Major depression is a common and impairing illness, often exhibiting a recurrent and progressive course. Depression ranks among the most common psychiatric disorders worldwide, with a 12-month prevalence between 10% and 17% and a lifetime prevalence between 17% and 40% (1, 2). Because the lifetime prevalence of depression is at most four times its 12-month prevalence, depressive episodes are not independent events uniformly distributed in the population but rather tend to cluster in the same individuals. Consistent with this evidence, 60% of individuals who have recovered from a depressive episode will have a recurrence within 5 years (3). The rate of recurrence may be even higher in individuals who experience more persistent depressive episodes and in those who have a poor treatment outcome with residual subthreshold symptoms (3, 4). Because of the large health impact and economic burden associated with a poor longitudinal course of depressive illness, it is important to identify factors that predict risk of developing recurrent and persistent depressive episodes and lack of remission or response during treatment for depression.
Childhood maltreatment may predict an unfavorable course of illness and treatment outcome. Compared with individuals who have not been maltreated, those with a history of childhood maltreatment are at greater risk of meeting criteria for a depressive episode at any point in life (5). Maltreated individuals are also at greater risk of enduring cognitive (6) and biological (7) vulnerabilities associated with heightened stress sensitivity (8), which might predispose them to an unfavorable course of illness and treatment outcome. For example, maltreated individuals tend to have elevated inflammation levels (7, 9), which have been associated with recurrent depressive episodes (10) and poor treatment response (11).
To test whether individuals with a history of childhood maltreatment are at elevated risk of an unfavorable depression course and treatment outcome, we performed meta-analyses of epidemiological studies investigating the association between childhood maltreatment and depression recurrence or persistence and of clinical trials investigating the association between childhood maltreatment and outcome of psychological, pharmacological, or combined treatment. We also explored the effects of various possible sources of artifact or bias on the results of the meta-analyses.
We identified studies satisfying the following criteria: definition of childhood adversities consistent with childhood maltreatment (physical abuse, sexual abuse, neglect, or family conflict or violence); diagnosis of depressive disorder, ascertained either in population-based or in clinical samples; and evaluation of relevant depression measures (12). Depression recurrence was defined in terms of number of depressive episodes over the period of observation or as risk of a new depressive episode in individuals with prior history of depression. Depression persistence was defined as prolonged duration of uninterrupted illness, including dysthymia, over the period of observation. Treatment outcome was defined as a change in depression scores between the beginning and the end of treatment, either as a continuous measure (i.e., improvement, defined as percentage improvement on a depression rating scale) or as a categorical outcome (i.e., response, defined as a reduction of depression rating scale score by 50% or more; or remission, defined as a decrease of depressive symptoms below a predetermined clinical significance level).
+
Identification of Studies
We searched MEDLINE, PsycINFO, and Embase databases for articles describing the relationship between childhood maltreatment (search terms: child* maltreatment, child* abuse, child* neglect, early experience) and relevant depression measures (search terms: depress*, mood disorder, MDD, recurrence, persistence, chronic, duration, length, improvement, response, remission, treatment, psychotherapy, CBT, pharmacotherapy, antidepressant, SSRI) using human subjects, written in English, and published by December 31, 2010.
Two authors independently extracted data from eligible articles. Inconsistencies were resolved in consensus meetings and confirmed with the authors of the primary studies when necessary.
Extracted data were converted to odds ratio effect sizes (13) reflecting the probability of unfavorable outcomes, with odds ratios above 1 reflecting a greater likelihood of recurrence, persistence, or poor treatment outcome in individuals with a history of childhood maltreatment compared to those without. The meta-analysis of clinical trials examined the difference in outcome between depressed patients with a history of childhood maltreatment and those without in the same active treatment arm. Where only continuous outcomes were reported, the risk of unfavorable outcomes was derived using validated methods (14). Heterogeneity between studies was tested with Cochran's Q test (15). In the presence of a homogeneous distribution, we carried out meta-analyses using fixed-effects models weighting each study by a measure of sampling error, such as the inverse variance. If the dispersion of effect sizes was greater than expected from sampling error alone (i.e., heterogeneity), we carried out meta-analyses using random-effects models, which include both sampling and study-level errors. Fixed- and random-effects meta-analyses were performed using the meta and metan programs applied in the Stata software package (StataCorp, College Station, Tex.). Additional analyses explored the effects of various possible sources of artifact or bias on the results.
First, we assessed the presence of publication bias visually by funnel plot (16) and formally by its direct statistical analogue, Begg's adjusted-rank correlation test (17), using the metabias program applied in Stata. In the presence of significant rank correlation tests, we adopted a nonparametric trim-and-fill method (18) using the metatrim program applied in Stata.
