Burden and determinants of perinatal depression: population-based estimates from linked healthcare utilization data and EPDS screening Authors: Eliana Gabellini, Maria Teresa Greco, Raffaella Ferrari, Federica Rolli, Silvano Casazza, Antonio Giampiero Russo Published: 2026 ABSTRACT PurposePerinatal depression is a frequent disorder affecting about 10–15% of women. Despite its significance, only a proportion of the female population is screened or intercepted by professional services. The aim of this study was to estimate its occurrence and to examine potential risk and protective factors.MethodsEvents attributable to perinatal depression were identified through an algorithm applied to Health Care Utilization (HCU) data in a cohort of 164,767 resident mothers giving birth between 2017 and 2023 in Agenzia di Tutela della Salute (ATS) della Città Metropolitana di Milano. Two multivariate logistic models were fitted, including prior psychological distress (12 months before pregnancy), sociodemographic variables (age, citizenship, education, employment, marital status) and parity.ResultsAccording to the HCU algorithm, 16.0% of mothers experienced perinatal depression–related events. Among screened mothers, 23.6% met criteria for high psychological vulnerability (EPDS ≥ 8). Risk factors included prior psychological distress (OR 8.1; 95% CI 7.7–8.6), age > 40 years (OR 1.3; 95% CI 1.2–1.4), migrant background (OR 1.1; 95% CI 1.1–1.2) and primiparity (OR 1.4; 95% CI 1.4–1.5). Lower education and employment appeared as modest protective factors.ConclusionsPerinatal depression affects approximately one in six women within the territory of ATS Milan. Prior psychological distress, advanced maternal age, migrant background, unmarried status and primiparity emerged as key risk factors. The identification of key risk and protective factors underscores the need for targeted, equity-oriented policies and early intervention strategies, strengthening the integration between primary care, mental health services, social services, and community-based resources to effectively reach and support vulnerable women. Introduction Perinatal depression (PD) is the most frequent mental-health complication of pregnancy and the postpartum period, with global point‑prevalence estimates ranging from 10% to 18% and large geographical variability (Caffieri et al. 2024, Doyle et al. 2015) Consequences span the entire life‑course: poorer obstetric outcomes, reduced breastfeeding, impaired early bonding and, in the long term, cognitive, emotional, and behavioural difficulties in offspring (Dagher et al. 2021). Maternal mortality audits also identify suicide as a leading cause of direct deaths up to one year after birth (Hagatulah et al. 2024). Given the impact of this disorder on several levels, accurate preventive interventions are essential. This is why effective prevention work can be done by intervening on known biological, psychological and socio-economic risk factors (Waqas et al. 2022). Screening programmes are vital for early identification of women at higher risk, enabling timely interventions to support both mother–child relationships and mental health. Across Europe, screening for perinatal depression is increasingly recommended but remains unevenly implemented. A recent scoping review across the WHO European region (Horakova et al. 2024) showed that only 10 out of 53 countries have established perinatal mental health screening programmes, while less than half have specific policies or guidelines in this area. As a result, screening is not systematically delivered, and a substantial proportion of women is not routinely assessed. In Italy, screening initiatives are mainly locally organized, resulting in variable coverage and leaving a significant share of the population unscreened. A range of standardized screening tools is used in clinical practice, including brief case-finding tools such as the Whooley questions and standardized instruments such as the Edinburgh Postnatal Depression Scale (EPDS) and the Patient Health Questionnaire (PHQ‑9). Among these, the EPDS is the most widely used and recommended screening instrument in perinatal care. The Edinburgh Postnatal Depression Scale (EPDS) is a 10‑item self-report questionnaire designed to identify depressive symptoms during the postpartum period, based on a series of items exploring mood, anxiety, and anhedonia over the previous week. It has been extensively validated across different populations and settings, demonstrating good psychometric properties and clinical utility in both antenatal and postnatal care (Levis et al., 2020). Regional health policy in Lombardy reaffirmed the importance of identifying vulnerable women and families through validated screening tools or spontaneous request. In the Lombardy Region, specific actions for the prevention, diagnosis and early intervention of perinatal depression were formally introduced with Regional Resolution No. X/5954 of 5 