Adults receiving treatment for behavioral health problems such as intimate partner violence, child abuse, substance abuse problems, and suicidality represent only a small fraction of those affected (e.g., Brown et al. 1998; Demyttenaere et al. 2004). These problems are stigmatized, and individuals frequently do not let others know of their difficulties, leading us to label such troubles as “secretive problems” (Heyman et al. 2011). Secretive problems are prevalent in military populations. In anonymous surveys of the Air Force (AF), about 35% of active duty members reported substance abuse, intimate partner violence (IPV), child abuse, or suicidality at a clinical level, yet only 1 in 13 of those reporting a secretive problem indicated that someone in uniform was aware of it (Heyman et al. 2011). These problems also exist at subclinical, high-risk levels for a larger segment of the population (e.g., Lorber et al. 2018). Given the breadth of need and the lack of voluntary revelation to either formal or informal help networks, a prevention science approach to implementing effective, efficient interventions widely is needed (e.g., Damschroder et al. 2009).

One challenge when designing and implementing prevention strategies is the need to be efficient. It is difficult to engage target populations in prevention activities, so, ideally, interventions would be both effective and impact numerous outcomes simultaneously. It is unlikely, for example, that broad swaths of the population would be willing to participate in a series of curricula, each seeking to prevent a different adverse outcome. One approach, taken by Communities that Care (CTC; Hawkins and Catalano 2002), is to target cross-cutting risk and protective factors (RPFs). CTC is not a specific prevention program. Rather, it is a framework targeting an interconnected net of youth problems. Community and school-based action planning teams are guided through conducting a needs assessment that assesses risk and protective factors, identifies priorities, and selects evidence-based prevention activities that they implement. With ongoing evaluation and refinement of plans and implementation strategies, CTC offers schools and communities a standardized framework for flexibly using evidence-based programs.

Research over the last few decades has made it clear that not only are secretive problems interconnected (e.g., Foran et al. 2014; Lorber et al. 2018), but they also appear to share a variety of RPFs (e.g., depressive symptoms, social support) that appear in the separate literatures for each problem (e.g., Foran et al. 2012). Thus, these RPFs may offer efficient intervention targets for integrated, community-level prevention. A community-based, public health approach focused on RPFs offers the advantage of not requiring high-risk individuals to be identified and referred to potentially stigmatized services. A risk factor–focused approach can be more efficient by focusing on RPFs that have impacts on multiple secretive problems (e.g., Hawkins et al. 2012).

We developed NORTH STAR (New Orientation for Reducing Threats to Health from Secretive Problems that Affect Readiness), a prevention planning and implementation system for adult problems in the AF, as a parallel to CTC (Hawkins and Catalano 2002), which targets adolescent problems. Like CTC, NORTH STAR is a system rather than a program, stepping community prevention teams through implementing a local community assessment, using data to select RPFs with multiple impacts, implementing evidence-based interventions to affect the selected RPFs broadly, and evaluating their impact. This approach is compatible with a limited resource context, where efficiency and sustainability are critical, making use of light-touch interventions with broad reach to make community-level changes.

Although population trials of preventative interventions are uncommon, Prinz et al. (2009) completed such a trial on the Triple P parenting program in South Carolina. Counties were randomized to treatment or control, and public records were used to conduct the evaluation. The intervention itself targets parenting at several levels of intensity that community members self-select. The goal is not solely individual-level change (from directly receiving the intervention), but rather population-level change. This requires that enough of the target population receive some dose of the intervention directly, or indirectly through social networks, for the overall rate of maltreatment to be lower in treated communities.

Although evaluating a program rather than a system, Prinz et al.’s trial shares several theoretical underpinnings and methodological characteristics with the current study. First, the theory behind the population-level intervention supposes that people can directly benefit from participation in an intervention, but may still be affected by an intervention in which they do not participate. Instead, if intervention penetration is sufficient, people will benefit indirectly. Although these interventions do not specify a mechanism through which this might occur, some possibilities include modeling of healthier behavior (e.g., Latkin and Knowlton 2015), social contagion originating from those whose behavior was affected by the intervention (e.g., Perkins et al. 2015), or a shift in the social norms for healthy and unhealthy behaviors (e.g., Sheeran et al. 2016). Thus, the premise is that if an approach is effective at the population level, effects should be apparent in population parameters. Geographical areas are the units of randomization, and outcomes are tested with cross-sectional population parameters regardless of the degree to which the individuals captured directly participated in any interventions. This is arguably a high bar to hold a prevention approach to, for it assumes that (a) specific component interventions will be effective when implemented in real-world settings under real-world conditions and (b) efforts to disseminate interventions will be effective in generating sufficient participation rates that the impact will be detectable at a population level.