Second, we assessed the sensitivity of meta-analysis results to different outcome measures through subgroup analyses.
Third, we assessed the sensitivity of the results to different sampling strategies through subgroup analyses. Studies in clinical (or prevalence) samples and those in population (or incidence) samples could lead to different results because of ascertainment bias (19). The potential ascertainment bias could not have affected the results of clinical trials, which were restricted to clinical samples.
Fourth, we assessed the impact of study quality on the meta-analysis results through meta-regression using the metareg program applied in Stata. The quality of epidemiological studies was assessed with the Newcastle-Ottawa Scale (20), which has been recommended by the Cochrane collaboration (21) and has been used in previous studies (22). The quality of clinical trials was assessed by an adapted version of the Jadad scale (23) based on randomization, consideration of dropouts, and blinding. Because the contrast of interest was between individuals with a history of maltreatment and those without in the same active treatment arm, we considered blinding to maltreatment status rather than to treatment allocation.
Fifth, we assessed the presence of age effects through meta-regression analyses. Age effects could have influenced the meta-analysis results in two ways. With regard to the epidemiological studies, it was possible that because maltreated individuals are younger at the onset of depression than individuals without a history of maltreatment (24), they could spend more time at risk of recurrence and show greater depression recurrence and persistence for this reason. It was not possible to extract information on age at illness onset from the studies examined. However, we reasoned that the relative contribution of earlier age at onset to the total time that maltreated individuals spend at risk decays as the age of the sample increases. Therefore, if age at onset explained the association between childhood maltreatment and depression recurrence and persistence, we should have observed a significant negative correlation between mean age of the samples and effect sizes. With regard to the clinical trials, previous research has suggested that individuals at different ages could have a different response to the same antidepressant treatment (25). Therefore, it was possible that the age of the sample could have explained between-study differences in the association between childhood maltreatment and treatment outcomes in depression.
Finally, we assessed the presence of recall bias through meta-regression analyses. Most of the studies we examined were based on retrospective recall of childhood maltreatment. The reliability of retrospective recall of childhood maltreatment may be limited in individuals with a lifetime history of depression (26, 27) because of biases in autobiographical memory (28, 29). Therefore, to the extent that lifetime history of depression was different between the studies, it was possible that recall bias could have explained between-study differences in the association between childhood maltreatment and depression course. Recall bias was not an issue for the meta-analysis of the clinical trials, because all participants were depressed at the time the assessment of maltreatment history was carried out.
The study selection procedure is summarized in Figure 1.
+
Epidemiological Studies
The association between childhood maltreatment and poor longitudinal course of depressive disorder (i.e., recurrence or persistence) was tested in 16 studies with a total of 23,544 participants (Table 1) (7, 30–44). We identified significant heterogeneity across studies (Q=46.1, df=15, p<0.001) and therefore performed an analysis using a random-effects model. A meta-analysis of these studies showed that compared to those without a history of childhood maltreatment, maltreated individuals were twice as likely to have an unfavorable longitudinal course of depression (odds ratio=2.27, 95% confidence interval [CI]=1.80–2.87) (Figure 2). We performed additional analyses to explore the effects of possible sources of artifact or bias on these results.
The funnel plot showed asymmetrical distribution of the studies, and results of Begg's adjusted-rank correlation test were significant (p=0.027), suggesting the possibility of publication bias. However, the trim-and-fill procedure achieved a similar combined effect size (odds ratio=1.78, 95% CI=1.38–2.31) to the nonfilled analysis, suggesting that potential publication bias did not significantly affect the results.
First, we performed sensitivity analyses to examine whether the association between childhood maltreatment and course of illness was comparable across different measures of depression course (Figure 2). The association between childhood maltreatment and depression recurrence was tested in seven studies. We identified significant heterogeneity across studies (Q=15.1, df=6, p=0.020) and therefore performed an analysis using a random-effects model. A meta-analysis of these studies showed that maltreated individuals were twice as likely as those without a history of childhood maltreatment to develop recurrent depressive episodes (odds ratio=2.24, 95% CI=1.62–3.10). The association between childhood maltreatment and depression persistence was tested in nine studies. We identified significant heterogeneity across the studies (Q=30.60, df=8, p<0.001) and therefore performed an analysis using a random-effects model. A meta-analysis of these studies showed that maltreated individuals were twice as likely as those without a history of childhood maltreatment to develop persistent depressive episodes (odds ratio=2.34, 95% CI=1.65–3.32).
