Author’s Accepted Manuscript Depression in patients with alcohol use disorders: Systematic review and meta-analysis of outcomes for independent and substance-induced disorders James A Foulds, Simon J Adamson, Joseph M Boden, Jonathan A Williman, Roger T Mulder www.elsevier.com/locate/jad
PII: DOI: Reference:
S0165-0327(15)00398-5 http://dx.doi.org/10.1016/j.jad.2015.06.024 JAD7530
To appear in: Journal of Affective Disorders Received date: 20 February 2015 Revised date: 11 June 2015 Accepted date: 14 June 2015 Cite this article as: James A Foulds, Simon J Adamson, Joseph M Boden, Jonathan A Williman and Roger T Mulder, Depression in patients with alcohol use disorders: Systematic review and meta-analysis of outcomes for independent and substance-induced disorders, Journal of Affective Disorders, http://dx.doi.org/10.1016/j.jad.2015.06.024 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Depression in patients with alcohol use disorders: systematic review and metaanalysis of outcomes for independent and substance-induced disorders
James A Foulds 1 Simon J Adamson1 Joseph M Boden2 Jonathan A Williman3 Roger T Mulder 1
1 Department of Psychological Medicine, University of Otago Christchurch, New Zealand 2 Christchurch Health and Development Study, University of Otago Christchurch, New Zealand 3 Biostatistics and Computational Biology, Department of Public Health and General Practice, University of Otago, Christchurch, New Zealand
Corresponding Author:
James Foulds Department of Psychological Medicine University of Otago, Christchurch PO Box 4345 Christchurch 8140 New Zealand Tel: ++64 3 3720 400 Fax: ++64 3 3720 407 Email: [email protected]
1
Abstract
Background: In patients with an alcohol use disorder, depression is commonly categorised as independent (ID) or substance-induced (SID). It is not established whether these conditions respond differently to treatment.
Methods: MEDLINE, Embase and Cochrane databases from 1980 to 2014 were searched for studies on alcohol use disorders with coexisting depressive symptoms. Meta-analyses were conducted using random effects models, to derive pooled effect estimates of the change in depression during treatment and the effect of antidepressant therapy.
Results: Twenty-two studies met inclusion criteria for the review, of which 11/22 were included in the meta-analysis. All studies reported a large improvement in depression symptom score, most of which occurred within the first 3 to 6 weeks of treatment. The amount of improvement during follow up was similar in studies on ID in comparison to those in undifferentiated depression. Evidence on the outcome for SID was limited.
The effect size of antidepressant therapy compared to placebo was 0.25 (0.06, 0.44) for ID and 0.08 (-0.31, 0.47) for SID or undifferentiated depression.
Limitations: Few studies examined the natural history and treatment response of SID. There was heterogeneity between studies, which was partly explained by baseline depression severity.
2
Conclusions: Treatment for depression co-occurring with an alcohol use disorder is associated with a large early improvement in depression, even if depression is believed to be independent of drinking. The effect of antidepressant therapy on depression in patients with alcohol use disorders is modest, with stronger evidence in ID.
Keywords: alcohol drinking; alcohol-induced disorder; alcohol-related disorders; depressive disorder; depression; antidepressive agents
3
1. INTRODUCTION Patients entering treatment for an alcohol use disorder often have high levels of depressive symptoms (Davidson, 1995; M. A. Schuckit, 1995). These symptoms typically improve rapidly with treatment (Amanda L Baker et al., 2013; S. A. Brown, 1988; Davidson, 1995; Kiefer & Barocka, 1999; M. A. Schuckit, 1985, 1995) but in spite of this, comorbid depression predicts worse outcomes in alcohol treatment (Lejoyeux & Lehert, 2011; Pettinati, 2013) and heavy drinkers have an increased risk of future depression even if they cut down (Deborah S. Hasin & Grant, 2002). Knowing which groups of patients are more or less likely to improve during treatment would allow scarce treatment resources to be allocated more effectively, but to date the ability to predict patient outcomes accurately has been limited.
