The strength of this study is that it has a randomized, controlled design and uses well-established and validated outcome measures for social support and depression. Its main weaknesses are the high percentage of dropouts, which may have led to a selection bias by making it difficult to do a completely fair comparison between the groups (cf. the note above about the intention-to-treat analysis).
One possible explanation for the loss of participants in the intervention group after the randomization could be that when it came closer to the start of the programme, the participants started to have second thoughts and the obligation to weekly meetings for a year seemed too much to fulfil. As a result, the easiest way out was to withdraw at an early stage. The controls accepted three home visits during the year and a coming programme the next year, which did not imply that much personal commitment. To cushion this effect, we could have followed up the participants more closely by, e.g. phone calls.
The low number of participants, which leaves the study with a low statistical power, is also a problem. Both of these problems were difficult to avoid given the limited resources available in the intervention project, which calls for great care when interpreting the findings.
There were some differences between the intervention and control groups, both with respect to size and various socio-economic variables, and at baseline and the end of the intervention, thereby suggesting that the observed results might have been due to the selection mechanisms on one or more variable(s). However, including the variables under consideration as predictors in regression analyses, did not significantly affect the results, which suggest that the observed results in the present paper were neither due to selection nor group differences.
This was also confirmed by using various raking techniques
[35, 36]. More specifically, we first assumed that both the intervention and control group were identically distributed with respect to the different variables from Table
1: gender, age, income, education, marital status and geography at baseline, although this did not significantly affect the BDI level. Moreover, since the dropout rate from baseline to the end of the intervention was different in the two groups, we also checked whether this could affect the change in BDI from baseline to the end of the intervention by applying similar techniques. However, this did not significantly affect the BDI levels.
The effect of dropouts is difficult to judge. There are arguments for assuming that the dropouts would have benefited less than those who stayed, so that the inclusion of the dropouts would have made the differences between the intervention group and the control group smaller. But the opposite is also conceivable. Dropouts reported higher levels of stress and illness, so perhaps socializing at senior centres would have been of particular help to these people.
Disregarding the possible biases related to the high dropout rate, we may draw some conclusions from the data. Since all 95% confidence intervals of effect estimates clearly overlap with zero, it is impossible to reject the null hypothesis (that the intervention did not have any effect). That does not in itself mean that the intervention cannot have been effective, as there may have been positive effects that just do not show up in a statistically significant way due to the small sample size. We therefore focused on the possibility of there being more than just small and perhaps clinically insignificant effects, i.e. the possibility of effects that one would reasonably consider to be of clinical and policymaking interest. From this perspective, we suggested specific target values for effect sizes which the intervention should have been able to meet. According to our data, the intervention in all probability failed to meet optimistic targets, but possibly met quite modest ones. The latter possibility is supported by the positive reporting from participants with respect to satisfaction with the intervention. There was also a tendency towards a ‘dose–response’ effect, although this was not significant.
The very modest effect observed on BDI was somewhat surprising. An important concern is whether BDI is an inappropriate instrument in this context, in which the majority were over 80 years old. This inventory was chosen because it is widely used among older adults, with a well-documented high reliability, internal consistency and validity. The BDI also demonstrates a good discrimination between patients with varying degrees of depression, and accurately reflects changes in the intensity of depression over time
[30–33]. Still, it may be difficult to separate depression and cognitive impairments, even in a diagnostic evaluation
. Since the majority of participants were over the age of 80 years, and knowing that the incidence of cognitive decline increases sharply in this age group, stratification in age groups over and less than 80 years was conducted. There were no significant differences in BDI scores between the intervention and control groups in either of the two age groups. Furthermore, no significant differences were found in memory impairment.
There was a relatively large but non-significant decrease in BDI score among persons aged less than 80 years, and a significant increase in BDI score among persons over 80 years. It could well be that participants from 65 to 79 experienced a process of awareness and optimism as a result of being surveyed. In contrast, the oldest group might have experienced no awareness and optimism due to their age and future expectations from the staring point to the end of the intervention.
Another possible explanation for the modest effect on BDI could be that the level of depression in the sample was too low (a mean BDI score at baseline of approximately 10) for a substantial effect to be expected from the intervention, though this explanation is not supported by a subgroup analysis of the data. When we split the material into those with a BDI score equal to or less than 10 and those with a BDI score higher than 10 (not reported in the tables), we observe that the positive effect is almost statistically significantly smaller in the high BDI group than in the low BDI group.
An alternative explanation is therefore that an intervention of this type does not so much serve to improve the condition of those who already have considerable depression, but rather to avert development of more severe depression in those who have only mild symptoms.
With regard to the fact that the intervention programme at most had a modest effect upon depression, the frequency of meetings and the level of competence of the group leaders must be taken into consideration in the evaluation of this programme. The leaders were volunteers and had no health professional or social work background that qualified them to address mental health problems, and most of them had no prior experience with conducting group programmes. If this programme is meant to address users’ special mental challenges, more experienced and professional group leaders are needed. The substance of the programme must then be developed towards a treatment course, such as a Coping With Depression Course (CWD) for the elderly. An effectiveness trial proved the CWD course to be effective for older people with subclinical depression, as well as for those with a current major depression