Systematic review of interventions for reducing occupational stress in health care workers.

OBJECTIVES
This study evaluated the effectiveness of interventions in reducing stress at work among health care workers.


METHODS
A systematic search was conducted of the literature on reducing stress or burnout in health care workers. The quality of the studies found was then appraised and the results combined. A meta-analysis was performed when appropriate.


RESULTS
Altogether 14 randomized controlled trials, three cluster-randomized trials, and two crossover trials, comprising 2812 participants, were included. Only two trials were of high quality. The following comparisons were possible: person-directed interventions versus no intervention, person-work interface interventions versus no intervention, and organizational interventions versus no intervention. Person-directed interventions can reduce stress [standardized mean difference (SMD) -0.85, 95% confidence interval (95% CI) -1.21 - -0.49] and burnout, measured as emotional exhaustion [weighted mean difference (WMD) -5.82, 95% CI -11.02 - -0.63) and lack of personal accomplishment (WMD -3.61; 95% CI -4.65 - -2.58). They also reduce anxiety, measured as state anxiety (WMD -9.42, 95% CI -16.92 - -1.93) and trait anxiety (WMD -6.91, 95% CI -12.80 - -1.01). Person-work interface interventions can reduce burnout, measured as depersonalization [mean difference (MD) -1.14, 95% CI -2.18 - -0.10]. Organizational interventions can also reduce stress symptoms (MD -0.34; 95% CI -0.62 - -0.06) and general symptoms (MD -2.90, 95% CI -5.16 - -0.64). No harmful effects were reported.


CONCLUSIONS
Limited evidence is available for a small, but probably relevant reduction in stress levels from person-directed, person-work interface, and organizational interventions among health care workers. This finding should lead to a more-active stress management policy in health care institutions. Before large-scale implementation can be advised, larger and better quality trials are needed.

Changes in global economic realities are progressively transforming the very nature of work from physical tasks to more mental and emotional endeavors. The prevention of high stress levels in the work environment is thereby imperative in efforts to improve the quality of worklife, even in the face of increased job insecurity. Stress can be defined as a subjective psychophysiological state characterized by a combination of high arousal with displeasure. Although there is some controversy about the exact mechanism of work-related or occupational stress, the most extensively used models, demand-control (1) and effort-reward (2), explain it as an imbalance between particular factors. The resultant effects of stress on individual persons are mediated by personal factors like age, experience, health, coping skills, and the like (3)(4)(5).
Health care workers are no exception when it comes to suffering from work-related stress. Under some conditions, their work-related stress can lead to anxiety and depression, burnout, or psychosomatic diseases and a resultant deterioration in quality of life and service provision (6)(7)(8)(9). Burnout has been defined as "a persistent, negative, work-related state of mind in 'normal' individuals that is primarily characterized by exhaustion, which is accompanied by distress, a sense of reduced effectiveness, decreased motivation, and the development of dysfunctional attitudes and behavior at work. This psychological condition develops gradually but may remain unnoticed for a long time for the individual involved. It results from a misfit between intentions and reality at the job. Often burnout is self-perpetuating because of inadequate coping strategies associated with the syndrome [p 388]" (4). Burnout is considered in this review as a form of psychological stress and not as a clinical diagnosis (10). The economic impact of such conditions is high, as can be inferred from data on absenteeism and turnover (11,12).
There are many stress factors in the workplace of health care workers that have been shown to increase the risk of distress and burnout, for example, an increasing administrative workload, contact with suffering and dying patients, verbal and physical abuse by patients, bullying by colleagues, the need to hide negative emotional responses, risk of litigation, role conflicts between professions, and organizational changes (10,(13)(14)(15)). In addition, many studies have shown that levels of dissatisfaction, distress, and burnout at work are high among health care workers (12,(16)(17)(18)(19)(20) and may even be higher than among workers in other occupations (7). There are numerous obstacles to conducting effective stress-related interventions at workplaces; therefore, it is all the more important to be systematic when determining what really works and what does not (21). Because health care workers form a relatively homogeneous and specific population, stress management interventions can be tailored to their specific needs, and consequently the results of a review concerning this occupational group may have a higher generalizability than synthesizing studies of various occupational groups. Therefore, we thought there was a need for a systematic review especially targeting health care workers.
Several reviews have been published on the effectiveness of interventions in preventing or treating stress (22)(23)(24)(25). However, there is only one that has focused specifically on health care workers, and it did not reach clear conclusions about the evidence (26). The aim of this review was to ascertain the effectiveness of interventions in reducing stress in health care workers.