1.1. Subtyping depression in patients with alcohol use disorders
An early approach to guide treatment and predict depression outcome in patients with an alcohol disorder was to classify depression as primary or secondary according to whether it developed before or after the onset of heavy drinking (M. A. Schuckit, 1985). In the 1990s, the typology evolved to also incorporate information about a past history of depression during abstinence (M A Schuckit et al., 1997). The term independent was used for depression that began before the onset of alcohol dependence or during sustained abstinence while depressive syndromes occurring only during a period of active alcohol dependence were labelled substance-induced (M A Schuckit, et al., 1997). Structured clinical assessment tools such as the Structured Clinical Interview for DSM (SCID) (Spitzer et al., 1992) and the Psychiatric Research Interview for Substance and Mental Disorders (PRISM) (Deborah S Hasin et al., 1996) have helped to operationalise these definitions and introduce greater diagnostic reliability.
1.2. Implications of subtypes
It has been argued that the distinction between independent depression (ID) and substance-induced depression (SID) has important implications for treatment and
4
prognosis: SID is considered a self-limiting condition that remits with abstinence while ID requires specific depression treatment (Pettinati, 2013; M. A. Schuckit, 1985; M A Schuckit, et al., 1997). In particular the view has been put forward that ID, but not SID, responds to antidepressant medication (M. A. Schuckit, 2006) but the evidence for this has not been systematically evaluated.
1.3. Aims of the review and meta-analysis
This review aimed to answer the following questions: i) In depressed patients with an alcohol use disorder, how much do depressive symptoms improve during treatment for depression? ii) Do patients with ID and SID have different patterns of treatment response? iii) Does antidepressant efficacy differ between patients with ID and SID?
2. METHODS 2.1. Selection Criteria
Studies were chosen according to the following criteria: 1. Studies reported longitudinal data on alcohol use and depression in adults over at least 8 weeks. 2. Change in mean score on a validated depression scale was reported. 3. Subjects had an active alcohol use disorder (alcohol dependence or alcohol abuse) diagnosed according to DSM or ICD criteria. 4. Mean baseline score ≥ 10 on the 17-item Hamilton Depression Rating Scale (HDRS) (Hamilton, 1960), or equivalent severity on another depression rating scale. Given the uncertain nosology of depression in alcohol use disorders, subjects were not required to have a diagnosis of a depressive disorder. The low severity threshold was chosen to avoid excluding studies in which baseline measures were taken after a short period of abstinence. A published score comparison table available at http://www.ids-qids.org/index2.html (University of Pittsburgh, 2014) was used to calibrate severity across scales.
5
5. Treatment for depressive symptoms was provided. A liberal definition of such treatment was applied, and it could include any form of medication or psychosocial treatment with plausible efficacy in treating depressive symptoms.
Unpublished and incomplete studies were not assessed for inclusion. Studies were excluded if they were conducted in special sub-populations (for example, prisoners) or those with another psychiatric or substance use disorder (for example, borderline personality disorder or nicotine dependence). Studies exclusively comprised of subjects with dysthymic disorder were also excluded as this condition has a different natural history compared to other depressive syndromes (Judd et al., 2002).
2.2. Information sources
Information sources were MEDLINE, Embase and Cochrane databases, searched on 9 November 2014. The reference lists from review articles (Amanda L. Baker, 2012; Nunes & Levin, 2004; Ostacher, 2007; Pettinati, 2004; Salloum & Jones, 2008; Sullivan et al., 2005) were used to hand-search for additional articles. Search items were alcohol drinking; alcohol-induced disorder; alcohol-related disorder; alcoholics; alcoholism AND depression; antidepressive agents. English-language publications on clinical trials from 1980 onwards were considered.
2.3. Study selection
The PRISMA diagram in Figure 1 illustrates the flow of studies during the selection process.
