The Efficacy of Resiliency Training Programs: A Systematic Review and Meta-Analysis of Randomized Trials

Importance Poor mental health places a burden on individuals and populations. Resilient persons are able to adapt to life’s challenges and maintain high quality of life and function. Finding effective strategies to bolster resilience in individuals and populations is of interest to many stakeholders. Objectives To synthesize the evidence for resiliency training programs in improving mental health and capacity in 1) diverse adult populations and 2) persons with chronic diseases. Data Sources Electronic databases, clinical trial registries, and bibliographies. We also contacted study authors and field experts. Study Selection Randomized trials assessing the efficacy of any program intended to enhance resilience in adults and published after 1990. No restrictions were made based on outcome measured or comparator used. Data Extraction and Synthesis Reviewers worked independently and in duplicate to extract study characteristics and data. These were confirmed with authors. We conducted a random effects meta-analysis on available data and tested for interaction in planned subgroups. Main Outcomes The standardized mean difference (SMD) effect of resiliency training programs on 1) resilience/hardiness, 2) quality of life/well-being, 3) self-efficacy/activation, 4) depression, 5) stress, and 6) anxiety. Results We found 25 small trials at moderate to high risk of bias. Interventions varied in format and theoretical approach. Random effects meta-analysis showed a moderate effect of generalized stress-directed programs on enhancing resilience [pooled SMD 0.37 (95% CI 0.18, 0.57) p = .0002; I2 = 41%] within 3 months of follow up. Improvement in other outcomes was favorable to the interventions and reached statistical significance after removing two studies at high risk of bias. Trauma-induced stress-directed programs significantly improved stress [−0.53 (−1.04, −0.03) p = .03; I2 = 73%] and depression [−0.51 (−0.92, −0.10) p = .04; I2 = 61%]. Conclusions We found evidence warranting low confidence that resiliency training programs have a small to moderate effect at improving resilience and other mental health outcomes. Further study is needed to better define the resilience construct and to design interventions specific to it. Registration Number PROSPERO #CRD42014007185


Rationale
Resilience has been defined as the ability of individuals to absorb life's challenges and to carry on and persevere in the face of adversity. [1] Overlapping extensively with the concept of hardiness, psychological resilience personifies and reflects characteristics of toughness, elasticity, and the ability to recover. Although the term has been used in many disciplines and applied to many contexts, a recent concept analysis defined resilience as the ''process of effectively negotiating, adapting to, or managing significant sources of stress or trauma.'' [2].
When conceptualized in this way (i.e. as a response to stress or trauma), it is practically helpful to briefly consider the position resilience holds within a relevant stress model, such as Lazarus' Transactional Model of Stress and Coping. According to this model, [3] many of the events that comprise the experience of life (i.e. illness, loss, trauma, new jobs or demands) can be considered ''stressors.'' In the absence of the resources needed to cope with and manage these stressors, people experience their effects in the form of reduced mental-and to a lesser extent physical-health. According to Lazarus' model, then, the value of personal resilience lies in its potential as an internal resource for mitigating the negative effects of stress and for maintaining mental health through adversity [4].
Indeed, poor mental health places major constraints on the wellbeing, productivity, and prosperity of individuals, communities, and nations. [5] As such, there is widespread interest in better understanding and applying the mechanism by which resilience is able to avoid these constraints and promote health. [6][7][8][9] The predictors and effects of resilience have been examined among those living with chronic illness, overcoming traumatic experiences, and prospering in stressful work environments. Overall, research suggests that resilience is a modifiable construct and not an inherent, immovable trait of individuals. To the extent this is true, the potential public health impact of identifying and translating a reliable and efficacious method of achieving resilience in people is great.
Resiliency can be thought of as the process of achieving resilience. Clinicians, researchers, patients, public health agencies, governments, and others are investing heavily in mechanisms aimed at facilitating resiliency. Key among these, ''resiliency training programs'' are a loosely defined group of interventions that systematically seek to enhance resilience in individuals or groups. To our knowledge, no single accepted theoretical framework or consensus statement exists to guide the development or application of these programs. Furthermore, despite international use and testing, there remains little clarity related to what is fundamentally required for a program to be considered resiliency training, let alone for it to be considered effective. Indeed, one could argue that, without more guidance and understanding, the field runs the risk of overtranslating and/or diffusing its efforts.
To better understand the efficacy of resiliency training programs and to provide information that can benefit decision makers in directing future study, we sought to conduct a systematic review and meta-analysis. Clinically, we were particularly interested in the role resiliency training might play in improving the lives and health of patients with chronic conditions.

