1 INTRODUCTION

The strong protective effect of breastfeeding for neonatal and maternal health is well recognized (Eidelman & Schanler, 2012; Victora et al., 2016). The World Health Organization (WHO) recommends exclusive breastfeeding during the first 6 months of a newborn's life, complemented with other appropriate foods, at least until the age of two (World Health Organization, 2019). Far from the WHO recommendations, the exclusive breastfeeding rates at 6 months postpartum are only around 39% in Spain (Ministerio de Sanidad Consumo y Bienestar Social, 2017).

Premature breastfeeding cessation has varied and complex origins. On an individual level, psychological factors such as postpartum depression, anxiety, maternal intention to breastfeed during pregnancy, and breastfeeding self-efficacy are critical (De Jager et al., 2013). Dennis defines breastfeeding self-efficacy as the confidence in, or self-perception of, a woman's ability to breastfeed (Dennis, 2003). Breastfeeding self-efficacy is mainly influenced by the mother's previous breastfeeding experience, learning by observation, physiological and affective conditions, and verbal persuasion (Dennis & Faux, 1999). In addition, breastfeeding self-efficacy is an important predictor of breastfeeding behavior during the first 6 months of a newborn's life. Women with low self-efficacy levels are less likely to start breastfeeding and to continue when they encounter difficulties (Minas & Ganga-Limando, 2016; Tuthill et al., 2016).

The 20-item Prenatal Breastfeeding Self-Efficacy Scale (PBSES) was developed as a tool for determining breastfeeding self-efficacy during pregnancy (Wells et al., 2006), when most women decide on their infant feeding method (Johnson et al., 2019). Thus, this tool can be useful for identifying risk groups and evaluating educational programs to promote breastfeeding. The PBSES considers the pregnant woman's confidence in getting information about breastfeeding from her social and health environment, in contrast to other tools designed for use during the postnatal period (Tuthill et al., 2016). The PBSES has been adapted into Spanish (Piñeiro-Albero et al., 2013) and Turkish (Aydin & Pasinlioglu, 2018; Hazar & Akca, 2018), demonstrating adequate internal consistency and evidence of discriminant and predictive validity. The PBSES scores are related to previous breastfeeding experience, breastfeeding intention during pregnancy, and breastfeeding initiation after childbirth (Piñeiro-Albero et al., 2013; Robinson & VandeVusse, 2011; Wells et al., 2006).

There is limited evidence regarding the structure of PBSES. A first exploratory analysis suggested a four-factor scale (Wells et al., 2006). However, later confirmatory factor analyses indicated marginal fit statistics for this structure even after modifying the composition of the factors or deleting one of the items (Hazar & Akca, 2018; Piñeiro-Albero et al., 2013). Further studies providing new evidence on the PBSES structure are recommended. The continuous improvement of existing measurement tools can be achieved using different approaches depending on the specific measurement objective. By choosing the most pertinent items to assure predictive validity, a criterion-keying strategy can be used for developing existing tools when the authors expect to have a high predictive value regarding a well-defined criterion (Smith et al., 2003), as in the case of the PBSES (Wells et al., 2006). Moreover, shorter tools are helpful in studies that require respondent adherence, such as those that use multiple measurement tools or require follow-up (Stanton et al., 2002). Thus, developing a more parsimonious version of the PBSES, that is, with the least amount of information necessary to achieve its measurement goal (Wieland et al., 2017), could increase response rates by reducing the burden on the respondent (Edwards et al., 2009). Therefore, in the current study, we aimed to assess the structural validity and psychometric characteristics of the PBSES after reducing its items according to the criteria for incremental validity and parsimony.

2 METHODS 2.1 Participants

This cross-sectional instrumental study with a longitudinal component was part of more extensive research on infant feeding conducted in six hospitals in eastern Spain. Two participating hospitals were designated as Baby Friendly Hospitals—a multilevel intervention aimed at promoting breastfeeding (World Health Organization, 2009).

A convenience sample of 1218 pregnant women in their third trimester, who completed the PBSES, was included. They all participated voluntarily and provided informed consent. All participants were able to read and speak Spanish and had no problems that could seriously complicate or contraindicate breastfeeding, such as previous breast surgery, human immunodeficiency virus infection, or congenital fetal pathology diagnosed during pregnancy. During the postpartum follow-up, women who had preterm or multiple births, or who had medical problems that prevented or seriously hindered breastfeeding, were excluded. Approval was obtained from the research ethics committees of the participating hospitals.

