Evaluation of antenatal risk factors for postpartum depression: a secondary cohort analysis of the cluster-randomised GeliS trial

The GeliS study is a prospective, multicentre, cluster-randomised, controlled, open intervention trial that primarily aimed to reduce the proportion of women with excessive GWG as defined by the US Institute of Medicine (IOM) [28]. Secondary aims were to reduce the risk for adverse perinatal and postpartum complications, such as PPD, and to improve behavioural outcomes, such as physical activity, dietary, and breastfeeding behaviour [23]. Details about the design, setting, population, and randomisation process have been described elsewhere [23].

In brief, women with (1) a pre-pregnancy BMI between ≥ 18.5 kg/m² and ≤ 40.0 kg/m², (2) a singleton pregnancy, (3) age between 18 and 43 years, (4) sufficient German language skills, and (5) stage of pregnancy before the end of the 12^th week of gestation were recruited between 2013 and 2015. The recruitment was conducted in gynaecological and midwifery practices in five administrative regions of Bavaria (Germany) depicting the ‘real-life’ setting of antenatal routine care. All participants gave their written informed consent for participation.

Participants in the control group obtained routine antenatal care and additionally general information on a healthy antenatal lifestyle by means of a flyer. Participants in the intervention group received a comprehensive lifestyle intervention programme. This programme consisted of three antenatal and one postpartum face-to-face counselling sessions on a healthy pre- and postnatal lifestyle according to current recommendations for the antenatal and postpartum period [29‐31]. The counselling sessions were given by previously trained midwives, medical personnel, or gynaecologists alongside routine care visits. Details on the counselling content have already been reported [23].

The study was performed in accordance with the current local regulatory requirements and the Declaration of Helsinki. The Ethics Commission of the Technical University of Munich approved the study protocol. The trial is registered at the ClinicalTrials.​gov Protocol Registration System (NCT01958307).

All baseline characteristics including sociodemographic information were collected at the time of recruitment (before the end of the 12^th week of gestation) using a screening questionnaire. Pre-pregnancy BMI was calculated based on the self-reported pre-conception weight. Having a BMI between 18.5 and 24.9 kg/m² was defined as being normal weight, between 25.0 and 29.9 kg/m² as being overweight, and between 30.0 and 40.0 kg/m² as having obesity. Based on details on the educational level collected via the screening questionnaire, participants were grouped into having a ‘lower educational level’ if they at least completed high school and into the ‘higher educational level’ category if they held a university degree.

During the course of pregnancy, maternal weight data were collected by means of routinely used maternity records. GWG was defined as the difference between the last measured weight before delivery and the first measured weight at the time of recruitment. Excessive GWG was defined according to the thresholds provided by the IOM [28] considering the woman’s pre-pregnancy BMI category. The optimal GWG ranges for women with normal weight were 11.5–16.0 kg, for women with overweight 7.0–11.5 kg, and for women with obesity 5.0–9.0 kg. Gaining weight above these thresholds was defined as excessive GWG [28]. Between the 24 and 28 weeks of gestation, a 2-h oral glucose tolerance test was performed for the screening and diagnosis of gestational diabetes mellitus. According to national and international recommendations [32, 33], gestational diabetes mellitus was diagnosed if one of the following thresholds was equalled or exceeded: fasting plasma glucose, 92 mg/dL (5.1 mmol/L); 1-h value, 180 mg/dL (10.0 mmol/L); and 2-h value, 153 mg/dL (8.5 mmol/L). Pre-pregnancy or early pregnancy lifestyle factors, such as smoking status, physical activity level, intake of alcohol, and mental health state, were inquired in a set of questionnaires that was answered by participants directly after inclusion (before the end of the 12^th week of gestation). This set of questionnaires contained a slightly modified version of the validated food frequency questionnaire developed for the ‘German Health Examination Survey for Adults’ (DEGS) study by the Robert Koch Institute, Berlin, Germany [34], which was used to group women according to ‘any alcohol consumption’ and ‘no alcohol consumption’. Moreover, it comprised the validated ‘Pregnancy Physical Activity Questionnaire’ (PPAQ) [35] that was slightly adapted to German habits. Thereby, participants had to estimate the mean time spent engaging in 32 activities in the past month. As described in the evaluation instructions of this questionnaire [35], calculated average weekly energy expenditure in MET-h/week was summed up into the category ‘total physical activity of light intensity and above’. The median MET-h/week value of this variable was used to group participants into having a ‘low level of physical activity’ or a ‘high level of physical activity’. Furthermore, the set of questionnaires comprised questions of anxiety and depressive symptoms by using validated ‘Patient Health Questionnaire for Depression and Anxiety’ (PHQ-4). It comprised four items with a 4-point scale to screen for depression and anxiety. The composite PHQ-4 total score ranges from 0 to 12, and scale scores of ≥ 3 were suggested as cut-off points of probable cases of depression or anxiety [36].