In addition, the NORTH STAR approach, as a system for community-based efforts, left the implementation of programs in the hands of the communities themselves (with support), which further necessitates that communities be effective in their implementation of dissemination efforts. Although the emerging field of implementation science has begun to research, systematically, how to best disseminate empirically supported prevention approaches (e.g., McHugh and Barlow 2010), this field is young enough that these real-world efforts are based more on experience and anecdote than science. When these implementation challenges are coupled with the logistic necessity of working with small numbers of geographic units, it becomes apparent why so few population trials for IPV, child abuse, substance problems, or suicidality have been conducted. That said, the need to develop effective community-based prevention approaches for these problems is clear.

Given the implementation challenges we anticipated, we expected that NORTH STAR’s effects would be moderated by AF base Community Action Teams’ attitudes, context, and the quality of their implementation plans (i.e., CAT process factors). In the implementation science literature, these factors are determinants that serve as barriers and facilitators of prevention effort impacts and thus can interact with prevention approaches to affect change (see Damschroder et al. 2009). Given that NORTH STAR is a system of planning, selecting, and implementing effective activities that organizes and directs the actions of already existing community teams, we reasoned that the system would be most effective when determinants were supporting effective action (e.g., poor CAT collaboration, high barriers to implementation, and poor community support). In contrast, when determinants align to undermine effective prevention, the AF’s systems would be less affected by the extra structure and tools within NORTH STAR.

Military services are ideal organizations within which to study community-based prevention. First, military installations are semipermeable systems that are both part of, and separate from, their surrounding communities making them well suited for comprehensive, multiproblem, focused prevention. Second, the AF had already (a) committed to preventing all of the targeted problems and (b) created an infrastructure to coordinate prevention activities among relevant agencies. Third, the AF conducted a biennial community assessment (CA), comprising theory-derived, psychometrically sound measures of individual, family, workplace, and community functioning that could serve as a data source regarding both RPFs and outcomes, reducing NORTH STAR’s financial and time burden and increasing disseminability.

The RCT included evaluations of both outcome and CAT process. We hypothesized that bases assigned to NORTH STAR, compared with those assigned to the control condition, would show reduced rates of suicidality, alcohol and drug problems, IPV, child abuse, and cumulative risk (from the outcome data). Furthermore, we reasoned that, because prevention teams’ action plans had to be approved by base leadership, improvements in problem prevalences would be more pronounced when installations had climates more supportive of prevention.

Method

Participants

Twenty-four of the 79 AF installations with CATs (approximately one-third of all AF installations with teams) volunteered and enrolled in the study. To recruit sites, AF prevention leadership met with behavioral health points of contact from AF Major Commands, who then approached their installation-level counterparts. Those who were interested met with installation commanders to determine if the project was of interest. We sought installations from each Major Command; within Major Commands, volunteers were accepted until the study had the required number of installations. Commanders at each base signed a Memorandum of Agreement indicating their approval for participation. Data for study outcomes (i.e., secretive problems and RPFs) at the participating bases were from the 2006 and 2008 AF CA. CAs were administered at all AF installations to representative samples of Active Duty (AD) members and to all spouses. Participation at the 24 bases was 16,020 AD members and 4833 spouses from April to June 2006 and 16,998 AD members and 3410 spouses from April to June 2008. AD response rates were excellent for long general population surveys with no payment (2006 44.7%, 2008 49.0%); spouse rates were considerably lower, perhaps in part because they were invited by mailed postcard (2006 12.3%, 2008 10.8%). Analyses of individual outcomes were restricted to the AD members (n = 33,018). Analyses of family outcomes were limited to individuals who were in a romantic relationship or had children; this sample included both AD members and spouses (n = 35,297). In cases in which multiple demographic variables indicated that both an AD member and his/her spouse participated, the spouse was selected for analysis. The individual outcome (AD-only) sample had the following characteristics: 73.8% male, M = 31.63 years of age (SD = 7.65), 67.1% married, 53.2% parents, and 21.5% officers. The family outcomes (AD member or spouse) sample had the following characteristics: 58.3% male; M = 32.61 years of age (SD = 7.65), 84.8% married, 64.5% parents, and 23.9% officers.

