Cluster randomized controlled trial protocol: addressing reproductive coercion in health settings (ARCHES)

This study evaluates a family planning clinic-based IPV and reproductive coercion (RC) intervention developed by a team of researchers, victim service advocates, and reproductive health practitioners [61]. Figure 3 describes the conceptual model of the ARCHES Intervention and Hypothesized Outcomes. One innovation of the ARCHES intervention is the focus on training not only clinicians, but also para-medical providers (i.e., medical assistants, health educators, and family planning counselors working in these settings) to discuss IPV and RC when counseling around contraception, pregnancy or STI testing. Additionally, the emphasis on universal provision of IPV/RC information recognizes that women often do not recognize IPV or RC and may not define sexual coercion as abuse, [62, 63] particularly when the perpetrator is known to them [63]. The lack of recognition of abusive behaviors in relationships [42, 43] has been associated with decreased IPV help-seeking, [63‐65] highlighting need for universal IPV/RC education and enhanced assessment.

ARCHES provides universal IPV/RC education and enhanced assessment through FP provider discussion of IPV/RC with their patients in a way that highlights the prevalence of such abuse among women seen at the clinic and educates patients about the reproductive health impact of such abuse. The enhanced assessment for IPV/RC integrates into the reproductive health visit, for example, by asking a woman seeking pregnancy testing whether her partner might be pressuring her to get pregnant. This education and assessment is facilitated by the use of a palm-sized brochure which describes healthy and unhealthy relationships, offers information about harm reduction, and provides IPV related resources. Evidence that clinic-based IPV assessment can be the first step in recognizing abuse, particularly when done in a context that normalizes such abuse experiences, [42, 66] strengthens the rationale for locating IPV and RC assessment within the context of supportive education for all women seeking FP services.

ARCHES also counsels women on harm reduction strategies. Harm reduction, originally used within substance abuse treatment, has been effective in managing a range of health risk behaviors [67] by ‘meeting clients where they are’ and assisting them with identifying strategies to decrease harm, including harms related to sexual health [68‐71]. IPV interventions appear to increase safety planning and harm reduction behaviors among victimized women, e.g., increase ability to refuse sex, [72] reduce substance use in dating contexts, [73] and advance preparation for safe escape should violence escalate [74]. Harm reduction behaviors have also been shown to protect against violence victimization among high-risk groups (i.e., women in prostitution), [75] and to reduce revictimization among college women [73]. Thus, ARCHES is designed to reduce women’s risk for violence victimization and unintended pregnancy via education regarding non-partner dependent contraceptives (longer acting reversible contraceptives such as the intrauterine device), access to emergency contraception, and provision of harm reduction strategies.

Finally, supported “warm” referrals (i.e., provider facilitation of referral to a victim service advocate via phone or in person) can assist clients in overcoming common barriers to accessing services, including self-blame, [43, 76] lack of recognition of abuse, [63, 64] lack of knowledge of services, [43, 76, 77] and perception that services are limited in scope (e.g., solely crisis oriented) [76]. Articulating the scope of services available to all women (regardless of disclosure of IPV or RC experiences), and normalizing use of these services may facilitate awareness and use of IPV services, improve mental health symptoms, [78‐81] and reduce revictimization [82‐84].

Nine clusters (11 clinics) were randomized to the intervention. In these clinics, family planning clinic staff (family planning counselors, medical assistants, and clinicians) were trained to educate clients about the links between reproductive health concerns and intimate partner violence/sexual assault (“enhanced screening”); counsel clients on harm reduction behaviors to reduce risk for IPV/RC (“harm reduction”); and provide clients with information on violence victimization support services (“supported ‘warm’ referrals”) with the goal of reducing IPV/RC and unintended pregnancy. The training, led by one of the PIs and the lead trainer from Futures without Violence (the non-profit violence prevention organization collaborating on this study), involved a half-day (approximately 3 h) introduction on the prevalence of IPV/RC, instruction on how to conduct universal IPV/RC client education with the aid of the palm-sized brochure, and role playing to practice IPV/RC assessment and provide harm reduction counseling for specific clinical scenarios (such as pregnancy testing). Training materials included copies of the reproductive health palm-sized brochure, reproductive health care setting medical guidelines, slides describing the prevalence of IPV and RC and relevance for family planning services, and video vignettes illustrating common reproductive health visits demonstrating with actors how to implement this universal education and brief counseling intervention.

