Improving the healthcare response to domestic violence and abuse in UK primary care: interrupted time series evaluation of a system-level training and support programme

Domestic violence and abuse (DVA) is a global health and societal concern [1]. It is a violation of human rights which damages health, posing a challenge to public health and clinical practice [2, 3]. World Health Organization [4], National Institute for Health and Care Excellence guidelines [5] and the most recent UK Government Department of Health & Social Care guidance [6] all support greater health sector involvement, to improve the healthcare of those affected by DVA. Most recently, the Guttmacher-Lancet Commission placed detection and management of gender-based violence, central to delivering sexual and reproductive health for all [7]. In most settings, including high-income countries, healthcare is still not responding adequately to violence against women, which will most commonly be DVA [8]. This response’s contribution to delivering better healthcare for all has not been realised [9]. Ten years following the completion of our cluster randomised controlled trial demonstrating the effectiveness of IRIS—Identification and Referral to Improve Safety of women affected by DVA—a complex, system-level, training, support and referral programme, designed to improve the primary healthcare response to DVA [10], we now report an analysis of a population-wide implementation evaluation of IRIS, over four years (2012 to 2017).

A previous systematic review of interventions to improve clinicians’ response to DVA [11], updated and developed to focus only on high-quality, low risk of bias trials, measuring outcomes changing either clinicians’ behaviour (identifying and referring cases) or patient outcomes, found three studies [ 10, 12, 13], two of which [10, 12] were identified in the original systematic review [11]. These higher-quality studies supported the review’s overall conclusion that the best strategy to improve clinicians’ responses to DVA is training clinicians, face-to-face (as opposed to online interventions that have not been evaluated for actual behaviour change)—with optimum duration and precise elements of training undefined—whilst increasing all staff’s understanding of DVA (not just awareness). Two of these studies supported a clinical case-finding approach [10, 12], with one intervention including system-supportive interventions–incorporating clinical reminders on identifying DVA within consultations and direct access for women to DVA services within the safe space of healthcare settings [10]. The most recent third study did not support a system of care for postal screening, followed by general practice invites to women affected by DVA [13]. Out of the three trials, the IRIS trial intervention was unique as it nurtured and established greater health services’ participation by linking NHS statutory primary care to the multi-sectoral response to violence against women, via the DVA agency-employed advocate educator [10]. The IRIS trial, based on 24 intervention general practices with controls, in two English cities (London and Bristol), over one year, reported an increase in recorded DVA identification (three-fold), referrals made for advocacy (seven-fold) and referrals’ discussion (22-fold). The intervention consisted of training sessions for the whole practice team, a focus on clinically relevant DVA case finding, using the accurate Humiliation, Afraid, Rape, Kick (HARK) tool [14, 15], embedded in the electronic medical record (EMR), reminding clinicians to listen for and consider DVA. Qualitative analysis nested within the original IRIS trial showed that women had positive experiences about being asked about abuse, when clinically indicated by health professionals and contact with DVA advocates [16]. Health professionals viewed IRIS as an acceptable intervention but were concerned that their four hours of training was lengthy [17]. Modelling with trial data showed that the IRIS intervention was cost-effective [18].

Following the trial, the IRIS intervention has been funded in 41 English and Welsh, urban and rural sites. In one quarter of sites, funding has stopped. Secure longer-term funding allowing more stable healthcare services’ involvement, at a much larger scale than at present, requires more evidence that the intervention is effective outside the trial context. We hypothesised that this system-level programme would be effective and sustained outside of a trial.

The aim of this study was to determine whether IRIS is effective when implemented in primary care outside a trial setting, in multiple sites, over four years. For full details of study design, participants, procedures, interventions, and data sources, see our protocol paper [19]. We have used the revised Standards for QUality Improvement Reporting Excellence (SQUIRE 2.0) [20] for reporting this evaluation (see Additional file 1).

Table 1 shows largely comparable characteristics, including population size, of the participating five boroughs (A, B, C, D and E). The general practices that were included in the ITS analysis were consistently less deprived than all general practices in each of the five boroughs. All the boroughs have diverse populations with substantial proportions of non-white ethnicity, with the highest proportion in the comparison borough.

