All procedures of the study were approved by a local (Tuscany) Ethics Committee ‘Comitato Etico Area Vasta Nord-Ovest’ (ref:-CEAVNO N12749/2018) and by the University of Liverpool Research Ethics Committee (ref:-IPH/3964). Women who chose to participate in the study gave their informed consent. All procedures used in human subjects research were in accordance with the ethical requirements of the institutional and/or national research committee and the 1964 Declaration of Helsinki and its subsequent revisions or comparable ethical standards.

Participants were mothers (N = 457) of infants aged between birth and six months postpartum. Of the 830 who took part, 373 were excluded from the final analyses as they had missing data on the PSAS. Women were selected for the study using non-probability (opportunity) sampling among those who had consented to participate in the study, understood Italian, and could communicate in Italian. The procedure for recruitment can be found in Figure 1.

The inclusion criteria were: women over the age of 18 years; good knowledge of the Italian language; no cognitive impairment; and no psychiatric disorders. Participants were excluded if they self-reported a history of mental illness and/or a traumatic event in the family history in the last six months. In addition, women with psychotic disorders or post-traumatic stress disorder [PTSD] were excluded because clinically diagnosed and/or managed mental illness and PTSD may alter postpartum-specific anxiety symptoms.

We calculated the minimum sample size to detect a clinically significant difference, assuming an effect size of 0.8 and a power of 0.95. The sample size for an effect size d is 0.8, an alpha error probability of 0.05 (2-tailed), and a power (1-probability of type II beta error) of 0.95 is 200 participants. Further, according to the literature, a sample size of 5 to 10 participants per item is necessary for factor analysis (38). With 51 items and 5 participants per item, a minimum of 255 women were required for an effectively powered study.

After obtaining the necessary approvals from the PSAS Working Group, the original text was translated from English into Italian using accepted translation methods (30). This was done by three native Italian speakers with an excellent command of both languages and knowledge of the PSAS and perinatal mental health. The Italian versions were then back-translated from Italian into English by two experienced back-translators who were not involved in the previous step and who are experts in psychology and midwifery research respectively, but not familiar with the scale. They selected the most eloquent version of the Italian to translate. Finally, two further people familiar with the specific topic area evaluated the translated and back-translated versions to check for any errors. The most eloquent translations were chosen to back-translate, meaning the resultant PSAS-IT was made up of items from all three original translators.

In addition to the PSAS-IT, participants were asked for relevant personal details and presented with a battery of psychometric scales. Demographic information and psychometric scales used in this study emulated those which would be collected and used, respectively, in Italian healthcare service settings.

Through on-line adverts with a link to Qualtrics software, mothers of infants between birth and six months postpartum were recruited for a study, between March and April 2022. Adverts were placed on parenting discussion forums in Italy, and social media sites. Each response was linked to a unique ID integrated into the survey program to maintain anonymity. The link to access the online survey remained active throughout the duration of the study until the follow-up phase.

An information sheet on the first page explained the objectives and methods of the study to participants. Mothers could withdraw from the study at any time without giving a reason, as participation was voluntary. Any distressing circumstance identified or request for support was forwarded to the study’s local Chief Investigator for discussion or referral to therapy. To this end, participants’ contact details were requested at the end of the survey. No link was made between participant data and contact information, and data were processed anonymously and identified only by a unique ID number. The database was stored on a secure server, and access to the information was limited to the research team members.

The statistical analysis of the study was performed with R version 4.0. A four-step procedure was used to assess the measurement invariance of the PSAS. For all analyses, data were fitted with a diagonally weighted least squares estimator because the PSAS is ordinal (1–4 points score), as recommended by Mîndrilã (43) and in accordance with the analysis strategy of Davies et al. (14). First, configuration invariance (i.e., whether the factor structure holds for both samples) was tested by fitting the PSAS factor structure with a country grouping variable. This model was evaluated using a series of fit indices following Hu and Bentler (44). The comparative fit index (CFI) and the Tucker-Lewis Index (TLI), with values of >0.95, were found to be good. For the root mean square error of approximation (RMSEA), values <0.06 were considered good, and for the standardized residual mean square error (SRMR), values <0.08 were considered a good fit (44). The configural model was then compared to the metric invariance model, which was the same model but with factor loadings fixed across groups (with intercepts allowed to vary). The validity of the metric model was assessed using the cut-off points described in Chen (45) given that we had equally sized groups and some mixed invariance. The criteria for metric invariance were CFI differences (∆CFI) < −0.01, ∆RMSEA<0.015 and ∆SRMR<0.03. The metric invariance model was then compared to the scalar invariance model, which assumes that the factor loadings and axis intercepts are assumed to be the same across our groups. The assessment was the same as for metric invariance, except that the SRMR cut-off was stricter with ∆SRMR<0.015 to indicate scalar invariance. Finally, we also tested strict invariance where both residuals and slopes and intercepts were assumed to be constant (comparing with the metric invariance model) using the same limit as in the previous model comparison.

Finally, we tested the internal reliability of each of the subscales in both the Italian and the British samples (separately). We used McDonald’s omega (ω), which, unlike the more commonly used Cronbach’s alpha, neither assumes tau equivalence nor represents a lower bound on estimates [see (46)]. In addition, we also calculate split-half reliability [Guttman λ4, see (47)]. In both cases, the values should be above 0.7.