Objectives
Our primary objective was to synthesize the evidence of resiliency training programs in improving resilience, quality of life, and self-efficacy and in reducing depression, stress, and anxiety in adults. A secondary aim was to determine the efficacy of these programs in patients with chronic conditions.

Methods
A published protocol [10] (PROSPERO registration number CRD42014007185) guided the conduct of this review, which we report in adherence to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) Statement [11].

Eligibility Criteria
Eligible studies were randomized controlled trials published in any language assessing the efficacy of any program designed to develop or enhance resilience (or a related construct, ''hardiness'') in adults. Eligible studies had to describe an intention to impact resilience or hardiness in their rationale or design. No eligibility restrictions were made based on the type of comparator used, the length of follow-up, or the outcomes measured. Studies that only evaluated dissemination and/or implementation of resiliency training programs were ineligible.

Information Sources
In conjunction with an experienced research librarian (PJE), we searched the following electronic databases from 1990 to January 14, 2014: PubMed, Scopus, EBSCO CINAHL, Ovid MEDLINE, Ovid EMBASE, Ovid Cochrane Library, Web of Science, and Ovid PsycINFO. The complete electronic search strategy is available in Supplement S1. We also searched clinical trial registries, contacted experts and study authors, and hand searched bibliographies.

Study Selection
After receiving formal instruction and piloting a small sample, a team of 7 reviewers (ALL, PRB, MRG, KRB, MMD, JBP, CZP) worked in duplicate and independently to screen out clearly ineligible papers by reading titles and abstracts and using a webbased software (Distiller SR). To aid in the identification of ongoing studies, reviewers were instructed to include study protocols of potentially eligible trials during this phase. Any conflicts warranted retrieval of a full text copy of the article and inclusion into the second phase of screening. During this phase, two reviewers (ALL, PRB) independently examined full text versions of candidate papers to determine final eligibility (kappa = 0.78). Study protocols were excluded at this stage after extraction of relevant author contact information; all conflicts were resolved by consensus.

Data Collection
After piloting a standardized data extraction form, two reviewers (ALL, PRB) worked independently and in duplicate to extract details about the included trials' participants, interventions, controls, outcomes, and risks of bias. Specific data extracted included the trial author, year of publication, setting, study objective, and type (patients, students, workforce, other) and demographics (age, gender, race) of participants. We extracted descriptions of the format and theoretical basis of the intervention and comparator, particularly noting whether the comparator was a well-matched attention control vs. not. We extracted information on the number of participants approached, enrolled, randomized, and analyzed when this was available. We extracted postintervention means and standard deviations for six, a priori determined patient-reported outcome domains at both short (longest follow up#3 months) and long (longest follow up $6 months) durations of follow-up.
The outcomes collected were patient-reported measures within the domains of 1) resilience, hardiness, or ability to cope; 2) quality of life or well-being; 3) patient activation, self-efficacy, or confidence for disease management; 4) depression; 5) stress; and 6) anxiety. A consensus of the authors was used to determine whether outcomes measured were appropriate for inclusion within a given domain. Each outcome was assigned a rating of ''appropriate,'' ''inappropriate,'' or ''questionable'' (see Appendix D). Only a single outcome was accepted within each domain for a given trial; when multiple outcomes existed within a single domain, a hierarchy was used that prioritized validated and frequently reported measures. When not reported, we calculated standard deviations from confidence intervals and standard errors and, when necessary, we estimated sample sizes from reported degrees of freedom. We imputed standard deviations in three cases [12][13][14] by using reported standard deviations from other trials using the same measure. To remain conservative, we used the largest standard deviation for each measure that we could find, prioritizing studies in comparable populations [15][16][17].
After extracting data, we emailed a standardized, pre-populated spreadsheet to all study authors to 1) confirm the accuracy of our extraction, 2) ascertain any missing information and, 3) inquire about other potentially eligible trials. Authors were given 10 days to respond before a second email was sent. If no response was received after the second email, we conducted an internet search to identify an alternative email or method of contact; if fruitful, a final contact attempt was made before declaring the author unreachable.