December 2016 (Regione Lombardia, 2016). This investigation combines survey data with multiple routine HCU databases to quantify the burden of perinatal depression, delineate risk and protective factors, and generate evidence to guide prevention and service planning across ATS Milano. Materials and methods To conduct the research, administrative Health Care Utilization (HCU) data from healthcare and social-healthcare facilities and the Perinatal Depression Survey of the Health Protection Agency of the Metropolitan City of Milan were used. The Agenzia di Tutela della Salute (ATS) della Città Metropolitana di Milano is an administrative unit of the Lombardy Regional Health System with responsibility for planning, regulating and guaranteeing appropriate provision of hospital and community services. ATS Milano includes 193 municipalities across the provinces of Milan and Lodi. Since 2016, ATS Milano has promoted an Action Programme for the Prevention, Diagnosis and Early Intervention of Perinatal Depression. Within this framework, the Perinatal Depression (PD) survey was developed, aimed at identifying women at risk using the Edinburgh Postnatal Depression Scale (EPDS). An electronic application was created for questionnaire entry and management. It records EPDS results, enables access across Aziende Socio-Sanitarie Territoriali (ASST) facilities, provides mapping of services, and allows printing of awareness materials. All data are handled under strict privacy protections, ensuring anonymity. Screening activities involved multiple services: Maternal and Child Departments, Family Counseling, and Mental Health Departments. A shared protocol with ASSTs guarantees permanent screening activity in hospitals and community facilities. A cohort of resident mothers giving birth 2017–2023 (n = 164,767) in ATS Milano hospitals was constructed using anonymized Birth Assistance Certificates (CEDAP) and Hospital Discharge Records (SDO). Table 1 presents the main characteristics of the cohort of mothers. Table 1Characteristics of the cohort of resident mothers who gave birth in public hospitalsVariableCategoryN.%Year of delivery201725,78815.7%201825,23215.3%201923,98814.6%202022,53113.7%202122,99314.0%202222,45713.6%202321,77813.2%Age group14–172820.2%18–2411,4386.9%25–3486,05152.2%35–4465,10939.5%45–541,8441.1%> 55150.0% Missing values 28NationalityItalian117,78371.5%Foreign46,98428.5%Education levelUniversity degree75,13945.7%High school diploma62,21937.9%Lower secondary school24,93315.2%Primary school2,0061.2% Missing values 470Marital statusMarried97,29359.2%Single63,71438.8%Separated/Divorced/Widowed3,2262.0% Missing values 534Employment statusEmployed116,20970.5%Unemployed48,55829.5%PrimiparityYes63,88439.0%No100,03061.0%Total164,767 Table 1Characteristics of the cohort of resident mothers who gave birth in public hospitalsVariableCategoryN.%Year of delivery201725,78815.7%201825,23215.3%201923,98814.6%202022,53113.7%202122,99314.0%202222,45713.6%202321,77813.2%Age group14–172820.2%18–2411,4386.9%25–3486,05152.2%35–4465,10939.5%45–541,8441.1%> 55150.0% Missing values 28NationalityItalian117,78371.5%Foreign46,98428.5%Education levelUniversity degree75,13945.7%High school diploma62,21937.9%Lower secondary school24,93315.2%Primary school2,0061.2% Missing values 470Marital statusMarried97,29359.2%Single63,71438.8%Separated/Divorced/Widowed3,2262.0% Missing values 534Employment statusEmployed116,20970.5%Unemployed48,55829.5%PrimiparityYes63,88439.0%No100,03061.0%Total164,767 Characteristics of the cohort of resident mothers who gave birth in public hospitals The identification algorithm was adapted from the ARS Toscana model, which uses record linkage across multiple administrative healthcare databases to identify women at risk of perinatal mental health conditions (Puglia et al. 2020). In the original framework, women giving birth are identified through birth registries and linked, via a unique anonymous identifier, to multiple healthcare utilization data sources. These include hospital discharge records, emergency department visits, mental health service contacts, outpatient psychological and psychiatric services, and pharmaceutical prescriptions. Based on these sources, the algorithm defines a set of mental health–related events, including hospitalizations or emergency visits with psychiatric diagnoses, contacts with mental health services, and prescriptions of psychotropic medications. Women are classified as at risk if at least one of these events occurs within a predefined observation window. This approach enables population-level identification of clinically relevant cases based on routine healthcare utilization data and reflects a pragmatic method for identifying mental health vulnerability through real-world service utilization. For the present study, the algorithm was adapted to the ATS Milan context in several respects. First, the set of data sources was aligned with locally available healthcare utilization