Second, we performed sensitivity analyses to examine whether the association between childhood maltreatment and course of illness was comparable across clinical and population samples. The association between childhood maltreatment and depression course was tested in seven clinical (prevalence) samples. Heterogeneity approached significance across studies (Q=10.69, df=6, p=0.098), so we performed an analysis using a random-effects model. A meta-analysis of the studies with clinical samples showed that maltreated individuals were nearly twice as likely as those without a history of childhood maltreatment to develop recurrent or persistant depressive episodes (odds ratio=1.78, 95% CI=1.36–2.34). The association between childhood maltreatment and depression course was tested in nine population (incidence) samples. We identified significant heterogeneity across the studies (Q=30.12, df=8, p<0.001) and therefore performed an analysis using a random-effects model. A meta-analysis of the studies with population samples showed that maltreated individuals were more than twice as likely as those without a history of childhood maltreatment to develop recurrent or persistent depressive episodes (odds ratio=2.75, 95% CI=1.94–3.91).
+
Bias owing to study quality.
Meta-regression analysis showed that between-study differences in quality ratings based on the Newcastle-Ottawa Scale did not explain the between-study variability in the association of childhood maltreatment with depression course (p=0.611), suggesting that bias due to study quality was unlikely. (The study quality assessment is reported in the data supplement that accompanies the online edition of this article.)
+
Effect of age at illness onset.
This analysis was based on proxy measures of mean age of the samples. We extracted data on mean age from 15 studies. We did not observe the negative correlation between age of the samples and effect sizes that was expected in the presence of an age-at-onset effect (r=–0.075, 95% CI=–0.541 to 0.392). In addition, meta-regression analysis showed that between-study differences in age of the sample did not explain the between-study variability in the association of childhood maltreatment with depression course (p=0.819), suggesting that confounding by age at onset was unlikely.
This analysis was based on proxy measures of lifetime history of depression. We extracted data on lifetime prevalence of depression from 14 studies. Meta-regression analysis showed that between-study differences in lifetime prevalence of depression did not explain the between-study variability in the association of childhood maltreatment with depression course (p=0.878), suggesting that recall bias was unlikely.
The association between childhood maltreatment and the outcome of treatment with psychological therapy, antidepressant medication, or combined treatment was tested in 10 studies (15 active treatment arms) with a total of 3,098 participants (Table 2) (45–54). We identified significant heterogeneity across studies (Q=24.77, df=14, p=0.037) and therefore performed an analysis using a random-effects model. A meta-analysis of these studies revealed that maltreated individuals were more likely than those without a history of childhood maltreatment to show poor treatment outcome (odds ratio=1.43, 95% CI=1.11–1.83) (Figure 3). We performed additional analyses to explore the effect of possible sources of artifact or bias on these results.
The funnel plot showed symmetrical distribution of the studies and Begg's adjusted rank correlation test was nonsignificant (p=0.553), suggesting that publication bias was unlikely.
We performed sensitivity analyses to examine whether the association between childhood maltreatment and treatment outcome was comparable across different treatment modalities (Figure 3). The association between childhood maltreatment and the outcome of psychotherapy was tested in four studies. We did not identify significant heterogeneity across studies (Q=6.03, df=3, p=0.110) and therefore performed an analysis using a fixed-effects model. A meta-analysis of these studies revealed that compared to those without a history of childhood maltreatment, maltreated individuals showed a nonsignificantly higher risk for poor outcome of psychotherapy (odds ratio=1.12, 95% CI=0.68–1.85). The association between childhood maltreatment and the outcome of pharmacological treatment was tested in six studies. We did not identify significant heterogeneity across studies (Q=8.41, df=5, p=0.135) and therefore performed an analysis using a fixed-effects model. A meta-analysis of these studies revealed that maltreated individuals showed a poorer outcome of pharmacological treatment compared to those without a history of childhood maltreatment (odds ratio=1.26, 95% CI=1.01–1.56). The association between childhood maltreatment and the outcome of combined treatment (psychotherapy and antidepressant medications) was tested in five studies. We did not identify significant heterogeneity across studies (Q=4.84, df=4, p=0.304) and therefore performed an analysis using a fixed-effects model. A meta-analysis of these studies revealed that maltreated individuals showed poorer outcome of combined treatment compared to those without a history of childhood maltreatment (odds ratio=1.90, 95% CI=1.40–2.58).
+
Bias owing to study quality.
Meta-regression analysis showed that between-study differences in quality ratings based on the adapted Jadad scale did not explain the between-study variability in the association of childhood maltreatment with treatment outcome (p=0.383), suggesting that bias due to study quality was unlikely. (The study quality assessment is reported in the online data supplement.)
Meta-regression analysis showed that between-study differences in mean age of the sample did not explain the between-study variability in the association between childhood maltreatment and poor response to antidepressant treatment (p=0.644). Furthermore, a meta-regression analysis showed that classification into pediatric (<18 years) compared with adult samples did not explain the variability in the association between childhood maltreatment and poor response to antidepressant treatment (p=0.498).
