Introduction

Intimate partner violence (IPV) is well recognized as a major public health problem [1]. IPV prevalence in Europe is estimated to exceed 20% in women aged 15 years and older and is higher in women of reproductive age [2]. Pregnancy and perinatal periods are significant risk factors of IPV that could begin or intensify during these periods. In 2013, a meta-analysis reported a mean IPV prevalence rate among pregnant women of 19.8% for 23 countries; the mean reported prevalence of abuse was 28.4% for emotional abuse, 13.8% for physical abuse and 8.0% for sexual abuse [3]. Acts of violence during pregnancy affects directly and indirectly the mortality and the morbidity of the child and the mother. IPV is associated to unwanted pregnancies, termination of pregnancy [4] and adverse maternal and neonatal outcomes [5–7]. Moreover, the prevalence of maternal death by homicide was estimated between 13 and 24%, considering all causes of deaths during pregnancy [8].

Given these serious health consequences, healthcare professionals specializing in the perinatal period (perinatal care providers), have a major role in the identification and the management of abused women. The French National Health Authority recommended in 2019 that perinatal care providers screen all women for IPV during the first prenatal visit and continue at least once per trimester and during postpartum visits [9], in accordance with international guidelines [10]. Pregnancy is recognized as a privileged moment to detect past or current violence suffered by a woman. Indeed, the frequency of prenatal consultations and the bond of trust that is established between the care provider and the women offer the opportunity to disclose IPV.

However, several studies shown that IPV screening was not always done by healthcare professionals, often in reason of insufficient preparedness [11–13]. In 2014, the French inter-ministerial mission for the protection of women victims of violence and the fight against human trafficking conducted a national study to evaluate the midwives’ IPV knowledge and their ability to screen IPV with a non-validated questionnaire [14]. They reported insufficient education on IPV and limited knowledge. Almost one third had never assessed their patients’ exposure to IPV and in case of IPV, 81% felt insufficiently prepared. Nonetheless, 90% considered that midwives had a role to play in the disclosure and management of IPV. The reasons given for not screening were the lack of training, the absence of clear referral procedures and the ignorance of appropriate response or local specialist IPV services. Similar results were found among Italian midwives [15]. An American study identified in 2015 that 31.2% of obstetricians-gynecologists screen all pregnant patients at specific times of their pregnancy and 36.8% report insufficient training to assist their patients in addressing IPV [16].

Thus, the creation of specific IPV education and training programs for perinatal care providers, teaching how to recognize and respond to IPV is primordial [17,18], as well as assessing the effectiveness of these educational interventions with reliable outcomes. Among them, the PREMIS (Physician Readiness to Manage Intimate Partner Violence Survey), developed by Short et al. in 2006, is a valid and reliable self-reported instrument designed to measure the knowledge, attitudes and preparedness of physician to manage IPV [19]. They demonstrated evidence for its construct validity, internal consistency (with Cronbach’s alpha higher than 0.65 for reliable subscales), predictive validity, external validity (correlations between self-reported practices and observed clinical behaviors) and temporal stability over 12 months in the absence of IPV education or intervention. The developers shown that the PREMIS could discriminate between trained and untrained physicians [20].

This instrument has been adapted and widely used in multiple populations, cultures and languages [16,21–33]. It was also used to measure the effectiveness of IPV training programs proposed to physicians [20], practicing general practitioners and residents of general practice [34], nurses [35], and healthcare providers working in fracture clinics [36].

To our knowledge, no study in France assessed with a valid and reliable tool neither IPV knowledge of perinatal care providers nor their ability to identify and address IPV victims. Thus, we performed a study to cross-culturally adapt and evaluate the psychometric properties of a French version of the PREMIS tool.

Methods

Results

Psychometric evaluation of the PREMIS-French

Participants.

The online survey was answered by 360 perinatal care providers. The characteristics of these respondents are described in Table 1 and S1 Table. They were predominantly women with a mean age of 43.3 years (SD 10.7) and practicing for 17.9 years (SD 10.8) mainly in gynecology and obstetrics. Less than 4% had no actual clinical activity. About one quarter of the care providers never had training about IPV issues. The mean amount of previous IPV training was 15.0 hours (SD 19.3) and the more frequent type of training was to attend a lecture or talk. In the last 6 months, 63.7% made at least one new diagnose of IPV (picked up an acute case, uncovered ongoing abuse, or had a patient disclose a past history).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Sociodemographic and professional characteristics of participants.

https://doi.org/10.1371/journal.pone.0258943.t001

Data completeness.