Members of the CATs (then known then in the AF as the “integrated delivery system”) at participating bases also participated. CATs were introduced in the late 1990s to plan and execute integrated, cross-problem efforts to address community needs. By AF regulations, each base was required to have a CAT, comprising representatives from agencies involved in health and wellness. CATs included representatives from Family Advocacy (IPV and child abuse); Alcohol and Drug Abuse Prevention and Treatment; Health and Wellness Center; Airman and Family Readiness Center; Chapel; Wellness; and often the base comptroller’s office. Although CATs did not have budgets, they worked directly with base leadership and were required to use CA data to identify needs and create a biennial community action plan (CAP). CAT members at all participating bases were invited to participate in the CAT process evaluation surveys; N = 205 participated (M per base = 8.91, SD = 5.04), ns = 205 (pre-action planning), 116 (post-action planning), and 136 (follow-up).

Procedure

Design

We conducted an RCT, with outcomes assessed using repeated, cross-sectional surveys, randomly assigning the 24 bases to NORTH STAR versus control with a 1:1 allocation ratio, assigned in a single block using Microsoft Excel’s random number generator. Dr. Heyman conducted the randomization; bases were informed of assignment by Col. Linkh. Repeated cross-sectional surveys sample each participating community at multiple time points, but sample separates individuals each time. The outcomes, secretive problems, and cumulative risk were measured in independent samples of people within each base in 2006 and 2008. As described in Atienza and King (2002) and Murray et al. (2004), the repeated cross-sectional surveys within an RCT suit the goal of community-based interventions: to change health at the community level. Also, repeated cross-sectional surveys are unaffected by attrition.

Community Assessments

AD members and spouses anonymously completed the online CA in the springs of 2006 and 2008 (see Snarr et al. 2007). The CA included all study RPFs and outcomes, as well as other constructs that are not of present focus.

Experimental Conditions

Bases were randomly assigned to NORTH STAR (the intervention condition; n = 12) or enhanced feedback (the control condition; n = 12). An activity-as-usual control was not an option given the interest in NORTH STAR and the need for randomization. Demographic variables are reported by group in Table S1 and did not differ.

NORTH STAR

NORTH STAR is a data-driven system for planning and evaluating the implementation of EBIs targeting RPFs at a community level. To accomplish this, we taught CATs to use data in an actionable way to select and implement evidence-based strategies that, over time, should improve the RPF profile of the targeted community. Sustainability was built in, in that the data were presented in an easy-to-understand feedback report that guided action planning steps. An online tool supported implementation planning, and ongoing coaching and support was provided. Although this trial consisted only of a single “round” of planning and implementation, the notion is that with successive rounds, CATs would learn the system and resources (i.e., feedback report, guidebook, and implementation and evaluation planning toolkit) and would be able, ultimately, to implement NORTH STAR without outside support. Following the 2006 CA, bases in the NORTH STAR condition received a 1.5-day on-site CAT training from the investigative team, who accompanied CAT leaders to pre- and post-training briefings with base leadership. The training reviewed the results of the base’s feedback report and assisted the CAT in developing an action plan. The feedback report provided (a) base prevalences of secretive problems and (b) their relationships with cross-cutting, malleable RPFs (e.g., depressive symptoms, parenting satisfaction). The individual, family, workplace, and community RPFs were selected from the literature and based on the AF’s Community Readiness Consultant Model (Bowen et al. 2009). This model formed the basis of the CA and was developed through an iterative process of working groups with key stakeholders (e.g., leaders of AF health and wellness entities, research partners) to select RPFs at different levels of the social ecology that were (a) consistent with existing theory and literature and (b) viewed as important and actionable by the stakeholders. NORTH STAR’s feedback report identified interrelations among secretive problems and RPFs, to identify risky RPFs that had relations to multiple problems. Once RPFs were prioritized, the CAT turned to the NORTH STAR Guidebook (Slep and Heyman 2006), comprising programs that were (a) empirically supported to improve one or more RPF, (b) implementable on a large scale, and (c) available for implementation. The Guidebook was compiled via extensive literature searching, coupled with contacting developers directly to (a) identify programs that were effective and disseminable, but were not yet in the literature and (b) understand the disseminability of programs that were in the literature but not systematically disseminated.