INSERT FIGURE 1 ABOUT HERE
Authors 1 (JF) and 2 (SA) independently screened titles and abstracts to exclude obviously irrelevant articles. Duplicate citations were deleted. The full text of articles in the screened sample was extracted and reviewed by authors 1 and 2. Studies meeting inclusion criteria were then compiled into the final sample by consensus.
6
2.4. Data items and summary measures
For analyses on the change in depression during treatment, results for each treatment arm of studies were aggregated. The mean depression score at baseline (taken as the earliest reported timepoint) and study completion, and their respective standard deviations were recorded. Study completion was chosen as the endpoint rather than a fixed time interval because studies did not report on outcomes at the same time points, and most change was expected to occur early in treatment (Amanda L Baker et al., 2014). The main summary measure was the standardised mean change in depression score from baseline to study completion, using the pooled standard deviation of baseline and follow up scores.
In two studies (Hernandez-Avila et al., 2004; Moak et al., 2003) the baseline standard deviation was very low, probably due to scores clustering just above the inclusion threshold. To avoid introducing bias, the baseline standard deviations were not used for these two studies and the follow-up standard deviations alone were used instead.
Depression type was defined according to whether studies were in subjects with ID, SID or a mixture of both types. Information on demographics, duration and adjunctive treatment was also extracted as shown in Table 1.
2.5. Assessment of bias and quality
Individual study bias was addressed by recording follow-up completion rates, and including in the meta-analysis only studies where reported outcomes were clearly from the intention-to-treat sample. External funding sources were noted. Study quality in randomized trials was assessed according to the adequacy of sequence generation, allocation concealment, blinding and completeness of reporting (Higgins & Green, 2008; Jadad et al., 1996). For observational studies, a quality checklist was used (Tooth et al., 2005). A funnel plot screened for publication bias.
2.6. Synthesis of results
7
A pooled estimate of the standardised mean change in depression score from baseline to follow-up was derived using a random effects model fitted with RevMan software (Cochrane Collaboration, 2008). This approach uses the pooled standard deviation of baseline and follow-up scores as the denominator of the standardized mean difference calculation, but ignores the correlation between these scores. However it is likely baseline and follow-up scores are in fact positively correlated, and ignoring this correlation can bias the estimation of effect sizes (Dunlap et al., 1996). Therefore a sensitivity analysis was performed to test the effect of correlated scores on the pooled effect size estimate. A plausible correlation figure of 0.4 was imputed, based on raw clinical trial data from a study conducted in our research unit (Adamson et al., 2015). The sensitivity analyses were performed with SAS v9.2 PROC MIXED (SAS Institute, Cary NC, USA) employing a method for random effects meta-analysis outlined by Sheu and Suzuki (2001). Variance estimates for each individual study outcome were obtained by the following formula (Dunlap, et al., 1996):
i.
Var (d) = [2(1 - r)/n] + [d2/(2n - 2)]
Where d is the standardized mean change for each study, r is the correlation between baseline and follow up scores within subjects and n is the sample size. Sensitivity analyses also investigated the influence of individual studies on the pooled effect size estimate.
Exploration of moderators of change in depression during treatment was then performed as a meta-regression analysis using PROC MIXED. Due to the small number of studies, moderator analyses were limited to baseline HDRS score, depression type, gender composition of the sample and manualisation of adjunctive psychological treatment.
2.7. Antidepressant efficacy
For the studies that compared an antidepressant to placebo, a pooled estimate of the treatment effect was derived using a random effects model in RevMan. In this analysis, the mean difference in change scores between antidepressant and placebo groups was calculated. This was then standardised using the pooled standard deviation 8
of follow-up scores. If follow-up standard deviations were not quoted, the standardised mean difference was derived from the F statistic from ANCOVA models. This approach to analysing change scores followed guidelines outlined in the Cochrane Handbook for Systematic Reviews of Interventions, page 269-270 (Higgins & Green, 2008) and it maintained consistency with a previous meta-analysis in this population (Nunes & Levin, 2004). A subgroup analysis then compared the pooled effect size between ID and SID / mixed studies. Sensitivity analyses investigated the influence of individual studies on the pooled effect size estimate.