Material and methods
We performed a systematic literature search up to May 2005 to locate studies in electronic databases, including MEDLINE, PsychINFO, the Cochrane Depression, Anxiety and Neurosis Group specialized registry, and the Cochrane Occupational Health Field database. References from articles and reviews were also reviewed, and all issues of Work & Stress between January 1987 and May 2005 were hand-searched (27). [See the appendix for the MEDLINE search strategy that we employed. ] We included studies with interventions that were directed at workers who had not actively sought help for stress, burnout, depression, or anxiety disorder and in which interventions were compared with nonintervention controls or with alternative interventions. As outcomes, we considered all validated self-report measures of stress or burnout and all measures of the detrimental effects of stress or burnout.
Two reviewers independently checked each identified trial, determined inclusion, and graded the methodological quality with a previously validated checklist (28). Disagreements were resolved by consensus. We needed an instrument that could also assess the quality of the nonrandomized studies of organizational interventions that we wanted to include. The checklist's scales of internal validity were used for rating study quality, scores higher than 75% of the maximum of the two scales combined indicating high internal validity.
If interventions, participants, and outcomes were comparable, we pooled the results of individual studies. If sufficient numerical data were available, we performed a meta-analysis of outcomes by combining trials using the Mantel-Haenszel method (RevMan 4.2.8, The Nordic Cochrane Centre, Copenhagen, Denmark, 2003). Statistical heterogeneity was evaluated and I 2 >50% was considered significant. Outcomes were summarized as standardized (SMD) or weighted (WMD) mean differences. A weighted mean difference can be calculated if all trials have measured the outcome on the same scale (eg, burnout with the Maslach Burnout Inventory). If the scales used differ but measure the same thing (eg, stress), then a standardized mean difference (MD) can be calculated. Individual trials affect both summary measures in proportion to their sample sizes so that the weight of a study is equal to the inverse of its variance. When there was significant heterogeneity, we applied a random-effects model; otherwise we used a fixed-effects model. For the remaining studies that did not report sufficient numerical data for pooling, we performed a qualitative synthesis (29). See table 1 for the system used to grade the evidence.
If a study used a cross-over design (two groups take turns being the intervention and the control group), we used the results from just after the first implementation of the intervention. One study (30) compared more than one active intervention; therefore, we entered the intervention that we judged to be the most intense into the meta-analysis to avoid having to input the same study more than once into the meta-analysis and thus Ruotsalainen et al compound its effect on the summary score. Using a fairly large assumed intracluster correlation of 0.10, we calculated the design effect for cluster-randomized studies that had not considered it. Where necessary, missing statistics data were sought from authors. Since all of the studies used questionnaires as outcome measures, we were unable to conduct an intention-to-treat analysis. We considered the influence of publication bias, but we did not use funnel plots due to the small number of trials in each comparison group. Figure 1 shows the details of the exclusion and inclusion of studies. Table 2 shows the characteristics of the 19 included studies, of which 14 were randomized controlled trials, three were cluster-randomized trials, and two were cross-over studies. All three of the clusterrandomized trials (31-33) had a unit of analysis error; in other words, they analyzed the results at the individual level without taking the effect of the cluster design into account. The reasons for excluding nine studies from the meta-analyses were unique interventions like recreational music-making (34), therapeutic massage (35), support and advice from a psychologist (31,36), and "primary nursing" (37), or insufficient reporting of outcome data (33,(38)(39)(40)(41).