The age of the final sample of 457 mothers ranged from 18 to 33 (M = 21.46; SD = 3.02). Participants were predominately women from Italy (97.4%), married (55.1%), primiparous (69.4%), and in administrative professions (24.3%). Very few women had a current clinical diagnosis of anxiety (5.0%) or depression (1.8%) at the time of participation. The babies’ ages ranged from 1 to 27 weeks (M = 15.32; SD = 7.38). See Table 1 for full demographic details.

Confirmatory factor analysis for multiple groups (Italy and United Kingdom) showed that the factor structure of the PSAS was valid for both groups (χ² (2436) = 4679.481, p < 0.001, TLI = 0.969, CFI =0.972, RMSEA = 0.045, SRMR =0.064). See Table 2 for factor loadings and Figure 2. See Table 3 for characteristics of the Italian study sample compared to the characteristics of the UK study sample.

The configural invariance model was compared to the metric invariance model (assuming equal loadings in all groups). The difference between the two models did not exceed the thresholds for ∆RMSEA = 0.013 or ∆SRMR =0.009, although it was just above the thresholds for ∆CFI = −0.02. Because two of the fit indices showed metric invariance and one was just above the threshold, we did not consider cross group loading, although a discussion of the items that must vary across groups to produce a ∆CFI < −0.01 can be found in S1.

The metric invariance model was then compared to the scalar invariance model (assuming equal loadings and intercepts between groups). The difference between the two models did not exceed the cut offs for ∆RMSEA = 0.005, ∆SRMR =0.004, although it was just above the limit for ∆CFI = −0.011. Again, two of the fit indices found were scalar invariant and the CFI was just above the limit, so we did not allow for intercept variation when comparing with strict invariance. In particular, variation of the intercept of PSAS14 resulted in a ∆CFI < −0.01.

The scalar invariance model was then compared to the strict invariance model (assuming equal loadings, intercepts and residuals in all groups). The difference between the models did not exceed the limits for ∆RMSEA = 0.001, ∆SRMR =0.002 or ∆CFI = −0.003. See Table 4 for McDonald’s Omega and Guttman’s lambda.

To assess the convergent validity of the PSAS-IT, Pearson rank correlation analyses were performed with other recognized anxiety and depression screening instruments (EDPS and GAD-7). The correlation coefficients obtained showed a strong and favorable relationship between the total scores of PSAS-IT and the anxiety scores of the GAD-7 and with the EPDS scale scores, demonstrating high convergent validity – see Table 5.

The original PSAS was adapted to the characteristics of the Italian population using a pilot study and psychometric evaluations which assessed the translated instrument’s acceptability, validity, and measurement invariance. As a result, the psychometric properties of the PSAS-IT were similar to those of the original English version, showing acceptable acceptability, construct and convergent validity, and internal consistency.

Our results demonstrate the data fit the expected factorial structure well and the model is entirely invariant between the two languages (evidence of generalizability). Configurable invariance means the theoretical construct has the same subscales and item-factor configurations in both groups. In addition, two of the fit indices used to test metric invariance showed no significant differences between the two groups, suggesting each item represents the construct similarly in Italy and the United Kingdom.

As with metric invariance, two of the fit indices used to test scalar invariance showed no significant differences between the two groups, indicating participants in the two groups tended to score the same on the items. In particular, the variation in the intercept of PSAS14 (“Mi sono preoccupata di far avere al mio bambino la sua routine”) yielded a ∆CFI < −0.01. Finally, strict invariance suggests that the two groups have the same variance in item errors.

The results are also evidence of good reliability. In particular, we checked whether the items within the same factor were coherent in the Italian and British samples (internal consistency). As expected, all four factors had values above 0.7 in both samples.

The PSAS-IT showed a positive relationship with previously validated and commonly used measures of anxiety and postpartum depression, including the GAD-7 and EPDS scales, indicating high convergent validity, in a similar fashion to Fallon et al. (13) and Silverio et al. (30), as well as the other PSAS validations [i.e. (32–37)]. Overall, we conclude the PSAS-IT results are valid and reliable.

As for internal consistency and reliability, previous validation studies have shown that both internal consistency and reliability are good. Indeed, the internal consistency of the four factors in the French version was good to excellent (Cronbach’s alpha coefficients ranged from 0.77 to 0.88), indicating a high reliability of the PSAS-FR. In the Iranian version, the reliability of the instrument was almost identical to the British version, showing that the types of anxiety experienced by women are the same across income levels.

Compared with other validated measures of general anxiety, our results demonstrate the utility of using PSAS-IT, an instrument that identifies clinical anxiety in new mothers that occurs only in the postnatal period.

This is the first study to investigate the psychometric capabilities of the PSAS in an Italian context. Including women who gave birth vaginally or by cesarean section and the random selection are two significant advantages of this study. In this study, data were collected online, as in Fallon et al. (3), which means less control of the sampling process. For future studies, it would be necessary to compare the performance of PSAS-IT with other samples of mothers, including those with low social support, high social complexity, and multiple disadvantages, or with a personal or family history of mental illness, as our sample is not sufficiently representative of at-risk populations. In addition, because women with a history of mental illness were not included in the study, the results are less generalizable to clinical populations. Finally, the sampling for this study took place during the COVID-19 pandemic, which could affect the results.

The PSAS constitutes to date the only reliable measure for postpartum-specific anxiety also in the Italian context. The tool is quick and easy to implement, which makes its use recommendable for early identification of postpartum anxiety. Given the adverse maternal and neonatal outcomes due to postpartum anxiety, clinicians and researchers should use PSAS in order to provide early intervention.