In the “Respondent profile” section, the percentage of missing values per item varied between 0 and 16.1% (for the item on the number of practitioners that have participated in an IPV training course in the past 6 months at the work site). In the “Background” section, the percentage of missing values was 11.4% for the item on the amount of previous IPV training and ranged from 0 to 1.7% for items of the ‘Perceived preparation’ subscale and from 0.6 to 2.2% for items of the ‘Perceived knowledge’ subscale. The percentage of missing values ranged from 0.3 to 2.8% for items of the “Actual IPV knowledge” section and from 1.9 to 7.8% for items of the “Opinion” section. In the “Practice issues” section, with items on the screening and management of IPV victims in the last 6 months, the percentage of missing values per item were higher and ranged between 4.2 and 16.4%.

Factor analysis.

Factor analysis with an oblique promax rotation were performed to study the distribution of the 36 items of the “Opinions” section in subscales. The significance value of Bartlett’s test of sphericity was <0.0001 (χ² = 3606.5, df = 630) and KMO measures of sampling adequacy was 0.846, indicating that the data were suitable for factor analysis. An initial Maximum Likelihood Factor (MLF) analysis identified a 11-factor solution that was statistically sound (χ² = 323.7, df = 289, p = 0.0783). However, four factors contained only two items, ten items cross-loaded on two or more factors, indicating a complex solution that lacked a good theoretical basis. Thus, new factor analysis were performed after discarding items with loadings lower than 0.32, with loadings greater than 0.32 on two or more factors or items that significantly lowered internal consistency. The final factor analysis identified six factors with eigenvalues greater than one and accounting for 59.1% of the total variance (Table 2). All items loaded higher than 0.40 on its subscale. The first factor comprised five items and was named ‘Preparation’. It had four items in common with the original ‘Preparation’ subscale (items 6, 10a, 10b and 10c) and one item from the original ‘Self-efficacy’ subscale (item 9). The second factor named ‘Workplace issues’ comprised six items, four items were in common with the original subscale (items 3, 4, 19, 25), item 2 was from the original ‘Self-efficacy’ subscale and item 18 was from the original ‘Victim understanding’ subscale. The third factor comprised the four items of the original ‘Legal requirements’ subscale (items 12a, 12b, 12c, 17) and was labeled identically. Item 17 loaded higher than 0.32 on two factors: 0.55 on its factor and 0.38 on factor 2. The fourth factor comprised four items (items 5, 14, 26, 32) and was labeled ‘Self-efficacy’. It had three items in common with the original subscale (items 5, 14, 32) and one item from the original ‘Workplace issues’ subscale (item 26). The fifth factor comprised the three items of the original ‘Alcohol/drugs’ subscale (items 7, 21, 31) and was named identically. The sixth factor comprised three items of the original ‘Victim understanding’ subscales (items 11, 15, 16) and was labeled identically. Item 15 loaded higher than 0.32 on two factors: 0.71 on its factor and 0.33 on factor 4.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Factor loadings from the factor analysis of the PREMIS-French “Opinions” items.

https://doi.org/10.1371/journal.pone.0258943.t002

Two original “Opinions” subscales were not found in the current model. The items of ‘Victim autonomy’ (items 8, 22, 30) and ‘Constraints’ (items 13 and 23) subscales were discarded during the selection process of the most adequate factor structure.

Descriptive statistics, score distribution, floor and ceiling effects.

The descriptive statistics and score distributions of the PREMIS-French subscales are presented in Table 3. The percentage of missing values were less than 7%, except for the ‘Actual knowledge’ subscale with 25% of missing values. Neither floor nor ceiling effects were found for the “Background”, “Actual IPV knowledge” and “Opinions” subscales.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Descriptive statistics and score distributions of the PREMIS-French subscales.

https://doi.org/10.1371/journal.pone.0258943.t003

Internal consistency and item-total correlations.

The “Background” subscales showed good internal consistency, with Cronbach’s α equal to 0.95 and 0.97 for ‘Perceived preparation’ and ‘Perceived knowledge’ subscales respectively. All corrected item-total correlations were higher than the required 0.40, ranging from 0.54 to 0.87 for ‘Perceived preparation’ subscale and from 0.50 to 0.90 for ‘Perceived knowledge’ subscale.