flows, including family counselling services and territorial psychiatry, which are not fully comparable with the original Toscana data structure. Second, diagnostic classifications and service codes were harmonized with the Lombardy administrative coding system. Third, the observation window for events was defined from pregnancy to 12 months after delivery, in line with the study objectives, whereas the original model primarily focuses on the post-partum period rather than explicitly covering the full perinatal timeframe. Despite these adaptations, the core logic of the algorithm—based on the occurrence of at least one mental health–related event—was maintained. These events were identified within a cohort of mothers residing in the ATS Milano territory who gave birth between 2017 and 2023. The process relied on de-identified keys, ensuring that individual privacy was fully protected and no personal identification was possible. In order to consider a woman at risk of perinatal depression, the following conditions were defined (Table 2) and were researched in the period from pregnancy to 12 months after childbirth. Table 2Definition of event related to risk of perinatal depression presence of at least one of the following conditions:Condition TypeDescriptionHospitalizationAt least one hospital admission with any of the following ICD-9 diagnoses:− 295: Schizophrenic psychoses− 296: Episodic mood disorders− 297: Delusional disorders− 298: Other non-organic psychoses− 300.3: Obsessive-compulsive disorders− 300.4: Dysthymic disorder− 308: Acute stress reaction− 309: Adjustment reaction− 311: Depressive disorder not elsewhere classified− 648.4: Mental disorders antepartum/postpartum- Or any contact with the Department of Mental HealthEmergency Room AccessAt least one ER visit with one of the diagnoses listed aboveOutpatient ServicesAt least one visit for:- Psychotherapy sessions- Clinical psychological interviews- Administration of psychodiagnostic tests- Psychiatric consultationsFamily Counseling ServicesAt least one visit for:- Psychological evaluation interview- Support interviewa- Group psychotherapy- Individual or couple psychotherapy- Family psychotherapyDrug PrescriptionsAt least one prescription for:- Psycholeptics (ATC code N06)- Antipsychotics (N05)- Antiepileptics (N03)aIt is important to highlight that among the services provided by the counseling center, Supportive Interviews were also considered. According to the Regional Resolution DGR 6131/2017 of Lombardy, these interventions are aimed at individuals, couples, or families to provide assistance during crisis situations and to address developmental and transitional tasks throughout the life cycle, which require enhancing individual and relational skills. We observed that including or excluding these interviews led to a significant variation in the estimation of the phenomenonThe choice of these conditions is consistent with the ARS Toscana algorithm and reflects a pragmatic approach to identifying perinatal mental health vulnerability using routinely collected healthcare data. While these indicators do not correspond to a formal clinical diagnosis, they enable the identification of women experiencing significant psychological distress or accessing mental health-related care. Table 2Definition of event related to risk of perinatal depression presence of at least one of the following conditions:Condition TypeDescriptionHospitalizationAt least one hospital admission with any of the following ICD-9 diagnoses:− 295: Schizophrenic psychoses− 296: Episodic mood disorders− 297: Delusional disorders− 298: Other non-organic psychoses− 300.3: Obsessive-compulsive disorders− 300.4: Dysthymic disorder− 308: Acute stress reaction− 309: Adjustment reaction− 311: Depressive disorder not elsewhere classified− 648.4: Mental disorders antepartum/postpartum- Or any contact with the Department of Mental HealthEmergency Room AccessAt least one ER visit with one of the diagnoses listed aboveOutpatient ServicesAt least one visit for:- Psychotherapy sessions- Clinical psychological interviews- Administration of psychodiagnostic tests- Psychiatric consultationsFamily Counseling ServicesAt least one visit for:- Psychological evaluation interview- Support interviewa- Group psychotherapy- Individual or couple psychotherapy- Family psychotherapyDrug PrescriptionsAt least one prescription for:- Psycholeptics (ATC code N06)- Antipsychotics (N05)- Antiepileptics (N03)aIt is important to highlight that among the services provided by the counseling center, Supportive Interviews were also considered. According to the Regional Resolution DGR 6131/2017 of Lombardy, these interventions are aimed at individuals, couples, or families to provide assistance during crisis situations and to address developmental and transitional tasks throughout the life cycle, which require enhancing individual and relational skills. We observed that including or excluding these interviews led to a