CAT teams considered the strength of the effectiveness evidence of candidate programs, and fit with needs and available resources to create a final plan that included two to three RPFs targeted by up to two programs each. Once final interventions were identified, CATs completed a series of implementation planning activities, identifying the number and nature of target consumers, methods of delivery, responsible parties for each task, and timelines. This implementation plan was briefed to the base leadership for their approval after the meeting. Plans also included easy-to-use systems for tracking plan execution to provide the CAT with feedback about the quality of their implementation. These were designed with each CAT and tailored to their base and implementation strategy. We built an online toolkit that provided resources for tracking reach, fidelity, and proximal outcome indicators that included strategies, methods, and measures for each selected program. After the initial visit to each NORTH STAR base, continued implementation support was provided. This included regular implementation phone calls with designated CAT members, instrumental assistance (e.g., contacting intervention developers), quarterly conference calls with bases implementing a given program, a moderated listserv (so bases could share questions and ideas), and an electronic newsletter.

Selected EBIs are reported in Table S2 of the supplement. They targeted the following RPFs: depressive symptoms, personal and family coping, intimate and parent-child relationship satisfaction, and physical activity. One CAT also chose an additional problem-focused prevention program.

CAT participants in the intervention group were asked the extent to which the activities of their action plans had been fully implemented at the follow-up assessment, with answers using a 4-point response scale. Approximately one-third of CAT participants selected 1 (not at all; 1.4%) or 2 (a little bit; 31%), with 49.3% selecting 3 (somewhat), and 18.3% selecting 4 (a lot).

Control Condition

Control bases were sent the detailed feedback reports summarizing the results of the CA that was identical to that reviewed at the NORTH STAR bases. This included much more extensive analyses of RPFs than was typically provided to bases following the CA. However, no additional training or explanation of the report occurred. Rarely, a control CAT contacted the research team with questions about the report, and these were answered.

CAT Process Assessments

Each participating CAT member completed self-report assessments. These assessments occurred on three occasions: pre-action planning (before briefings on 2006 CA results), post-action planning (after briefings on 2006 CA results and CAPs were to have been made), and follow-up (before the 2008 CA).

Measures

Secretive Problems

Each secretive problem was scored as 1 or 0 (problem present/absent) based on thresholds denoting clinical significance. Because of the dichotomous scoring, internal consistency is not reported. All outcomes were considered primary outcomes.

Hazardous Drinking

Hazardous drinking was measured with the Alcohol Use Disorders Identification Test (Allen et al. 1997). The AUDIT is a 10-item self-report measure of alcohol dependence created by the World Health Organization. It has well-established sensitivity and specificity against clinical assessments (Reinert and Allen 2002; Saunders et al. 1993). Per Rumpf et al. (2002), individuals who scored ≥ 8 were classified as above the cutoff for hazardous drinking.

Controlled Prescription Drug Misuse

Participants completed a checklist of commonly abused controlled prescription medications (e.g., amphetamines and codeine; Heyman et al. 2011). For each drug checked, the respondent was asked the frequency of use (a) when s/he did not have a prescription and (b) at a dosage higher than prescribed. Prescription drug misuse was scored as present based on any positive response.

Suicidality

Suicidality (either serious ideation or attempts) during the previous year was assessed with four items from the Youth Risk Behavior Survey that have been used in nationally representative studies (Brener et al. 2002; Witte et al. 2008). Individuals were classified with suicidal ideation if they reported that they had (a) seriously considered attempting suicide rarely, sometimes, or frequently; (b) had thoughts of ending their lives sometimes or frequently; or (c) had planned a suicide. Suicidal behavior was indicated by a non-zero response to a single item reflecting the frequency of actual suicide attempts.

Clinically Significant IPV and Child Abuse [Emotional and Physical]

The Family Maltreatment measure (Heyman et al. 2020) was used to measure maltreatment that matches Department of Defense criteria (which have been adopted by the Diagnostic and Statistical Manual, 5th Edition and the International Classification of Diseases, 11th Edition) for clinically significant (CS) IPV and child abuse—non-accidental acts that cause harm (e.g., injury, fear) or have a high potential for harm (e.g., burning, using a weapon, choking). This measure has demonstrated content, concurrent, convergent, and response process validity (Heyman et al. 2020). The Family Maltreatment measure has four modules: (1) physical IPV perpetration and victimization, (2) emotional IPV victimization, (3) physical child abuse perpetration, and (4) emotional child abuse perpetration. Each asks about (a) 12-month occurrence of acts: partner emotional aggression (9 items), physical aggression (14 items) and child emotional aggression (9 items), and physical aggression (18 items); and (b) impacts of the acts (e.g., injury, fear). To meet the threshold of CS-IPV or CS-child abuse, individuals needed to report (a) one or more acts of aggression and (b) significant harm or high potential for harm. Physical CS-IPV perpetration and victimization were combined into a variable indicating physical CS-IPV in the household; child abuse perpetration was combined across children.

