Interview procedures and setting (Consolidated Criteria for Reporting Qualitative Research (COREQ): items 10–15, 17–23)

Interviews were conducted face-to-face in private space within the child welfare clinics at the three facilities, with only the participant and interviewer present; infants were present when breastfeeding was required (no other non-participants were present). After each survey, women with a Patient Health Questionnaire-9 (PHQ-9) score >9 (indicative of PPDS), as automatically identified by the electronic data capture tool, were asked about interest in a follow-up in-depth interview by the trained research assistants. The research assistant took contact and location details from interested women. Women signed an informed consent form before participating in the interview. We encouraged participants eligible for the interview to see a clinician or a nurse to discuss PPDS symptoms and to return to the clinic the next day for an in-depth interview (no repeat interviews). Participants were informed that their care would not be affected by their responses.

A semi-structured interview guide (developed from the literature and the study objectives) covered topics such as experiences during childbirth, postpartum emotional well-being, support from family and community, stressors and coping strategies (Supplementary Table 1 available at https://doi.org/10.1192/bjo.2025.10857). The guide was pilot-tested with two postpartum women from a non-study clinic to refine wording and flow. The average duration of the interview was 45 min. We recorded, with permission, all interviews with a Sony ICD digital voice audio recorder. Brief field notes were written immediately after each interview to capture contextual details and initial analytic impressions. We discussed data sufficiency at regular intervals and concluded that thematic saturation had been reached when no new themes emerged in the final interviews. Transcripts were not returned to participants for correction because of time constraints and to minimise participant burden. Instead, we prioritised clear in-interview summaries and probes to enhance credibility.

We gave GHC45 (approximately US$3) to all survey and interview participants as appreciation for their time. Their familiarity with the culture helped elicit the most accurate responses. ^{Reference Sibbald, Phelan, Beagan and Pride14}

Research team and reflexivity (COREQ: items 1–8)

Interviews were conducted by L.T. (female, PhD candidate in Mental Health), supported by two female research assistants fluent in Ewe and English. The core team comprised L.T. (PhD candidate), S.B. (PhD, biostatistics), A.S. (ScD, social science), A.D.J. (PhD, midwifery science) and J.H. (PhD, perinatal mental health). At the time of the study, team members were employed at academic or research institutions in Ghana, The Netherlands and Zambia. Interviewers received study-specific training on trauma-informed interviewing, confidentiality and reflexive practice. They conducted mock interviews and reviewed role-plays before data collection. No prior relationships existed with participants. During recruitment, the interviewer introduced herself as a researcher studying postpartum well-being, emphasising the voluntary nature of participation and that there were no right or wrong answers. We maintained a reflexive log to document positionality, assumptions and analytic decisions. The team’s clinical and public health backgrounds were acknowledged as potential sources of interpretive bias. To mitigate these, we used peer debriefs and multidisciplinary review of codes and themes.

Outcome and covariates

The PHQ-9 was used to measure our primary outcome, defined as screening positive for PPDS. The PHQ-9 is a nine-item scale with responses scored on a four-point scale (0 = not at all; 1 = several days; 2 = more than half the days; 3 = nearly every day). The total score ranges from 0 to 27. We used a PHQ-9 score >9 as the cut-off to indicate elevated depressive symptoms, based on evidence indicating this threshold provides strong predictive validity. ^{Reference Kroenke, Spitzer and Williams15} Although the Edinburgh Postnatal Depression Scale (EPDS) is a widely recognised tool specifically designed for PPDS, we selected the PHQ-9 scale because of its broader applicability, strong correlation with the EPDS and demonstrated reliability and validity for detecting PPDS, with a sensitivity of 0.84 and specificity of 0.81. ^{Reference Wang, Kroenke, Stump and Monahan16} Similarly, the PHQ-9 has been validated in Ghana (e.g. Kintampo district) with satisfactory reliability (Cronbach’s alpha of approximately 0.75) and concordant performance with the EPDS. ^{Reference Weobong, Akpalu, Doku, Owusu-Agyei, Hurt and Kirkwood17} Importantly, the PHQ-9 offers practical advantages in this context, being simple, easy to administer and well-suited to Ghanaian women, particularly those with limited formal education, where its straightforward format and accessibility enhance its acceptability and feasibility. ^{Reference Weobong, Akpalu, Doku, Owusu-Agyei, Hurt and Kirkwood17}

We collected sociodemographic data (e.g. age, education, marital status, employment, etc.), pregnancy and obstetric history (e.g. parity, pregnancy planning), and psychosocial factors (e.g. partner support, stressful life events, self-esteem) that are potentially related to PPDS based on the literature. ^{Reference Agrawal, Mehendale and Malhotra5,Reference Alshahrani, Al-Saleh, Siddiqui, Khalil, Alsaleem and AlHefdhi10} We defined ‘in union’ as being married or cohabiting, and ‘not in union’ as women who were single, divorced, separated or widowed. Cohabiting women were grouped with married women because of sociocultural similarity in support structures in Ghana. We asked mothers a structured question, ‘Has your husband/partner been supportive since delivery? (e.g. providing emotional, financial or practical help)’. We defined ‘partner support’ as ‘supportive’ if moderate/strong support was reported, and ‘unsupportive’ if the mother reported no or poor support from husband/partner. Pregnancy was classified as planned if the mother reported intending the pregnancy; otherwise, it was ‘unplanned’. Low birth weight was defined as <2.5 kg. We asked women if they had any paid work (employment or self-employment) at the time of participating in the survey. Those answering ‘yes’ to any paid work were coded as ‘employed or engaged in paid work’; others (including housewives without income) were coded as ‘unemployed or not working’.