For “Opinions” subscales, Cronbach’s α ranged from 0.54 to 0.88 (Table 4), showing good internal consistency for four subscales, and moderate internal consistency for the ‘Alcohol/drugs’ and ‘Victim understanding’ subscales, which did not obtain the minimum required coefficient of 0.70. All corrected item-total correlations were higher than the required 0.40, ranging from 0.40 to 0.85, except for item 21 from ‘Alcohol/drugs’ subscale with a value of 0.38, and for items 11 and 15 from ‘Victim understanding’ subscale with values of 0.32 and 0.34 (S2 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Internal consistency (Cronbach’s α) and inter-subscale correlations for the PREMIS-French “Opinions” subscales.

https://doi.org/10.1371/journal.pone.0258943.t004

In addition, each item of the “Opinions” section correlated better with its parent subscale (corrected for overlap) than with the other subscales, except for two items of the ‘Self-efficacy’ subscale (S2 Table). For item 14 (Able to gather the necessary information to identify IPV as the underlying cause of patient illnesses), the item-total correlation with its subscale was 0.57 and the correlation with the ‘Preparation’ subscale was also 0.57. For item 26 (Able to gather the necessary information to identify IPV as the underlying cause of patient injuries), the item-total correlation with its subscale was 0.54 and the correlation with the ‘Workplace issues’ subscale was 0.57.

Inter-subscale correlations.

Correlations between PREMIS-French “Opinions” subscales ranged from -0.003 to 0.54 (Table 4). The ‘Preparation’ subscale had a significant moderate correlation with the ‘Workplace issues’ subscale (r = 0.54). The ‘Legal requirements’ subscale had significant small correlations with ‘Preparation’ (r = 0.44), ‘Workplace issues’ (r = 0.45) and ‘Self-efficacy’ (r = 0.45) subscales. The ‘Self-efficacy’ subscale had significant moderate correlations with ‘Preparation’ (r = 0.53) and ‘Workplace issues’ (r = 0.50) subscales. The ‘Alcohol/drugs’ subscales had significant very small correlations with ‘Workplace issues’ (r = 0.14) and ‘Self-efficacy’ (r = 0.17) subscales, and no significant correlation with the others. The ‘Victim understanding’ subscale had only significant very small correlations with ‘Preparation’ (r = 0.21) and ‘Workplace issues’ (r = 0.13) subscales.

Convergent validity.

Correlations between the amount of previous IPV training, the “Background” subscales, the ‘Actual knowledge’ subscale and the “Opinions” subscales are shown in Table 5.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Spearman’s correlation coefficients between the amount of previous IPV training and the PREMIS-French “Background”, “Actual IPV knowledge”, “Opinions” subscales.

https://doi.org/10.1371/journal.pone.0258943.t005

The amount of previous IPV training was moderately correlated to ‘Perceived preparation’ and ‘Perceived knowledge’ subscales (r = 0.66 and r = 0.67, respectively) and very weakly to ‘Actual knowledge’ subscale (r = 0.28). The ‘Perceived knowledge’ and ‘Perceived preparation’ subscales were strongly correlated (r = 0.91). Correlations between the ‘Actual knowledge’ subscale and the “Background” subscales were significant but very small: r = 0.22 with ‘Perceived preparation’ and r = 0.27 with ‘Perceived knowledge’.

Hours of IPV training showed moderate correlations with ‘Preparation’ (r = 0.52) and ‘Workplace issues’ (r = 0.52) subscales, almost moderate correlations with ‘Legal requirements’ (r = 0.45) and ‘Self-efficacy’ (r = 0.46) subscales, and no significant correlation with the two other “Opinions” subscales. All “Opinions” subscales except for the ‘Alcohol/drugs’ one were significantly correlated to the ‘Perceived preparation’ and ‘Perceived knowledge’ subscales. Correlations were strong with the ‘Preparation’ subscale (r = 0.70 and 0.74, respectively), moderate with ‘Workplace issues’, ‘Legal requirements’ and ‘Self-efficacy’ subscales (ranging from 0.52 to 0.62 and from 0.54 to 0.64, respectively) and very small for the ‘Victim understanding’ subscale (r = 0.15 and 0.18, respectively). Correlations between “Opinions” subscales and the ‘Actual knowledge’ subscale were significant but very small, ranging from 0.14 to 0.23, except for the ‘Victim understanding’ subscale with a higher correlation of 0.33.

Reliability.

Of the fifty perinatal care providers randomly selected for the test-retest, 40 (80.0%) completed the test questionnaire. Among them, 24 (60.0%) completed the retest questionnaire. No one reported having a training course about IPV between test and retest. The characteristics of the test-retest participants are described in S1 Table. They were predominantly women (83.3%), with a mean age of 44.2 years (SD 11.6). Twenty of them practiced in gynecology and obstetrics, two in mother and child protection services, one in pediatrics and one in psychology.

