Our analysis failed to provide sufficient evidence to decision-makers that the intervention is cost-effective. In case of severe depression, the evidence even indicates that routine care is economically advantageous. These results are not in line with the findings by Härter et al, who observed for patients in the SCM a pronounced improvement of symptom burden as well as increased odds of response and remission (13). Nevertheless, some indicators for an impact of the intervention on health care delivery warrant further research on the intervention.
The development of the NCPG and hence of this intervention was triggered by the need to improve identification of depression cases and to bring patients into adequate treatment (10, 11). Several of our results can be interpreted as in accord with this aim. The NCPG recommends low intensity treatments for patients with mild depression (10). In our study, these interventional measures (bibliotherapy, web-based self-management, telephone psychotherapy) showed the highest incremental costs in this group of patients in comparison to other degrees of severity. Furthermore, the NCPG lays a strong emphasis on treatment in the outpatient sector by mental health professionals (10). In the complete sample, we found that the costs for psychiatric outpatient services were significantly increased in the IG. The same trend was found for all subgroups and the psychotherapeutic services. Considering the share of patients utilizing these services, we found that these services were more often utilized in the IG, especially in the mildly and moderately affected subgroups.
A different pattern appeared with respect to psychiatric inpatient services. For these services, we found a trend for cost reductions (moderate and severe depression) and negligibly higher costs (moderately severe depression) in the IG. As the NCPG fosters outpatient treatment, this can be interpreted as in line with the NCPG. However, there are two intuitively contradicting observations. IG patients in the group with mild depression showed the largest increase in costs for psychiatric hospital services and those in the group with severe depression the largest decrease of these costs. It is not to expect that mildly affected patients are in need of a care-intensive inpatient treatment, while severely affected patients appear to be the main target population for these services. The observation in the severely affected group could be interpreted as an intervention effect. It is possible that the SCM provided the coordinating physician with a variety of different treatment options making inpatient treatment a last line treatment. The observation in the group of mildly affected patients is part of a larger utilization pattern. This group exhibits not only the highest cost difference between IG and CG in psychiatric hospital services but also in drugs and hospital services in general. On the one hand, it is possible that this result was affected by the small sample size in this group (IG: 63; CG 30) and is hence not generalizable for the group of all patients with mild depression. On the other hand, there could be a systematic effect. For example, the SCM could increase the availability of mental health care services. Furthermore, by including the patient in the process of decision-making the demand for health care services could grow. This could be an explanation for the higher costs for drugs and inpatient psychiatric services. The trend towards higher costs for hospital services in general is not only a phenomenon observable in the group with mild depression but also in other subgroups. These findings and the corresponding observations for outpatient services (a trend for an increase in mental health unspecific services) can be interpreted as indicators of an activating effect of the SCM. Depressive disorders are associated with a lack of motivation and energy and strong feelings of hope- and helplessness (37). Furthermore, they have a relationship of mutual influence with somatic disorders (3, 38, 39). The analysis by Härter et al showed that the symptom status of patients in the SCM group improved significantly (13). Additionally, this program is based on regular monitoring and frequent contacts to health care professionals. This results in a higher probability of the treatment of previously untreated health issues and is an explanation for the higher utilization of mental health unspecific services.
The last aspect of the results that need to be discussed are the significantly increased costs for absenteeism from work caused by participants with severe depression in the IG. As only the cost difference between T0 and T2 was significant, we can conclude that physicians tend to attest long-term sick leave right at the beginning of the intervention to facilitate the systematic treatment of the severely affected patients. After six months and successfully initiating treatment, physicians might start to reintegrate patients into their daily routines again.
In view of the results and the discussion it becomes obvious that follow-up time is the most limiting factor in our study. The time horizon of our study was one year. This implies that it is possible that not all effects and health care consequences of the intervention were observed. Due to the natural course of depression, the duration and number of episodes, the duration of remission and the risk of relapses, one year might be too short to observe all differences between the interventions (40, 41). According to our results, the costs go up in the first year and there is no notable effect on HRQL. However, we cannot rule out that the increase in costs is an investment in the future and will result in a decrease in later years. On the effects side, we know that the intervention reduces symptom severity and leads to more remissions, but not to better HRQL. The conceptual literature on patient reported outcomes shows that symptom status and HRQL are related and part of the same outcome continuum (40, 41). Symptom status can affect HRQL directly, if it is especially burdensome, or via its influence in functioning and general health perception. Supposing that the change in symptom status was not pronounced enough to affect HRQL directly, it has to influence functioning and general health perception. Considering this, we can assume that the time horizon of one year could not have been long enough to lead to changes in functioning and perception and hence to a substantial change in HRQL.
Furthermore, we have to consider that the use of patient questionnaires is associated with a risk of missing values and recall bias. The degree of missing values was manageable and was handled by an elaborated approach. The presence of a recall bias cannot be ruled out or controlled. Additionally, in the interpretation of the results, we have to keep in mind that we identified mostly trends and that the randomization was not stratified for the subgroups. This means that the composition of the subgroups was not necessarily randomized. For this reason, we adjusted the analyses in the subgroups for the group specific significant baseline differences.