Validation of the Wijma delivery expectancy/experience questionnaire for pregnant women in Malawi: a descriptive, cross-sectional study

A descriptive, cross-sectional study of pregnant women (N = 212) in their third trimester was conducted at a district hospital in Malawi from August to September 2018, and was used for the exploratory factor analysis. For confirmatory factor analysis, a secondary analysis was conducted using data from the baseline time point (N = 264) of a prospective cohort study which followed pregnant women from 34-weeks’ gestation, postpartum 1 day, and 1 week, to examine the changes of childbirth fear and quality of life over time. The study was approved by the institutional review board of Taipei Medical University, Taiwan, and the National Health Science Research Committee, Ministry of Health, Malawi. Each participant provided written informed consent before participating in the study. Investigators read out the consent form and assisted participants to fill it out if participants were unable to read and write.

Pregnant women receiving prenatal care at the antenatal clinic were screened for their eligibility and recruited by the investigators. The study site was a primary hospital for a catchment area including 538,345 people, with approximately 5712 deliveries per year [23], and covers 22 health centers for medical care referral [24]. According to the 2015 Malawi Demographic Health Survey (MDHS), estimates of hospital deliveries in Malawi are around 91% [25]. Being the primary hospital that provides comprehensive maternity care, the study hospital is comparable to other district hospitals where childbirth services are provided.

A convenience sampling method was used in this study. The inclusion criteria of participants were as follows: aged 20 ~ 50 years, spoke and understood Chichewa, in their third trimester (≥34 weeks) of primipara or multipara, and with a singleton pregnancy. The exclusion criteria were women who had obstetric complications, including preeclampsia or hemorrhage, or had a history of medical or mental illness. Women who met the inclusion criteria and agreed to participate were included in the study. We used medical health records to confirm the eligibility of participants.

Considering the ratio of five participants per item for the factor analysis [26, 27], at least 165 women were required to complete the W-DEQ A scale. As the W-DEQ A has 33 items [1], a sample of 330 was needed for both the exploratory and confirmatory factor analysis. Taking into account the potential attrition rate and missing data of 30%, a sample size of approximately 470 was needed.

The Malawian version of the W-DEQ was assessed in two steps including translation process and psychometric analytical procedures (Table 1). In step 1, we translated the English version of the W-DEQ A to Chichewa language following recommended guidelines by Wild et al. [28]. In brief, the original English version of the W-DEQ A was translated to Chichewa by a professional bilingual translator. Malawian-speaking experts were then asked to review the translated version for appropriate wording and clarity. Back translation of the Chichewa version into English was subsequently completed by three independent bilingual translators. The final Malawian version of the W-DEQ A was obtained after comparing the original and back-translated questionnaires. A pilot test of 15 pregnant women was conducted to confirm the wording and adequacy of the final version at a prenatal clinic. The content validity of the Malawi W-DEQ A scale in areas of relevance and clarity of the instrument was evaluated by an expert group of four nurses and midwives in Malawi. A content validity index (CVI) of > 0.80 was considered as valid [29]. We then conducted the psychometric analysis of the W-DEQ A scale at the step 2 (Table 1).

Data were collected using the paper-based questionnaire with seventy-four questions and it approximately took 30 min to complete. For women who were unable to read and write, the investigator read out the questionnaire in Chichewa language and assisted them to fill it out.

Data analysis were conducted using the Statistical Package for the Social Sciences (SPSS) vers. 21 (SPSS, Chicago, IL, USA), and AMOS vers. 25 (SPSS, Chicago, IL, USA). We used descriptive statistics, including frequency, percentage, mean, and standard deviation (SD) to summarize participants’ characteristics. Validation of the W-DEQ A scale was assessed using exploratory (EFA) and confirmatory factor analysis (CFA) [37]. For EFA, the adequacy of the data was confirmed using the Kaiser-Meyer-Olkin (KMO) method with > 0.6 considered acceptable [38], and a significant Bartlett’s test of sphericity [39]. Principal component analysis and varimax rotation were used to extract factors and improve the interpretability of the solution. The number of factors retained was based on an eigenvalue of > 1 and an assessment of the scree plot [40]. A factor loading of ≥0.35 was used to identify whether an item satisfactorily represented its factor [41]. Items were eliminated one at a time, rerunning the analysis at each step to achieve an optimal solution [41]. For cross-loadings, if any, the rule of retaining an item to its factor was based on (a) whether the item’s loadings in the main factor are higher than loadings in others and (b) at least a difference of 0.20 between loadings [42].

