Factors influencing mental health improvements in school teachers

Objective To identify changes in work-related psychological attitudes that influence mental health improvement in school teachers after participation in a psychological group program. Methods In an exploratory study with N = 544 matched cases we combined a screening instrument for general mental health (GHQ) with measures of work-related behavioral and experiential patterns (AVEM). We compared four GHQ change types pre and post intervention with regard to their performance on eleven sub-scales that figure as professional resources. Factors that showed significant relative changes and thus (likely) contributed to improved health status were identified by means of pairwise t-tests and corresponding effect sizes. Results Decreases in willingness to work to exhaustion (VB), in striving for perfection (PS), and in the tendency for resignation in the face of failure (RT), as well as an increase of distancing ability (DF) and of inner calm and balance (IR) appear to be the main factors influencing health improvement in the intervention. Simultaneously, an increase of satisfaction with life (LZ) is observed. Conclusions The balanced use of professional resources is a critical ingredient in maintaining teachers' health. Adjusting the balance between commitment and resistance through factors found in this analysis help teachers in maintaining and strengthening resilience. The coaching program addresses these factors by focusing on personal attitudes and good interpersonal relationships in the school environment.

The change below the cut-off is highly significant with the following statistic: Pearson s Chi-squared test with Yates continuity correction data: cutoff_pre and cutoff_post X-squared = 26.64, df = 1, p-value = 2.451e-07 Note that it is not the aim of this study to establish the effectiveness of the intervention. So this result only confirms what has been found before in a more rigorous (RCT) fashion [2].

GHQ effect
Effect size for the reduction of impaired GHQ status is calculated from the proportions above cut-off pre and post intervention (using arc-sine approximation [4]):  Table 2 shows that the reduction from 266/544=48.90% to 118/544=21.7% of the proportion of impaired health condition corresponds to a medium effect with Cohen's h=0.58.
The reduction is reflected in the GHQ score distribution showing significant changes pre/post intervention (Fig 1).

AVEM parameters
The AVEM inventory [5] has 11 sub-scales listed as follows (Table 3) We use stanine (standard-nine) scaled values for the AVEM parameters, normalized to a representative German school teacher population with N=18095 [5].

Replacement for missing values in AVEM parameters
In the GHQ score we required already complete cases, i.e. subjects must have valid GHQ status in pre (T1) and post (T2) measurement. Eliminating even more subjects, however, for missing values in one or more of the AVEM parameters is too expensive a method of data sanitizing. Therefore, we decided for a conservative missing replacement method, involving two steps: a) missing in pre measure is replaced by the median (most probable value) for that parameter and b) a missing in post measures is replaced by the pre value, thus will stay the same. In 7 cases we found 10 or more parameters missing (i.e. AVEM questionnaire was considered incomplete), those subjects are discarded entirely. The remaining cases have at most 3 missings per data record and are replaced accordingly (Table 4).

Groupwise paired t-tests on AVEM differences
AVEM parameters are measures in pre (T1) and post (T2), paired by anonymous match code. As they are not independent, we perform grouped pair-wise t-tests on the stanine values for each AVEM parameter and over all p-values applied the Holm-adjustment [6] method for multiple testing. The groups are defined by our four types of change in GHQ status, before and after the intervention, labelled as stable healthy, improvers, worseners, and stable at risk. GHQ status is defined as logical cut-off condition ghq ≥ 4, where ghq is the GHQ score on a scale from 0 to 12 (Fig 2).
We are mainly interested in effects and calculate effect sizes (Cohen's d) for correlated samples according to [7]  t-statistic, r is the correlation of the stanine values in each pair of AVEM parameters, and n df is the number of degrees of freedom in the t-test.   As can be seen in Fig 3, highly significant effects only show in the improvers type, whereas in the other types effects are either too small or insignificant. In the worseners type it can be argued that the decrease in VB (the willingness to work to exhaustion) is not a result of the intervention but a consequence of exhaustion itself. The change in AVEM feature LZ (satisfaction with life) in general is a consequence of health improvement or decline.

Limitations
We are aware that the changes in attitude and experience on health improvement are to be seen as correlates of the four GHQ change types. A deeper understanding of their mediating influence must be gained through further investigation. The forgoing analysis focuses on effect sizes rather than significance of changes. The effect sizes found are based on the t-test statistic and consider the correlation between pre and post measures of the AVEM features. The p-values are merely used as indicators. However, they have been adjusted by an appropriate adjustment method (Holm-correction) and underline the reliability of the result. The significance level is an aid to support our claim that the effects in the improvers group are to be considered as relevant. Correlation between AVEM features is assumed and the relevant factors may contribute in an orchestrated manner. Nevertheless, an analysis of correlation between AVEM features remains to be done and should be performed in a sequel to this paper. The missing replacement by median values in T1 and copy in T2 is simple and conservative, and a more sophisticated method could be sought, such as replacement on the item level of the questionnaire. Since missing pertains mainly to the feature SU (experience of social support), which is a resource that does not change much under the intervention, we consider our choice of replacement as safe.