Approximately 1–3 months after the initial training, in conjunction with starting data collection, a 30-minute refresher training was conducted by the study coordinator, to review key points of the training and provide opportunity for providers to ask clarifying questions about the intervention to increase fidelity.

To maintain quality of intervention delivery and track fidelity to the intervention, data collected from client exit surveys (brief surveys completed by clients as soon as their clinic visit was over) were summarized on a weekly basis, and sent via email to the office manager of each intervention site. The data highlighted client responses to two questions: “Today, did your health care provider talk with you about healthy and unhealthy relationships?” and “Today, did your health care provider give you a card called” Do You Know Your Relationship Affects Your Health?“talking about partner’s control of pregnancy and your reproductive health?” The proportion of clients reporting they had had this discussion and had received the palm-sized brochure was reported weekly to each office manager, with the study coordinator offering suggestions on strategies to increase provider discussions and card delivery.

Eight clusters (14 clinics) were randomized to the standard-of-care (‘control’) condition, meaning they continued diagnostic and treatment procedures as clinic staff typically would for their patients with no additional training. In discussions of sexual health or pregnancy risk, IPV may be discussed, but no training on or provision of protocols related to IPV education or referral were included as standard-of-care. Apart from such standard care, at the conclusion of the exit survey at the baseline clinic visit, control group participants received written materials regarding community resources which included IPV resources mixed in with other information around housing, food, and social services. Assigned to a wait-list condition, a control site could opt to receive the ARCHES intervention training following the completion of follow-up data collection.

A scale of 9 items was developed by the investigators to assess participants’ perceptions of how abusive behaviors specific to sexual and reproductive coercion are (e.g., “trying to make someone have an abortion when they don’t want one” and “pressuring someone to have sex when they’ve said no”). Participants are asked to rate the behaviors on a 4-point Likert scale from “not abusive” to “extremely abusive” (Cronbach alpha = 0.86). The outcome is measured as a mean score of the 9 responses with a higher score indicating greater perception of abusiveness.

Participants are presented with four statements that are designed to assess level of confidence in implementing behaviors to reduce the impact of reproductive and sexual coercion (e.g., “If my sexual partner is keeping me from using birth control, I am confident I could ask a healthcare provider about other kinds of birth control that my partner can’t mess with.”). Responses are rated on a 5-point Likert scale from “strongly agree” to “strongly disagree” with the outcome modeled as a mean response to the 4 items (Cronbach alpha = 0.72). To reduce measurement effect (i.e., the survey items themselves increasing women’s knowledge of harm reduction strategies), this construct is assessed at follow-up visits only (T2 and T3).

We also collected a general self-efficacy measurement at baseline, using The New General Self-Efficacy scale (NGES) developed by Chen, Gully, and Eden [89] to acquire the statistical power-enhancing advantages of a baseline covariate that is correlated with the above self-efficacy outcome, but whose assessment would not lead to a “measurement effect.” The NGES consists of 8 items (e.g., “Even when things are tough, I can perform quite well.”) with response options on a 5 point Likert scale (“strongly agree” to “strongly disagree”) (Cronbach alpha = 0.89). General self-efficacy is modeled as the mean response to these items and the baseline value is included as a covariate in the models assessing self-efficacy to implement harm reduction strategies.

Use of 6 harm reduction strategies to decrease the occurrence and impact of reproductive and sexual coercion and unsafe relationships was assessed at both follow-up visits. Participants are asked to endorse whether they had used these strategies (e.g., “have you hidden your birth control from your partner because you were afraid he would get upset with you for using it” and “have you tried to protect yourself from your partner by asking family or friends to call or text you when you were with that partner”) in the past 3 months (Cronbach alpha = 0.39). The outcome is modeled as a summary score with a maximum value of 6 (one point per each strategy endorsed). To prevent measurement bias, this construct is assessed at follow-up visits only (T2 and T3).

To assess knowledge of IPV-related resources and services, participants are presented a list of resources and asked to endorse each one for which they “had received information on in the past 3 months” (Cronbach alpha = 0.80). Listed resources included Domestic Violence Advocacy Services, Rape Crisis Center, National Domestic Violence Hotline, Teen Dating Abuse Helpline, and Loveisrespect.org. A summary score is calculated with one point per resource selected for a maximum score of 5. To reduce the possibility of participants endorsing items because they had seen the resources identified in earlier surveys, this measure was not included in the baseline survey.