Intervention Categorization
Early in the review process, it became clear to us that study authors used diverse conceptual approaches when applying their training programs. For example, we found a particular dichotomizing distinction between programs based on the type of stress they sought to mitigate. Specifically, programs intending to impact trauma-induced stress (i.e. as might occur in individuals with posttraumatic stress disorder after a major catastrophe or tragic event) were very different in terms of approach used and outcomes evaluated from those intending to impact more generalized, everyday stresses. To aid in the organization, conceptualization, and analysis of the programs, we developed an ad hoc classification framework ( Figure 1). This framework broadly classified training programs based on 1) whether they sought to mitigate generalized or trauma-induced stress, 2) whether they focused on developing resilience as an end goal or as a mediating variable, 3) whether they were designed to be used in single/specific or multiple/ general populations, and 4) whether they were intended to be administered universally or in a targeted fashion (i.e. only ''as needed'').

Risk of Bias Within Studies
Risk of bias was assessed for each trial independently by two team members (AL, PB) using the Cochrane Collaboration's Tool. [18] Specifically, we considered the quality of the randomization sequence generation; whether treatment arm allocation was concealed; the type and quality of blinding of participants, personnel, and outcome assessors; the degree and potential impact of missing data; the likelihood of incomplete reporting; and the potential role of conflicting interests. In cases where the intervention was explicitly intended to impact resilience and no measure of resilience was reported, we considered the study to be at high risk of selective reporting. We judged the potential impact of all biases on a given study's reported outcomes and identified those studies at highest risk of bias. Particular weight was given to the impact of missing data, which was a well-distributed variable across studies. Conflicts in judgment were resolved through discussion and consensus.

Data Synthesis
To permit pooling of effects across different measures of similar constructs, we converted the differences in post-intervention means to standardized mean differences (SMDs). Because of differences in the conceptual approaches of resiliency training programs designed to mitigate generalized stress compared to those specifically designed to impact post-traumatic stress-and in differences in the underlying psychobiology of these states-we elected, before looking at the data, to analyze these categories of programs separately. For both types of programs, when possible, we conducted a random effects meta-analysis of the SMDs within each of the six outcome domains collected. We assessed for between trial heterogeneity in excess of chance by calculating the I 2 statistic. [19] We used RevMan Version 5.2 statistical software [20] for all analyses. Studies not reporting outcomes within the a priori domains or not reporting them at the level of the randomized participants (e.g. reporting changes in team or group culture as measured in different post-intervention samples) were not included in the meta-analyses.

Risk of Bias Across Studies
Because included trials were small in size and few in number, it was inappropriate to assess for publication bias through planned funnel plot analyses. [21] Rather, we used global assessments of the body of evidence to postulate on its impact.

Additional Analyses
We conducted planned subgroup analyses based on whether 1) the study participants had a chronic disease and 2) whether the trial had an attention control comparator. Because of heterogeneity in the format, structure, and theoretical approaches of programs, and the small number of trials for a given outcome, we were unable to formally assess the effects of intervention characteristics on outcomes.
We conducted sensitivity analyses based on the appropriateness of the included outcome (i.e. whether the outcome was rated as ''questionable'' for inclusion within a given domain), whether the study was judged at high risk of bias, and whether any required data was imputed.

Study Selection
The study flow diagram is presented in Figure 2. The electronic database search generated 516 candidate citations. Through title and abstract screening, we identified 68 potentially eligible trial reports or protocols. For these, we retrieved and reviewed full text versions, resulting in the inclusion of 22 trials. A complete list of full text papers reviewed and rationale for exclusion is provided in Supplement S1. Two additional trials were obtained through protocol author contact and one ongoing, eligible trial was identified through expert contact. Thus, the final sample consisted of 25 randomized trials ( [13,14,; Sharma, unpublished data; and Burton, unpublished data). Authors responded to contact for 17 of the included studies but were often unable to provide additional data or information. A method of contact could not be identified for one study author [13].