Between 6 and 8 weeks postpartum, symptoms of PPD were assessed using the validated German version of the ‘Edinburgh Postnatal Depression Scale’ (EPDS) [37]. Showing symptoms of PPD was defined as having an EPDS score ≥ 13 and is in the following described as ‘having PPD’.

A power calculation was performed based on the primary study outcome excessive GWG and was described elsewhere [23]. We found no between-group differences in the history of anxiety and depressive symptoms in early pregnancy or in the history of PPD (Additional file 1). Therefore, we made the post hoc decision to pool data from both the intervention and control groups and considered the group assignment as a covariate.

We performed complete-case analyses as defined a priori [24] and included participants with GWG data, EPDS data, and covariate data available. We excluded those participants who had a preterm delivery (< 37^th delivery). All statistical analyses were performed in SAS version 9.4 (SAS Institute Inc., Cary, NC), and p values below 0.05 were considered as statistically significant.

Overall, 2286 women were enrolled in the GeliS study. Among them, 1684 were eligible for the present analyses and 1583 of them provided information on all covariates (Fig. 1). Excluded participants differed in terms of mean GWG, parity, history of gestational diabetes, marriage status, smoking, and physical activity habits from the study sample as outlined in Additional file 2.

The baseline study population had a mean age of 30.4 ± 4.4 years (Table 1). In total, 1047 (66.1%) women were in the normal weight category, 352 (22.2%) had a BMI between 25.0 and 29.9 kg/m² and thus overweight, and 184 (11.6%) had a BMI ≥ 30.0 kg/m² and thus obesity (Table 1). In the postpartum period, 7.9% (n = 138) participants had PPD among whom 16.7% (n = 23) had obesity and 53.6% (n = 74) had excessive GWG. The prevalence of overweight and obesity was higher in the subgroup of women with PPD compared to women without PPD (Table 1). Moreover, the rate of excessive GWG tended to be higher in women with PPD (Table 1). Table 1 summarises the sociodemographic, lifestyle, metabolic, and psychological characteristics of the participants according to PPD status. Participants who experienced PPD were more likely to have a lower educational level, be unmarried, smoke during early pregnancy, and suffer from antenatal anxiety and depressive symptoms. Moreover, the proportion of women with a university degree was lower in the subgroup of women with PPD. There were no significant differences in age, parity, alcohol consumption, living conditions, physical activity level, and gestational diabetes mellitus status between women with and without PPD.

Women with obesity had a lower GWG in comparison to women with normal weight or overweight (mean ± SD: obesity, 11.0 ± 6.7 kg; overweight, 14.0 ± 5.7 kg; normal weight, 14.7 ± 4.5 kg; p < .001). Additional file 3 shows the characteristics of women according to their pre-pregnancy BMI category.

Overall, 45.6% (n = 722) of participants showed excessive GWG according to the IOM criteria. The proportion of overweight and obesity was higher in the subgroup of women with excessive GWG compared to non-excessive GWG counterparts. This also applied for having a history of antenatal anxiety and depressive symptoms (Additional file 4).

Table 2 shows the multivariable regression models for the association between pre-pregnancy BMI and PPD presented with the odds ratios (ORs) and 95% CIs. Pre-pregnancy BMI (per 5-unit increment) was positively associated with the odds of experiencing PPD (model 1: OR = 1.25, 95% CI = 1.10–1.44, p = 0.03), indicating that a 5-kg/m² increase in BMI corresponded to a 25% increase in the odds of having PPD. The association between pre-pregnancy BMI and PPD remained stable after adjusting for concurrent risk factors (full model OR = 1.23, 1.08–1.41, p = 0.002). Being married significantly decreased the odds of PPD (model 4: OR = 0.70, 95% CI = 0.54–0.91, p = 0.04), whereas a low educational level was positively associated with the odds of PPD (model 4: OR = 1.41, 95% CI = 1.02–1.94, p = 0.01). Among all, history of anxiety or depressive symptoms led to the highest odds of experiencing PPD (full model OR = 3.42, 95% CI = 2.42–4.82, p < .001). There was no significant evidence for associations between additional sociodemographic and lifestyle factors or GDM status and the odds of experiencing PPD symptoms. The QIC statistics revealed that the final model was the preferred model fit indicated by the smallest QIC values (data not shown).

In multivariable logistic regression analyses, the odds of experiencing PPD significantly increased with increasing BMI category. Compared to the reference weight category (normal weight), being in the overweight and obesity weight category was associated with increasing odds of PPD (Table 3; overweight: OR = 1.72, 95% CI = 1.15–2.57, p < 0.01; obesity: OR = 1.91, 95% CI = 1.16–3.14, p = 0.01). This association remained significant after adjustment for further sociodemographic or lifestyle factors, and GDM (model 3:overweight, BMI 25.0–29.9 kg/m²: OR = 1.78, 95% CI = 1.18–2.70, p < 0.01; obesity, BMI 30.0–40.0 kg/m²: OR = 1.80, 95% CI = 1.07–3.05, p = 0.03). Moreover, this association remained significant after full adjustment for antenatal history of anxiety or depressive symptoms (full model or model 4: overweight: OR = 1.72, 95% CI = 1.13–2.62, p = 0.01; obesity: OR = 1.76, 95% CI = 1.04–2.99, p = 0.04).

