Results
Searches and Inclusion of Studies
The literature search resulted in a total of 7447 articles found in PubMed (n = 2006), Cochrane Central Register of Controlled Trials (n = 2707), PsychInfo (n = 932) and EMBASE (n = 1802). We removed duplicates, leaving 4591 articles to be examined. We retrieved a total of 235 full-text articles that potentially met our inclusion criteria. Of these, 203 were excluded. Most (n = 135) were excluded because they lacked a diagnosis at the baseline and/or the follow up. Another reason for not including studies was lack of randomization (n = 18). All reasons for exclusion are noted in Figure 1. Control groups primarily consisted of care as usual, with some exceptions such as: placebo pill, booklet or no intervention (Table 1).
(Enlarge Image)
Figure 1.
Flow chart of included studies
Characteristics of Included Studies
A total of 32 studies with 6214 participants (3312 in the prevention groups and 2902 in the control groups) met all inclusion criteria. In one study three different intervention groups were examined so we were able to include 34 comparisons between preventive interventions and control groups. Table 1 shows selected characteristics of the included studies. Sheffield et al. (2006) investigated an universal preventive intervention and two indicated preventive interventions. One other study examined universal prevention, whereas indicated prevention and selected prevention were each investigated by 15 other studies.
The majority of studies focused on preventing MDD, 9 studies aimed at postpartum depression (PMDD) and 4 dealt with mood mixed disorder (e.g. a combination of MDD, dysthymia and/or minor depression). These were diagnosed by diagnostic instruments, such as the Structured Clinical Interview for DSM-IV (SCID) (Table 1), which use DSM-III-R or DSM-IV criteria. Most studies did not inform whether they excluded or included participants with a history of depressive disorders (n = 20). Four studies reported using participants with first episode of depression. Eight studies reported including participants with a history of depression, however participants did not experience a depressive disorder at the time of the baseline measure. Eight studies focused on adults in general, 1 study focused on adults with diabetes, 6 studies on pregnant women and 3 studies on (new) mothers, but most studies focused on adolescents or students (n = 14). Fifteen interventions were based on the principle of cognitive behavioural therapy. Some studies based their intervention on other psychological approaches, such as problem-solving therapy (n = 2) or interpersonal group therapy (n = 5). The number of sessions ranged from 4 to 15. Most studies used interventions which consisted of 12 sessions (n = 7), 2 studies used preventive interventions which consisted of 4 sessions and 2 studies used preventive interventions consisting of 15 sessions.
Eleven studies were conducted in Europe, 14 in the USA and 9 elsewhere. The follow-up periods of these studies varied between 2 and 60 months (median = 9 months). Only one study reported a follow-up of 5 years and one study reported a follow-up of 36 months. One study reported a follow-up period of 2 months and 8 studies reported a follow-up period of 3 months. Most studies, however, also reported a follow-up period of 6 or 12 months (n = 28). Drop-out rates in the studies varied between 2% and 64%. Intention-to-treat-analyses were done by most studies (n = 19).
Quality of the articles was relatively high. Quality of studies was assessed on four criteria: allocation concealment, incomplete outcome data, blinding of outcome assessors, and sequence generation. Sixteen studies reported that blinding of the allocation of interventions was done adequately. Eight studies met all four criteria, 18 studies met two or three criteria and six studies met no or only one criterion.
Overall Incidence Rate Ratios
We calculated the mean IRR by combining the IRRs at different follow-up times into a single estimate. When looking at the fixed-effects model, the IRR for all 34 comparisons from the 32 studies was 0.82 [95% confidence interval (CI) = 0.73–0.91; P = 0.000]. Focusing on the random-effects model, the IRR for all 34 comparisons from the 32 studies was 0.79 (95% CI = 0.69–0.91; P = 0.001). Heterogeneity was low (I = 24%). Because the differences between the fixed- and the random-effects models were small, we only report the results for the random-effects model (Table 2 and Figure 2).
(Enlarge Image)
Figure 2.
The effects of preventive interventions on the incidence of depressive disorders; incidence rate ratios and numbers needed to treat. Lines represent IRR and 95% CI; the size of the square indicates the weight of each study
There was one study that compared three interventions with one control group. Since these comparisons were not independent from each other, we examined whether removal of these comparisons would increase heterogeneity. The overall analyses of 32 studies resulted in a mean IRR of 0.77 (95% CI: 0.66–0.90, P = 0.005), with low heterogeneity (I = 29%). This was comparable to the mean IRR found in the total sample.
Since the IRR could differ at varying follow-up periods, we conducted several sensitivity analyses. We examined the IRR for each follow-up period separately (<5 months; 6 months, 7–12 months, ≥13 months; Table 2). We also conducted a separate analysis in which we used only the last follow-up period reported in each study (0.78; 95% CI: 0.68–0.89; P = 0.001; I = 29), and another analysis with only the first follow-up period of each study (0.79; 95% CI: 0.69–0.92; P = 0.002; I = 29). As can be seen in Table 2, we found few indications that the outcomes differed very much from the IRR in which all follow-up periods were pooled.
We also conducted meta-regression analyses to see whether there was any effect decay over time. First, we examined the association between IRR and the first follow-up period reported in the study. We did not find an association between IRR and first follow-up period (the point estimate of the slope was 0.003; 95% CI: −0.007 to 0.013), although there was a trend (P = 0.06) suggesting that the effects of the interventions are lower at longer follow-up periods (median: 7.5 months; range: 2–60 months). In the second meta-regression analysis we used the last follow-up period reported in the studies. Again, these results did not show an association between IRR and last-follow-up occasion (P = 0.06; the point estimate of the slope was −0.000; 95% CI: −0.01 to 0.01), suggesting that the longer it takes before the last follow-up period, the lower the incidence of depression is (median = 12; range = 3–60 months).
Inspection of the funnel plot (Supplementary Appendix C, available as Supplementarydata at IJE online) and Duval and Tweedie's trim-and-fill procedure attested to the possible presence of publication bias. After adjustment for publication bias, the effect size was increased from 0.82 to 0.86 (95% CI: 0.74–1.00; number of trimmed studies: 10). The Egger's test also indicated an asymmetric funnel plot [intercept: −1.24, 95% CI: −1.95 to −0.53, degree of freedom (df) 32, P = 0.001]. The fail-safe n was 175, indicating that 175 studies with an effect size of 0 would have to be included to not find a publication bias.
Subgroup Analyses
We conducted a series of subgroup analyses (Table 2). We examined whether the IRR differed according to type of prevention (indicated, universal or selective), type of intervention (CBT, IPT or other), age group (adolescent, adults or elderly), number of sessions (1–7, 8–11, ≥12; one study did not report the number of intervention sessions), country of publication (USA, EU or other), and target group (school-based, general medical, perinatal or other).
The IRR did not differ in any of the subgroups (Table 2). The difference between CBT and IPT interventions, found by Cuijpers et al. in 2008, could not be replicated in the current meta-analyses. This null finding might be caused by the low number of studies using IPT as an intervention (n = 5).
However, when looking at NNT, as indicated in Table 2, there was a difference between number needed to treat of CBT (NNT = 71), IPT (NNT = 7) and other (NNT = 12) interventions (P = 0.003), suggesting that preventive interventions using IPT are more effective than preventive interventions using CBT.
In most subgroup analyses the heterogeneity was low to moderate. No heterogeneity was found in several subgroups of studies: subgroups using CBT, those focusing on the elderly and those using a target population of general medical patients. Also, no heterogeneity was found in subgroups having 8–11 sessions, having a publication score of 3 or 4 or from studies not published in Europe.