This meta-analysis addressed the possible developmental origins of heterogeneity in clinically relevant measures of depression such as course of illness and treatment outcome. The epidemiological studies suggested that maltreated individuals were twice as likely as those without a history of childhood maltreatment to develop both recurrent and persistent depressive episodes. Findings from clinical trials were consistent with the epidemiological observations. Compared with depressed individuals without a history of childhood maltreatment, depressed and maltreated individuals appeared to benefit less from treatment (and particularly from combined treatment), thereby incurring greater risk of recurrent and persistent depressive episodes (4).
These results should be evaluated in the context of several potential limitations. First, selective publication of studies reporting significant associations between childhood maltreatment and clinically relevant depression measures may influence results. We identified evidence of publication bias in epidemiological studies, but the trim-and-fill procedure suggested that publication bias was unlikely to significantly change the meta-analysis results. In contrast, both graphical and statistical methods suggested that the presence of significant publication bias was unlikely in clinical trials. Second, the results could also be limited to particular outcome measures. However, subgroup analyses appeared to be insensitive to the outcome measures examined. Third, results from epidemiological studies could be limited to clinical or population samples. However, subgroup analyses appeared to be overall insensitive to the sample type examined. Fourth, study quality could have influenced the results. However, differences in study quality could not explain between-study heterogeneity in the association between childhood maltreatment and depression course or treatment outcome. Fifth, the earlier age at onset for depression in maltreated individuals could influence epidemiological study results, leading to spurious findings of greater depression recurrence in maltreated individuals compared with individuals with no history of maltreatment by increasing the amount of time that maltreated individuals spent at risk. However, we could not observe the negative correlation between effect sizes and mean age of the samples that was expected in the presence of age-at-onset effect. In addition, the differences in mean age could not explain between-study heterogeneity in the association between childhood maltreatment and depression course. Sixth, results from epidemiological studies could also be influenced by recall bias owing to between-study differences in lifetime history of depression. However, differences in lifetime history of depression could not explain between-study heterogeneity in the association between childhood maltreatment and depression course. Finally, the available data did not enable us to examine the relative contribution of different maltreatment subtypes or the effect of other potential confounders.
Despite the potential limitations, the study results have important implications in several areas.
Childhood maltreatment may be conceptualized as a developmental risk factor triggering a chain of risks (55) such as subsequent depressive episodes that might progressively potentiate the vulnerability to poor course of illness (56, 57). In order to understand the origins of this chain of risks, studies should explore the cognitive and biological correlates of maltreatment in childhood before the accumulation of multiple depressive episodes (58). Furthermore, it will be important to characterize the gene-environment interplay underlying the effects of childhood maltreatment on depression outcomes (59–61). Childhood maltreatment may be conceptualized as an environmental risk factor for poor depression course and a moderator of treatment outcome (62), complementing the emerging genetic markers of vulnerability to recurrent depression (63) and poor treatment response (64, 65).
Information about a history of childhood maltreatment helps identify individuals who are at high risk of developing a recurrent and persistent subtype of depression and those who will respond poorly to treatment. Because of the lack of placebo-controlled studies, we were unable to test whether depressed and maltreated individuals benefit more or less from treatment (versus placebo) than depressed individuals who were not maltreated. In contrast, we observed that depressed and maltreated individuals have a poorer outcome compared with depressed individuals who were not maltreated (i.e., a poor prognosis) within treatment groups. Clinicians may consider that the routine inquiry about childhood maltreatment is not harmful (66) and could add important prognostic information to their assessment. Clinicians may also consider more intensive and alternative treatment options for depressed individuals with a history of childhood maltreatment. The meta-analytical evidence that maltreated and depressed individuals have a poor response to combined treatment with structured psychological therapy and antidepressant medications indicates that simply combining these two common options is not sufficient. It will be important to explore the response of maltreated and depressed individuals to new treatments targeting the biological vulnerabilities described in this subgroup (7, 60), including elevated inflammation levels (67, 68).
Interventions aimed at reducing childhood maltreatment could help prevent the large health and economic burden linked to poor depression course. Childhood years are thought to be a sensitive developmental window for the maturation of emotion regulation (69). Therefore, in the same way that childhood educational interventions have greater returns in human capital formation than interventions at later ages (70), early preventive and therapeutic interventions may be more effective (and cost-effective) in preventing a poor longitudinal course of depression than interventions at later ages, when harmful developmental trajectories have already been established.
The authors thank Professor Sir Michael Rutter and Professor George W. Brown for the helpful comments on a previous version of the manuscript.