Test-retest reliability was almost perfect for the “Background” subscales, with ICCs equal to 0.92 for ‘Perceived preparation’ and 0.88 for ‘Perceived knowledge’ (Table 6). Agreement was substantial for ‘Actual knowledge’ subscale with an ICC of 0.69. Reliability was also substantial for “Opinions” subscales, with ICCs ranging from 0.65 to 0.83, except for ‘Alcohol/drugs’ subscale with a moderate ICC of 0.46.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Test-retest reliability for the PREMIS-French “Background”, “Actual IPV knowledge” and “Opinions” subscales.

https://doi.org/10.1371/journal.pone.0258943.t006

Discussion

The present study describes the cross-cultural adaptation and the evaluation of the psychometric properties of the French version of the PREMIS, named the PREMIS-French, in a national sample of perinatal care providers.

The PREMIS was successfully translated and cross-culturally adapted from English to French. The PREMIS-French had good acceptability with low percentages of missing values per item, except for some items from the “Practice issues” section, evaluating practices or workplace specificities that could be not relevant for some participants. It also had good response distribution, neither floor nor ceiling effect were found for the subscales, indicating that the instrument was adapted to the studied population.

The six-factor structure of the PREMIS-French “Opinions” section differs slightly from the eight-factor structure of the original version of the PREMIS [19]. Only the two original ‘Victim autonomy’ and ‘Constraints’ subscales were not found in our validation. However, in the original PREMIS, this subscales were not sufficiently reliable and only kept for future testing.

The ‘Preparation’ subscale groups together four items from the original ‘Preparation’ subscale and one item from the original ‘Self-efficacy’ subscale. This item (Feel comfortable discussing IPV with patients) was more related to the feeling of being prepared in the French context. The PREMIS-French ‘Workplace issues’ subscale comprised items from the original subscale, and two items related to the systematic abuse screening of patients from the original ‘Self-efficacy’ and ‘Victim understanding’ subscales. IPV screening in France depends on the institutions’ policies where health care providers work and not all encourage IPV identification unfortunately. The ‘Self-efficacy’ subscale comprised one item from the original ‘Workplace issues’ subscale. Being able to gather the necessary information to identify IPV as the underlying cause of patient injuries is more related to self-efficacy than to work site. The ‘Victim understanding’ subscale comprised three items on the seven items of the original subscale. Lastly, ‘Legal requirements’ and ‘Alcohol/drugs’ subscales were identical in the French and the original versions [19].

The other PREMIS versions also presented factor structures that were somewhat different from the original version. Two subscales were identified in all other studies, ‘Preparation’ and ‘Alcohol/drugs’ subscales. The Greek physicians’ version [23] identified a 2-item ‘IPV screening’ subscale. The ‘Legal requirements’ subscale was not found in this version where the items were excluded during the adaptation of the PREMIS [23], as the ‘Victim autonomy’ subscale. The Spanish version for physicians and nurses identified a ‘Barriers’ subscale [24]. The American pharmacists’ version identified one single subscale for self-efficacy and workplace-efficacy [22]. The American health care students’ version identified a 2-item ‘IPV screening’ subscale [21]. The ‘Workplace issues’ and the ‘Constraints’ subscales were not identified, as it was not appropriate for students [21]. The Australian paramedic and nursing students’ version did not find the ‘Workplace issues’ and ‘Constraints’ subscales [25]. The ‘Legal requirements’ subscale was not found in this version as the items were excluded during the adaptation of the PREMIS [25].

Internal consistency of the “Background” subscales was high and comparable to those of the original subscales with Cronbach’s α superior to 0.90 [19]. Same results were found for the physicians’ Greek and Spanish versions [23,24], the American and Australian healthcare students’ versions [21,25] and the American pharmacists’ version [22]. For “Opinions” section, ‘Preparation’, ‘Workplace issues’, ‘Legal requirements’ and ‘Self-efficacy’ subscales demonstrated good internal consistency (α = 0.88, α = 0.77, α = 0.76, and α = 0.72), comparable to those reported in the original version (α = 0.85, α = 0.82, α = 0.79, and α = 0.69 respectively) [19]. ‘Alcohol/drugs’ and ‘Victim understanding’ subscales displayed moderate internal consistency (α = 0.59 and α = 0.54), lower than those reported in the original version (α = 0.70 and α = 0.69 respectively) [19]. This moderate internal consistency was also found in the Greek physicians’ version (α<0.50 and α = 0.63 respectively) [23], in the American healthcare students’ version (α = 0.48 and α = 0.46 respectively) [21], and in the Australian paramedic and nursing students’ version (α = 0.57 for the ‘Alcohol/drugs’ subscale) [25]. The Spanish physicians and the American pharmacists’ versions reported higher internal consistency for the ‘Alcohol/drugs’ subscale (α = 0.66 and α = 0.80 respectively) [22,24].