Cumulative Risk

We counted additive risk across 22 RPFs, following Sameroff et al. (1993). All CA RPFs have adequate-to-strong internal consistency (see online supplement) and indications of construct validity in the current samples (see Langhinrichsen-Rohling et al. 2011; Foran et al. 2012). RPFs are grouped in four ecological levels: individual (economic stress, physical health, personal coping, spirituality/religiosity, depressive symptoms, and personal deployment preparedness), family (parent-child relationship satisfaction, intimate relationship satisfaction, family coping, career support from a significant other, and partner readiness for deployment), workplace (workgroup cohesion, workplace relationship satisfaction, and satisfaction with the AF), and community (community safety, satisfaction with community resources, community cohesion, support from neighbors, support from formal agencies, social support, community support for youth, and support from AF leadership). Each RPF was dichotomously scored; individuals who fell into the least adaptive one-fourth of each variable’s distribution (i.e., the top 25% for risk factors; the bottom 25% for protective factors) received a 1; the remaining three-fourths received a 0. The cumulative risk index was calculated by summing across these 22 dichotomous scores (range = 0–22).

CAT Process

Responses to four scales were combined to create composite measures of CAT process factors. The online supplement provides expanded descriptions and psychometrics for the scales: the (a) Prevention Programming and Implementation Questionnaire (PPIQ), (b) the Community Readiness Factors Questionnaire (CRFQ), (c) Efficacy and Outcome Expectancy Questionnaire (EOEQ), and (d) the Community Action Plan Questionnaire (CAPQ). Based on conceptual and empirical criteria (i.e., correlations ≥ .50 among the CAT process variables), we calculated four multiscale composite scores by first standardizing and then averaging constituent subscales’ scores. Orientation toward empirical prevention comprised the PPIQ use of data, criteria influence, and risk and protective factor framework use subscales’ scores. Community support comprised the CRFQ community support for prevention, base leadership support for prevention, effective base leadership, and community/CAT resistance to change (reversed) subscales’ scores. CAP development comprised the CRFQ community action goal development, action plan development, and action plan specificity subscale’ scores. Barriers to implementation constituted the CRFQ barriers to implementation and EOEQ-positive program-related expectancy subscales’ scores. The final set of eight baseline (i.e., measured at the pre-action planning assessment) CAT process variables used in our analyses included the above four composite variables and the PPIQ attitude toward community mental health data, CRFQ CAT collaboration, EOEQ present efficacy, and EOEQ program-related efficacy scores.

Change in CAT process factors was calculated for each of the eight above variables via linear slope scores across the pre-action planning, post-action planning, and follow-up assessments (coded 1, 2, and 3, respectively) at the base level for each variable. For composites, slopes were calculated for each constituent variable, then standardized and averaged. For the two variables not assessed at follow-up (both belonging to the orientation toward empirical prevention composite), slopes were equivalent to change scores.

Analytic Strategy

Hypotheses were tested with multilevel analysis with robust maximum likelihood estimation in Mplus (Muthén and Muthén 1998-2017): “type = TWOLEVEL RANDOM” in Mplus settings. Individual outcomes were analyzed in the AD sample (n = 33,018). CS-IPV was analyzed for those in the family sample with intimate partners (n = 34,314); likewise, CS-child abuse was analyzed for those in the family sample with children (n = 22,755). Multiple imputation estimated missing data, using IVEware (Raghunathan et al. 2002). For each sample, five datasets were imputed, analyzed, and results combined according to Rubin’s rules (Schafer and Graham 2002).

Data for both active duty members and spouses were weighted to their respective AF population level for rank (of the military member) and sex.

Baseline Differences Between Groups

In the AD and family datasets, each of 10 demographic variables was examined for time 1 group differences that might confound intervention effects. This was accomplished via multilevel models that regressed each of these variables on group. The CAT process variables were compared in the CAT process data set via independent samples t tests, with accompanying ds.

