A cross-sectional study was conducted in rural areas of one prefecture in Sichuan Province. According to the National Bureau of Statistics of China, 48% of Sichuan’s population are rural residents. The average per capita disposable income in the rural areas of Sichuan is 13,331 RMB (1906 US dollars), far lower than the national average of 28,228 RMB (4033 US dollars) [27]. The study area can be considered relatively representative of rural southwestern China.

The study sample was drawn from the baseline sample of an interventional study assessing the impacts of a community health worker program on the health and nutrition of pregnant women, new mothers, and infants. The sampling strategy for this study was determined by power requirements to detect impacts of the intervention on infant hemoglobin concentrations, exclusive breastfeeding among infants under 6 months, and dietary diversity among infants older than 6 months (not reported in this paper). Using parameters from previous research, we assumed an intra-cluster correlation of .01 and a cluster size of 12 samples per township. Based on this, we calculated an overall sample frame of 80 townships with an average of 12 mother-child dyads per township.

The research team then implemented a 3-step sampling protocol to select households for the study. First, from the nine counties within the sample prefecture, the four nationally-designated “poverty counties” (pinkun xian) were included, which are counties that have more than 2% of their population living under the national poverty line of 3000 RMB (458 US dollars) [28]. These counties were selected in order to sample the mental health and breastfeeding outcomes of low-income rural mothers. Second, sample townships were chosen within each sample county. The sampling frame excluded non-rural townships and rural townships with populations less than 10,000. Of the remaining townships, 20 townships per county were randomly selected, resulting in a total of 80 rural townships. Third, the research team obtained a list of all households with pregnant women beyond their second trimester or with infants under 6 months old from the local county-level Maternal and Child Hospital in each sample county and selected a maximum of 25 eligible mother-child dyads per township. In townships with less than 25 eligible samples, the research team traveled to surrounding villages to enroll eligible participants until all eligible samples had been enrolled or a maximum of 25 samples was reached. In townships with more than 25 eligible samples, participants were randomly selected for enrollment until the maximum of 25 was reached.

Following this strategy, 1296 participants were sampled and 1233 agreed to participate in the survey. Because this study focuses on postnatal mental health and breastfeeding, the analytical sample excluded 353 pregnant women, 41 non-mother caregivers and 89 non-breastfeeding mothers. Eight respondents did not complete the survey and were also excluded. This resulted in a final sample of 742 breastfeeding mothers in 80 rural townships.

Data were collected from November to December 2019 by trained survey enumerators. During the training, enumerators were taught how to implement the survey instruments for each of the study’s main components. Once the training was completed, enumerators conducted one-on-one survey interviews at each sampled household. The survey collected data on postnatal mental health (symptoms of depression, anxiety, and stress), breastfeeding beliefs (attitudes and self-efficacy), breastfeeding practices, and demographic characteristics.

To measure the mental health of sample mothers, the research team used the Depression, Anxiety, and Stress Scale-21 (DASS-21). This is a 21-item short-form version of the DASS-42 that was originally created by Lovibond and Lovibond [29]. The DASS-21 scale has been validated in China [30, 31]. To complete the DASS-21, enumerators asked mothers to rank each statement (7 for each subsection) from 0 to 3, depending on how much the statement applied to them in the past week. Each individual’s DASS-21 score was then calculated by adding up the ranking for every sub-question in a specific section and multiplying the sum by 2. Hence, for each section, the total sum score could range from 0 to 42. Mothers who scored greater than or equal to 9 for depression, 7 for anxiety, and 14 for stress were considered to have symptoms of each respective mental health problem. It is important to note that the resulting score of the DASS-21 scale is not a clinical diagnosis but only a reasonable measurement of the severity of depression, anxiety, and stress symptoms.

Two measures for breastfeeding beliefs were collected: breastfeeding attitudes and breastfeeding self-efficacy. To measure breastfeeding attitudes, enumerators administered the Iowa Infant Feeding Attitude Scale [32], which has been used in Turkey [12], Ethiopia [33], the United Kingdom [34], Russia [34], and the United States [35]. It has also been validated in China [36]. Breastfeeding mothers were read 18 statements, such as “breastfeeding is more convenient than formula feeding” and asked to use a 5-point Likert scale to express if they agreed or disagreed with the statement (1 being strongly disagree and 5 being strongly agree). Total scores range from 18 to 90. Statements that were representative of worse attitudes were reverse coded so that a higher score correlates to a more positive attitude about breastfeeding.