The CFA was undertaken to determine the one-dimensional fit of the original W-DEQ A scale and to confirm the EFA structure solution identified in this study. A maximum-likelihood (ML) estimator was used to achieve robust results [43]. Goodness-of-fit indices were used to confirm the model fit, namely, a Chi-squared test (χ²) and its p-value (≥0.05 = acceptable fit; ≥0.10 = good fit), which is sensitive to large sample size. Therefore, the Chi-squared test divided by its degrees of freedom (χ²/df: ≤5 = acceptable fit; ≤3 = good fit) was considered [44, 45]. Other indices included root mean square error of approximation (RMSEA: ≤0.08 = acceptable fit; ≤0.05 = good fit), comparative fit index (CFI: ≥0.85 = acceptable fit; ≥0.95 = good fit), Tucker-Lewis index (TLI: ≥0.80 = acceptable fit; ≥0.95 = good fit), and the Akaike information criterion (AIC: the smaller, the better) [44, 45]. At least three adequacy indices with values within the acceptable ranges were considered in analyzing the goodness of fit of data [46, 47]. Internal consistency reliability was assessed using Cronbach’s α coefficient, and a level of ≥0.70 was acceptable [48]. Pearson’s correlation coefficient was used to assess the convergent and divergent validities. We tested the extent to which the W-DEQ A scale and its factors were correlated with other concepts of depressive symptoms and the QOL. We expected the W-DEQ A scale and its factors to be positively correlated with depressive symptoms and negatively correlated with domains of the QOL.

In total, 476 pregnant women were included in this study, with 212 women for EFA, 264 women for CFA, and a response rate of 100% for both analyses (Table 2). For EFA, the mean age of participants was 28.8 (SD = 7.3) years. The majority of women were older (≥25 years, 61.8%), married (97.6%), and illiterate or had a primary school education (78.3%) (Table 2). Most of the women were employed (96.2%), and more than half of them earned less than Malawi Kwacha (MK) 20,000/month (56.1%) (US$1 ≈ MK719). For CFA, the mean age of participants was 27.7 (SD = 6.4) years. Most women were older (≥25 years; 58.7%), married (95.5%), and illiterate or had a primary school education (70.5%) (Table 2). The majority of the women were employed (93.9%), and more than half of them earned less than MK20,000/month (53.0%). The t-test or χ² test showed EFA and CFA samples had similar characteristics (p = 0.05 ~ 0.50). The mean score of the 33 items of the W-DEQ ranged 0.57 ~ 2.92, with an absolute skewness of < 1.66 and absolute kurtosis of < 4.2, indicating little deviation from a normal distribution. The average CVI of the 33-item W-DEQ was 0.95.

In EFA, we identified a 26-item 3-factor solution of fear (10-item, e.g., ‘frightful’, ‘hopelessness’), negative appraisal (8-item, e.g., ‘not relaxed’, ‘not fantastic’), and lack of self-efficacy (8-item, e.g., ‘not longing for the child’, ‘not strong’) (Table 3). The sampling adequacy of EFA was confirmed by a KMO of 0.80 and a significant Bartlett’s sphericity (χ² = 1499.4, p < 0.001). Principal component analysis of the 33-item of the W-DEQ revealed ten factors with eigenvalues of > 1, which accounted for 60.98% of the variance. However, visual inspection of the scree plot revealed three factors that were appropriate for retention. Varimax rotation of the 3-factor solution was optimal and interpretable with seven items failing to load at ≥0.35 (i.e., item 19, panic; item 24, pain; item 27, loss of self-control; item 29, not natural; item 30, not self-evident; item 32, child will die; and item 33, child will be injured). 3 items (i.e., item 7, deserted; item 11, desolate; item 15, abandoned) loaded on a factor of fear and negative appraisal simultaneously and were retained to the factor which had the highest factor loading, and there was at least a 0.2 difference between the loadings. Likewise, item 4 (not strong), and 10 (not independent) loaded on negative appraisal and lack of self-efficacy factors, while item 31 (dangerous) loaded on fear and lack of self-efficacy factors at the same time. These three items (items 4, 10, 31) were appropriately retained to its factors. After eliminating items one at a time, rerunning the analysis at each step, the final 26-item 3-factor solution was obtained and accounted for 38.44% of the variance. The first factor accounted for 22.27% of the variance, with the second and third factors accounting for 8.50 and 7.67% of the variance, respectively. The reliability of the internal consistency for the total scale and the three factors were: total scale = 0.85; fear = 0.78; negative appraisal = 0.69; and lack of self-efficacy = 0.73.