significant variation in the estimation of the phenomenon Definition of event related to risk of perinatal depression presence of at least one of the following conditions: aIt is important to highlight that among the services provided by the counseling center, Supportive Interviews were also considered. According to the Regional Resolution DGR 6131/2017 of Lombardy, these interventions are aimed at individuals, couples, or families to provide assistance during crisis situations and to address developmental and transitional tasks throughout the life cycle, which require enhancing individual and relational skills. We observed that including or excluding these interviews led to a significant variation in the estimation of the phenomenon SAS Software was used to conduct statistical analyses. As stated before, in the computation of the prevalence estimates, the cohort of mothers residing in the ATS who gave birth between 2017 and 2023 (n = 164,767) was used as a starting point to cross-reference the data on conditions associated with a risk of perinatal depression found in the HCU data. At the same time, data from the questionnaires administered as part of the Perinatal Depression survey were used to construct the cohort of mothers identified through the screening activity (n = 20,871). This cohort served both for estimating prevalence and for conducting the sensitivity analysis of the applied regression model. A flow diagram (Fig. 1) was developed to clarify the study population and the distinct analytical use of the full cohort and the EPDS-screened subgroup. Fig. 1Flow diagram study population Flow diagram study population The analysis work with regressions started by performing simple logistic analyses. Using a multivariate logistic model, the risk of developing events attributable to perinatal depression was studied by including also the presence of events associated with psychological distress in the 12 months preceding the start of pregnancy. Socio-demographic variables (age, citizenship, educational qualification, employment status, marital status) and pregnancy-related information (primiparity) were also considered in the analysis. Maternal age was categorized into three groups based on the distribution of the study population and to reflect clinically relevant thresholds used in the literature, particularly distinguishing younger and advanced maternal age groups. For regression analyses, maternal age was categorized as < 25, 26–39, and ≥ 40 years. Obstetric variables such as delivery type, fetal outcomes, and length of hospital stay were not included in the analysis. The model was intentionally restricted to a limited set of socio-demographic and prior mental health–related variables to identify broad population-level determinants of risk and support the development of targeted public health interventions. To validate and compare the findings on risk and protective factors, a second regression model was developed using the cohort of mothers identified by the PD survey as the reference group. The outcome variable was the result of the Edinburgh Test (score ≥ 8 vs. score < 8) and the independent variables were the same as the first model. The backward method was used to select the set of variables entered in both models and an automatic option was implemented to exclude records with missing values in any of the independent variables, given that the number of affected cases was minimal. Potential multicollinearity among predictors was assessed by examining correlations between variables and the stability of model estimates. Backward selection was applied to reduce redundancy among predictors, and no evidence of problematic collinearity was observed. Several model performance metrics were computed, acknowledging that the primary objective of the analysis was exploratory, aimed at identifying risk and protective factors for intervention rather than constructing a predictive model. Results The algorithm applied to the six administrative archives identified 30,330 episodes compatible with perinatal psychological distress within the 2017–2023 maternity cohort. Table 3 presents the distribution of PPD-related events across data sources. The majority of events were identified through family counselling services (n = 19,770; 65.2%), followed by pharmaceutical records (n = 3,714; 12.2%) and outpatient services (n = 3,659; 12.1%). Table 3Events by data sourceData SourceN.%Outpatient services (AMB)3,65912.1%Hospital Discharge Records (SDO)1,8856.2%Emergency Department (ED)1710.6%Pharmaceutical records (FARMA)3,71412.2%Family Counseling Services (CONS)19,77065.2%Territorial Psychiatry Services (46-SAN)1,1313.7%Total30,330100.0%Table 4 presents the characteristics of mothers stratified by socio-demographic variables and pregnancy-related information with and without PPD-related events. Table 3Events by data sourceData SourceN.