Mothers’ breastfeeding self-efficacy was assessed using the short form of the Breastfeeding Self-Efficacy Scale [37]. It has previously been used in Kenya [38], Turkey [39], and Malaysia [40], and was validated in Chinese by Ip et al. [41]. Breastfeeding mothers were asked to rate how confident they were about 16 breastfeeding-related items (e.g., “I manage to keep up with my infant’s breastfeeding demands”) using a 5-point Likert scale with 1 being not at all confident and 5 being always confident. Total scores range from 16 to 80, with higher scores corresponding to higher levels of self-efficacy.

Breastfeeding practices were determined through a 24-h dietary recall questionnaire, in which enumerators asked mothers to list all the foods and liquids they fed their infant in the last 24 h. Based on this data, mothers were then divided into two feeding categories: exclusive breastfeeding and non-exclusive breastfeeding. As defined by the WHO, exclusive breastfeeding is breastfeeding while giving no other foods or liquids [42]. Non-exclusive breastfeeding in our study refers to any other feeding practices than exclusive breastfeeding, including predominant breastfeeding (feeding breastmilk along with other water or water-based liquids), and mixed breastfeeding (feeding breastmilk along with formula or animal milk, other liquids or solids).

Data on demographic characteristics of each sample family and infant were also collected by enumerators. Family characteristics included mother’s age, mother’s and father’s education levels, and family yearly income. Demographic characteristics of the infant were collected from the birth certificate, and included the infant’s gender, age in months, whether the infant had low birth weight, and whether the infant was born prematurely.

The statistical analysis for the study is composed of four parts. First, the research team calculated the overall prevalence of maternal depression, anxiety, and stress symptoms, as well as the overall prevalence of breastfeeding outcomes, which include breastfeeding practice and beliefs. Second, an adjusted ordinary least squares (OLS) regression was used to identify the correlations between mental health and breastfeeding beliefs, while controlling for other potential confounders. The specification of the adjusted model is:

$$}_=_0+_1\mathrm\_}_+}_+_$$

(1)

where Beliefi refers to breastfeeding beliefs (breastfeeding attitude or breastfeeding self-efficacy). Mental _ healthi is the dummy variable for mental health (depression, anxiety, or stress symptoms), which takes the value of 1 when motheri showed symptoms of mental health problems and takes the value of 0 when motheri did not. Controli is a set of control variables, including infant gender and age, whether the infant was born prematurely, whether the infant has low birth weight, maternal age and education level, paternal education level, and family income.

Third, a multivariate logistic regression was run to examine the associations between mental health problems and breastfeeding practices. The specification of the regression model is:

$$}_}=_0+_1\mathrm\_}_}+}_}+}_}$$

(2)

where Practicei takes the value of 1 when motheri was fully breastfeeding and takes the value of 0 when motheri was mixed breastfeeding. In Eq. (2), the definitions of the variables, Mental _ healthi and Controli are the same as those in Eq. (1).

Finally, a heterogeneous analysis was conducted to measure the effects of symptoms of maternal mental health problems on breastfeeding practice by five demographic characteristics, including three socio-economic status (SES) characteristics (education levels of each parent and family income), maternal age, and infant age. The specification of the adjusted model is:

$$}_}=_0+_1\mathrm\_}_}+_2\mathrm\_}_}+_3\mathrm\_}_}+}_}+}_}$$

(3)

where Mp _ lowi, Mp _ highi, and Mh _ lowi are three binary variables. Mp _ lowi takes the value of 1 when motheri shows symptoms of mental health problems and is of low SES/low maternal age/low infant age, and 0 otherwise. Mp _ highi takes the value of 1 when motheri shows symptoms of mental health problems and is of high SES/high maternal age/high infant age, and 0 otherwise. Mh _ lowi takes the value of 1 when motheri does not have symptoms of mental health problems and is of low SES/low maternal age/low infant age, and 0 otherwise. Practicei and Controli are the same as those in Eq. (1).