The efficacy of antidepressant medication has been shown empirically to be overestimated due to publication bias, but this has only been inferred statistically with regard to psychological treatment for depression. We assessed directly the extent of study publication bias in trials examining the efficacy of psychological treatment for depression.
Methods and Findings
We identified US National Institutes of Health grants awarded to fund randomized clinical trials comparing psychological treatment to control conditions or other treatments in patients diagnosed with major depressive disorder for the period 1972–2008, and we determined whether those grants led to publications. For studies that were not published, data were requested from investigators and included in the meta-analyses. Thirteen (23.6%) of the 55 funded grants that began trials did not result in publications, and two others never started. Among comparisons to control conditions, adding unpublished studies (Hedges’ g = 0.20; CI95% -0.11~0.51; k = 6) to published studies (g = 0.52; 0.37~0.68; k = 20) reduced the psychotherapy effect size point estimate (g = 0.39; 0.08~0.70) by 25%. Moreover, these findings may overestimate the "true" effect of psychological treatment for depression as outcome reporting bias could not be examined quantitatively.
The efficacy of psychological interventions for depression has been overestimated in the published literature, just as it has been for pharmacotherapy. Both are efficacious but not to the extent that the published literature would suggest. Funding agencies and journals should archive both original protocols and raw data from treatment trials to allow the detection and correction of outcome reporting bias. Clinicians, guidelines developers, and decision makers should be aware that the published literature overestimates the effects of the predominant treatments for depression.
Citation: Driessen E, Hollon SD, Bockting CLH, Cuijpers P, Turner EH (2015) Does Publication Bias Inflate the Apparent Efficacy of Psychological Treatment for Major Depressive Disorder? A Systematic Review and Meta-Analysis of US National Institutes of Health-Funded Trials. PLoS ONE 10(9): e0137864. https://doi.org/10.1371/journal.pone.0137864
Editor: Lin Lu, Peking University, CHINA
Received: February 17, 2015; Accepted: August 22, 2015; Published: September 30, 2015
Copyright: © 2015 Driessen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited
Data Availability: All relevant data from published studies and a number of the unpublished studies are within the paper and its Supporting Information files. We cannot make the data from all unpublished studies publicly available as we obtained it this data from third parties (the original investigators of these studies), who did not provide us permission to do so. However, others can request this data from the relevant investigators. Contact information can be retrieved from "http://projectreporter.nih.gov/reporter.cfm" by entering the relevant grant number listed in Table 1 of this manuscript.
Funding: The authors have no support or funding to report.
Competing interests: Claudi L. H. Bockting is an academic editor for PLOS ONE. This does not alter the authors' adherence to PLOS ONE Editorial policies and criteria. The other authors have declared that no competing interests exist.
Publication bias has been defined as the tendency for authors to submit, or journals to accept, manuscripts for publication based on the direction or strength of the study’s findings . It has been long recognized that publication bias can lead to an overestimation of treatment effects  that threatens the validity of evidence-based decisions.
Major depression is a highly prevalent and disabling disorder associated with major personal and societal costs [3–5]. It is the fourth leading cause of disease burden worldwide and is expected to rank first in high-income countries by the year 2030 . Antidepressant medication is recommended as a first-line treatment for major depressive disorder in most treatment guidelines [7–8] and the majority of depressed patients are now so treated in primary care . However, the efficacy of pharmacotherapy for depression has been overestimated due to publication bias. Turner and colleagues  found that publication was strongly linked to outcome in 74 placebo-controlled antidepressant studies submitted to the US Food and Drug Administration (FDA). Among 38 studies judged positive by the FDA, all but one were published in a way that agreed with those judgments. By contrast, among 36 studies judged negative or questionable by the FDA, all but 3 were subjected to publication bias in one of two forms: (1) the studies were not published (study publication bias, 61%), or (2) negative results were “spun” to make them look positive (outcome reporting bias, 31%) through selective reporting of sites, subjects (especially dropouts), and measures. Comparing the published literature with the original FDA data resulted in a 24% reduction in pooled mean effect size for pharmacotherapy versus pill-placebo, from 0.41 (CI95% 0.36~0.45) to 0.31 (0.27~0.35).
The above-mentioned treatment guidelines also recommend the use of a depression-focused psychotherapy as an alternative to medications for patients with mild to moderate major depressive disorder [7–8]. However, one might ask whether the effects of psychological treatments for depression could also be overestimated due to publication bias. Cuijpers and colleagues  examined 117 published trials, including 175 comparisons between psychological treatments and control conditions, and using statistical procedures [12–13], found evidence for 51 “missing” studies; when their results were imputed, the overall effect size was reduced by 37%, from 0.67 (CI95% 0.60~0.75) to 0.42 (0.33~0.51).
