Efficacy and Safety of SSRIs and SNRis in Older Adults
Randomized controlled trials (RCTs) comparing SSRIs or SNRIs with placebo or another active antidepressant were included. The eligible population included individuals aged 60 and older with a primary diagnosis of MDD. RCTs had to assess the active interventions for at least 6 weeks.
In consultation with a medical librarian, a comprehensive systematic literature search was conducted in the following databases: Medline, EMBASE, Cochrane Central Register of Controlled Trials, PsycINFO, and Web of Science (from inception to December 2013). This search used terms related to each intervention of interest and terms reflecting depressive disorders. Manual searches of clinicaltrials.gov and conference proceedings of major psychiatric conferences for the past 2 years were also performed to identify RCTs that have not been published in a full manuscript but are potentially eligible for inclusion. Two investigators (ED, PW) independently conducted the searches.
Two investigators (ED, PW) independently scanned all abstracts and proceedings identified in the literature search and reviewed potentially relevant abstracts and proceedings in full text. If any discrepancies occurred between the studies that the two investigators selected, a third investigator (KT) provided arbitration.
Efficacy and safety outcomes were assessed. For efficacy, partial response to treatment (defined as ≥50% reduction in Hamilton Depression Rating Scale (HDRS) score or Montgomery–Asberg Depression Rating Scale (MADRS) score from baseline) was considered. Preference was given to the HDRS over MADRS because it is most universally reported in clinical trials of antidepressants. As such, if a trial reported HDRS and MADRS, the HDRS was taken. Specific safety outcomes included dizziness, syncope, vertigo, loss of consciousness, and falls. These safety outcomes were chosen because they are related to treatment tolerability and can be linked to more-serious side effects.
Two reviewers (ED, PW) independently extracted all data and recorded it in a spreadsheet. A third reviewer (KT) checked all data extraction for consistency. Clinical data were extracted for each trial on number of centers, region(s) of origin, dose and duration of treatment, duration of follow-up, number of major adverse events reported, criteria used to diagnose depression, proportion of women, proportion of individuals of various races, and proportion of individuals excluded from analysis (e.g., in modified intention-to-treat analysis). Data were extracted per intention-to-treat analyses.
This was a network meta-analysis, in which interest centers on the comparison of the treatment effects of interventions that are not studied in a head-to-head fashion. Geometric networks were first plotted, making direct comparisons between treatments and control, for each outcome. Second, consistency between the direct and indirect comparisons evaluated for networks that consisted of closed loops was assessed. For each of the comparisons that were part of a closed loop made up of more than one RCT, the available trials were split into direct and indirect information. For each comparison in question, two (pooled) relative treatment effect estimates were obtained, one with independent-means models using only the trials providing direct comparisons and one based on a network meta-analysis of the remaining trials providing only indirect evidence. The difference in estimates that the two sets of evidence generated was evaluated using the Bucher test for inconsistency.
The network meta-analysis was conducted within a Bayesian framework. Analyses within the Bayesian framework involve data, a likelihood distribution, a model with parameters, and prior distributions. The model compares the data from the individual trials to basic parameters reflecting the (pooled) relative treatment effect of each intervention with an overall reference treatment. Based on these basic parameters, the relative efficacy or safety of each of the competing interventions is obtained.
Because data are binary, the statistical model followed a logistic regression setup under which comparative odds ratios and associated 95% credible intervals (CrIs, the Bayesian equivalent to conventional 95% confidence intervals) were obtained. Under the Markov Chain Monte Carlo sampling for the Bayesian model, odds ratios were transformed into relative risks (RRs) with 95% (CrIs) using the average control group response from the included placebo control trials. Because there was one trial per comparison for all but one comparison, estimation of heterogeneity was not possible, so a fixed-effect model was used. In order not to influence the observed results by the prior distribution, noninformative prior distributions were used for all analyses. All Bayesian network meta-analyses were carried out in WinBUGS version 1.4.3 (Cambridge Institute of Public Health, Cambridge, UK).
