Introduction
Up to 20 % of women in developed countries experience mental health problems postnatally (World Health Organisation [WHO]
2016). Several decades of research have focused on postpartum depression, while symptoms of anxiety have been largely overlooked. However, postpartum anxiety has become a condition of interest to perinatal researchers and practitioners in recognition of high prevalence rates and impact on maternal and infant outcomes (Lonstein
2007; Glasheen et al.
2010; Paul et al.
2013). Studies of postpartum anxiety reveal incidence estimates ranging from 3 to 43 %, with evidence suggesting that it may occur independently and at a higher rate than postpartum depression (PPD) does (Wenzel et al.
2005; Britton
2008; Glasheen et al.
2010; Paul et al.
2013).
The postpartum period involves a series of temporally unique transitions which are often experienced as stressful and overwhelming. This can lead to specific postpartum concerns such as personal appearance and postpartum weight gain (Walker and Freeland‐Graves
1998), health and wellbeing of the infant (Lugina et al.
2004), interpersonal relationships (Hiser
1991) and general infant care (Warren
2005). Comprehensive reviews by Lonstein (
2007) and Glasheen et al. (
2010) also link postpartum anxiety to a range of adverse developmental, somatic and psychological outcomes in the infant. The interpretation of these results, however, is limited by the use of general scales of anxiety such as the State Trait Anxiety Inventory (STAI) (Spielberger et al.
1970) and/or scales that focus predominantly on postpartum depression (i.e. the Edinburgh Postnatal Depression Scale; EPDS; Cox et al.
1987).
General measures of anxiety are relied upon in a large majority of studies examining postpartum anxiety
1 (Glasheen et al.
2010; Lonstein
2007; Meades and Ayers
2011) and may be psychometrically problematic. Many commonly used general measures include somatic items which may occur naturally in the postpartum (e.g. STAI: ‘I feel rested’ or ‘I feel comfortable’). These may inflate anxiety scores in postpartum populations (Meades and Ayers
2011) and increase the likelihood of false positives (Swallow et al.
2003). Furthermore, symptoms of anxiety occurring in the postpartum may have distinct presentations which are not encompassed by items in general scales (Meades and Ayers
2011; Phillips et al.
2009); this limitation has been addressed when examining anxieties occurring in pregnancy (Van den Bergh
1990; Levin
1991; Wadwha et al.
1993; Huizink et al.
2002).
A variety of self-report questionnaires has been developed to assess specific anxieties relating to the gestational period that would not bear relevance in general scales. These include the Pregnancy Anxiety Scale (PAS; Levin
1991), the Pregnancy-Related Anxiety Questionnaire (PRAQ; Van Den Bergh
1990), the PRAQ-R (Huizink et al.
2004) and the Pregnancy-Related Anxiety Scale (PRAS; Wadwha et al.
1993). These measures include constructs such as fear of childbirth, foetal health and wellbeing, bearing a physically or mentally handicapped child, the mother-infant relationship, relationship changes and changes in appearance. Two key findings have been observed by studies using these measures: (a) that they predict perinatal outcomes more effectively than general measures of anxiety do (Wadwha et al.
1993; Sandman
1997) and (b) that they are qualitatively and quantitatively distinct from general indices of anxiety and depression (Huizink et al.
2002). This has led researchers to regard pregnancy-specific anxiety as a distinct entity to anxiety experienced at other times of life (Huizink et al.
2004).
In a similar manner, postpartum-specific scales have been designed to measure depression. These include the EPDS (Cox et al.
1987) and the Postpartum Depression Screening Scale (Beck and Gable
2000). Given high comorbidity with anxiety in some postpartum samples, some researchers have argued that they may be utilised to screen for both anxiety and depression concurrently (Stuart et al.
1998; Ross et al.
2003; Reck et al.
2008). While three items of the EPDS have been found to cluster together on an anxiety factor in postpartum women in several studies (Ross et al.
2003; Matthey
2008; Phillips et al.
2009; Matthey et al.
2013), the authors maintain that the scale does not measure anxiety (Cox et al.
1987). Furthermore, the EPDS does not distinguish whether anxiety scores on these three items are a feature of depression or a distinct entity (Matthey et al.
2003; Ross et al.
2003). This limits the utility of such tools considering work which finds that anxiety occurs more frequently (Muzik et al.
2000; Wenzel et al.
2005; Paul et al.
2013) and independently (Muzik et al.
2000; Matthey et al.
2003; Wenzel et al.