This suggests that psychological treatment, like pharmacologic treatment, may not be as efficacious as the published literature would indicate. However, the statistical procedures used by Cuijpers et al.  relied on the assumption that any asymmetry in the funnel plots, which depict the association between sample size and effect size in meta-analyses, stems from a failure to publish small studies with small effect sizes. Nonetheless, small studies may show disproportionately large effects for reasons other than publication bias [14–15]. Thus, as Borenstein and colleagues  state: “the only true test for publication bias is to compare effects in the published studies formally with effects in the unpublished studies” (p. 280).
In the present study, we sought to estimate the “true” effect of study publication bias by examining a cohort of trials funded by the US National Institutes of Health (NIH). The NIH is an agency of the US Department of Health and Human Services and is the primary agency of the United States government responsible for biomedical and health-related research, in which it invests over $30.9 billion annually. The majority of NIH funding is awarded through competitive grants to researchers at universities, medical schools, and other research institutions in the US and abroad. We identified grants awarded for randomized clinical trials examining the efficacy of psychological treatments, compared to control or alternative interventions, in adult or geriatric patients with major depressive disorder. We ascertained the frequency with which these studies were conducted but not published. In addition, we sought to better estimate the effect of psychological treatment on major depressive disorder by pooling unpublished with published findings.
No review protocol exists beyond what is described in our methods section.
Defining the set of relevant funded grants
We searched NIH databases (see below) for grants, funding randomized clinical trials testing the effects of psychological treatments on major depressive disorder for the years 1972 through May 2008. NIH records do not go back before 1972; and the cutoff date was established in order to allow at least 60 months (5 years) for publication of trial results before our search started in June 2013. This duration was derived from a study of time to publication among 635 NIH-funded clinical trials, which showed that about 95% of to-be-published trials were published within 60 months of trial completion .
We used http://crisp.cit.nih.gov/crisp_query.generate_screen to search for relevant grants through 1984 and http://projectreporter.nih.gov/reporter.cfm for trials completed from 1985 on. These databases provide summary information on the grants and often include abstracts. We searched using all possible combinations of terms for (1) “depression” (depression, depressive, major depressive disorder, mood disorder, affective disorder, melancholic, melancholia) and (2) “psychological treatment” (cognitive therapy, behavior therapy, behavioral therapy, interpersonal therapy, psychodynamic therapy, dynamic therapy, humanistic therapy, therapy, supportive therapy, experiential therapy, [self-] control therapy, [problem-] solving therapy, [supportive-] expressive therapy, family therapy, group therapy, marital therapy, couples therapy, aversive therapy, exposure therapy, psychotherapy, psychotherapies, psychotherapeutic, counseling, disease management, psychoanalytic, behavioral activation, cognitive behavioral analysis system, desensitization, relaxation techniques, and progressive muscle relaxation). We deleted duplicate listings and multiple years for the same grants.
We then screened the grant titles and available abstracts for possible inclusion. We followed the methods used by Turner and colleagues  as closely as possible in order to facilitate comparison of our results to those with pharmacological trials. We included all grants that proposed to conduct (1) a randomized clinical trial examining (2) psychological treatment for (3) acute depression in (4) adults. We used measures of depressive symptoms as the sole outcome for this meta-analysis. Following Cuijpers and colleagues , we defined psychological treatment as any intervention in which verbal communication between therapist and client was the core element. We also included bibliotherapy if it was supplemented with personal support from a therapist. We included treatments aimed at decreasing depressive symptoms and excluded treatments aimed solely at increasing adherence to antidepressant medication. We only included studies in which the differential efficacy of psychological treatment could be ascertained. Thus, we included studies that examined the efficacy of psychological treatment combined with antidepressant medication versus that of medication alone, but we excluded studies in which all patients received the same psychological treatment and studies in which not all patients in a condition received psychological treatment.
Depression was defined as meeting diagnostic criteria for major depressive disorder (MDD) or its equivalent in an earlier nosology (e.g., Research Diagnostic Criteria  or Feighner criteria ). We included studies that sampled patients who did not meet criteria for MDD only if outcomes were reported separately for those who did (since those were the patients included in our analyses). We excluded studies on children and adolescents younger than 18, but we included studies with geriatric patients since more recent studies of adults and elderly patients have found that psychological treatment efficacy does not differ with age .
Matching grants to published articles
We consulted www.evidencebasedpsychotherapies.org, a publicly available database of 352 randomized clinical trials regarding psychological treatment for adult depression , in order to identify published articles. This database is updated annually through comprehensive literature searches in Pubmed, PsycINFO, Embase and the Cochrane Central Register of Controlled Trials. We consider this database as defining the published literature on randomized clinical trials for the psychological treatment of depression.