2.8 Psychological treatments A brief narrative review of psychological treatment studies was performed. Due to the small number of randomised trials on psychological treatments that met inclusion criteria, no attempt was made to summarise results in a meta-analysis.
2.9 Predictors of treatment outcome In addition to the meta-regression analyses described above, a narrative review of predictors of treatment outcome at a within-study level was performed. This section focused on gender and alcohol use, as these predictors were most commonly studied and reported.
3. RESULTS 3.1. Search results and quality of studies
The search yielded 22 individual studies included in the systematic review. Table 1 presents a summary of these studies.
INSERT TABLE 1 ABOUT HERE
Twenty studies were randomized trials, of which eighteen were pharmacotherapy studies and two investigated non-pharmacological treatments. The remaining two studies were open label pharmacotherapy trials.
9
Of the 22 studies, 16 studies clearly presented data at baseline and follow-up for the intention-to-treat sample. Follow-up completion rates in these studies ranged from 48 to 93% (median 70%). The remaining 6 studies (Altintoprak et al., 2008; Dorus et al., 1989; Janiri et al., 1996; Mason et al., 1996; Paparrigopoulos et al., 2010; Petrakis et al., 2007) were excluded from the meta-analysis because their reported outcomes were not clearly identified as relating to the intention-to-treat sample.
The quality of randomized trials selected for the meta-analysis sample was good as judged by the Jadad quality score (range 3-5, median 4). For the two open-label studies, quality was acceptable to good.
3.2. Defining and measuring depression in subjects with an alcohol use disorder
Depression and alcohol use disorders were diagnosed with the Structured Clinical Interview for DSM (SCID) (Spitzer, et al., 1992) in fourteen studies; the Psychiatric Research Interview for Substance and Mental Disorders (PRISM) (Deborah S Hasin, et al., 1996) in two studies; the Schedule for Clinical Assessment in Neuropsychiatry (SCAN) (Wing et al., 1990) in one study; and the Diagnostic Interview Schedule for DSM (Robins et al., 1981) in one study. The remaining studies used unstructured or semi-structured interviews.
Sixteen studies, all pharmacotherapy trials, used versions of the Hamilton Depression Rating Scale (HDRS) (Hamilton, 1960) as a primary outcome. The standard 17-item version was used in 10/16 studies, the 24-item version in 3/16, the 21-item version in 2/16 and the remaining study did not state the version used. Three studies used the Beck Depression Inventory (BDI), (Beck et al., 1961). Two studies used the Montgomery-Asberg Depression Rating Scale (MADRS), (Montgomery & Asberg, 1979) and the remaining study used the QIDS-SR (Rush et al., 2003).
As shown in Table 1, there was variation between studies in the timing of baseline depression measures and the inclusion or exclusion of patients who responded during the initial assessment phase of the study. Most studies had a mean baseline depression 10
score in the mild to moderate depression range (Zimmerman et al., 2013); for example, among the studies using the 17-item HDRS, the score at baseline ranged from 12.6 to 23.9 (median 19.6).
Operational definitions of depression types varied between studies, and these are summarised in Table 1. For simplicity, the terms independent and substance-induced are used in the present review while studies on primary and secondary depression are grouped with those on ID and SID respectively.
Eleven studies determined depression type at baseline. Of these, five studies included only subjects with ID (Cornelius et al., 1997; McGrath et al., 1996; Moak, et al., 2003; Pettinati et al., 2010; Roy-Byrne et al., 2000). Three studies (Adamson, et al., 2015; R. A. Brown et al., 2011; Oslin, 2005) explicitly identified whether subjects had ID or SID, and included both types. The remaining three studies were on subjects with SID (Gual et al., 2003; Mason, et al., 1996; Paparrigopoulos, et al., 2010).
