Implementation of the Identification and Referral to Improve Safety programme for patients with experience of domestic violence and abuse: A theory‐based mixed‐method process evaluation

Abstract Identification and Referral to Improve Safety (IRIS) is a training and support programme to improve the response to domestic violence and abuse (DVA) in general practice. Following a pragmatic cluster‐randomised trial, IRIS has been implemented in over 30 administrative localities in the UK. The trial and local evaluations of the IRIS implementation showed an increase in referrals from general practice to third sector DVA services with a variation in the referral rates within and across practices. Using Normalisation Process Theory (NPT), we aimed to understand the reasons for such variability by identifying factors that influenced the implementation of IRIS in the National Health Service (NHS). We conducted a mixed‐method process evaluation which included: (a) a case study (100 hr of participant observation, 19 interviews); (b) a survey (n = 118); (c) qualitative analysis of free‐text comments from the survey; (d) qualitative interviews (n = 8); (e) document review (n = 44). Data were collected from NHS and third sector staff across five London boroughs from August 2015 to December 2017, analysed descriptively and thematically and triangulated using the NPT constructs coherence, cognitive participation, collection action and reflexive monitoring. The survey showed wide variation in the extent to which practice staff saw IRIS as a normal part of their daily work. Qualitative data and documents illuminated drivers of DVA work, implementation barriers and suggested solutions. The drivers were related to individual professional's characteristics and relationships. The barriers were linked to the differing sense‐making and legitimisation of DVA work and differing contexts between the NHS and third sector. Solutions were adaptations to IRIS relative to these contextual differences. The suggested solutions can be used to update IRIS commissioning guidance, training for trainers and training for general practice. The updates should reflect the importance of ongoing support of IRIS from practice leads and commissioners, extended funding periods for IRIS and continuity of the IRIS team.


| INTRODUC TI ON
Domestic violence and abuse (DVA) is a global public health and clinical problem (Department of Health, 2005;NICE, 2014;WHO, 2013) that causes significant morbidity and disability among women (Feder & Howarth, 2014). In the UK, the largest cost associated with DVA is to the National Health Service (NHS): £1.7 billion per year with the major cost borne by acute trusts and primary care (Walby & Olive, 2014). Almost all women with experience of DVA access the NHS regularly, either as the first or only point of contact with professionals (Department of Health, 2010). Although healthcare practitioners cannot meet all the needs of patients affected by DVA, they potentially play a pivotal role in the multisector response through identifying such patients and referring them to local DVA services (Garcia-Moreno et al., 2015) which are largely based in the third sector.
The Identification and Referral to Improve Safety (IRIS) intervention is a programme of training and support to improve the response to DVA in general practice. The programme focuses on the identification of women patients affected by DVA, an appropriate response by clinicians, and referral to a specialist, named IRIS advocate educator (AE) (Gregory et al., 2010) leading to increased safety and improvement in women's health and well-being (Rivas et al., 2015). The IRIS model as intended is described in detail elsewhere . Box summarises the implementation of the IRIS intervention as intended.
A pragmatic cluster-randomised controlled trial showed that the IRIS intervention increased the rate of referrals to DVA services sevenfold (Feder et al.,2011). IRIS has also been found to be cost-effective (Devine, Spencer, Eldridge, Norman, & Feder,2012) and acceptable to clinicians (Yeung, Chowdhury, Malpass, & Feder,2012) and patients (Malpass et al., 2014). Following the success of the trial, IRIS has been implemented in over 30 administrative localities in the UK. In line with the trial, local evaluations of IRIS implementation showed an increase in K E Y W O R D S domestic violence, general practice, implementation research, primary health and social care interface, primary healthcare, process evaluation What is known about this topic • IRIS is an intervention for general practice addressing domestic violence and abuse (DVA).
• After a trial showing increased rates of referrals from general practice to third sector domestic violence services, IRIS was widely implemented.
• As in the trial, local evaluations of the implementation found wide variation in referral rates within and across practices.

What this paper adds
• IRIS helped initiate and maintain the work of identifying and referring patients affected by DVA from general practice to the third sector.
• Variations in referral rates can be due to differing understanding of DVA among clinicians, which is influenced by both individual and practice level factors, with system-level barriers acting as an additional challenge.