%Outpatient services (AMB)3,65912.1%Hospital Discharge Records (SDO)1,8856.2%Emergency Department (ED)1710.6%Pharmaceutical records (FARMA)3,71412.2%Family Counseling Services (CONS)19,77065.2%Territorial Psychiatry Services (46-SAN)1,1313.7%Total30,330100.0% Events by data source Table 4Characteristics of mothers with and without PPD-related eventsVariableMothers with PPD-related eventMothers without PPD-related eventN.%N.%Age group 14–171060.4%1760.1% 18–241,8367.0%9,6026.9% 25–3412,86248.7%73,18952.9% 35–4411,10642.1%54,00339.0% 45–544781.8%1,3661.0% > 5540.0%110.0% Missing values = 28Nationality Italian19,21572.8%98,56871.2% Foreign7,17927.2%39,80528.8%Education level University degree12,89748.9%62,24245.0% High school diploma9,17634.8%53,04338.3% Lower secondary school3,94815.0%20,98515.2% Primary school2841.1%1,7221.2% Missing values = 89Marital status Married14,50855.0%82,78559.8% Single11,17442.3%52,54038.0% Separated/Divorced/Widowed6122.3%2,6141.9% Missing values = 89Employment status Employed18,62770.6%97,58270.5% Unemployed7,76729.4%40,79129.5%First-time mother Yes11,74944.5%52,13537.7% No14,49154.9%85,53961.8% Missing values = 853Total26,394100%138,373100% Table 4Characteristics of mothers with and without PPD-related eventsVariableMothers with PPD-related eventMothers without PPD-related eventN.%N.%Age group 14–171060.4%1760.1% 18–241,8367.0%9,6026.9% 25–3412,86248.7%73,18952.9% 35–4411,10642.1%54,00339.0% 45–544781.8%1,3661.0% > 5540.0%110.0% Missing values = 28Nationality Italian19,21572.8%98,56871.2% Foreign7,17927.2%39,80528.8%Education level University degree12,89748.9%62,24245.0% High school diploma9,17634.8%53,04338.3% Lower secondary school3,94815.0%20,98515.2% Primary school2841.1%1,7221.2% Missing values = 89Marital status Married14,50855.0%82,78559.8% Single11,17442.3%52,54038.0% Separated/Divorced/Widowed6122.3%2,6141.9% Missing values = 89Employment status Employed18,62770.6%97,58270.5% Unemployed7,76729.4%40,79129.5%First-time mother Yes11,74944.5%52,13537.7% No14,49154.9%85,53961.8% Missing values = 853Total26,394100%138,373100% Characteristics of mothers with and without PPD-related events Overall, 26,394 mothers—corresponding to 16.0% of the total cohort of 164,767—had at least one perinatal depression–related event. This estimate should be interpreted conservatively, since women seeking help only in the private sector are not captured in HCU data. The socio-demographic breakdown presented in Table 4 shows a heterogeneous distribution across groups. Mothers with PPD-related events were more frequently single (42.3% vs. 38.0%) and primiparous (44.5% vs. 37.7%). Screening data from the Perinatal Depression Survey provided an independent picture. Among 20,871 mothers who completed at least one EPDS during the study period, 23.6% scored eight or more, the Lombardy threshold for high vulnerability. Using the cutoff of nine, the proportion was 18.5%, while with a score above fourteen—indicating probable major depressive disorder—the estimate was 5.6%. To examine individual associations more formally, unadjusted logistic regressions were first undertaken (Table 5). Table 5Simple logistic regressionsSimple logistic regressionOR95% CIp-valueAge 26–391.0# Under 251.030.98–1.070.3 Over 401.291.24–1.34< 0.0001Psychological distress Yes8.017.63–8.41< 0.0001 No1.0#Origin Italian1.0# Migratory background0.940.92–0.97< 0.0001Education Bachelor’s degree1.0# High school diploma0.830.81–0.86< 0.0001 Middle/Primary school0.900.87–0.93< 0.0001Marital status Married1.0# Unmarried1.221.18–1.25< 0.0001Employed Yes1.000.97–1.030.9 No1.0#Primiparity Yes1.331.29–1.37< 0.0001 No1.0## reference categoryPsychological distress recorded in the twelve months preceding conception emerged as the single strongest predictor, multiplying the odds of a perinatal depressive-related event by approximately eight (OR = 8.01; 9% CI 7.63–8.41; p < 0.0001). Maternal age above forty raised the odds by 29% (OR = 1.29) whereas age under twenty-five was not significant after grouping. Migratory background showed a small but statistically significant protective effect in the unadjusted analysis (OR = 0.94), an observation that will be revisited in the multivariable context. Both educational levels below a bachelor’s degree and being unmarried were associated with elevated odds, while employment status did not reach significance. Primiparity was associated with higher odds by one-third (OR = 1.33). Table 5Simple logistic regressionsSimple logistic regressionOR95% CIp-valueAge 26–391.0# Under 251.030.98–1.070.3 Over 401.291.24–1.34< 0.0001Psychological distress Yes8.017.63–8.41< 0.0001 No1.0#Origin Italian1.0# Migratory background0.940.92–0.97< 0.0001Education Bachelor’s degree1.0# High school diploma0.830.81–0.86< 0.0001 Middle/Primary school0.900.87–0.93< 0.0001Marital status Married1.0# Unmarried1.221.18–1.25< 0.0001Employed Yes1.000.97–1.030.9 No1.0#Primiparity Yes1.331.29–1.37< 0.0001 No1.0## reference category Simple logistic regressions # reference category Below we report the results of the