2.2 Measures

To assess the relationships of the PBSES with theoretically related external criteria, as recommended for validation studies (Stanton et al., 2002), we obtained data from the variables and scales related to the construct measured. In addition to the PBSES, two additional measures of self-efficacy—one general and one specific—were included. In addition, we obtained data on maternal history related to breastfeeding and information on actual breastfeeding behavior after childbirth.

The PBSES is a scale designed to assess a woman's judgment of her capacity to organize and execute the actions required to perform breastfeeding behavior and to explore the role of prenatal self-efficacy in predicting breastfeeding (Wells et al., 2006). It includes 20 items, assessed with a 5-point Likert scale, ranging from 1 (not at all sure) to 5 (completely sure), and a total score ranging between 20 and 100. The structure proposed for the Spanish version of the scale (PBSES-e), used in this study, grouped the PBSES items into four dimensions with a second-order factor: Skills and demands required for breastfeeding (eight items), gathering information about breastfeeding (five items), breastfeeding around other people and feelings of embarrassment during breastfeeding (four items), and social pressure when breastfeeding (three items). The scale had a Cronbach's α coefficient for internal consistency of 0.91 (Piñeiro-Albero et al., 2013). In this current work, the scale had internal consistency reliability of 0.89.

The Breastfeeding Self-Efficacy Scale-Short Form (BSES-SF; Dennis, 2003) is a one-dimensional 14-item scale developed to measure the mother's confidence in her ability to breastfeed after birth. All items are written in a positive wording and are completed using a 5-point Likert scale, where 1 indicates “not confident at all” and 5 indicates “very confident.” In its Spanish version, the scale has good internal consistency with a Cronbach's α of 0.92 (Oliver-Roig et al., 2012). In this current work, the scale had internal consistency reliability of 0.93.

General self-efficacy was assessed using the Spanish version of the General Self-Efficacy Scale (GSES-e). The GSES-e is a 10-item scale that determines participants' personal perception of adequately managing stressful situations in everyday life, with a Cronbach's α of 0.87 (Sanjuán et al., 2000). In this current work, the scale had internal consistency reliability of 0.88.

Sociodemographic and pregnancy variables concerning breastfeeding were obtained from a questionnaire that included age, educational level, marital status, family income, parity, previous experience of breastfeeding, duration of the previous breastfeeding, breastfeeding decision during pregnancy, and intention to breastfeed during pregnancy.

In addition, data were obtained on the moment when breastfed babies first received additional water-based fluids, breast milk substitutes, and complementary foods at least once a day, and on the moment of breastfeeding cessation. We used the following definitions of breastfeeding proposed by the WHO: Exclusive breastfeeding occurs when infants are fed solely on breast milk, although oral medications or rehydration solutions are allowed. Any breastfeeding is when infants receive any amount of breast milk, whether accompanied by other foods or liquids or not (World Health Organization, 2008).

2.3 Data collection

Data were collected at four different time periods. First, during a pregnancy checkup at 28–42 weeks, participants were given a series of self-report questionnaires, including the PBSES-e and the GSES-e, along with the sociodemographic and pregnancy questionnaires.

Second, during postpartum hospitalization, participants completed a self-administered questionnaire between 2 and 4 days after birth, including the BSES-SF and data on the initiation and type of breastfeeding during their hospitalization. Information on obstetric variables was also obtained from clinical records.

Furthermore, to obtain breastfeeding data after discharge, we collected data at 5 and 12 months postpartum. Five months after delivery, a self-administered free postal questionnaire was sent to mothers breastfeeding their newborns at discharge. The questionnaire was sent up to three times, at 3-week intervals, to those participants who did not respond. Finally, at 12 months postpartum, we conducted a telephone survey with mothers who said they were breastfeeding in the 5-month questionnaire. The self-administered questionnaires and the telephone survey included questions about the time in months from birth that their babies were first offered nonnutritive liquids or foods other than breast milk at least once a day. If mothers were not breastfeeding on the survey day, we asked them when they had stopped breastfeeding.