The CFA was conducted to test the factor structure identified by the EFA in our study, and the one-factor solution of the original W-DEQA scale. Specifically, we examined the 33-item 1-factor dimension, 26-item 3-factor structure, and 23-item 3-factor structure (Table 4). The results confirmed that a 23-item 3-factor structure of fear (10-item, e.g. ‘deserted’, ‘weak’), negative appraisal (5-item, e.g. ‘not proud’, ‘not being composed’), and lack of self-efficacy (8-item, e.g. ‘not surrendering my body’, ‘not independent’), instead of the 33-item 1-factor dimension (Table 4). The CFA confirmed that the entire scale of the W-DEQA with a one-factor solution yielded a very poor model fit (χ²/df = 2.43, RMSEA = 0.08, CFI = 0.54, TLI = 0.51, and AIC = 1336). In checking the structure identified by the EFA, the initial three-factor model consisting of 26 items showed a poor fit (χ²/df = 2.41, RMSEA = 0.07, CFI = 0.71, TLI = 0.68, and AIC = 823). Due to low standardized coefficients for item 1 = 0.17, item 9 = − 0.01, and item 31 = 0.22 in a factor of negative appraisal, these three items were deleted one at a time. The fitting index of the three-factor solution consisting of 23 items was improved (χ²/df = 2.52, RMSEA = 0.07, CFI = 0.75, TLI = 0.72, and AIC = 669). These results showed that our model was satisfactory based on the indices of χ²/df, RMSEA, and AIC. Although the CFI and TLI were not of a good fit, they were close to acceptable fit values. The modified 23-item 3-factor solution adopted in this study is illustrated in Fig. 1. The reliability of the internal consistency of the total scale and the three factors were: total scale = 0.84; fear = 0.78; negative appraisal = 0.65; and lack of self-efficacy = 0.73.

Correlation analysis of the 23-item W-DEQ A scale with depressive symptoms and subscales of the QOL was conducted to test the convergent and divergent validities of the instrument (Table 5). The 23-item W-DEQ A scale was significantly positively correlated with the total of depressive symptoms measured by the EPDS (r = 0.23, p < 0.001), and was significantly negatively correlated with domains of QOL measured by the WHOQoL-BREF (physical health: r = − 0.26, p < 0.001; psychological health: r = − 0.23, p < 0.001; social relationships: r = − 0.17, p < 0.01; environmental: r = − 0.25, p < 0.001).

The following limitations need to be taken into account when interpreting the study results. First, we included low-risk pregnant women, and thus the results might not apply to women with high-risk pregnancies. Furthermore, women included in this study were consecutive attendees at prenatal clinics in the study hospital, the generalizability to women with home delivery was limited. Second, the study was conducted in a rural setting, and therefore, the use of this scale in urban samples should be further explored. Third, larger sample size was required for the measurement invariance analysis in CFA, which is important for examining whether the factor structure identified is equivalent in different groups of participants [61]. It is estimated a sample size of 200 each group is required for this analysis [62]. Despite these limitations, this study is the first attempt to evaluate an instrument that measures fear during pregnancy in Malawi and Africa. The instrument will provide healthcare professionals with a basis for understanding the phenomenon and can lead to further development of interventions that will improve the quality of health care and childbirth outcomes for Malawian women.