Table 4 shows associations between GWG or excessive GWG and the odds of experiencing PPD in relation to concurrent risk factors. A 1-kg increase in total GWG increased the odds of experiencing PPD with a borderline statistical significance (Table 4, model 1: OR = 1.19, 95% CI = 1.00–1.43, p = 0.05). However, GWG (per 1-kg increase) was not significantly associated with PPD after adjustments for potential confounders (Table 4, model 4: OR = 1.16, 95% CI 0.94–1.44, p > 0.05). Further analyses disclosed that excessive GWG was significantly and positively associated with the odds of experiencing PPD when adjusted for age and group allocation (OR = 1.39, 95% CI = 1.10–1.76, p = 0.006). The association remained significant after adjustment for sociodemographic factors, parity, lifestyle factors, gestational diabetes mellitus, and history of anxiety or depressive symptoms during pregnancy (OR = 1.31, 95% CI = 1.06–1.61, p = 0.01) with history of anxiety or depressive symptoms being the most prominent determinant (OR = 3.36, 95% CI 2.46–4.58, p < .0001). However, excessive GWG was not independently associated with the odds of PPD when accounting for the interaction between GWG and BMI. Thus, excessive GWG was not significantly associated with PPD in the final model (Table 4, model 4: OR = 3.48, 95% CI 0.35–34.94, p > 0.05). The full model had the smallest QIC statistics (data not shown).

There was no significant evidence of a non-linear association between GWG (continuous) and PPD as calculated in the logistic regression by using restricted cubic splines (ß, standard error, p values: knot 1, − 0.05, 0.04, 0.17; knot 2, 0.02, 0.03, 0.42; knot 3, − 0.03, 0.05, 0.60). When GWG was assessed in three subgroups (inadequate or excessive vs. adequate (reference group)), there was a dose-response increment for the odds of having PPD, albeit non-significant (Additional file 5).

We further investigated the modifying role of antenatal history of anxiety or depressive symptoms on the association between pre-pregnancy obesity and PPD, as shown in Fig. 2. The fully adjusted logistic regression model demonstrated a significant statistical interaction between antenatal history of anxiety or depressive symptoms and pre-pregnancy obesity on PPD (p value for interaction term = 0.03). No significant interaction of antenatal history of anxiety or depressive symptoms and GWG on PPD was observed (p = 0.22).

Figure 2 shows results of the logistic regression analyses on the association between pre-pregnancy BMI category and PPD stratified by antenatal history of anxiety or depressive symptoms. Pre-pregnancy overweight and obesity significantly increased the odds of experiencing PPD, but only in the subgroup of women with history of anxiety or depressive symptoms. In this subpopulation, the odds for developing PPD amounted up to 1.93 in women with overweight and 2.11 in women with obesity (fully adjusted model, overweight: OR = 1.93, 95% CI = 1.15–3.22, p = 0.01; obesity: OR = 2.11, 95% CI = 1.13–3.96, p = 0.02). Thus, both women with overweight and obesity also having a history of anxiety or depressive symptoms during pregnancy had an approximately 2-fold increased risk of experiencing PPD compared with women with normal pre-pregnancy weight and antenatal history of distress.

Our study was limited by the recruitment criteria excluding women with underweight. The self-reported pre-pregnancy BMI might have led women to underreport their initial weight [53]. We acknowledge that quantitative analyses might reveal the potential contribution of bias introduced by the self-reports of pre-pregnancy weight [38, 54]. Weight during the course of pregnancy was measured in several study centres which might have introduced some inaccuracies. Through our approach of defining pregnancy weight gain with two measures (at inclusion and at birth), we did not consider trimester-specific pattern of GWG and the definite timing of exceeding IOM criteria during the course of pregnancy [38]. We acknowledge that assessing the contribution of longitudinal weight gain may provide further insights into the interplay between GWG and occurrence of PPD and might be valuable to derive concrete implications for primary care. Although the EPDS is a validated questionnaire, we are aware that estimating the rate of women with a history of depressive symptoms using the EPDS might partly underestimate the actual incidence of PPD. Despite statistical significance, our modest OR values (below 2.0) may be of moderate clinical significance and thus should be interpreted with caution.

The strength of our study is based on the trial design. Data were collected within the routine antenatal care system and thus under real-life conditions. We longitudinally collected data over the course of pregnancy and were thus able to consider the contribution of various determinants to the development of PPD beyond crude weight data. We were also able to reach a sample of participants in both urban and rural regions. The relatively large sample size provides a comprehensive and valuable assessment of early predictors of PPD. Data are robust to adjustment for an appropriate set of covariates. By employing the EPDS, we used a validated, easily applicable, and widely used screening tool for PPD symptoms.