We used this database as one source for matching published articles to the grants. We matched a published article to a grant based on the grant number acknowledged in the paper, making sure that it also matched with respect to investigator, type of psychological treatment, and comparator. When we could not find a matching publication within the database, we searched PubMed for all articles published by the principal investigator. If we still could not find a matching publication, we contacted the investigator to ask whether those findings had ever been published.
We included articles that reported outcomes for the short-term treatment of depression; we excluded articles reporting the prevention of subsequent relapse (continuation treatment) or recurrence (maintenance treatment). When multiple sequential articles were found reporting acute response on the same outcomes, we selected the one that was most complete in terms of sample size. We excluded articles that reported secondary analyses in favor of articles reporting primary data.
In order to ensure that we had not missed relevant grants with our grant search strategy, we checked all the published studies in the database just described to see if they acknowledged NIH grant funding. Whenever we found an article that acknowledged a grant that was not in our list, we resolved the discrepancy, if necessary by contacting the principle investigator. All published articles matched to the NIH grants were read independently by two authors to ensure that they met inclusion criteria. Disagreements were resolved by consensus.
Assessment of study publication bias
We calculated: (1) the proportion of funded grants not resulting in publication and (2) the proportion of funded grants resulting in publication but not reporting data sufficient for meta-analysis. These proportions provide a direct assessment of the extent of study publication bias. In cases of non-publication, we contacted the investigators to request the unpublished data and to ask why they had not been published.
Two raters independently extracted data. Boolean formulas in Excel were used to compare and flag any mismatches between the two sets of data extractions. Discrepancies were resolved by consensus after checking the primary data. We calculated post-treatment effect sizes for four different types of comparisons: (1) psychological treatment versus control conditions, (2) psychological treatment versus another psychological treatment, (3) psychological treatment versus antidepressant medication, and (4) psychological treatment combined with antidepressant medication versus medication alone. We distinguished between no-treatment control conditions and control conditions in which some kind of treatment was provided; we subclassified treatment control conditions into (a) treatment-as-usual, (b) pill-placebo, and (c) psychological-placebo conditions intended to control for time and contact (nonspecific controls).
Within each study, effect size was calculated by subtracting the mean post-treatment score for one condition from the mean for the other, then dividing the difference by the pooled standard deviation, with positive signs indicating superiority of psychological treatment. For studies including multiple psychological treatment conditions, we ranked the different psychological treatment conditions in terms of the investigators’ presumed expected efficacy; effect sizes were then calculated in such a way that positive signs indicated superiority of the psychological treatment that we presumed the investigators expected to be more efficacious. This procedure is described in detail below.
If means and standard deviations were not reported, we used other data (e.g., event rates, reported effect sizes, t-statistics, or change scores, in that order) to compute the effect size. When no such data were available, the study was classified as not reporting sufficient data for meta-analysis. We calculated effect sizes as Hedges’ g, because it provides a less biased estimate than Cohen's d when sample sizes are small (, page 27). All meta-analyses were conducted using random effects models in Comprehensive Meta-analysis (version 2.2.064 ).
We used depressive severity as the sole outcome measure for this meta-analysis. Most commonly, depressive severity was measured with the Hamilton Depression Rating Scale  or the Beck Depression Inventory . When a study’s results were reported separately for multiple outcome measures, a mean effect size was calculated and this synthetic score was then included in the meta-analysis (, pages 226–233). When a study’s results were reported for independent subgroups, we performed a fixed-effects meta-analysis on the subgroups to obtain a composite study-level effect size. These study-level effect sizes were then used in the meta-analysis (, pages 219–221). When a study included multiple versions of the same comparison (e.g., psychological treatment 1 versus waitlist control condition and psychological treatment 2 versus waitlist control condition), the mean effect size was calculated and this combined effect size for the study was included in the meta-analysis (, pages 239–241). Heterogeneity was examined using the Q and I2 statistics. For the latter, we calculated the I2 95% confidence intervals  using the heterogi module  within Stata (Stata Statistical Software: Release 11), with the non-central chi-squared-based option .