Table 2 shows a summary of the data collected, with absolute numbers of referrals received by DVA service providers from general practices, new DVA cases identified and recorded by general practices and explanation of missing data. In the boroughs where IRIS was implemented (A, B, C and D), 144 general practices were funded to receive IRIS training. Twenty-two general practices did not receive it. Three practices had an unknown number of women registered. Two practices had missing training dates. One practice closed. Hence, 116 practices (81%) were included in the analysis. In borough E, 27 general practices out of 61 sent no staff to their DVA session, so 34 (56%) practices were included in the analysis. The IRIS programme was implemented in northeast London from November 2013. IRIS continued to be funded in all four implementation boroughs (A, B, C and D) at the end of the data collection on 31 March 2017. In the comparator borough E, the DVA information sessions were delivered from May 2014. The data collection period included the year preceding the implementation, in all five boroughs. This was a total of 53 months, from November 2012 to the end of March 2017.

The global IRR for referrals received in the four boroughs where IRIS was implemented was 30.24 (95% CI 20.55 to 44.77, p < 0.001). Table 3 shows the IRRs for referrals received in all five boroughs with the delivery of their separate DVA sessions. The IRRs for the individual boroughs (A, B, C and D) are consistent with the global IRR, representing a large increase in the number of referrals received with the first training session delivery, in contrast to borough E that has no improvement in referrals received with its session delivery (IRR 0.95, 95% CI 0.13 to 6.84, p = 0.959). Figure 1 shows the average daily referral rate per 1000 women registered in the IRIS boroughs’ general practices (A, B, C and D), plotted against time centred around the day of delivery of their first IRIS training session, illustrating the large effect of IRIS implementation on referrals, and post-implementation shows that IRIS has a sustained effect on the increased numbers of referrals received. This scatter diagram is consistent with our chosen step change impact model, supporting that this model fitted the data well across all four boroughs. Figure 2 shows that the average daily referral rate per 1000 women registered in borough E, before and after the day of delivery of their DVA session, did not significantly differ. The result from the sensitivity analysis with clustered standard errors was IRR 28.78 (95% CI 16.38 to 50.56, p < 0.001). These results allow for potential autocorrelation and are consistent with our main findings. The estimated effect of IRIS’ first training session delivery on referrals received is not significantly different in the four implementation boroughs (see Fig. 3), nor affected by socioeconomic status, as reflected by the deprivation level of each practice (see Fig. 4): low deprivation (score 10 to ≤ 30, medium (score > 30 to < 30.65) and high (score > 30.65). Cochran’s heterogeneity test shows there is no significant heterogeneity between boroughs or deprivation score level, producing p values of 0.460 and 0.850 respectively. IRISi reports that nationally, IRIS advocate educators have contact with and support 85% of the patients that are referred to them from general practice (IRIS National Report 2018).

Table 4 shows IRRs for new DVA cases identified and recorded by general practices in boroughs C and E with the delivery of DVA sessions. The IRR in borough C was 1.27 (95% CI 1.09 to 1.48, p < 0.002), demonstrating that there is a significant increase in the number of new DVA cases identified and recorded by general practices, on average with the first IRIS training session delivery. This increase is not seen with the other DVA model’s information session delivery in borough E (IRR = 1.05, 95% CI 0.82 to 1.34, p = 0.699).

Our interrupted time series analysis of data from five London boroughs shows sustained effectiveness of IRIS implementation over four years, with clinicians significantly increasing referrals to DVA service providers. A different DVA education model, focussed on conveying information, in a comparable neighbouring borough, more representative of usual care in the UK, did not increase referrals. The difference between the boroughs where IRIS was implemented and the other comparator borough was substantial, with the global IRR in the former being 30.24 whilst the IRR in the latter 0.95.

In the boroughs where IRIS was implemented, the preceding time trend confirms that this large step-wise increase in referrals with IRIS training delivery was not due to an underlying temporal trend. The other borough’s time trend before and after the delivery of its DVA session showed no equivalent step-wise change. Following the step-wise increase in the boroughs where IRIS was implemented, the higher referral rate continued, with no significant differences across the four boroughs, and did not vary by deprivation status of the practice populations.