2.4 Data analysis

A descriptive analysis of the study variables was performed using frequencies and percentages for discrete variables, arithmetic means, and standard deviations for continuous variables. In addition, we explored the item reduction of the PBSES according to common statistical criteria. Finally, as recommended by Stanton et al. (2002), we developed a reliability assessment and a revalidation of the short form to assess the extent to which it replicated the relation patterns referred to in previous studies.

2.4.1 Item reduction and the incremental validity strategy

We focused the item reduction on evaluating both internal and external item qualities (Stanton et al., 2002). We did not consider judgmental item quality criteria such as clarity, relevance, required reading level, and items' “face” validity, because these aspects were evaluated previously (Piñeiro-Albero et al., 2013).

The internal item qualities were assessed with reference to the scale itself (Stanton et al., 2002). We chose the ceiling effect (the proportion of responses for the highest score of 5) as a distributional criterion to increase variability and interitem correlations as a correlational criterion to decrease redundancy (Nunnally & Berstein, 1994; Terwee et al., 2007). A ceiling effect occurs when more than 15% of responses are for the highest score of a test or measurement (Terwee et al., 2007). Although there was no ceiling effect for the total PBSES scores, all items, except item 9, individually demonstrated a high ceiling effect (Piñeiro-Albero et al., 2013). Thus, as a norm-referenced criterion (Glass & Stanley, 1970), we eliminated items with a ceiling effect ≥60% above the upper quartile of the proportion of responses in the higher score for all items. Furthermore, we deleted one item from item pairs with interitem correlations above 0.80—a criterion indicating redundant content (Nunnally & Berstein, 1994). Finally, the usual criteria for maximizing internal consistency, such as corrected item-total correlations or Cronbach's α if an item was deleted (Stanton et al., 2002), were not applied, since no reliability problems were previously described for the PBSES.

In addition, regarding the external item qualities, as an incremental validity strategy, we chose the most pertinent items to maximize the PBSES's ability to predict relevant breastfeeding outcomes (Smith et al., 2003). Thus, we deleted those items whose scores did not demonstrate statistically significant differences in relevant breastfeeding status variables indicating risk peaks for breastfeeding cessation in Spain (Oliver-Roig et al., 2008): initiation of breastfeeding, exclusive breastfeeding at discharge, exclusive breastfeeding at 1 month, exclusive breastfeeding at 5 months, any breastfeeding at 6 months, and any breastfeeding at 1 year.

2.4.2 Factor analysis and structural validity

The sample was divided into two random cohorts for the analysis of the first reduced version structure. In the first half of the sample (n = 609), we performed an exploratory factorial analysis (EFA) using the unweighted LS method with a matrix of polychoric correlations and applying a promin rotation to achieve factor simplicity (Lorenzo-Seva, 1999). We also included a parallel analysis to determine the number of factors (Timmerman & Lorenzo-Seva, 2011). After performing the EFA, we deleted items with loadings less than 0.3, indicating a weak relationship between the item and the assigned factor (Stanton et al., 2002). Thus, we obtained a second reduced version of the scale (PBSES-SF), for which we conducted all of the following reliability and validity analyses.

In the second half of the sample (n = 609), a confirmatory factorial analysis (CFA) was performed using the least squares (LS) normal theory estimator based on the covariance matrix (Bentler, 2004). Four options were considered in the search for the model that best fitted the data: (1) the first model examined the four-factor structure with a second-order factor of the Spanish PBSES, including all the original items (Piñeiro-Albero et al., 2013); (2) the second model assessed the structure suggested by the parallel analysis including only the PBSES-SF items; (3) the third model explored the PBSES structure including only the PBSES-SF items; (4) and the fourth model explored the PBSES structure with a second-order factor, including only the PBSES-SF items. To calculate the fit indices of the CFA, we considered the factors as correlated. The models were assessed using the nonnormed fit index, goodness-of-fit index, comparative fit index, root mean squared error of approximation, and standardized root mean square residual (Hu & Bentler, 1999; Kline, 2015).