Box 1 IRIS implementation as intended
Identification and Referral to Improve Safety (IRIS) integrates National Health Service (local health commissioners, general practices) and the third sector (local providers of domestic violence and abuse [DVA] services) in a multi-sector response to DVA. A social enterprise IRIS Interventions (IRISi) supports the local commissioning, implementation, maintenance and growth of the IRIS programme (IRISi, 2017).
IRISi managers raise awareness of the model across the UK and respond to interest from local health commissioners wanting to implement it. Commissioners appoint a third sector organisation (IRIS host) through a tendering process and identify a general practitioner (GP) interested in DVA to act as a clinical lead (CL). The host organisation recruits a specialist DVA worker. The GP and DVA worker receive training from IRISi in how to deliver the service and become the IRIS CL and IRIS advocate educator (AE) respectively. The commissioner covers the costs associated with these posts and pays an annual fee to IRISi. A local steering group of stakeholders is set up to monitor the implementation of IRIS. The CL and AE identify general practices interested in accessing the IRIS service and work with up to 25 practices to provide in-house training, patient and professional resources and referral pathways for all patients affected by DVA.
Resources include: (a) training pack; (b) referral forms; (c) care pathways; (d) electronic prompt in the medical record triggered by clinical presentation associated with DVA (Humiliate Afraid Rape Kick Safety [HARK] template); (e) DVA posters; (f) wallet size cards for patients.
The AE is the named contact for patient referrals; she provides DVA advocacy to the patients, and DVA consultancy and ongoing support to practice staff. The practice is often used as a safe setting where the AE meets with referred patients, though meetings also happen within the community. referrals from general practice to IRIS AEs with a variation in the referral rates within and across general practices (Howell, Johnson, Goddard, & Harrison, 2016;Johnson, Downes, Howell, Goddard, & Harrison, 2018). We hypothesised that the variability in the practice level outcome may reflect influential implementation factors. This study is aimed at understanding the reasons for the outcome variability by identifying factors that influenced the implementation of IRIS in the real-world NHS.

| Design
Informed by the Medical Research Council guidance on evaluating complex interventions (Moore et al., 2015;MRC, 2008), we conducted a theory-based mixed-method process evaluation of the implementation of IRIS which included: (a) a case study; (b) a survey; (c) qualitative analysis of free-text comments from the survey; (d) qualitative interviews; (e) document review. This process evaluation was carried out alongside the evaluation of the outcomes  and cost-effectiveness . The choice of the theoretical and analytical frameworks, study design and methods were influenced by the complexity of the IRIS intervention, the target audience of people involved in implementation of DVA programmes and the experience of the research team.
We conceptualised IRIS as a complex intervention (MRC, 2008) because the model: (a) includes several components; (b) requires changes in professional behaviour and ways of working at individual, organisation and inter-organisation levels; (c) involves co-ordinated work across NHS and third sector; (d) permits some adaptions to local context. The authors, with backgrounds in health services research (GF, CG, NL, AS), implementation science (CG, AD), health psychology (NL) and social science (AD), have approached the study with a paradigmatic perspective of critical realism (Shannon-Baker, 2016).

This study was informed by Normalisation Process Theory (NPT)
-a middle-range socio-behavioural theory (May & Finch, 2009) which has been most commonly used to assist understanding of interventions as part of feasibility studies and process evaluations . NPT offers a framework with four constructs and 16 sub-constructs to assess the behaviour change and work that individuals and teams do to implement a new practice into their daily routine (Finch et al., 2013). In the context of IRIS, we conceptualised 'practice' (synonym 'DVA work') as the change in professionals' behaviour and ways of working leading to identification of patients with experience of DVA and referral to the IRIS service. We used the NPT framework to formulate propositions for the successful embedding of DVA work in the daily routine of general practice (see Table 1). We then interrogated our data against these propositions to identify implementation factors that could promote and inhibit their effectiveness and therefore explain the variation in referral rates within and across practices.
The choice of a mixed-method approach was informed by prior research in the field of DVA (Bacchus, Buller, Ferrari, Brzank, & Feder, 2018;Hooker, Small, Humphreys, Hegarty, & Taft, 2015;Hooker, Small, & Taft, 2016) and was based on several assumptions. First, it allowed us to capture the complexity of the IRIS intervention and of the implementation context (Greene, 2007). Second, it helped to draw a more complete picture of the implementation process through answering

NPT construct
Application to the normalisation of IRIS 1. Coherence -sense-making work DVA work should make sense to the general practice team and third sector organisation team (communal specification) and the individuals (individual specification); DVA work should match norms and values of NHS and third sector staff (internalisation); it should be distinct from other work and comprehensible to all the actors (differentiation).