The very small to moderate correlations between the “Opinions” subscales imply that they measures related but relatively different constructs. ‘Alcohol/drugs’ subscale displayed the lowest correlations with the other subscales, results found in the original development of the PREMIS [19] and in the other studies [21–25].

Convergent validity was explored by assessing the inter-subscales correlations. The majority of the correlations replicated findings of previous studies. Only the Spanish validation did not explore these correlations [24]. ‘Perceived preparation’ and ‘Perceived knowledge’ subscales were strongly correlated (r = 0.91). In the previous studies, correlations ranged from 0.74 to 0.89 [19,21–25]. These subscales were not significantly correlated to ‘Alcohol/drugs’ subscale, as found in the original version [19], the Greek physicians’ version [23] and the Australian healthcare students’ version [25]. Correlations between the amount of previous IPV training and the “Background” subscales were moderate and higher than those found in all previous studies [19,21–25]. ‘Actual knowledge’ subscale had small correlations with the amount of previous training, the “Background” and “Opinions” subscales. The higher correlation was found with the ‘Victim understanding’ subscale, which was the only significant correlation in the Greek physicians [23] and American healthcare students’ versions [21]. Short et al in the original version found significant but small correlations between ‘Actual knowledge’ and ‘Perceived knowledge’, and between ‘Actual knowledge’ and “Opinions” subscales, except ‘Preparation’ and ‘Legal requirements’ subscales [19]. In the American pharmacists’ version, ‘Actual knowledge’ was only significantly correlated to ‘Legal requirements’ subscale [22].

Test-retest reliability, which is an essential property [53], showed substantial to almost perfect ICCs, except for the ‘Alcohol/drugs’ opinions subscale, with moderate reliability. Only the Greek physicians validation reported ICCs for test-retest reliability over a 3- to 4-week period [23]. In our study, ICCs for the ‘Perceived preparation’, ‘Perceived knowledge’, ‘Actual knowledge’ and ‘Self-efficacy’ subscales were higher (0.92, 0.88, 0.69, and 0.83 versus 0.84, 0.78, 0.68, and.078 respectively). For the other similar “Opinions” subscales, our ICCs were lower: 0.65 versus 0.83 for ‘Preparation’ subscale, 0.78 versus 0.93 for ‘Workplace issues’, 0.74 versus 0.81 for ‘Victim understanding’, 0.46 versus 0.79 for ‘Alcohol/drugs’.

Our study has some limitations. First, participants were mainly women. This could be explained by the fact that midwives are widely represented in perinatal care occupations. In France in 2017, only 2.6% of midwives were men. As women, these participants certainly felt more concerned about this topic and this could have affected their responses. Almost one quarter of the participants reported no previous IPV training. In the other cross-cultural validations, this proportion varied from 45.7% [21] to 95% [23]. Nevertheless, Short et al reported that 13.5% of their first sample and 31.3% of their second sample used to validate the original PREMIS reported no previous training, which was closer to our results [19]. Besides, it was not possible to distinguish between town and clinic caregivers and those who are regular employees of the hospitals or health care institutions. Practice variations could be explained by the work site, in particular by the presence of protocol for dealing with abused women, IPV referral resources or institutional policies. For the reliability test, 24 respondents completed the retest, a number lower than the 50 subjects recommended by Terwee et al. [38], but higher than in the other validation studies:10 subjects for the Spanish version [24], 18 subjects for the Australian version [25] and 20 subjects for the Greek version [23].

Further studies are needed on sample with more men to confirm the psychometric properties of the questionnaire, and in other settings, like in primary health care physicians. The responsiveness to change, that is the ability of the PREMIS-French to detect changes after IPV training, had to be evaluated, as well as convergent validity, as no validated instruments was validated in French to serve for comparison.

Conclusion

This study provides evidence of the good psychometric properties of the PREMIS-French when delivered to perinatal care providers to assess their readiness to manage IPV. As the original and the other versions of the PREMIS, the PREMIS-French could be used in several ways: to assess knowledge, attitudes, beliefs, behaviors and skills in order to assess needs that could be addressed during IPV education program; as a pre and post-test to measure the changes over time (spontaneous evolution of practices) or after IPV trainings or interventions; to compare perinatal care providers who had received training and those who did not have it. This is a valid and easy instrument to use which will help to understand perinatal care providers’ barriers to IPV screening and management and will help to focus on specific lacks of knowledge for developing IPV education programs. This measure of educational outcome would also allow the evaluation of IPV training courses that will be developed in the near future, in accordance with the recent recommendations of the French National Health Authority.

Supporting information