3.3. Change in depression during treatment
A meta-analysis was initially planned on the 16 studies that reported on change in depression from baseline to study completion for an intention-to-treat sample. However there was very high heterogeneity in this sample (Q=152.21, df=15, p
PII: DOI: Reference:
S0165-0327(15)00398-5 http://dx.doi.org/10.1016/j.jad.2015.06.024 JAD7530
To appear in: Journal of Affective Disorders Received date: 20 February 2015 Revised date: 11 June 2015 Accepted date: 14 June 2015 Cite this article as: James A Foulds, Simon J Adamson, Joseph M Boden, Jonathan A Williman and Roger T Mulder, Depression in patients with alcohol use disorders: Systematic review and meta-analysis of outcomes for independent and substance-induced disorders, Journal of Affective Disorders, http://dx.doi.org/10.1016/j.jad.2015.06.024 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Depression in patients with alcohol use disorders: systematic review and metaanalysis of outcomes for independent and substance-induced disorders
James A Foulds 1 Simon J Adamson1 Joseph M Boden2 Jonathan A Williman3 Roger T Mulder 1
1 Department of Psychological Medicine, University of Otago Christchurch, New Zealand 2 Christchurch Health and Development Study, University of Otago Christchurch, New Zealand 3 Biostatistics and Computational Biology, Department of Public Health and General Practice, University of Otago, Christchurch, New Zealand
Corresponding Author:
James Foulds Department of Psychological Medicine University of Otago, Christchurch PO Box 4345 Christchurch 8140 New Zealand Tel: ++64 3 3720 400 Fax: ++64 3 3720 407 Email: [email protected]
1
Abstract
Background: In patients with an alcohol use disorder, depression is commonly categorised as independent (ID) or substance-induced (SID). It is not established whether these conditions respond differently to treatment.
Methods: MEDLINE, Embase and Cochrane databases from 1980 to 2014 were searched for studies on alcohol use disorders with coexisting depressive symptoms. Meta-analyses were conducted using random effects models, to derive pooled effect estimates of the change in depression during treatment and the effect of antidepressant therapy.
Results: Twenty-two studies met inclusion criteria for the review, of which 11/22 were included in the meta-analysis. All studies reported a large improvement in depression symptom score, most of which occurred within the first 3 to 6 weeks of treatment. The amount of improvement during follow up was similar in studies on ID in comparison to those in undifferentiated depression. Evidence on the outcome for SID was limited.
The effect size of antidepressant therapy compared to placebo was 0.25 (0.06, 0.44) for ID and 0.08 (-0.31, 0.47) for SID or undifferentiated depression.
Limitations: Few studies examined the natural history and treatment response of SID. There was heterogeneity between studies, which was partly explained by baseline depression severity.
2
Conclusions: Treatment for depression co-occurring with an alcohol use disorder is associated with a large early improvement in depression, even if depression is believed to be independent of drinking. The effect of antidepressant therapy on depression in patients with alcohol use disorders is modest, with stronger evidence in ID.
Keywords: alcohol drinking; alcohol-induced disorder; alcohol-related disorders; depressive disorder; depression; antidepressive agents
3
1. INTRODUCTION Patients entering treatment for an alcohol use disorder often have high levels of depressive symptoms (Davidson, 1995; M. A. Schuckit, 1995). These symptoms typically improve rapidly with treatment (Amanda L Baker et al., 2013; S. A. Brown, 1988; Davidson, 1995; Kiefer & Barocka, 1999; M. A. Schuckit, 1985, 1995) but in spite of this, comorbid depression predicts worse outcomes in alcohol treatment (Lejoyeux & Lehert, 2011; Pettinati, 2013) and heavy drinkers have an increased risk of future depression even if they cut down (Deborah S. Hasin & Grant, 2002). Knowing which groups of patients are more or less likely to improve during treatment would allow scarce treatment resources to be allocated more effectively, but to date the ability to predict patient outcomes accurately has been limited.