Cognitive participation -relational work
NHS and third sector staff should work together to come to an agreement on DVA work (legitimisation); establish ways of working (enrolment); initiate DVA work with resources (initiation); and collectively establish ways to sustain it over time (activation).
3. Collective action -operational work in a given setting NHS and third sector staff should have access to IRIS resources to support DVA work and use these resources in the context (contextual integration) and the group (relational integration); they should develop ways to work with each other and the resources to accomplish the DVA work (interactional workability) and figure out a way to divide labour to identify and care for patients with experience of DVA (skill-set workability).

Reflexive monitoring -appraisal work
NHS and third sector staff should work out a system to define, collect and collate information about effects of IRIS (systematisation); work together and individually to appraise their DVA work and evaluate its worth (communal and individual appraisal); they should (if needed) modify IRIS for their context (reconfiguration).
Note. DVA: domestic violence and abuse; IRIS: Identification and Referral to Improve Safety; NHS: National Health Service; NPT: normalisation process theory.
TA B L E 1 Propositions for the successful embedding of DVA work in the daily routine of general practice mapped on the Normalisation Process Theory framework different research questions (see Figure 1). Finally, it enabled the gathering of a wider range of data from multiple sources. This was important to compensate for poor engagement of general practice staff in research known from previous studies (Lewis et al., 2017;Parkinson et al., 2015).
The use of NPT throughout the collection and analysis of quantitative and qualitative data helped to overcome some of the epistemological challenges of combining data (Farr et al., 2018). We followed a triangulation protocol as described by O'Cathain et al (O'Cathain, Murphy, & Nicholl, 2010). Four data sets were collected separately and analysed using the NPT framework. The triangulation took place at the analysis and interpretation stage through mapping quantitative and qualitative results onto the NPT constructs.

| Data collection
This study took place in five London boroughs (local government districts) which implemented IRIS between November 2010 (when the trial ended (Feder et al., 2011)  and their relevance to the wider study. This case study was selected based on it being an instrumental case (Stake, 1995), that is, an example of IRIS that was considered to be very similar to the model developed during the trial (Feder et al., 2011) and which could shed light on other cases. These data provided insights which informed the development of subsequent data collection procedures across the all wider study localities.
IRISi provided a shortlist of potential localities in the London area, excluding localities that had been involved in the original trial or delivered IRIS anomalously. One researcher (AD) approached three potential sites and one (locality III) consented to participate.
The researcher was permitted to conduct participant observation (Spradley, 1980) and interviews (Rapley, 2004). Participant observation was undertaken to gather insight into the organisation of the IRIS (attending organisational delivery and steering group meetings) and the delivery of training (attending general practice IRIS training sessions). NHS interview participants were purposively sampled for diversity in career stage (early-, mid-and senior) and familiarity with IRIS (referring/non-referring).
The customisation was informed by the case study and took place through consultations with instrument authors and IRIS providers, two pilots and three revisions.

| Qualitative interviews
Researchers (NL, AD) agreed to conduct up to 15 semi-structured interviews (Murphy, Dingwall, Greatbatch, Parker, & Watson, 1999) with general practice staff and AEs, following the Malterud et al (Malterud, Siersma, & Guassora, 2015) approach to determining the 'information power' required to generate sufficient insight for the study. The interview topic guide, which was informed by previous research (Hooker et al., 2015) and our case study, reflected the four NPT constructs. We piloted and refined the topic guide during the first three interviews.
To reduce research burden, we recruited interview participants from the survey sample across four of five participating boroughs (except locality III previously involved in the case study). We drew a purposive sample of NHS and third sector participants in relation to locality, professional roles (AE, clinical and non-clinical practice staff) and familiarity with IRIS (referring/non-referring). First, we invited all IRIS AEs working with survey practices to participate in the study. Then we selected one survey practice in each locality with a midlevel rate of referrals to IRIS and asked their AE to send our interview invite to practice manager, referring clinician and non-referring clinician. The researcher (NL, AD) obtained informed consent from professionals who expressed interest and arranged interviews at a convenient format, time and place. All interviews were audio recorded and transcribed. From the case study data set in locality III, we sampled four interviews transcripts meeting the above criteria.