Main Effects of the Intervention on Secretive Problems and Cumulative Risk

The intended analytic strategy was to model base level changes in secretive problems and cumulative risk as a function of group. However, although descriptively secretive problems seemed to show varying degrees of changes (level 2), the degree of variability among the bases was not statistically significant for secretive problems or cumulative risk. Thus, we conducted multilevel analyses with intervention effects estimated at level 1 (i.e., person). Each outcome was simultaneously regressed on time (2006 and 2008 cohorts treated as independent groups, given the repeated cross-sectional assessments), group, time × group, and five control variables (CAT collaboration, community action plan development, program-related efficacy, present efficacy, and barriers to implementation), all treated as level 1 covariates. Dichotomous predictors were effects coded (+ 1 vs. − 1); continuous predictors were grand mean centered. Level 2 variation in the outcomes was also allowed, as was a threshold/intercept. The time × group term (i.e., Does change over time in the secretive problem depend on group?) reflects the main effects of intervention, adjusted for covariates. We report the ICC for within-base nesting effects for cumulative risk only, as ICCs cannot be calculated for dichotomous outcomes. Sample syntax is included in the online supplement to this article.

Moderation of Intervention Effects by CAT Process Variables

We evaluated whether intervention effects were moderated by (a) baseline levels of each of the eight CAT process variables and (b) change in each of the CAT process variables. These effects were tested via multilevel models at level 1 (level 2 variation in the outcomes was also allowed), with each outcome regressed on time (2006 and 2008 cohorts treated as independent groups), group, moderator, time × group, time × moderator, group × moderator, time × group × moderator, and five control variables. Continuous predictors were mean centered; dichotomous variables (time and group) were centered with effects coding (− 1 and 1). Significant interactions were decomposed via simple slopes plotted at ± 1 SDs on the moderator (Preacher et al. 2006). To control for type I error, Bonferroni corrections were employed with familywise alpha set to .05. The adjusted criterion p value for the eight moderation tests per outcome for CAT process variables was .006 in both the CAT process baseline and change analyses.

Results

Baseline Differences Between Groups

None of the multilevel models found significant demographic differences between groups. As shown in Supplement Table S3, the largest group difference effect sizes (d = .11 and odds ratio [OR] = .84) also indicated differences were negligible. Among CAT process variables, only program-related efficacy significantly differed between groups (t = − 2.07, p = .039). However, five of the ds > .30, and thus were selected as covariates in outcome analyses.

Main Effects of Intervention on Secretive Problems and Cumulative Risk

None of the time × group effects were significant (Tables S4–5); thus, the main effect hypotheses were not supported. Within-base nesting effects were minimal (ICC = .01) for cumulative risk, the one outcome with a computable ICC; there was little evidence to suggest similarity among individuals due to shared membership in AF installations.

Moderation of Intervention Effects

Moderation by Baseline CAT Process Variables

CAT process variables significantly moderated intervention effects for suicidality (Supplement Table S6), emotional CS-IPV (Supplement Table S7), and physical CS-child abuse (Supplement Table S8) (Fig. 1).

Fig. 1
figure 1

CONSORT diagram of the NORTH STAR RCT

Full size image

Suicidality

The time × group × barrier-to-implementation interaction was significant (Fig. 2a). When barriers to implementation were low, NORTH STAR participants exhibited a significant decrease in suicidality (simple slope (B) = − 0.40, SE = 0.07, p < .001, 95% CI − 0.55, − 0.26), whereas control participants did not exhibit reliable change (B = − 0.07, SE = 0.05, p = .126, 95% CI − 0.17, 0.02). When barriers to implementation were high, neither NORTH STAR (B = 0.01, SE = 0.06, p = .919, 95% CI − 0.12, 0.13) nor control (B = − 0.06, SE = 0.05, p = .234, 95% CI − 0.16, 0.04) participants exhibited reliable change in suicidality.

Fig. 2
figure 2

Significant 3-way interactions

Full size image

Emotional CS-IPV

The time × group × barrier-to-implementation interaction was significant for emotional CS-IPV (Fig. 2b). When barriers to implementation were low, NORTH STAR participants exhibited a significant decrease in emotional CS-IPV (B = − 0.19, SE = 0.05, p < .001, 95% CI − 0.28, − 0.09), whereas control participants did not exhibit reliable change (B = 0.03, SE = 0.04, p = .404, 95% CI − 0.04, 0.10). When barriers to implementation were high, neither NORTH STAR (B = 0.04, SE = 0.04, p = .316, 95% CI − 0.04, 0.11) nor control (B = − 0.05, SE = 0.08, p = .516, 95% CI − 0.22, 0.11) participants exhibited reliable change in emotional CS-IPV.