1.1. Subtyping depression in patients with alcohol use disorders
An early approach to guide treatment and predict depression outcome in patients with an alcohol disorder was to classify depression as primary or secondary according to whether it developed before or after the onset of heavy drinking (M. A. Schuckit, 1985). In the 1990s, the typology evolved to also incorporate information about a past history of depression during abstinence (M A Schuckit et al., 1997). The term independent was used for depression that began before the onset of alcohol dependence or during sustained abstinence while depressive syndromes occurring only during a period of active alcohol dependence were labelled substance-induced (M A Schuckit, et al., 1997). Structured clinical assessment tools such as the Structured Clinical Interview for DSM (SCID) (Spitzer et al., 1992) and the Psychiatric Research Interview for Substance and Mental Disorders (PRISM) (Deborah S Hasin et al., 1996) have helped to operationalise these definitions and introduce greater diagnostic reliability.
1.2. Implications of subtypes
It has been argued that the distinction between independent depression (ID) and substance-induced depression (SID) has important implications for treatment and
4
prognosis: SID is considered a self-limiting condition that remits with abstinence while ID requires specific depression treatment (Pettinati, 2013; M. A. Schuckit, 1985; M A Schuckit, et al., 1997). In particular the view has been put forward that ID, but not SID, responds to antidepressant medication (M. A. Schuckit, 2006) but the evidence for this has not been systematically evaluated.
1.3. Aims of the review and meta-analysis
This review aimed to answer the following questions: i) In depressed patients with an alcohol use disorder, how much do depressive symptoms improve during treatment for depression? ii) Do patients with ID and SID have different patterns of treatment response? iii) Does antidepressant efficacy differ between patients with ID and SID?
2. METHODS 2.1. Selection Criteria
Studies were chosen according to the following criteria: 1. Studies reported longitudinal data on alcohol use and depression in adults over at least 8 weeks. 2. Change in mean score on a validated depression scale was reported. 3. Subjects had an active alcohol use disorder (alcohol dependence or alcohol abuse) diagnosed according to DSM or ICD criteria. 4. Mean baseline score ≥ 10 on the 17-item Hamilton Depression Rating Scale (HDRS) (Hamilton, 1960), or equivalent severity on another depression rating scale. Given the uncertain nosology of depression in alcohol use disorders, subjects were not required to have a diagnosis of a depressive disorder. The low severity threshold was chosen to avoid excluding studies in which baseline measures were taken after a short period of abstinence. A published score comparison table available at http://www.ids-qids.org/index2.html (University of Pittsburgh, 2014) was used to calibrate severity across scales.
5
5. Treatment for depressive symptoms was provided. A liberal definition of such treatment was applied, and it could include any form of medication or psychosocial treatment with plausible efficacy in treating depressive symptoms.
Unpublished and incomplete studies were not assessed for inclusion. Studies were excluded if they were conducted in special sub-populations (for example, prisoners) or those with another psychiatric or substance use disorder (for example, borderline personality disorder or nicotine dependence). Studies exclusively comprised of subjects with dysthymic disorder were also excluded as this condition has a different natural history compared to other depressive syndromes (Judd et al., 2002).
2.2. Information sources
Information sources were MEDLINE, Embase and Cochrane databases, searched on 9 November 2014. The reference lists from review articles (Amanda L. Baker, 2012; Nunes & Levin, 2004; Ostacher, 2007; Pettinati, 2004; Salloum & Jones, 2008; Sullivan et al., 2005) were used to hand-search for additional articles. Search items were alcohol drinking; alcohol-induced disorder; alcohol-related disorder; alcoholics; alcoholism AND depression; antidepressive agents. English-language publications on clinical trials from 1980 onwards were considered.
2.3. Study selection
The PRISMA diagram in Figure 1 illustrates the flow of studies during the selection process.
INSERT FIGURE 1 ABOUT HERE
Authors 1 (JF) and 2 (SA) independently screened titles and abstracts to exclude obviously irrelevant articles. Duplicate citations were deleted. The full text of articles in the screened sample was extracted and reviewed by authors 1 and 2. Studies meeting inclusion criteria were then compiled into the final sample by consensus.