| Document review
Researchers (NL, AS, AD) collected emails, meeting minutes, reports and other working documents from IRISi, NHS and third sector organisations in each locality (Shaw, Elston, & Abbott, 2006).

| Qualitative data
Interview audio recordings were professionally typed using in-  (4) 24 (22) 68 (63) 12 (11) 3.2. Relational integration Do they build accountability and maintain confidence in the DVA work and each other?
TA B L E 2 (Continued) and a deductive approach using NPT as the analytical framework (Finch et al., 2015). First, we coded inductively discussing findings regularly to ensure reliability. Then, we organised codes into themes summarising influential implementation factors and grouped them under the NPT constructs and sub-constructs. The NPT framework allowed us to frame the findings in the language of the theory and to provide a structure for combining quantitative and qualitative data.

| Documents review
Researchers (NL, AD) read and re-read each document, extracted data on core implementation information into a Word

| FINDING S
The flow of participants through the study is shown in Figure  The survey sample was overrepresented by experienced female clinicians, 60% of whom had attended IRIS training and nearly half had referred patients to IRIS service (see Table 3). The interview sample included six female AEs practising between 6 months and 4 years and six members of staff from general practices (one practice manager, one healthcare assistant and four general practitioners GPs) practising between 7 months and 21 years. Of six practice staff, four attended IRIS training and four referred patients to IRIS.
The document review showed that IRIS was funded by varied local health commissioners (one NHS Clinical Commissioning Group (CCG), two local authorities, two joint funders) and hosted by three third sector organisations (see Table 4). Two host organisations were charities specialising in DVA, based on an explicitly feminist perspective and one was a charity for peo-  Table 2). Another notable trend was a comparatively high proportion of neutral answers (3%-34%) indicating professional uncertainty about the service. Free-text comments provided possible explanations: no direct contact with patients (n = 7) and not aware of IRIS being implemented in the practice (n = 4).
Qualitative findings were organised into the three themes (drivers of DVA work, implementation barriers and suggested solutions) under the four core NPT constructs. The drivers were related to individual professional's characteristics and professional relationships.
The barriers were linked to the differing sense-making and legitimation of DVA work and differing implementation contexts between NHS and third sector. Solutions represented IRIS adaptations to these contextual differences. We report findings according to the NPT constructs.

| Coherence (making sense of DVA work)
The survey showed that DVA work was meaningful for most individuals and teams across general practices, although less than a half could differentiate it from routine practice (see Table 2). This can indicate either unawareness of the IRIS service or its normalisation to the extent that the service becomes part of routine practice.

| Cognitive participation (enrolling people into DVA work)
Most general practice staff were positive about participating in training and referring to IRIS, although less than 50% could identify DVA leads in their practice (see Table 2). The case study also found that while it was possible to initiate a new approach to DVA work through training, a barrier was having key people within practices supporting the ongoing activation of the work among the other demands of primary care environments.
Interviews and documents showed that AEs were the main driver for DVA work. can be just seen as slightly as an after, oh that's a really nice add on but we need to get the, I don't know, the diabetes training done first because that's sort of a requirement that's come in."