This study was sufficiently powered to detect a clinically important increase in referrals. In the four boroughs where IRIS was implemented, with the first IRIS training session delivery, each IRR increases (far excess of doubling), reflecting the large increase in referrals made by individual general practices with IRIS implementation—a large relative difference in referral numbers. The wide confidence intervals of these IRRs, in these four boroughs, reflects the uncertainty about the exact point estimate—but examining the lower limits of all four of these confidence intervals (lowest 7.51 in borough A), there is always much more than a doubling in the referrals received by the DVA agencies from general practices. The relative increase in referrals received with IRIS implementation was much larger than the relative increase in recorded identification of new DVA cases. The IRIS advocate educators’ on-going support, consultancy and interactions with the whole practice team, including refresher training, conceivably have a greater impact on referrals than on clinicians’ recording of new DVA cases. The absolute numbers of referrals made with IRIS implementation are much smaller than the numbers of new DVA cases recorded within the EMR (in borough C 394 referrals vs 2808 cases over 3½ years), as the latter is dependent on many external factors, including communication with other statutory organisations such as the police, with a child safeguarding focus. In turn, the numbers of new DVA cases recorded are smaller than the estimates of DVA prevalence in community or healthcare settings—using a one-year prevalence estimate of 4.2% [29], in borough C, there would be ~ 4418 cases. This difference is unsurprising, as most DVA is not shared outside of the home. This also indicates potential gains from continuing to improve the healthcare response to DVA.

Study strengths include no exclusion of potential borough sites, outcomes measured on a daily basis and sustainability examined robustly. Data collection in the four boroughs where IRIS was implemented surpassed the minimum of 24 months pre-specified in the sample size calculation. Collecting more data than was required, over a longer time period, gave us the capability to evaluate the sustainability of an effect over time, rather than only measuring the initial change in referral rate following delivery of IRIS training. Data accuracy was validated by coinciding sources if possible (for example, national data compared to local data). The primary outcome measure, referrals received by DVA service providers from general practices, was potentially susceptible to ascertainment bias, since the agencies knew they were part of the IRIS programme. Data collection was likely to be the most complete in the comparison borough, as all referrals were collected at a single site, the One-Stop Shop, ensuring ascertainment bias was less likely here than the IRIS implementation boroughs. Therefore, the comparison of effect differences was conservative. Referrals actually received by DVA service providers, as opposed to those reportedly made by clinicians, have proved to be the most reliable measure of a healthcare system responding to DVA [10]. This observational study’s community setting, in routine general practice, with borough-wide implementation, with the IRIS effect not significantly different in the four implementation boroughs with diverse populations, makes its results generalisable to similar settings in the UK and potentially to comparable primary care systems in high-income countries. For lower-income countries [30], this healthcare response to DVA is modifiable, if the health setting is safe, using existing community resources if DVA service infrastructure is absent.

Study limitations include not measuring direct patient-level outcomes, using exposure to DVA advocacy as a proxy measure, outcome assessors not blinded to data origin and no double entry of data. DVA service providers generously shared their data—but we did not have complete access to the entirety of all their data sets so we were unable to assess its completeness. In two of the IRIS implementation boroughs, smaller DVA service providers, receiving less than 20% referrals, did not routinely record their client’s registered general practice, precluding their data from the analysis.

Our study did not have any randomisation component. However, overall, the causal inference that IRIS causes clinicians’ behaviour to change, increasing the referrals received by DVA service providers, is strengthened, as it is now supported by both the preceding randomised trial, as well as this ITS. Furthermore, the ITS analysis of the comparator borough E without IRIS, showing no increase in the referral rate, is also evidence that it was IRIS that was responsible for the increase in referrals received of women affected by DVA. Unlike other DVA work, this study is linked to a cost-effectiveness analysis, estimating that IRIS outside the trial setting is cost-effective from a health service and societal perspective, moreover good value for the NHS and cost saving for society (the incremental net monetary benefit was £22 and £42 respectively) [31].

Our study provides evidence that a system-level programme that embeds direct referral pathways to specialist DVA agencies within health services, underpinned by training of clinicians and their teams, including on-going reinforcement strategies and processes in place, from the outset, improves the healthcare response to DVA. This can be successfully implemented, with continued effectiveness and sustained over four years simultaneously in multiple areas, outside of a trial setting. Global health policy is responding to the challenge of ensuring that health services understand their role in providing effective and compassionate healthcare in the context of DVA [4]. For health professionals to engage effectively, internationally as well as in the UK, further resources are required (within health and for specialist DVA agencies), best care reconsidered and new grass root adopters of DVA programmes respected [32]. As a healthcare response to DVA is implemented, the next step is translation research questions, directly assessing population health impact [33]. This study provides further evidence for funding of healthcare-based programmes that combine direct referral pathway to specialist DVA services with training and on-going reinforcement.