first multivariate logistic regression model, presenting as dependent variable having had an event associated with the risk of perinatal depression, derived from Health Care Utilization (HCU) data sources (Table 6). Table 6Estimated odds ratios and 95% confidence intervals – Model n.1EffectEstimate (OR)95% CIp-valuePsychological distress Yes vs. No8.137.668.62< 0.0001Over 40 vs. 26–391.311.251.38< 0.0001Under 25 vs. a 26–390.940.880.990.03Migratory Background vs. Italian1.121.071.16< 0.0001High school diploma vs. bachelor’s degree0.810.780.84< 0.0001Middle/Primary school vs. bachelor’s degree0.860.810.90< 0.0001Unmarried vs. Married1.201.161.24< 0.0001Employed vs. Unemployed0.920.880.960.0004Primiparous vs. Multiparous1.421.371.47< 0.0001 Table 6Estimated odds ratios and 95% confidence intervals – Model n.1EffectEstimate (OR)95% CIp-valuePsychological distress Yes vs. No8.137.668.62< 0.0001Over 40 vs. 26–391.311.251.38< 0.0001Under 25 vs. a 26–390.940.880.990.03Migratory Background vs. Italian1.121.071.16< 0.0001High school diploma vs. bachelor’s degree0.810.780.84< 0.0001Middle/Primary school vs. bachelor’s degree0.860.810.90< 0.0001Unmarried vs. Married1.201.161.24< 0.0001Employed vs. Unemployed0.920.880.960.0004Primiparous vs. Multiparous1.421.371.47< 0.0001 Estimated odds ratios and 95% confidence intervals – Model n.1 The Type III effects were found to be statistically significant based on Wald’s Chi-square test, as well as the analysis of maximum likelihood estimates. Firstly, it should be noted that events associated with psychological distress in the 12 months preceding the start of pregnancy increase by 8 times the risk of perinatal depression. Secondly, it is appropriate to consider the issue of age: particular attention should be paid to women over 40 as the odds of having events associated with perinatal depression rises by 31.3% compared with the 26–39 bracket. Migration background was significantly associated with perinatal depressive events. Women with a migrant background had 1.12 times higher odds (OR = 1.117; 95% CI: 1.073–1.164; p < 0.0001) of experiencing depressive symptoms compared to women of Italian origin. The last risk factor to be considered is parity: being primiparous compared with not being primiparous associated with higher odds for perinatal depression by 42%. Having lower educational qualifications seems to decrease the probability of developing such events. Employment appears to be a modest protective factor, with employed women showing 7.6% lower odds of experiencing perinatal depressive events compared to unemployed women (OR = 0.924; 95% CI: 0.885–0.965; p = 0.0004). The developed model shows a modest discrimination ability, with a c-statistic (AUC) around 0.62 in both the training and validation datasets. Goodness-of-fit statistics of the model, such as Somers’ D-value (0.24) and Gamma (0.25), confirm a positive but moderate association between predicted probabilities and observed responses. The error rate is around 15–16%, while the value of Brier’s Score (around 0.125) suggests a fair accuracy of the probabilistic predictions. A second regression model was also created by selecting only the cohort of mothers intercepted by the PD Survey. The Edinburgh Test Result (not at-risk EPDS < 8 /at risk EDPS > = 8) was used as the response variable and the same variables as in the previous model were included. The idea behind this was to verify that the profile of the woman at risk overlaps with that of the general cohort. Table 7 displays the results of Model n.2. Table 7Estimated odds ratios and 95% confidence intervals – Model n.2EffectEstimate (OR)95% CIp-valuePsychological distress Yes vs. No2.261.982.57< 0.0001Migratory Background vs. Italian1.431.311.55< 0.0001High school diploma vs. bachelor’s degree0.890.830.960.002Middle/Primary school vs. bachelor’s degree0.880.790.980.02Unmarried vs. Married1.191.111.27< 0.0001Employed vs. Unemployed0.910.840.990.03Primiparous vs. Multiparous1.151.081.23< 0.0001Therefore, the same risk and protective factors are highlighted, although with slightly different odds ratios. Previous events attributable to psychological distress proves to be crucial in predicting the outcome (OR:2.26), as well as having foreign origins (OR:1.43). Other risk characteristics concern being single (OR:1.19) and first-time mother (OR:1.15). Lower educational qualifications and having a job are confirmed as protective factors. Table 7Estimated odds ratios and 95% confidence intervals – Model n.2EffectEstimate (OR)95% CIp-valuePsychological distress Yes vs. No2.261.982.57< 0.0001Migratory Background vs. Italian1.431.311.55< 0.0001High school diploma vs. bachelor’s degree0.890.830.960.002Middle/Primary school vs. bachelor’s degree0.880.790.980.02Unmarried vs. Married1.191.111.27< 0.0001Employed vs. Unemployed0.910.840.990.03Primiparous vs. Multiparous1.151.081.23< 0.0001 Estimated odds ratios and 95% confidence intervals – Model n.2 The goodness-of-fit