Physical CS-Child Abuse

The time × group × CAT collaboration interaction was significant (Fig. 2c). Decomposition of this interaction indicated that none of the constituent simple slopes were significant. At high levels of CAT collaboration, the simple slope for NORTH STAR was positive (B = 0.12, SE = 0.08, p = .137, 95% CI − 0.04, 0.28) and the simple slope for control was negative (B = − 0.06, SE = 0.04, p = .127, 95% CI − 0.14, 0.02). At low levels of CAT collaboration, the simple slope for NORTH STAR was negative (B = − 0.08, SE = 0.05, p = .090, 95% CI − 0.18, 0.01) and the simple slope for control was positive (B = 0.06, SE = 0.05, p = .265, 95% CI − 0.04, 0.15).

Moderation by CAT Process Change Variables

CAT process change across time significantly moderated intervention effects for suicidality (Supplement Table S9) and physical CS-child abuse (Supplement Table S10).

Suicidality

The time × group × community support change interaction was significant (Fig. 2d). With decreasing community support, NORTH STAR participants exhibited a significant decrease in suicidality (B = − 0.17, SE = 0.04, p < .001, 95% CI − 0.25, − 0.10), whereas control participants did not exhibit reliable change (B = − 0.01, SE = 0.04, p = .855, 95% CI − 0.09, 0.08). With increasing community support, the pattern was reversed: control participants exhibited a significant decrease in suicidality (B = − 0.15, SE = 0.04, p = .001, 95% CI − 0.23, − 0.06), whereas NORTH STAR participants did not exhibit reliable change (B = − 0.05, SE = 0.07, p = .447, 95% CI − 0.18, 0.08).

Physical CS-Child Abuse

The time × group × CAT collaboration change interaction was significant (Fig. 2e). With increasing CAT collaboration, NORTH STAR participants exhibited a significant decrease in physical CS-child abuse (B = − 0.09, SE = 0.04, p = .035, 95% CI − 0.18, − 0.01), whereas control participants did not exhibit reliable change (B = 0.02, SE = 0.02, p = .298, 95% CI − 0.02, 0.07). With decreasing CAT collaboration, neither control (B = − 0.09, SE = 0.07, p = .176, 95% CI − 0.23, 0.04) nor NORTH STAR (B = 0.10, SE = 0.06, p = .117, 95% CI − 0.03, 0.22) participants exhibited statistically significant change in physical CS-child abuse.

Discussion

The impact of NORTH STAR, a prevention planning system for reducing rates of secretive problems, was tested in the US Air Force. We hypothesized that NORTH STAR would reduce rates of hazardous drinking, suicidality, CS-IPV, and CS-child abuse. However, no significant main effects emerged. We further hypothesized that NORTH STAR would be more effective when the climate for prevention was supportive. NORTH STAR significantly reduced emotional CS-IPV, physical CS-child abuse, and suicidality when the local environment for prevention was supportive, even if the environment became less supportive over time. There were no instances where NORTH STAR had iatrogenic effects, even when interactions with CAT process variables were examined. These results suggest that NORTH STAR is a promising approach to reducing hidden behavioral health problems such as suicide and family violence.

NORTH STAR would likely produce greater improvements if more consistent implementation could be achieved. One-third of the intervention bases did not implement any prevention strategies. The intent-to-treat analyses we conducted are appropriate, but provide a conservative estimate of the effects of the intervention under ideal conditions. It could be that working with entire installations as the unit of implementation was not optimal because the base population is diverse, and stakeholders on the prevention teams often had primary allegiances to their specific agencies and supervisors. A just-completed trial of NORTH STAR implemented action plans in military workgroups, and the unit commanders oversaw these efforts. On the one hand, these commanders had no expertise in prevention planning and had other pressing duties. On the other, their motivation to support the functioning of their members was high. It could be that a narrower focus with more invested implementers will result in stronger impacts.

It could also be that with repeated cycles of assessment and implementation, implementation would have grown successively stronger as CATs gained familiarity with the system and programs. Community-based interventions seem to take several years to take root to achieve full impact (e.g., Quinby et al. 2008).