2005; Miller et al.
2006) of depression in the postpartum.
Two recent endeavours have been made to create an anxiety scale relevant to postpartum women: the Perinatal Anxiety Screening Scale (PASS; Somerville et al.
2014) and the Postpartum Worry Scale-Revised (PWS-R; Moran et al.
2014). Both measures aim to detect clinically significant levels of anxiety which map onto existing diagnostic criteria for anxiety disorders, although the PWS-R focuses only on generalised anxiety disorder (Moran et al.
2014). Emerging evidence highlights a large number of postpartum women who do not meet diagnostic criteria for an existing anxiety disorder yet experience a clinically significant degree of ‘maternally focused worry’ (Wenzel et al.
2005; Phillips et al.
2007; Phillips et al.
2009). As such, items within these scales may not encompass the full range of symptoms of anxiety experienced postnatally and a scale with a more focused domain is necessary. Furthermore, the PASS was designed for use with both antenatal and postnatal women (Somerville et al.
2014) suggesting that symptoms are comparable across childbirth. Although an overlap between pregnancy and postpartum anxiety has been identified (Heron et al.
2004), a body of literature provides evidence for a temporally specific pregnancy anxiety (Van den Bergh
1990; Levin
1991; Wadwha et al.
1993; Huizink et al.
2004), which includes constructs such as ‘fear of childbirth’ (Huizink et al.
2004) that would not be applicable postpartum. Furthermore, some women may be more prone to developing postpartum anxiety as consequence of specific physiological and psychological processes associated with birth (Heron et al.
2004), which raises additional considerations for measurement. Finally, items from both the PASS and the PWS-R were generated from researcher/clinician experience (Moran et al.
2014; Somerville et al.
2014). Although clinicians may be the best observers of the outward manifestations of symptoms or disorder, only those who experience it can effectively capture the subjective elements (Streiner et al.
2015). This multifactorial rationale supports the development of an anxiety scale specific to the postpartum period which takes into account the limitations of the existing evidence base.
Stage 1: item generation
Items were predominately informed from interviews conducted with a group of 19 postpartum women at two time points (time one 4–8 weeks; time two 12–16 weeks) to ensure that an accurate, experiential representation of postpartum-specific anxieties was achieved. Responses to the open question ‘What are the main anxieties that women have at this stage of motherhood’ were digitally recorded, and a basic content analysis was performed to identify themes and develop items. The scale items were further developed by reviewing validated pregnancy and postpartum anxiety questionnaires (PASS: Somerville et al.
2014; PWS-R: Moran et al.
2014; PAS: Levin
1991; PRAQ: Van Den Bergh
1990; PRAQ-R: Huizink et al.
2004; PRAS: Wadwha et al.
1993) and the postpartum anxiety research literature. The item pool was developed to systematically encompass a broad range of anxieties that were temporally specific to the postpartum period.
Consistent with other validated scales in the field, the 51-item PSAS was formatted as a self-report questionnaire with a four-point Likert scale assessing the frequency of specific anxieties with consistent response options (from zero = ‘not at all’ to three = ‘almost always’). The order of 27 responses was randomly reversed in order to avoid ‘yea-saying’ bias and aid participant concentration (Streiner et al.
2015). The wording and amount of Likert-scale divisions were chosen based on best current practice in the psychometrics literature (Streiner et al.
2015) and careful review of the self-admitted limitations of already-validated anxiety scales (Somerville et al.
2014). The timeframe for rating frequency of anxieties was limited to over the past 7 days. This is congruent with pregnancy-specific anxiety scales and deemed necessary given the transient nature of anxieties occurring in the postpartum.
Stage 2: expert panel and face and content validity
The preliminary 51-item scale was reviewed and refined by a panel of 12 individuals, each reflecting distinct insights of scale development and/or postpartum anxiety. The panel included three experienced perinatal researchers, three senior community midwives, three research midwives (one senior), one statistician and two psychometricians. Each panel member (blind to the other members’ feedback) provided detailed comments on individual items and the overall suitability of the scale. Qualitative responses from the panel indicated that the preliminary scale appeared to be measuring postpartum-specific anxieties and was both clinically acceptable for perinatal women and psychometrically feasible, indicating adequate face validity. The panel members also evaluated each item on a four-point Likert scale (four = highly relevant; three = quite relevant or highly relevant but needs rewording; two = somewhat relevant; and one = not relevant). A content validation ratio (CVR; Streiner et al.