6
2.4. Data items and summary measures
For analyses on the change in depression during treatment, results for each treatment arm of studies were aggregated. The mean depression score at baseline (taken as the earliest reported timepoint) and study completion, and their respective standard deviations were recorded. Study completion was chosen as the endpoint rather than a fixed time interval because studies did not report on outcomes at the same time points, and most change was expected to occur early in treatment (Amanda L Baker et al., 2014). The main summary measure was the standardised mean change in depression score from baseline to study completion, using the pooled standard deviation of baseline and follow up scores.
In two studies (Hernandez-Avila et al., 2004; Moak et al., 2003) the baseline standard deviation was very low, probably due to scores clustering just above the inclusion threshold. To avoid introducing bias, the baseline standard deviations were not used for these two studies and the follow-up standard deviations alone were used instead.
Depression type was defined according to whether studies were in subjects with ID, SID or a mixture of both types. Information on demographics, duration and adjunctive treatment was also extracted as shown in Table 1.
2.5. Assessment of bias and quality
Individual study bias was addressed by recording follow-up completion rates, and including in the meta-analysis only studies where reported outcomes were clearly from the intention-to-treat sample. External funding sources were noted. Study quality in randomized trials was assessed according to the adequacy of sequence generation, allocation concealment, blinding and completeness of reporting (Higgins & Green, 2008; Jadad et al., 1996). For observational studies, a quality checklist was used (Tooth et al., 2005). A funnel plot screened for publication bias.
2.6. Synthesis of results
7
A pooled estimate of the standardised mean change in depression score from baseline to follow-up was derived using a random effects model fitted with RevMan software (Cochrane Collaboration, 2008). This approach uses the pooled standard deviation of baseline and follow-up scores as the denominator of the standardized mean difference calculation, but ignores the correlation between these scores. However it is likely baseline and follow-up scores are in fact positively correlated, and ignoring this correlation can bias the estimation of effect sizes (Dunlap et al., 1996). Therefore a sensitivity analysis was performed to test the effect of correlated scores on the pooled effect size estimate. A plausible correlation figure of 0.4 was imputed, based on raw clinical trial data from a study conducted in our research unit (Adamson et al., 2015). The sensitivity analyses were performed with SAS v9.2 PROC MIXED (SAS Institute, Cary NC, USA) employing a method for random effects meta-analysis outlined by Sheu and Suzuki (2001). Variance estimates for each individual study outcome were obtained by the following formula (Dunlap, et al., 1996):
i.
Var (d) = [2(1 - r)/n] + [d2/(2n - 2)]
Where d is the standardized mean change for each study, r is the correlation between baseline and follow up scores within subjects and n is the sample size. Sensitivity analyses also investigated the influence of individual studies on the pooled effect size estimate.
Exploration of moderators of change in depression during treatment was then performed as a meta-regression analysis using PROC MIXED. Due to the small number of studies, moderator analyses were limited to baseline HDRS score, depression type, gender composition of the sample and manualisation of adjunctive psychological treatment.
2.7. Antidepressant efficacy
For the studies that compared an antidepressant to placebo, a pooled estimate of the treatment effect was derived using a random effects model in RevMan. In this analysis, the mean difference in change scores between antidepressant and placebo groups was calculated. This was then standardised using the pooled standard deviation 8
of follow-up scores. If follow-up standard deviations were not quoted, the standardised mean difference was derived from the F statistic from ANCOVA models. This approach to analysing change scores followed guidelines outlined in the Cochrane Handbook for Systematic Reviews of Interventions, page 269-270 (Higgins & Green, 2008) and it maintained consistency with a previous meta-analysis in this population (Nunes & Levin, 2004). A subgroup analysis then compared the pooled effect size between ID and SID / mixed studies. Sensitivity analyses investigated the influence of individual studies on the pooled effect size estimate.