In order to contrast the mean effect of the published studies with the mean effect of the unpublished studies, we conducted subgroup analysis testing using Comprehensive Meta-analysis (version 2.2.064 ). This employs a fully random effects analysis and pools study-to-study variance across subgroups (, pages 171–178), as recommended when subgroups involve small numbers of studies (, page 163). The pooled mean effect sizes were contrasted by means of a Q-test based on analysis of variance (, pages 177–178). If the p-value for the Q-between value (the weighted sum of squares of the subgroup means around the grand mean) is <.05, the pooled mean effect sizes of the two subgroups differ significantly, suggesting that effect size depends on publication status. Finally, we calculated the overall pooled mean effect size by combining published and unpublished studies. We calculated the change between this overall pooled mean effect size point estimate and the pooled mean effect size point estimate for published studies only (both the raw difference in Hedges’ g and the percentage change) using 5 decimals.
Assigning ranks to comparisons of two different psychological treatments
For studies including multiple psychological treatment conditions, we ranked the different psychological treatment conditions in terms of the investigators’ presumed expected efficacy. Unlike other comparisons, such as psychological treatment versus control conditions (which can be assigned a positive sign if the psychological treatment is superior to the control condition and a negative sign if the control condition is superior to the psychological treatment), comparisons between different psychological treatments have no natural order. (For example, a hypothetical effect size comparing cognitive therapy with interpersonal therapy could be either -0.20 or +0.20.) Strategies used in earlier quantitative reviews to handle this matter include making all signs positive, which overestimates the pooled mean effect size, or assigning signs at random, which artificially biases the pooled mean effect size toward zero . Neither approach was suitable for our purposes. Publication bias might be more likely if the treatment the investigators hypothesized to be more efficacious proved to be inferior to the treatment hypothesized to be less efficacious. Therefore, positive effect size values indicate superiority of the psychological treatment that the investigators expected, in our opinion (see below), to be more efficacious, while negative signs indicated superiority of the psychological treatment that the investigators expected to be less efficacious. If publication bias were present in comparisons of different psychological treatments, we hoped to be able to show it by this method, since we maximized the difference in effect sizes between published comparisons (likely positive since the psychological treatment expected to be more efficacious was superior) and unpublished comparisons (likely negative since the psychological treatment expected to be more efficacious was inferior).
We based our rankings of expected efficacy among the treatments on explicit statements to that effect whenever possible. If such statements were not evident, we based our rankings on the order of presentation of the respective treatments in the title or method section (in that order). In the case of dismantling studies (studies seeking to isolate separate components of a larger psychological treatment package), conditions including all components were ranked highest in terms of expected efficacy, while conditions with fewer components were ranker lower. Two authors judged presumed expected efficacy independently and disagreements were resolved by consensus. When consensus could not be reached, sensitivity analyses were conducted for each of the different rankings.
Quality assessment of the included studies
The following four criteria of the Cochrane ‘risk of bias’ assessment tool were used to assess the quality of the included studies:
- Intention-to-treat analysis (ITT): The study reports that ITT analyses have been conducted (whether or not the reported data used for effect size calculations include the total population).
- Blind assessment of outcome: The outcome assessors were blinded and did not know to which condition the respondents were assigned. If only self-report measures were used to examine the effects, the study was coded as “no blind assessment”.
- Adequate sequence generation: The investigators describe a random component in the sequence generation process.
- Independent randomization: It is reported that randomization to conditions was conducted by an independent (third) party, with options including an independent person who conducted randomization or when a computer program or sealed envelopes were used to assign patients to conditions.
Unless the published article explicitly reported that a specific criterion had been met, we scored it as if it had not. We know from our communications with the investigators that these criteria were sometimes followed but not reported (on the assumption that they were standard methods that did not need to be made explicit) but we wanted to be consistent with other reviews that might be conducted solely on the basis of the published articles. Unpublished papers were rated from draft manuscripts whenever possible. When no drafts were available, the principal investigators of the unpublished studies were asked for quality criteria by email. Two raters independently rated these criteria, and disagreements were resolved by consensus.
As shown in Fig 1, 4073 grants were identified, of which 3841 were excluded based on titles and abstracts as not meeting inclusion criteria or as consecutive years of the same grant. Of the remaining 232 grants, 176 were excluded after a review of the published article or contacting the investigator, most often because the study population did not meet diagnostic criteria for MDD (n = 61). This left 56 grants that met our inclusion criteria.
Note: MDD = major depressive disorder; NIH = US National Institutes of Health; RCT = randomized clinical trial.
We examined the completeness of our grant search strategy by checking all the published studies in the above-mentioned database for NIH grant acknowledgements. This yielded 38 articles that acknowledged NIH support but were not discovered in our search of the NIH grants database. In 13 of these 38 articles, the study participants were not selected on the basis of MDD; in another 5 studies all patients received the same psychological treatment. Three of the publications acknowledged NIH support that was not directly used to fund the study. One study was funded by a grant from the military and fifteen studies were funded by NIH grants awarded outside our cut-off years. In all, only one of the above-mentioned 38 articles met our inclusion criteria. That article described the results of an acute treatment trial  that had not been mentioned in the abstract of the NIH grant that provided its support; the original award listed only a maintenance trial and thus fell outside of the criteria for our database search. Because the acute trial met our inclusion criteria, we included the grant in our dataset, bringing the total number to 57.