Study characteristics
There was only one study (42) that reported using an acceptable method of randomization and concealing group allocation until it was completed and irrevocable. There was mention in only one study (43) of blinding those rating the outcome measurements. In all but one of the included studies (35), the blinding of the participants with respect to the intervention would have been impossible due to the nature and aims of the intervention being self-evident. The loss of participants was low throughout, and only two studies (40, 44) suffered a loss exceeding 20% of the initial sample. Two reviewers (JR and JV) independently rated all of the included studies on a quality checklist (28) (table 3). The Cohen's kappa of agreement was 0.62 for the internal validity scales. There were only two studies (33,43) that were rated as being high in quality. Both of them were studies on organizational interventions.
Altogether 11 of the 19 included studies (30,35,37,38,(42)(43)(44)(45)(46)(47)(48) had interventions that were specifically directed towards nurses. In seven studies (32-34, 36, 39, 49, 50), interventions were directed at all of the staff of participating health care facilities, and, in one study, the intervention was directed towards respiratory therapists (40). We categorized interventions as (i) person-directed if they were aimed at changing personal characteristics without explicit reference to functioning at work, (ii) person-work interface intervention if they were aimed at improving the fit between the person and the organization (eg, role conflict-ambiguity, relationships, employee involvement in decision making), and (iii) organizational if they targeted organizational or social environments (eg, organizational restructuring, training, and job redesign) that may produce stress (22, 51). A total of 13 studies used person-directed interventions, including cognitive-behavioral training (45,47,50), relaxation training (30,44), music-making (34), therapeutic massage (35), and multicomponent intervention (31,36,38,40,42,49). One study employed a person-work interface intervention consisting of mobilizing support from colleagues and learning participatory problem solving and decision-making skills (39). In the remaining five studies, the intervention [consisting of psychological training programs to improve attitudes, communication skills, and occupational stress (43,46) or changes in work organization, knowledge, skills training, and support and advice from supervisors (32,33,37)] was directed towards improving the employees' functioning in worktasks.  (55). The summary effect of two subscales of the MBI favored the intervention, emotional exhaustion (WMD -5.82, 95% CI -11.02 --0.63) and lack of personal accomplishment (WMD -4.89, 95% CI -8.71 --1.07). There was considerable heterogeneity between the three studies in the meta-analysis ( figure 3). The results of one study (50) showed that, when compared with 6 weeks of cognitive-behavioral training, having refresher sessions at 5, 11, and 17 months led to significantly lower emotional exhaustion (MD -6.00, 95% CI -8.16 --3.84) and lack of personal accomplishment (MD -5.82, 95% CI -7.89 --3.75) after 2 years.    Organizational interventions. According to one study (37), support and advice given by nurse managers or quality-care coordinators reduced symptoms on one of the subscales of the MBI: depersonalization (MD -1.14, 95% CI -2.18 --0.10) when compared with no intervention.

General symptoms
Person-directed interventions. According to two studies (30,42), person-directed interventions did not reduce general symptoms significantly more than no intervention (WMD -11.87, 95% CI -27.24 -3.49) when measured with the General Health Questionnaire (57). The results of one study (30) showed that cognitive relaxation training maintained the reduction in scores of the General Health Questionnaire for at least 1 month, whereas stretch-release relaxation training did not (MD -7.10, 95% CI -10.58 --3.62).
Organizational interventions. One study (32) showed that a combination of training knowledge and skills and individual program-planning decreased general symptoms (MD -2.90, 95% CI -5.16 --0.64) when compared with no intervention and when measured with the General Health Questionnaire (57).

Levels of evidence
The qualitative analyses agreed with the results of the meta-analyses. [See table 4 for the results of our grading     (28) were considered high in quality.

