.97 (0.92‐1.01)0.96 (0.91‐1.01).325WilcoxonAlanine aminotransferase (U/L)17 (12.25‐28)18 (12-31).170WilcoxonAspartate aminotransferase (U/L)21 (18-27)21 (17-28).263WilcoxonTotal bile acid (µmol/L)2.3 (1.6‐3.5)2.5 (1.6‐3.7).091WilcoxonDirect bilirubin (µmol/L)2.1 (1.6‐2.8)2.1 (1.7‐2.7).771WilcoxonAlbumin (g/L)38.7 (36.3‐41.2)38.7 (36.3‐41.4).627WilcoxonGlobulin (g/L)27.6 (25.4‐30.1)27.3 (25.2‐29.9).207WilcoxonLactate dehydrogenase (U/L)179 (163-201)181 (164-204).161WilcoxonAlkaline phosphatase (U/L)84 (55-121)87 (55-128).423WilcoxonUrea nitrogen (µmol/L)3.5 (3.07‐4.3475)3.48 (3.08‐4.35).787WilcoxonCreatinine (µmol/L)44 (40-48)44 (40-48).561WilcoxonCystatin C (µmol/L)0.77 (0.64‐0.97)0.77 (0.64‐0.99).725WilcoxonUric acid (µmol/L)259 (217-309)254 (218-305).250WilcoxonThyroid-stimulating hormone (mIU/L)1.9695 (1.277‐2.8878)1.847 (1.176‐2.776).183WilcoxonFree thyroxine (pmol/L)14.56 (13.29‐16.22)14.55 (13.21‐16.02).522WilcoxonThyroid peroxidase antibody (U/mL)40.65 (30.4‐56.1)40.1 (30.2‐55.3).425WilcoxonVitamin D (nmol/L)23.9 (17.3‐31.3)22.9 (16.1‐29.8).011Wilcoxon

aPPD: postpartum depression.

Variable Screening

After LASSO regression, 18 variables with nonzero coefficients were identified as potential predictors of PPD. Among these variables, the 5-minute Apgar (appearance, pulse, grimace, activity, and respiration) score and the 10-minute Apgar score had VIF values over 10, indicating multicollinearity between them. Of these 2 variables, the 5-minute Apgar score had a lower coefficient in absolute value in the LASSO regression, suggesting its lower contribution to the outcome; therefore, it was excluded. Another round of VIF analysis was performed after excluding the 5-minute Apgar score, and the results showed that all remaining variables had a VIF below 10, indicating low multicollinearity. Finally, 17 variables including prenatal depression, ethnics, occupation, income, only child, family satisfaction, adverse marital status, amniotic fluid volume disorder, Rh negative, intrauterine death, fetal distress, age, fetal weight, 10-minute Apgar score, serum ferroprotein, thyroid-stimulating hormone (TSH), and thyroid peroxidase antibody (TPOAb) were identified as features to develop predictive models. The detailed results of the LASSO regression and VIF analyses are presented in .

Table 4. LASSO coefficients and VIF of screened variables after LASSO regression.VariableCoefficientVIFVIF (second round)Prenatal depression2.2451.0141.013Ethnics0.0551.0081.008Occupation−0.2151.0471.045Income0.0151.0431.043Only child0.0811.0131.012Family satisfaction0.1341.2871.286Adverse marital status0.4221.2961.293Amniotic fluid volume disorder0.4161.0111.011Rh negative−0.1631.0121.011Intrauterine