statistics of the model suggest that the estimated logistic model is statistically valid, although its predictive ability is low overall. The log-likelihood of the model is -11281.7, with an error rate of 23.6%. The information indices AIC (22579.5) and BIC (22643.1) are high, in line with the large sample size analyzed (n = 20,871), and are mainly useful for comparison with alternative models. The two analytical approaches—algorithmic surveillance in administrative sources and direct EPDS screening—converge on the same constellation of determinants, lending robustness to the observed associations despite the modest c-statistics that preclude individual prognostication based on these variables alone. Discussion and conclusions Perinatal depression is associated with adverse pregnancy outcomes and with long-term emotional, social, and cognitive difficulties in children. It is also linked to increased maternal morbidity and mortality, imposing a considerable burden on women, families, and society at large (Grigoriadis et al. 2013). Our prevalence estimates based on HCU data (16%) and on EPDS screening (18–24%) are consistent with international evidence. They align with pooled global postpartum prevalence of 17.4% and with the 16.3% reported in Southern Europe, as well as with Italian survey data showing 24.5% of women at risk when using the ≥ 9 threshold (Camoni et al. 2022; Pan et al. 2024). The Perinatal Depression Survey of ATS of Milan confirmed that 23.6% of the mothers screened were classified as being in a condition of high vulnerability (EPDS ≥ 8). Using the 9-point cutoff, 18.5% of women were identified as at risk of perinatal depression. When restricting the analysis to women with an EPDS score above 14, this estimate decreased to 5.6%. All these findings are particularly significant when considering the hidden cases- those who do not access services or who instead turn to private providers. According to a recent global systematic review and meta-analysis (Wang et al. 2021), the global pooled prevalence of post-partum depression is expected to be 17.4%, with considerable geographical variations. Subgroup analyses also showed that in Southern Europe this estimate is as high as 16.3%. In a survey conducted in Italy between 2021 and 2022 (Camoni et al. 2022), EPDS scores equal to or above the cutoff of 14—indicating a probable condition of depression—were considered. An overall prevalence of 6.9% was observed, with significant differences between the two periods analyzed. Women scoring ≥ 14 are 9.0% in the prepartum period and 4.2% in the postpartum period, respectively. Considering all mothers with Edinburgh Scale scores of 9 or more, it appears that nearly a quarter (24.5%) of the sample is at risk of experiencing depressive episodes during the perinatal period. Lastly, a recent meta-analysis conducted on prevalence studies of perinatal depression in Italy (Camoni et al. 2023) reports that in postpartum studies using EPDS cut-off scores ≥ 9 the overall prevalence estimate was 27.5% (95% CI 17.8–37.3), while the pooled prepartum risk of depression prevalence was 20.2% (CI 95% 15.3–24.5). The risk factor profile aligns neatly with international evidence. A history of psychological distress during the year before conception emerges as the most powerful predictor. This finding reinforces the case for integrating reproductive‑age mental‑health monitoring into routine primary care, thereby offering anticipatory guidance long before pregnancy is confirmed. Advanced maternal age, migrant background, unmarried status and primiparity each contribute additional risk. A recent umbrella review (Kim et al. 2022) highlighted antenatal anxiety, psychological and intimate partner violence, smoking during pregnancy, violence of any kind experienced, primiparity and unplanned pregnancies as risk factors. Another review (Hutchens and Kearney 2020) recognized prenatal depression, current or past abuse, high levels of life stress, lack of social support and dissatisfaction in the relationship with the partner as the most common risk factors. The association with migration merits specific attention because it intertwines clinical vulnerability with structural barriers. Languages, differing perceptions of mental health and fear of stigmatisation can impede access precisely where screening and early care would be most beneficial. The literature suggests that approximately 20% of immigrant women experience symptoms of postpartum depression, and that they are 1.5 to 2 times more likely to develop this condition than non-immigrant women. (Falah-Hassani et al. 2015) It is therefore essential to consider how cultural factors shape perceptions of perinatal depression. Due to cultural factors, stigma surrounding perinatal depression may hinder help-seeking, driven by fears of being seen as unfit mothers, social judgment, and norms discouraging emotional disclosure. (Dela Cruz et al. 2022) Two results appear counterintuitive