Discussion
From the meta-analyses and qualitative analyses, we can conclude that there is limited evidence that persondirected interventions among health care workers effectively reduce the levels of burnout, anxiety, and stress and that organizational interventions reduce the levels of stress, burnout, and general symptoms.
We could not subdivide the person-directed interventions further because almost all of the studies used several different components. Most of the studies claimed that it is possible to change the participants' cognitions about stressful elements at work. We refrained from further sensitivity analyses based on differences (i) in quality because most of the studies scored about mid-range on the quality scale and (ii) in the content of the interventions because most of the interventions were complex. It is interesting to note that the studies involving organizational interventions scored higher on the quality checklist even though it is more difficult to carry out interventions targeted at worktasks (21). This finding is surprising because the implementation of organizational interventions is dependent on more stakeholders than the patient alone and these types of interventions are therefore more difficult to organize and keep under control. The authors of the included organizational intervention studies not only succeeded in randomizing workplaces, they did it well. One would have not only expected these studies to have weaker study designs as is usually the case (21), but also to display the largest effects since  (52)].

Limitations of the review
In the meta-analyses of person-directed interventions there was considerable heterogeneity, two studies (30,50) showing less decrease in burnout and anxiety immediately after the intervention. However, in the studies in which follow-up lasted at least a month, these differences disappeared. We could not find a good explanation for the heterogeneity in the results of the studies with post-intervention measurements only. The quality of the evidence that we found was not very high. Some of the studies applied rigorous methods but contended with attrition problems. Most of the randomized controlled trials were small, and, in all but one (42) of the included studies, the method of randomization was not reported or not valid. It was also difficult to get a good impression of the concealment of allocation to the researchers. We assumed that outcomes that were measured by a questionnaire were reported blind to the researchers. Even though we found significant results, it is difficult to say how the results are related to the clinical relevance of the changes achieved. With the Maslach Burnout Inventory (55), there is no generally accepted change that would be regarded as clinically relevant (Wilmar Schaufeli, personal communication). Since most of the studies had only a small sample size and all of them reported positive outcomes, it is conceivable that there may have been publication bias. However, the extent of publication bias is impossible to assess, as all of the studies did not report outcomes that could be used for a statistical analysis.

Comparison with other reviews
This review used more rigorous inclusion criteria and found more and better-quality evidence than previous reviews about stress interventions directed towards health care workers (7,26). Another review involved a metaanalysis of interventions to prevent or treat stress in all occupations (25). The review synthesized all of the study outcomes available in primary studies. This approach makes it difficult to decipher the meaning of their findings. In contrast to our review, they concluded that there was no evidence for work-directed or person-work interface interventions (25). Due to the different samples of studies (our review contains 13 studies that theirs does not) and methods used, it is difficult to explain this contrasting finding. A recent review about interventions to improve the morale of staff was restricted to mental health care workers only (60).

Implications
The results of this review show that stress management interventions can lead to positive health effects among health care personnel. There is evidence from one trial (30) that interventions that contain cognitive elements yield better results than those with behavioral elements. However, before large-scale implementation can be advised, larger and better-quality trials are needed. It would also be good to know what the current prevalence of various stress management strategies is. According to our own experience, it is much more common to measure various indices of stress than it is to do something about it. None of the studies looked specifically at stress reduction among physicians, probably because this professional group is often more reluctant to participate. Since physicians usually have more decision latitude or control, autonomy, possibilities for development, and rewards than nurses do, it is logical to assume that they may also need especially tailored stress management interventions. It is therefore difficult to say whether our results can be generalized also to physicians. Studies are needed that contrast various stress-or burnout-reducing techniques with one another. Studies that contrast organizational and person-work interface interventions with person-directed interventions will show whether or not one type of intervention is more effective in reducing stress levels than the other.

Ruotsalainen et al
This systematic review was prepared under the aegis of The Cochrane Collaboration, an international organization that aims to help people make well-informed decisions about health care by preparing, maintaining, and promoting the accessibility of systematic reviews of the effects of health care interventions. The Collaboration's publication policy permits journals to publish reviews but also permits The Cochrane Collaboration also to publish and disseminate such reviews.