In addition to the above outcomes, two additional outcomes are included for analyses restricted to women who endorsed any physical or sexual violence or RC at baseline, as this group was hypothesized to receive the greatest benefit from the intervention. For convenience, we refer to this analysis subset as the “Aim 2” analysis, with the full population analyses described as “Aim 1” analyses.

Participants are asked to endorse all IPV-related resources that they used in the past 3 months from the same list presented in the measure of knowledge of IPV-related resources and services (above) (Cronbach alpha = 0.55). As with the knowledge measure, this question is presented at follow-up visits only (T2 and T3). The outcome is modeled as a summary score, tallying number of endorsed items.

The study was approved by the University of Pittsburgh Institutional Review Board (IRB) on June 21, 2011 by expedited review (Protocol Number PRO11050458). The University’s IRB approval included a waiver for the requirement to obtain a written informed consent to participate in the research, in order to reduce the risk for breach of confidentiality (considered the primary risk associated with this study); women were offered a copy of an information sheet that was reviewed verbally with all clients. For minors ages 16 and 17, a waiver of parental informed consent for research participation was also approved by the IRB to protect youth seeking confidential services at these clinics from breach of confidentiality. A federal Certificate of Confidentiality was also obtained for the study to protect research records from subpoena.

Additional safety precautions were in place to reduce risk of harm to women for participating in the study, consistent with international recommendations and best practices on violence-related research [90]. Women who were at the clinic with a male partner were only approached about the study when they were alone. All follow up contacts made about the study referred to the study as the “Women’s Health Study.” To mitigate risk for emotional distress potentially associated with taking a survey about IPV, all research assistants were trained to ask women about how they were feeling after taking the survey, offer a list of community resources (with IPV and sexual assault information embedded among other less sensitive information about housing and food), and to suggest women speak with the clinical staff on site if they wished to talk more about the survey questions. Additionally, if any woman disclosed to a research assistant during survey administration that she was concerned about her safety, research assistants were trained to connect women with clinical staff as well as the local victim service advocacy organization. Finally, for follow up surveys conducted via email, women were encouraged to find a computer in a private space to complete the survey and to erase the history after survey completion. For those completing a follow up survey by phone, the research coordinator would suggest a special ‘code phrase’ the woman could use if she needed to hang up because someone else could overhear the conversation.

Research assistants (RAs) approached women in the waiting room of FP clinics after they checked in for their clinic visit if they were not in the presence of a male partner and asked if they were here for a clinic visit today, and whether they would like to hear about a women’s health research study. If they were interested, the RA assessed for eligibility. If eligible, potential participants were brought to a private space in the clinic to complete the consent process and survey administration. RAs let clinic staff know that the patient may be interested in being in the study and would be taking the survey before her appointment if she consented.

RAs explained to potential participants that they would complete two surveys (one before and one immediately after their clinic visit today) and receive a $15 gift card, one survey in approximately 3 months for $25, and one survey in 12 months for $40 to thank them for their time and effort.

Before participants’ appointments, they provided their contact information (phone number and email address) and two alternate contacts to the RA. RAs also went over the verbal consent sheet, which provided details on why the study was being conducted, the data collection process at each of the three time points, the implications of the Certificate of Confidentiality that was obtained for the study, the voluntary nature of participating, and the schedule of compensation. RAs then answered any questions participants had and asked participants to sign an optional medical chart release form (provided by the clinic) for chart abstraction at the conclusion of the study. After being assigned a four-digit ID number for the survey data, participants completed an audio computer-assisted survey instrument (ACASI) on a laptop with the option to have questions read aloud (in English or Spanish) through headphones. When participants finished with the baseline survey, they either returned to the waiting room to wait to be seen by the clinician(s) or went directly to an exam room for their scheduled clinical visit. RAs communicated with clinic staff to indicate which participants had taken the baseline survey so that they could direct the patients (after their clinical visits) back to the RAs for the exit survey.

After the participants’ clinic visits, they were asked to return to the private room with the RA to complete the client exit survey, also completed using the ACASI program on a laptop. Upon finishing this survey, RAs issued $15 gift cards, local resource sheets, and appointment reminder cards for the next follow survey to all participants. In intervention clinics, RAs also offered extra palm-sized brochures to participants, explaining one was to keep, and one was to give away to either a friend or family member. Control clinic participants only received the one page community resource sheet. When distributing brochures and resource sheets to participants, RAs told participants that taking these materials was voluntary, if they felt safe taking the materials with them.