| Collective action (enacting DVA work within context)
Most respondents reported that they have received sufficient training and resources for enacting DVA work within their practice, although up to 25% were not sure if this work was assigned to those with appropriate skills (see Table 2). In line with the survey, most interview participants acknowledged the importance of the IRIS training and peer influence in initiating DVA work. However, differing organisational cultures of general practice and the third sector made it challenging to enact IRIS and sustain it over time.
AEs had to undertake a long period of invisible work establishing relationships and building trust to get into general practices before the intervention could begin. This demanded significant flexibility on the part of the service, as these periods of invisible work did not result in any referrals. After the service was launched, AEs used constant reminders and visits to practice to maintain the rates of referrals.
While there was confidence in the allocation of work between general practice and third sector organisations, short-term funding for the IRIS host organisation, resulting in professional uncertainty and staff turnover, made it difficult for participants to build confidence in one another and embed DVA work into their routine practice: And sector because if these type of service are just disappear I think, they don't have anyone to refer to, you know, they're not going to build those relationships. (AE04) The decommissioning and recommissioning of the service is a real hassle on the front line.
Further, half of AEs felt that NHS culture and processes were too slow and bureaucratic to match the nature of the DVA work which they conceptualised as an emergency: The CCG have been very … they're not being obstructive -they just go at their pace as though they have all the time in the world. They're very demanding when they want information, but they're not as forthcoming when you need it, so I find that can be difficult.
The overall emotional toll of general practice was mentioned as one of the factors preventing some clinicians from consistently asking

| Reflexive monitoring (monitoring and sustaining DVA work over time)
The survey showed that most NHS staff felt that IRIS is worthwhile and IRIS feedback can be used to improve the service. However, only a third of respondents were aware of any IRIS reports and only half could assess the effect of IRIS on their own practice (see Table 2). Q ualitative data also showed that the feedback component to general practices. Interestingly, good performance on all agreed metrics for monitoring did not protect the service from a period without funding in three of five localities (see Table 4).
According to AEs, monitoring made up a large part of their workload, which they felt took away from the time they could spend en-

| Strengths and limitations
The strengths of this process evaluation include a mixed-method approach and theory-informed analysis and interpretation of findings.
Quantitative tools included a reliable validated NoMAD instrument. Qualitative study with NHS and third sector participants gave equal voices to both professional groups and illuminated converging and conflicting perspectives within and between the groups.
Involvement of researchers with multidisciplinary backgrounds throughout data analysis broadened the interpretation of findings.
Study limitations are related to poor engagement of general practice staff in the research, which has been previously reported (Lewis et al., 2017;Parkinson et al., 2015).

| Practical implications
We demonstrated how the mixed-method approach and NPT framework can be used in the evaluation of a post-trial implementation of a complex intervention across healthcare and third sectors. Our findings are relevant to the implementation and sustainability of any complex intervention which involves multi-agency work when providing whole-person care to patients with medico-social problems.
Solutions to implementation barriers can be used to update IRIS commissioning guidance (Howell & Johnson, 2011), IRISi training for trainers and IRIS training for general practices, and could be of relevance more broadly for DVA interventions in healthcare. Updates should reflect the importance of leadership with regard to DVA both within individual practices and by commissioning bodies, and the vital role of effectively managed communication between NHS and third sector practitioners for building shared understanding of the service. The damage to confidence in the service that results from the uncertainty of short-term funding should also be emphasised.
Updates could consider how IRIS training could be locally adapted to fit into an over-burdened general practice, perhaps by blending face-to-face training with e-learning.

| Conclusion
The IRIS model facilitates behaviour change among general practice staff and collaboration between the NHS and third sector, with the aim of initiating and sustaining DVA work. The IRIS AE is the main driver of the IRIS model bridging the NHS and third sector, maintaining consistency of the core model components whilst adapting its delivery to fit into differing organisational contexts. Ongoing organisation and system-level support from the funder and practice leadership enable DVA work to be sustained.
Repeated training and the physical presence of the AE in the practice supports sustainability of that work. Continuous iterative evaluation and feedback acceptable to both NHS and third sector staff could improve appraisal of the DVA work. The approach taken in this paper demonstrates the value of conducting a theoretically informed process evaluation to further understanding of the implementation of complex interventions in real-world settings.

ACK N OWLED G EM ENTS
This research was funded by the National Institute for Health

Research (NIHR) Collaboration for Leadership in Applied Health
Research and Care North Thames at Bart's Health NHS Trust (NIHR CLAHRC North Thames). The views expressed in this article are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care. We thank data manager Farah El-Shoghri for helping with quantitative analysis and the team at IRISi and IRIS commissioners for providing documents for review.