We used the Childbirth Experience Questionnaire 2 (CEQ2) to measure mothers’ experiences during childbirth and labour. The CEQ2 consists of 22 items. For 19 items, we used a four-point Likert scale (1 = totally agree, 2 = mostly agree, 3 = mostly disagree, 4 = totally disagree). Higher scores indicate a negative experience. We assessed three additional items (i.e. labour pain, sense of security and control) using a visual analogue scale (VAS). We categorised VAS scores as follows: 0–40 = 1, 41–60 = 2, 61–80 = 3 and 81–100 = 4, with higher scores indicating a more positive experience. ^{Reference Anna Dencker, Bergqvist, Lilja and Berg18,Reference Walker, Dencker and Thornton19} We reversed items with increasing scores, indicating positive experience before calculating the total score, summing all individual four-point items, and expressed as a percentage. All 22 items (19 Likert items, three converted VAS items) were summed for a total score (minimum 22, maximum 88) and expressed as a percentage. Based on the CEQ2 manual and aligning with the four-point Likert scale, having a median of 50%, we classified CEQ2 scores >50% as negative childbirth experience and ≤50% as positive childbirth experience. ^{Reference Anna Dencker, Bergqvist, Lilja and Berg18,Reference Walker, Dencker and Thornton19}

We used the Stressful Life Event Screening Questionnaire to assess stressful life events experienced by mothers throughout their lives. This 13-item questionnaire includes binary (yes/no) responses, and a higher scores indicate greater exposure to stressful life events. ^{Reference Goodman, Corcoran, Turner, Yuan and Green20} Moderate-to-high stress was defined as scores ≥4. To measure self-esteem, we used the Rosenberg Self-Esteem Scale (RSE), a ten-item scale with four response options (strongly agree, agree, strongly disagree, disagree). We reversed negative items and calculated the total score by summing all individual four-point items, with higher scores indicating higher self-esteem. ^{Reference Rosenberg21} We defined low self-esteem as scores ≤14. ^{Reference Zabel, Goldbach, Fröhlich, Schneider, Görgülü and Oertel22} We pre-tested the instruments for comprehension; however, formal validation of the CEQ2 in Ghana was not performed.

Quantitative data analysis

We summarised background characteristics of mothers by using frequency and proportions for categorical variables, whereas median and interquartile interval (IQI) were used for continuous variables. We used the Pearson chi-squared test or Fisher’s exact test, as appropriate, to assess the association of potential factors associated with PPDS. To develop a parsimonious model of factors independently associated with PPDS, variables with a P-value less than or equal to 0.05 were considered as candidates for inclusion in the multivariable binomial generalised linear model (GLM). Variables were removed if P-values were larger than 0.1. We chose the binomial GLM with a log-link to estimate the prevalence ratio because the odds ratio would overestimate the effect owing to the high prevalence of PPDS in our sample. Missing responses were rare (i.e. one item-nonresponse) because of electronic validation checks; analyses were performed on complete cases. All analyses were performed using Stata 18 MP for macOS.

Qualitative analysis and trustworthiness (COREQ: items 24–28; reporting 29–32)

We used reflexive thematic analysis following Braun and Clarke’s six steps: (1) familiarisation with data, (2) generation of initial codes, (3) searching for themes, (4) reviewing themes, (5) defining and naming themes and (6) writing the report. ^{Reference Braun and Clarke23} Interviews (n = 2) were recorded in the local language (Ewe) and translated into English. All interviews were transcribed verbatim, and detailed notes were compiled, ensuring that all responses were labelled with unique participant identifiers. After familiarisation with the interview data, transcripts and notes, L.T. developed a coding framework. L.T. coded the data systematically, tagging relevant sections with descriptive codes, combining inductive development of data-driven codes with deductive attention to constructs identified a priori from the literature and the quantitative findings. These codes were then grouped into broader themes. A.S. independently reviewed a subset of the transcripts, and discrepancies were discussed with A.D.J., J.H. and S.B. until interpretive consensus was reached. Finally, L.T. synthesised the results into a coherent narrative, highlighting key themes illustrated by representative maternal quotes. We managed data and coding using Dedoose version 9.0 for macOS (Sociocultural Research Consultants, LLC, Los Angeles, California, USA; see https://www.dedoose.com).

We provided an excerpt of the coding tree with definitions and example quotations in Supplementary Table 2. We assessed credibility through integration with the quantitative results. We compared statistically significant factors with emergent qualitative themes, via a joint display and narrative, to interpret how these findings corroborate or extend one another. ^{Reference Fetters, Curry and Creswell12} Although member-checking of full findings with participants was not undertaken to reduce burden, we incorporated in-interview summaries and clarification probes, and we present diverse and negative cases alongside the major themes to support dependability and confirmability. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (Supplementary Table 3) in reporting the cross-sectional component, ^{Reference von Elm, Altman, Egger, Pocock, Gøtzsche and Vandenbroucke24} whereas the qualitative findings followed the COREQ checklist ^{Reference Booth, Hannes, Harden, Noyes, Harris and Tong25} (Supplementary Table 4).