2015) was calculated to provide a quantitative expression of content validity. The mean CVR across all items was 0.76 which is indicative of good content validity. The panel was also asked to comment on whether any items had been omitted to further establish content coverage.
Specific qualitative feedback was collated and analysis of this phase indicated a need to revise certain aspects of the scale. Thirty-two items were reworded based on the general consensus of the panel. Of particular importance was the rewording of 11 items to reflect the specificity of postpartum anxiety. For example, the item ‘I have worried about my relationship with my partner’ was reworded to ‘I have worried more about my relationship with my partner than before my baby was born’. Five items were deleted either due to repetition or because there was general agreement that they did not specifically relate to postpartum anxiety (e.g. low CVR). In addition, seven new items were included based on content coverage ideas provided by the panel.
The design and presentation of the final 53-item scale was then extensively reviewed to ensure it was streamlined and easy to respond to. The wording of the final items was subjected to a computer literacy check (Flesch-Kincaid Grade Level test) as being understandable for someone with 5 years of education or a 10-year-old child. A question understanding aid (QUAID) (Graesser et al.
2006) was also used, and no issues were found with the wording, syntax or semantics of the questions.
Stage 3: pilot study
An online pilot study was conducted via the Qualtrics survey software platform to assess comprehensibility of item wording, ease of responding, time taken to complete and any other implementation issues. Mothers of infants aged between 0 and 6 months (n = 146) were recruited via online forums (Mumsnet, Netmums) and social media platforms (Facebook groups and Twitter). Participants completed the 53-item scale and rated comprehensibility and ease of responding on two 10-point Likert scales (i.e. ‘not at all easy to understand’ [0] to ‘extremely easy to understand’ [10] and ‘not at all easy to complete’ [0] to ‘extremely easy to complete’ [10]). An optional free text box was provided at the end of the survey to allow qualitative comments on the questionnaire content and experience of completion to be made.
The acceptability of the PSAS was excellent. The mean scores for the comprehensibility item and the ease of completion item were 9.29 (±1.24) and 9.18 (±1.26), respectively. The mean time taken to complete the 53-item scale was 9 min (range 3 to 15 min). Based on the qualitative responses from 18 women, a ‘not applicable’ option was created for 7 items relating to partner, families and work to avoid response ambiguity for women who may not have these life domains. Positive comments about the scale design and items were also recorded by 36 women, which provided further evidence of its acceptability in this population.
A preliminary item analysis (endorsement frequency and item homogeneity) was also conducted on the pilot study data. The overall scale had excellent reliability (Cronbach’s α = 0.96). Inter-item correlations were between 0.15 and 0.50. Item-total correlations were between 0.30 and 0.70. No problematic items were identified at this stage.
Discussion
This study reports the development and initial validation of the PSAS, a 51-item measure of postpartum specific anxiety, using a large online sample of mothers in the first 6 months postpartum. The results suggest that the PSAS is an acceptable measure with sound psychometric properties. The low to moderate size correlations between factors indicated that they are not derived from a single underlying latent variable. It has a simple four-factor structure which showed good face and content validity and can be distinguished as (1) competence and attachment anxieties, (2) infant safety and welfare anxieties, (3) practical baby care anxieties and (4) psychosocial adjustment to motherhood.
Despite limited discussion about the qualitative nature of the symptoms of postpartum anxiety, these constructs are theoretically meaningful when examined in relation to some recent work. Brockington et al. (
2006) found qualitative themes of ‘fear of cot death’, ‘fear of the criticism of mothering skills’ and ‘fear of disordered maternal attachment’ in a sample of 129 women referred to psychiatric services. Similar symptoms were also found in a recent interview study (Highet et al.
2014) alongside a theme of ‘adjustment difficulties’ which included anxieties relating to changes in appearance, daily activities and social roles. Phillips et al. (
2009) investigated symptom presentations of postpartum women with an anxiety disorder not otherwise specified (ADNOS). They identified 65 % of women reporting anxieties in relation to infant health, safety and wellbeing; 53 % with anxieties concerning performance as a mother; and 18 % with anxieties relating to practical day-to-day care of the infant. This finding suggests that the PSAS, unlike existing measures, may possess constructs that are sensitive to postpartum women experiencing clinically significant maternally focused worry, yet failing to meet diagnostic criteria for an anxiety disorder (Phillips et al.
2009). Further examination of the construct validity of the PSAS is necessary to reexamine the proposed model and to provide further confirmation of these factors.