Study Characteristics
A summary of the included trials' characteristics, including the theoretical basis and operational format of all interventions is presented in Table 1. In general, studies were small and conducted at single centers in diverse populations. Interventions varied widely in format, duration, and theoretical basis. Selfdirected, electronic interventions; individual coaching or training sessions; and group courses and sessions were all tried with some efficacy across varying outcomes. Five studies evaluated programs designed to mitigate trauma-induced stress, while the remainder sought to impact stress more generally. Most trials were explicit in describing their intention to impact resilience, while three were less direct in describing this desire. [22,33,36] Two studies sought to impact resilience only as a mediator of a broader psychological construct. [30,39] The theoretical bases of the tested interventions ranged from the use and application of well-established and/or resilience specific models and frameworks (i.e. The 5 C's of Resilience, The Resilience Model, Lazarus' Stress Model, etc.) to less clear and/or combined theoretical approaches drawing on broadly applicable strategies of stress management, attention interpretation, coping, and/or cognitive behavioral therapy. Most studies were of a wait-list control design, although 10 used an attention control.  showed similar results. CHD risk indicators and physical activity also measured but not sought for this review.

Risk of Bias Within Studies
A summary of the risk of bias within each study is presented in Supplement S1. The risk of bias was judged to be moderate to high (agreement = 81%) for most studies. Unclear or incomplete reporting of methods and/or a high risk of missing data was frequently seen. In some cases, total numbers of subjects randomized and losses to follow-up were not reported and almost all studies conducted per protocol analyses. Seven studies were judged to have a particularly high risk of bias.( [13,26,[30][31][32]34] and Burton, unpublished) We could not rule out a potential conflict of interest in six studies [26,30,33,34,36,39].

Results of Individual Studies
In general, resiliency training showed benefit in a number of mental health domains across diverse populations at #3 months of follow-up. In a number of cases, key variables needed for metaanalysis were not reported and could not be reliably imputed or obtained through author contact. To ensure the comprehensiveness of this review, we have summarized the results of all included studies in Table 1. For any given outcome, there was never more than one study reporting at a follow-up time $6 months. This precluded planned meta-analyses of the long-term effectiveness of resiliency training programs.  Figure 3. Forest plots for all other analyses can be found in Supplement S1. The complete results of the a priori meta-analyses, summarized by effect size, are presented in Table 2.

Risk of Bias Across Studies
The potential for publication and reporting bias was judged to be high. Of the 22 studies explicitly describing a desire to impact personal resilience, 10 failed to report an outcome measuring this construct. This was characteristic of trauma-directed [24,29,31,35] and resilience-mediated [30,39] training programs, which may have been less focused on resilience as a primary outcome. One study explicitly described a resilience-directed intervention and reported a resilience outcome in one paper, [25] but described the intervention's purpose differently and reported different outcomes in other papers that were not captured by our initial database search. [12,43] Of the 6 studies judged to have a potential conflict of interest, 4 failed to report a resilience outcome. Although the overall risk of bias for included studies was judged to be high, it was somewhat lower among the 18 studies contributing to the meta-analyses.

Subgroup analyses
Among generalized stress-directed resiliency training programs, planned subgroup analyses based on whether an attention control was used or whether participants had a chronic disease failed to show a significant difference in intervention effect. Among studies evaluating trauma-directed resiliency training programs, both the non-attention-controlled and chronic disease subgroups comprised a single study conducted in patients with post-traumatic stress disorder (PTSD). [29] This study was significantly more effective at reducing depression (interaction p = .03), stress (interaction p, .01) and anxiety (interaction p = .02) than the other trauma-

Sensitivity Analyses
Sensitivity analyses based on whether an included outcome was rated as ''questionable'' for pooling appropriateness did not change interpretations. Of the seven studies judged to be at the highest risk of bias, three ( [13,34] and Burton, unpublished) contributed at least one outcome to the meta-analyses. Removal of the study by Sadow [13] did not change interpretation of the selfefficacy outcome. Removal of the studies by Abbott [34] and Burton (unpublished) however, independently resulted in increased estimates of the effect of resiliency training and reductions in heterogeneity across all included outcomes [resilience (Burton only), quality of life (Abbott only), and depression, stress, and anxiety (both Burton and Abbott)]. The study by Abbott lost about half of its sample to follow up and conducted an intention to treat (ITT) analysis; this likely underestimates the effectiveness of the intervention. The study by Burton used a cluster-randomized design that allocated participants by clusters according to type of employment and geographic location. The distribution of clusters was markedly unbalanced at baseline, however, and the treatment arms experienced different stressors at key points of data collection. Removing both of these studies from the analyses caused the estimated benefits in quality of life, depression, and stress to achieve statistical significance. The effects of their exclusion are summarized in Table 3.