NORTH STAR is innovative in several ways. First, it offers an integrated approach to behavioral health promotion by targeting RPFs shared among many outcomes. Second, within the context of a focal community, it can be implemented with relatively low costs. Third, it is a population-level prevention approach. Thus, it is a framework that complements traditional emotional or psychoeducational prevention formats and policy-based prevention initiatives, offering stakeholders a more comprehensive prevention strategy.

NORTH STAR—because it is a framework, not a specific set of programs—is inherently flexible. As evidence accumulates and prevention programs aimed at the included RPFs evolve, and innovations are made, the menu of prevention choices can be modified. In addition, if an RPF not targeted accumulates evidence that it is more powerful than one originally targeted, the framework can incorporate it. In this way, NORTH STAR is more sustainable than many fixed programs because it is flexible and adaptable to improvements in both the assessment and intervention components. Implementation science has not yet addressed how to optimally balance flexibility and fidelity in prevention systems to optimize sustained impact. This will be a critical area of inquiry as community-based prevention efforts mature.

This RCT had numerous limitations. The study was in the field a decade ago. Although we think it is unlikely that time, or the evolution of AF activities in the intervening years, affected the psychometrics of the measures or the results of the study, it is impossible to know that with certainty. Nevertheless, this is the only study of its kind, offering valuable insights into the potential of community-based, multipronged prevention systems. Additionally, the study did not have enough base-level variability in secretive problems to model treatment effects at the level of randomization. Measures were limited to self-report, and all the biases inherent in that. Additionally, because the data were repeated cross-sectional, rather than longitudinal within individual, we were unable to model change within person. The implementation challenges within the NORTH STAR condition suggest that despite the emphasis placed on making the system easy to use and selecting easy-to-implement activities, taking population-level action is inherently challenging and requires significant support to take hold. This experience was likely exacerbated by the relatively brief 2-year study period, as similar prevention systems take a minimum of 2 years for implementation to begin to affect outcomes (e.g., Quinby et al. 2008). We expect that if installation CATs completed successive cycles of data collection, planning, and implementation, they would build their skills and infrastructure and would gradually need less support to implement their programming choices effectively. Testing this hypothesis would require a longer research period than was feasible in this study. Also, the impact of a framework is inherently dependent on the effectiveness of the empirically supported interventions that are selected and implemented within it. When the effectiveness of available interventions is limited, it necessarily impacts the potential effectiveness of NORTH STAR. Finally, although one-third of all the AF installations worldwide participated in the trial, it is likely they are not fully representative of all installations. Participating installations were able to organize the actions necessary to volunteer for the study, for example. In addition, we did not have the resources necessary to execute the study at more than one-third of the bases in the AF simultaneously, and the intervention length was fixed by funding constraints and the timing of the CAs. Although additional approaches to gaining power were considered (e.g., randomizing timing of interventions within installation), this was not practical and cut against the design of NORTH STAR, which was to work within existing structures and real-world systems to increase ease of use and sustainability. The study has limited power at the level of the installation.

This study suggests several avenues for future research. First, this study highlights the importance of understanding the mechanisms driving uptake and implementation. Despite decisions to implement specific EBIs, and the availability of technical support, many bases did not implement their action plans. As implementation science grows, it will be important to understand mechanisms driving complete and efficient implementation in real-world contexts. Second, little is known about the mechanisms of change in population-based prevention studies such as this one. It is clear that there can be dynamics at different levels of the social ecology that can spread the reach of an individual receiving an intervention (or spread resistance to a particular program). Identifying mechanisms of “contagion” throughout the social milieu will help improve the impact of interventions seeking to achieve population-level change.

In summary, NORTH STAR has promise to complement existing prevention efforts that tend to be problem-specific (e.g., reducing hazardous drinking in junior enlisted AF members or preventing child abuse in at-risk parents). We did not observe significant main effects. Yet, in supportive AF environments, NORTH STAR appears to reduce clinically significant problems without targeting them directly. To reduce secretive problems, an approach that targets shared RPFs instead of the outcomes themselves might help improve health in ways that problem-specific programs cannot. Finally, NORTH STAR has the flexibility to incorporate advances and target emerging needs, boosting sustainability. Taken together, there is potential for NORTH STAR, and frameworks like it, to help promote empirically supported interventions in large systems to decrease problems and improve health.