2.8 Psychological treatments A brief narrative review of psychological treatment studies was performed. Due to the small number of randomised trials on psychological treatments that met inclusion criteria, no attempt was made to summarise results in a meta-analysis.
2.9 Predictors of treatment outcome In addition to the meta-regression analyses described above, a narrative review of predictors of treatment outcome at a within-study level was performed. This section focused on gender and alcohol use, as these predictors were most commonly studied and reported.
3. RESULTS 3.1. Search results and quality of studies
The search yielded 22 individual studies included in the systematic review. Table 1 presents a summary of these studies.
INSERT TABLE 1 ABOUT HERE
Twenty studies were randomized trials, of which eighteen were pharmacotherapy studies and two investigated non-pharmacological treatments. The remaining two studies were open label pharmacotherapy trials.
9
Of the 22 studies, 16 studies clearly presented data at baseline and follow-up for the intention-to-treat sample. Follow-up completion rates in these studies ranged from 48 to 93% (median 70%). The remaining 6 studies (Altintoprak et al., 2008; Dorus et al., 1989; Janiri et al., 1996; Mason et al., 1996; Paparrigopoulos et al., 2010; Petrakis et al., 2007) were excluded from the meta-analysis because their reported outcomes were not clearly identified as relating to the intention-to-treat sample.
The quality of randomized trials selected for the meta-analysis sample was good as judged by the Jadad quality score (range 3-5, median 4). For the two open-label studies, quality was acceptable to good.
3.2. Defining and measuring depression in subjects with an alcohol use disorder
Depression and alcohol use disorders were diagnosed with the Structured Clinical Interview for DSM (SCID) (Spitzer, et al., 1992) in fourteen studies; the Psychiatric Research Interview for Substance and Mental Disorders (PRISM) (Deborah S Hasin, et al., 1996) in two studies; the Schedule for Clinical Assessment in Neuropsychiatry (SCAN) (Wing et al., 1990) in one study; and the Diagnostic Interview Schedule for DSM (Robins et al., 1981) in one study. The remaining studies used unstructured or semi-structured interviews.
Sixteen studies, all pharmacotherapy trials, used versions of the Hamilton Depression Rating Scale (HDRS) (Hamilton, 1960) as a primary outcome. The standard 17-item version was used in 10/16 studies, the 24-item version in 3/16, the 21-item version in 2/16 and the remaining study did not state the version used. Three studies used the Beck Depression Inventory (BDI), (Beck et al., 1961). Two studies used the Montgomery-Asberg Depression Rating Scale (MADRS), (Montgomery & Asberg, 1979) and the remaining study used the QIDS-SR (Rush et al., 2003).
As shown in Table 1, there was variation between studies in the timing of baseline depression measures and the inclusion or exclusion of patients who responded during the initial assessment phase of the study. Most studies had a mean baseline depression 10
score in the mild to moderate depression range (Zimmerman et al., 2013); for example, among the studies using the 17-item HDRS, the score at baseline ranged from 12.6 to 23.9 (median 19.6).
Operational definitions of depression types varied between studies, and these are summarised in Table 1. For simplicity, the terms independent and substance-induced are used in the present review while studies on primary and secondary depression are grouped with those on ID and SID respectively.
Eleven studies determined depression type at baseline. Of these, five studies included only subjects with ID (Cornelius et al., 1997; McGrath et al., 1996; Moak, et al., 2003; Pettinati et al., 2010; Roy-Byrne et al., 2000). Three studies (Adamson, et al., 2015; R. A. Brown et al., 2011; Oslin, 2005) explicitly identified whether subjects had ID or SID, and included both types. The remaining three studies were on subjects with SID (Gual et al., 2003; Mason, et al., 1996; Paparrigopoulos, et al., 2010).
3.3. Change in depression during treatment
A meta-analysis was initially planned on the 16 studies that reported on change in depression from baseline to study completion for an intention-to-treat sample. However there was very high heterogeneity in this sample (Q=152.21, df=15, p