Methods
Eligibility Criteria
Randomized controlled trials (RCTs) comparing SSRIs or SNRIs with placebo or another active antidepressant were included. The eligible population included individuals aged 60 and older with a primary diagnosis of MDD. RCTs had to assess the active interventions for at least 6 weeks.
Search Strategy
In consultation with a medical librarian, a comprehensive systematic literature search was conducted in the following databases: Medline, EMBASE, Cochrane Central Register of Controlled Trials, PsycINFO, and Web of Science (from inception to December 2013). This search used terms related to each intervention of interest and terms reflecting depressive disorders. Manual searches of clinicaltrials.gov and conference proceedings of major psychiatric conferences for the past 2 years were also performed to identify RCTs that have not been published in a full manuscript but are potentially eligible for inclusion. Two investigators (ED, PW) independently conducted the searches.
Study Selection
Two investigators (ED, PW) independently scanned all abstracts and proceedings identified in the literature search and reviewed potentially relevant abstracts and proceedings in full text. If any discrepancies occurred between the studies that the two investigators selected, a third investigator (KT) provided arbitration.
Outcomes
Efficacy and safety outcomes were assessed. For efficacy, partial response to treatment (defined as ≥50% reduction in Hamilton Depression Rating Scale (HDRS) score or Montgomery–Asberg Depression Rating Scale (MADRS) score from baseline) was considered. Preference was given to the HDRS over MADRS because it is most universally reported in clinical trials of antidepressants. As such, if a trial reported HDRS and MADRS, the HDRS was taken. Specific safety outcomes included dizziness, syncope, vertigo, loss of consciousness, and falls. These safety outcomes were chosen because they are related to treatment tolerability and can be linked to more-serious side effects.
Data Extraction
Two reviewers (ED, PW) independently extracted all data and recorded it in a spreadsheet. A third reviewer (KT) checked all data extraction for consistency. Clinical data were extracted for each trial on number of centers, region(s) of origin, dose and duration of treatment, duration of follow-up, number of major adverse events reported, criteria used to diagnose depression, proportion of women, proportion of individuals of various races, and proportion of individuals excluded from analysis (e.g., in modified intention-to-treat analysis). Data were extracted per intention-to-treat analyses.
Analysis
This was a network meta-analysis, in which interest centers on the comparison of the treatment effects of interventions that are not studied in a head-to-head fashion. Geometric networks were first plotted, making direct comparisons between treatments and control, for each outcome. Second, consistency between the direct and indirect comparisons evaluated for networks that consisted of closed loops was assessed. For each of the comparisons that were part of a closed loop made up of more than one RCT, the available trials were split into direct and indirect information. For each comparison in question, two (pooled) relative treatment effect estimates were obtained, one with independent-means models using only the trials providing direct comparisons and one based on a network meta-analysis of the remaining trials providing only indirect evidence. The difference in estimates that the two sets of evidence generated was evaluated using the Bucher test for inconsistency.
The network meta-analysis was conducted within a Bayesian framework. Analyses within the Bayesian framework involve data, a likelihood distribution, a model with parameters, and prior distributions. The model compares the data from the individual trials to basic parameters reflecting the (pooled) relative treatment effect of each intervention with an overall reference treatment. Based on these basic parameters, the relative efficacy or safety of each of the competing interventions is obtained.
Because data are binary, the statistical model followed a logistic regression setup under which comparative odds ratios and associated 95% credible intervals (CrIs, the Bayesian equivalent to conventional 95% confidence intervals) were obtained. Under the Markov Chain Monte Carlo sampling for the Bayesian model, odds ratios were transformed into relative risks (RRs) with 95% (CrIs) using the average control group response from the included placebo control trials. Because there was one trial per comparison for all but one comparison, estimation of heterogeneity was not possible, so a fixed-effect model was used. In order not to influence the observed results by the prior distribution, noninformative prior distributions were used for all analyses. All Bayesian network meta-analyses were carried out in WinBUGS version 1.4.3 (Cambridge Institute of Public Health, Cambridge, UK).
SHARE