at first glance. Lower educational attainment retained a mild protective coefficient, while employment status also showed a small but significant protective effect. A plausible explanation is that highly educated, or professionally active women are more likely to use public mental health services, generating higher detection, while women with fewer resources may rely on informal networks. International studies have reported similar patterns. At national level (Camoni et al. 2022) found that women with lower than secondary education reported less anxiety compared to those with higher education. Regarding employment, precarious economic status has often been identified as a risk factor (Leigh and Milgrom 2008; Yang et al. 2022; Milgrom et al. 2019). Both logistic regression models showed statistically significant associations between the risk of perinatal depression and the variables included in the models. The models yielded consistent results in terms of direction and statistical significance of the covariates, suggesting that the associations identified between maternal characteristics and the risk of depression are robust across different outcome measures. With respect to model evaluation metrics, some limitations of the study should be acknowledged. From a predictive performance perspective, both models showed relatively low area under the curve (AUC) values, as well as low McFadden’s R². However, this limited predictive performance does not constitute a substantial limitation in the context of the analysis, which aims not at individual prediction, but rather at identifying systematic associations between socio-demographic determinants and depressive outcomes. The use of the same set of covariates also enabled a methodologically rigorous comparison between the two outcomes, highlighting the consistency of the results obtained from different data sources. This consistency strengthens the validity of the identified factors and contributes to a deeper understanding of the risk dynamics of perinatal depression. This study has some limitations that should be considered when interpreting the findings. The use of healthcare utilization (HCU) data may capture only part of the actual burden of perinatal depression, as it does not include women who seek care in private settings or those who do not access services at all. At the same time, it may also include events not fully attributable to perinatal depression, potentially leading to some misclassification. Second, the EPDS screening data were available only for a subgroup of women, reflecting incomplete coverage of perinatal mental health screening. This may limit the generalizability of the survey-based findings and introduce potential selection bias. Finally, the exclusion of obstetric variables represents a limitation of the present analysis, as clinical factors related to pregnancy and delivery may contribute to the risk of perinatal depression. Future studies could benefit from integrating this information to better characterize clinical pathways and refine risk stratification. In conclusion, perinatal depression in the territory of ATS Milano affects at least one woman in six and therefore constitutes a major population‑level challenge. The condition is not randomly distributed; it clusters around identifiable determinants – a recent history of psychological distress, older maternal age, migrant origin, absence of a stable partner, and first‑time motherhood – each of which indicates specific entry points for preventive action. Health policy should translate this knowledge into tangible measures integrated throughout the continuum of maternal and childcare. Systematic screening needs to be embedded in every routine contact, from the first antenatal visit to infant vaccination and paediatric well‑baby checks. Stepped pathways that begin with brief psycho‑educational or digital interventions for mild symptomatology and escalate to specialised perinatal mental‑health teams for complex presentations provide a framework that is both evidence‑based and scalable. Equity must guide all decisions. Outreach strategies employing cultural mediators, multilingual resources and peer mentors can bridge linguistic and social barriers that currently hamper migrant women’s access. Data integration is equally vital. Extending the Perinatal Depression survey to interface with electronic medical records, social‑service information systems and child‑development databases would allow real‑time monitoring of outcomes and resource utilisation, supporting learning loops that refine practice considering emerging evidence. Future work will build on this foundation by incorporating additional psychosocial variables and exploring longitudinal trajectories of maternal mental health beyond the first postpartum year. The goal is to inform policy choices that protect the wellbeing of mothers and children, reduce avoidable morbidity and foster healthier communities across Lombardy.