2.4.3 Reliability and construct and predictive validity assessment

Reliability was evaluated through internal consistency analysis using Cronbach's α. We also explored discriminant and convergent validity as two subtypes of construct validity. We expected similar or better validity evidence than the full-length scale. First, to assess discriminant validity, known group comparisons were conducted with the following hypotheses: (1) women who had previous breastfeeding experience and (2) those who chose to breastfeed their baby during pregnancy would have greater breastfeeding self-efficacy than women without experience or those who decided not to breastfeed their baby, respectively. Second, to assess convergent validity, we explored the relationship between PBSES-SF scores, (3) the expected time of breastfeeding, (4) the GSES scores, and (5) the BSES-SF scores at discharge.

Finally, predictive validity was evaluated by comparing the PBSES-SF scores of participants according to their breastfeeding status at follow-up cross-sections proposed for the incremental validity strategy (breastfeeding initiation, exclusive breastfeeding at discharge, exclusive breastfeeding at 1 month, exclusive breastfeeding at 5 months, any breastfeeding at 6 months, and any breastfeeding at 1 year postpartum). Additionally, we used receiver operating characteristic (ROC) curves to determine the overall predictive performance of the PBSES-SF and compare it to that of the PBSES, differentiating participants who wanted to breastfeed during pregnancy and those who actually breastfed after childbirth from those who did not. Youden's J statistic was also used to identify the optimal cutoff points when the area under the ROC curve (AUC) of the PBSES-SF or its factors were approximately 0.70, the value from which a classifier was deemed good (Staffa & Zurakowski, 2019).

We used Student's t-test and Pearson's correlations to evaluate the hypotheses and examine their assumptions. In the case of the Student's t-test, if the variances were unequal, we calculated the Welch method to adjust for degrees of freedom and repeated the analysis with Mann–Whitney's U test. In addition to Pearson's correlations, we calculated Spearman's correlations. In the case of discrepancies in the results, we described the nonparametric test results as recommended by Altman (1991). We calculated ROC curves based on a binomial distribution using maximum likelihood estimation and compared the AUC parameters using the χ2 test.

3 RESULTS 3.1 Sample

Of the 1218 participants recruited during pregnancy, 35 (2.9%) were excluded due to preterm deliveries. Figure 1 illustrates the evolution of the follow-up period. Table 1 illustrates the characteristics of the initial sample and the information obtained regarding breastfeeding status at the babies' different ages. We obtained complete information during the first year postpartum on the total duration of exclusive breastfeeding for 993 (83.9%) participants and any breastfeeding for 659 (55.7%) participants, including women who did not initiate breastfeeding. The mean age of the sample was 31.74 years (SD = 4.77). The main characteristics of the study participants are listed in Table 1. Descriptive data of the PBSES-e are presented in Table 2.