Following a sequence of preliminary analyses to characterize the sample participants and outcome measures, statistical methods for cluster-randomized experimental studies will be used to assess study hypotheses. The anticipated study data is multilevel, with baseline and follow-up data collected from patients, the units of analysis, who are nested within clinics, the units of randomization. The primary assessment of intervention effects will be based on intent-to-treat estimates. As-treated effect parameters will be estimated in secondary analyses and reported as exploratory. Following preliminary phases of descriptive, bivariate and multivariate analyses to characterize the sample and to refine and characterize scale-based outcome measures, estimates and hypothesis testing will be conducted in a multilevel generalized linear mixed (regression) modeling framework that will specify within-clinic and within-patient cluster effects. Regression models will include a parsimonious set of covariates to adjust for the restricted stratified randomization of clinics, to reduce the potential for confounding arising from between-arm imbalances in patient and clinic characteristics, and to increase the precision of intervention effects, with the specification of link and variance functions tailored to the type of outcome variable (e.g., logistic links for binary outcomes).

Two types of biases will be specifically explored using statistical testing. Participation bias will be assessed by comparison of age, race/ethnicity, and other demographics of clients participating in the study with that of the demographics of the overall family planning clinic clients (at participating sites) during the period of data collection, with significant differences noted as potential threats to generalizability. Descriptive statistics and bivariate and multivariate analyses will characterize patient demographics and compare groups with respect to key baseline measures. Variables differing across treatment groups at the p < .20 level will be included as covariates in subsequent adjusted outcome analyses. A similar set of analyses will be conducted to compare those who complete all three visits to those who do not to assess attrition bias. If between-arm differences in attrition are found, this potential validity threat will be noted.

We will use generalized linear mixed effects models for evaluating the effect of the intervention on follow-up outcomes. Separate models will be created for each outcome, and predictors will include a parsimonious set of important covariates, including potential confounders identified in preliminary analyses, the baseline measure of the outcome (when available), and a clinic-level binary indicator for whether the patient was in a treatment (vs. control) clinic. Random effects will be specified to account for the nesting of patients within clinic and, where appropriate, for repeated follow-up measurements. Recognition of sexual coercion, reproductive coercion, and physical and sexual partner violence baseline measures will be available for all time points and thus baseline measures will be included as covariates in the respective models to improve the precision of our estimated intervention effects. Within-clinic and within-patient cluster effects will be specified as random effects in the models.

For Aim 2 hypotheses, the analysis will be restricted to the patients who report at baseline having recently experienced physical or sexual partner violence or reproductive coercion.

Following the definitive assessments of key study hypotheses, a series of exploratory analyses will be conducted. In particular, the potential for differential impact of the intervention based on patient characteristics will be evaluated, with a view toward identifying factors (e.g., race/ethnicity, age, immigrant status, involvement in commercial sex) that may facilitate or impede the efficacy of the intervention, and to identify patient subgroups most receptive to the intervention or that may require further assistance.

For each outcome, we will use all eligible records at both follow-up visits to model timepoint-specific intervention effects, with the first model specifying (TIMEPOINT x INTERVENTION) interaction terms to permit testing for homogeneity, using Wald-tests with a statistical significance threshold of 0.05. For outcomes without statistically significantly heterogeneous intervention effects, the primary assessment of the intervention will involve the overall main effect of the intervention (versus the control), pooled across both follow-up occasions. Otherwise, timepoint specific contrasts will be used for the primary assessment of the intervention on the outcome. Each outcome will be assessed with a two-tailed alpha = 0.05. No adjustments for multiple testing will be made, as we feel that each of these hypotheses is of independent scientific interest. Graphical and analytical methods will be used to examine model fit and to evaluate assumptions used to justify inferences.

The eligible population for these analyses is restricted to measurements from women who report (i.e., at the given timepoint) having heterosexual sexual intercourse (vaginal, oral, or anal). At each timepoint, this eligibility criterion was reassessed and further survey data necessary for defining outcome variables at that timepoint were collected only when this criterion was met. For women with follow-up data for whom the baseline survey did not collect responses needed to compute baseline covariates to be used in the mixed-effects models, baseline covariate values will be imputed.

In addition, for intermittently missing values (e.g., when the respondent selected “don’t know” or “refuse to answer”) for items in eligible follow-up visits, values will be imputed using multiple imputation (Markov Chain Monte Carlo algorithm) procedures in SAS PROC MI and PROC MIANALYSE for the imputation and analyses steps, respectively.