As hypothesised, the PSAS was significantly positively correlated with theoretically related measures of anxiety, which demonstrates initial evidence of convergent validity. The PSAS was also significantly associated with measures of depression, which was anticipated given the high comorbidity identified in previous work (Stuart et al.
1998; Ross et al.
2003; Reck et al.
2008) and provides further convergent support. It has been suggested that the overlap between depression and anxiety reflects the co-occurrence of phenomenologically distinct constructs (Beck
1976; Beck et al.
1979; Burns and Eidelson
1998). As such, Burns and Eidelson (
1998) contend that any valid and reliable measure of anxiety and depression should correlate approximately at the 0.70 level; the PSAS exceeded this benchmark. In addition, the internal consistency of the overall PSAS scale and four factors was good to excellent (George and Mallery
2003; Ponterotto and Ruckdeschel
2007). Test-retest reliability also indicated better stability over time than other recent endeavours (Somerville et al.
2014).
A preliminary ROC analysis demonstrated that the PSAS performed well at identifying women with a current clinical diagnosis of anxiety and/or depression. At the optimal cut-off score of 112, 75 % of women with a diagnosis were detected, which surpasses other recent efforts (Somerville et al.
2014). Furthermore, the PSAS performed better than did other general (i.e. STAI-S) and postpartum-specific (i.e. EPDS-A) measures of anxiety. However, determining the case finding abilities of the PSAS was not a primary aim of the research and it is acknowledged that the self-report methods used to ascertain a current, clinical diagnosis of anxiety and/or depression in the sample are crude compared to other work (Somerville et al.
2014). Furthermore, the design precluded the differentiation of anxiety and depression within the sample. Interestingly, trait anxiety had the best case-finding abilities and previous work has suggested that the trait scale may examine depression, as well as anxiety (Bieling et al.
1998; Julian
2011), which could explain the high area under the curve (AUC) observed in this sample. Despite these limitations, the analysis suggests that the PSAS may be a useful screening tool for postpartum women and future work in clinical samples across the full spectrum of anxiety disorders is necessary to confirm this.
In the interim, the PSAS can be used to capture a range of anxieties relating to both mother and infant, which are specific to the postpartum period. Other potential avenues for research use include examining the prevalence of postpartum-specific anxiety and examining how this varies in different populations (e.g. those with high-risk pregnancies, mothers of premature infants, mothers who have experienced previous miscarriage or stillbirth). Administering the PSAS in samples of postpartum women with non-comorbid anxiety and depression will allow examination of whether the PSAS measures ‘pure’ anxiety and can differentiate anxiety from depression. A comparison of scores on the PSAS in women with ADNOS and other anxiety disorders (e.g. GAD, OCD) would be particularly interesting given recent findings concerning maternally focused worry in samples of postpartum women with ADNOS (Phillips et al.
2007; Phillips et al.
2009).
Validation of a measure is an iterative process and there are several areas for future work that are necessary to continue the development and evaluation of the PSAS. Firstly, the study used an online convenience sample which provided an appropriate sample size for the analyses conducted (in particular PCA) but lacked sampling control. The samples were predominately married, professional women from the UK. Thus, the psychometric properties of the PSAS may vary in other populations, and it will be important to replicate these findings in diverse samples, particularly those at risk of developing postpartum anxiety. Second, the pilot study demonstrated excellent acceptability to postpartum women in its current form, which probably reflects the qualitative inquiry used to inform its development. However, the item analyses (inter-item, item total) displayed psychometric potential for the development of a short form, which may increase its utility in both clinical and research settings.
Finally, the pregnancy anxiety literature provides findings that differentiate pregnancy-specific anxiety from general measures of anxiety and depression (Huizink et al.
2004) and highlights that temporally specific measures may be more efficacious at predicting perinatal outcomes than the more commonly used general measures (Wadwha et al.
1993; Huizink et al.
2002; Huizink et al.
2003). Further research should attempt to replicate this work with the PSAS. Isolation of child-bearing-related anxiety from symptoms of general anxiety and depression may allow clinicians and researchers to address issues of identification, prediction and prevention more precisely (Huizink et al.
2004). Associations between postpartum anxiety and maternal attachment (Mertesacker et al.
2004), infant feeding (Britton
2007; Paul et al.
2013) and infant temperament (Coplan et al.
2005) have been previously identified and warrant examination to ascertain the predictive value of the PSAS for maternal and infant outcomes and determine whether it may be a more effective predictor of perinatal outcomes than general measures of anxiety.