Flow chart of the study

Table 1. Characteristics of the validation sample: Qualitative variables (N = 1218) Variable n % Educational status Primary school or lower 568 46.9 High school 256 21.1 University degree 388 32.4 Marital status De facto or married 1081 89.2 Single/divorced 130 10.7 Widow 1 0.1 Family annual income <18.000 €/year 648 57.4 >18.000 €/year 480 42.6 First child Yes 648 53.4 No 569 46.8 Previous breastfeeding experience Yes 479 39.5 No (including participants without a previous child) 733 60.5 Duration of the previous breastfeeding 5 months or less 178 37.2 More than 5 months 300 62.8 Decision on infant feeding method during pregnancy Breastfeeding 1094 90.6 Bottle feeding 61 5.0 Not decided 53 4.4 Time planned to breastfeed during pregnancyb 5 months or less 122 21.0 6–11 months 252 43.4 12 months or more 202 35.6 Attended in a Baby Friendly Hospital Yes 283 23.2 No 935 76.8 Birth methodc Vaginal 753 75.8 Cesarean 241 24.2 Initiation of breastfeeding Yes 1054 91.4 No 99 8.6 Type of breastfeeding at postpartum dischargec Exclusive breastfeedingd 618 64.9 Predominant breastfeedinge 9 0.9 Partially breastfeedingf 223 23.4 Formula feeding 102 10.7 Exclusive breastfeeding at 1 monthc,g Yes 783 82.5 No 166 17.5 Exclusive breastfeeding at 5 monthsc,g Yes 338 35.7 No 608 64.3 Any breastfeeding at 6 monthsc,g Yes 432 56.5 No 332 43.5 Any breastfeeding at 12 monthsc,g Yes 180 25.6 No 523 74.4 aNumbers may not add up to total because of missing data. bOnly if decided to breastfeed. cExcluding recruited participants with premature childbirths (n = 1183). dExclusive breastfeeding was considered if infants had only received human milk during the follow-up. ePredominant breastfeeding was considered if infants had only received human milk and rehydration solutions, drops, or syrups, during the follow-up. fBreast milk and formula feeding. gData after discharge period, excluding participants that not initiated breastfeeding (n = 1054). Table 2. Floor and ceiling effects of the PBSES, M, and SD (n = 1218) Ítem Floora (%1) Ceilingb (%5) M SD 1. I can find the information I need about problems I have breastfeeding my baby 0.9 25.1 3.61 0.98 2. I can find out what I need to know about breastfeeding my baby 0.2 29.9 3.79 0.94 3. I know who to ask if I have any questions about breastfeeding my baby 1.0 35.3 3.89 1.00 4. I can talk to my partner about the importance of breastfeeding my baby 1.6 65.5 4.40 0.96 5. I can talk to my health care provider about breastfeeding my baby 0.2 58.0 4.37 0.85 6. I can schedule my day around the breastfeeding of my baby 4.4 21.0 3.37 1.12 7. I can make the time to breastfeed my baby even when I feel busy 2.5 31.6 3.66 1.14 8. I can breastfeed my baby even when I am tired 1.1 37.1 3.85 1.07 9. I can breastfeed my baby when I am upset 10.7 12.2 2.72 1.17 10. I can use a breast pump to obtain milk 4.2 29.7 3.60 1.15 11. I can prepare breast milk so others can breastfeed my baby 8.9 25.4 3.36 1.27 12. I can breastfeed my baby even if it causes mild discomfort 3.3 25.6 3.51 1.13 13. I can breastfeed my baby without feeling embarrassed 3.0 46.4 3.93 1.19 14. I can breastfeed my baby when my partner is with me 0.4 80.5 4.72 0.64 15. I can breastfeed my baby when my family or friends are with me 3.5 50.7 3.99 1.21 16. I can breastfeed my baby around people I do not know 15.4 29.5 3.18 1.46 17. I can call a lactation counselor if I have problems breastfeeding 7.6 25.0 3.30 1.26 18. I can choose to breastfeed my baby even if my partner does not want me to 0.8 64.4 4.44 0.85 19. I can choose to breastfeed my baby even if my family does not want me to 0.7 70.3 4.54 0.79 20. I can breastfeed my baby for 1 year 10.6 32.8 3.32 1.41 PBSESc 0.1 1.2 75.54 12.57 PBSES-SFd 0.1 1.6 43.37 8.58 Abbreviations: M, mean; PBSES, Prenatal Breastfeeding Self-Efficacy Scale; PBSES-SF, short form of the PBSES; SD, standard deviation. aProportion of women responding to the item with 1 (not at all sure). bProportion of women responding to the item with 5 (completely). c20-Item Spanish version of the Prenatal Breastfeeding Self-Efficacy Scale. d12-Item Spanish version of the Prenatal Breastfeeding Self-Efficacy Scale.

Statistically significant differences were observed between women who participated and those who did not participate in the follow-up after delivery. Women who did not participate were younger, t(220,7) = 4.26, p < 0.01; had lower educational status, χ2(1, N = 1078) = 17.032, p < 0.01; and decided not to breastfeed during the third trimester of pregnancy to a greater extent, χ2(2, N = 1074) = 14.29, p < 0.01. No statistically significant differences were observed between those participating and those not participating in the follow-up regarding having previous children, χ2(1, N = 1083) = 0.59; marital status, χ2(1, N = 1077) = 2.07; birth method, χ2(1, N = 900) = 1.78; having babies born in a Baby Friendly Hospital, χ2(1, N = 1084) = 0.90; and GSES (U = 73,804.0), and PBSES-e scores (U = 73,006.0).

3.2 Item reduction

We eliminated items 4, 14, 18, and 19 due to a ceiling effect ≥ 60% (Table 2) and items 10 and 11 because there was no relationship between any of the proposed breastfeeding status variables after childbirth (Table 3). We also found an interitem correlation above 0.80 between items 1 and 2 and eliminated item 2 because it was less predictive.

Table 3. Prenatal Breastfeeding Self-Efficacy Scale and its Short-Form