Summary of Findings
In general, the body of randomized trial evidence supports a modest but consistent benefit of resiliency training programs in improving a number of mental health outcomes within three months of follow-up. When excluding studies rated at high risk of bias, the estimated benefits are larger, more consistent, and more significant. Still, the overall methodological quality of included trials was low and several were poorly reported. We found no interaction with effect based on whether participants had chronic medical conditions. Although not statistically significant, we did identify a reduction in measured benefit in attention-controlled trials. Included studies were also small in number and size, which limits our ability to draw conclusions in high confidence.
There remains a lack of clarity related to what critically defines a resiliency training program. Programs are operationalized in diverse ways and lack a common theoretical or scientific specificity. The field also lacks a consistent approach to measurement [44] and it is often unclear whether outcomes chosen are sufficiently specific to the intervention. We developed a training program framework that helps to organize the operational approaches that have been taken in intervention design.

Comparison With Prior Research
To our knowledge, this is the first systematic review and metaanalysis of resiliency training programs in adults, although a prior meta-analysis of a particular resiliency training program for children showed a similar effect in improving depression. [45] Our findings are also consistent with recent meta-analyses of meditation and mindfulness-based programs that showed efficacy in improving stress, depression, and well-being outcomes in clinical populations. [46][47][48] The effect sizes in these studies were comparable to those seen in our review, and may suggest similar value for resiliency training in patients with chronic conditions. Our subgroup analyses support this conclusion.

Strengths and Limitations
We conducted this study according to a pre-defined and published protocol. To accumulate a high quality body of evidence, we restricted our inclusion to randomized trials and we searched databases and registries and contacted authors and experts to identify unpublished work. Still, this study has a number of limitations. First, our criteria for determining whether an intervention was a resiliency training program relied on our interpretations of the authors' descriptions. We also combined a number of measures within construct domains. Despite efforts to account for the appropriateness of this approach, some uncertainty is inherent. The populations studied were heterogeneous and a normal distribution of outcomes was assumed in most cases; if this assumption were shown to be incorrect it would limit the validity of the pooled SMD estimates. Finally, we combined all outcomes reported within 3 months of follow-up. This approach gives a general impression of short-term program effectiveness but may overestimate the effect seen by excluding studies reporting outcomes immediately post-intervention.

Implications
Clinicians, researchers, health policymakers, and governments are intrigued by the concept of resilience and the role it may play in promoting health and well-being. Finding reliable and effective ways to bolster resilience in individuals and populations is thus a key area of investigation. We have summarized the randomized trial evidence of programs designed to impact personal resilience.

Future Study
To date, most studies related to resilience have been observational in nature. This may be an appropriate approach to further define the resilience construct and purposefully and scientifically design interventions to impact it. Research should focus on identifying a consistent and specific strategy for targeting resilience and a corresponding approach to measurement. When programs have clear scientific and theoretical rationale for effectiveness, they should be evaluated in larger, randomized controlled trials. In the future, comparative effectiveness studies will be needed to assess the specific and incremental value of resiliency training as compared to alternative programs (e.g. traditional cognitive behavioral therapy, mindfulness-based interventions, etc.). These trials should also have longer durations of follow-up to fully evaluate their effectiveness.

Conclusions
Resiliency training programs seem to have benefit in improving mental health and well-being in diverse adult populations, although the quality of the randomized trial evidence precludes conclusions based in high confidence. There is no specific format, structure, or theoretical basis that defines a resiliency training program. In addition, no gold standard method of evaluation or measurement exists. Significant stakeholder interest in the potential of resiliency training programs warrants further study in this area. Such study should be rationally and scientifically organized, however, to achieve maximal value and fill key gaps in knowledge.

Supporting Information
Checklist S1 PRISMA checklist for this review. (DOC) Data S1 Supplementary spreadsheet of all raw data used in analyses.

(XLSX)
Supplement S1 Supplementary file that includes the complete search strategy, a summary of excluded studies, the risk of bias assessments, a summary of pooled measures, and forest plots for all analyses. (DOCX)