Extent of study publication bias
The 57 grants awarded are described in Table 1. Multiple grants that funded the same study are counted as a single grant, as are each of two grants that generated multiple published randomized clinical trials. Two other grant-funded studies were never started, one because of difficulty recruiting patients (S. Chisholm-Stockard, personal communication, March 3, 2011) and the other because of difficulty finding psychodynamic therapists willing to participate in clinical research with Hispanic elders (J. Szapocznik, personal communication, August 30, 2010). These last two grants were excluded from further consideration.
Of the 55 grants that started studies, we were able to locate published articles corresponding to 42 (76.4%) grants [28–73], but not for the other 13 (23.6%). (If the two studies that weren’t started had not been excluded from consideration, the proportion of grants that did and did not lead to publications would have been 73.7% (42/57) and 26.3% (15/57), respectively). These 13 grants met our definition of unpublished studies. We were able to obtain the original data from 11 of these studies (84,6%). With respect to the remaining studies, the twelfth investigator was not yet ready to share her data (R. Clark). The thirteenth investigator expressed willingness to share data, but the data had been collected over a quarter of a century earlier and had not been retained, though he recalled that the sample was small (no more than a dozen patients per condition) and that the differences were negligible (G. Gottlieb, personal communication, June 10, 2012). Therefore, we excluded this study from the main analyses, but conducted sensitivity analyses, including this study in the relevant comparisons (psychological treatment alone or in combination versus antidepressant medication monotherapy) and estimating the study’s effect size to be g = 0.00 with n = 10 per condition. The total number of participants over the 42 published studies (4581) and 11 unpublished studies for which we had data (839) was 5420.
All but one of the published studies reported outcome data sufficient for calculating effect size. The remaining study  reported the grant-funded trial’s outcomes only in aggregate with outcomes of other trials. We requested the data for the grant-funded trial from the principal investigator. We counted the grant as published, because it resulted in a publication, but also conducted a sensitivity analysis coding it as unpublished, which resulted in published and unpublished rates of 74.5% (41/55) and 25.5% (14/55), respectively.
Adjusting for study publication bias
Note: Not all results of the unpublished studies are presented at study level, because we did not have permission of the investigators to do so. BDI = Beck Depression Inventory; CTRL-NS = non-specific control condition (psychological placebo); CTRL-NS(PLAC) = pill-placebo control condition; CTRL-NT = no-treatment control condition; CTRL-TAU = treatment as usual control condition; HAMD = Hamilton Depression Rating Scale; PT = psychological treatment; SCL-90-D = Symptom Checklist—90 item, depression subscale.
Note: Not all results of the unpublished studies are presented at study level, because we did not have permission of the investigators to do so. CTRL-NT = no-treatment control condition; PT = psychological treatment; SCL-90-D = Symptom Checklist—90 item, depression subscale.
Note: Not all results of the unpublished studies are presented at study level, because we did not have permission of the investigators to do so. BDI = Beck Depression Inventory; CTRL-NS = non-specific control condition (psychological placebo); CTRL-NS(PLAC) = pill-placebo control condition; CTRL-TAU = treatment as usual control condition; HAMD = Hamilton Depression Rating Scale; PT = psychological treatment.
Note: Not all results of the unpublished studies are presented at study level, because we did not have permission of the investigators to do so. BDI = Beck Depression Inventory; HAMD = Hamilton Depression Rating Scale; PT = psychological treatment.
Note: Not all results of the unpublished studies are presented at study level, because we did not have permission of the investigators to do so. ADM = antidepressant medication; HAMD = Hamilton Depression Rating Scale; PT = psychological treatment.
Note: Not all results of the unpublished studies are presented at study level, because we did not have permission of the investigators to do so. ADM = antidepressant medication; HAMD = Hamilton Depression Rating Scale; IDS-C30 = Inventory of Depressive Symptomatology– 30-item Clinician Rated version; PT = psychological treatment.
Psychological treatment versus control conditions.
Among the studies that compared psychological treatments to control conditions (of any type), the overall pooled mean effect size was g = 0.20 (CI95% -0.11~0.51) in the 6 unpublished studies (n = 336) versus g = 0.52 (0.37~0.68) in the 20 published studies (n = 1767). Adding unpublished studies to published studies resulted in a 25% decrease in effect size point estimate to g = 0.39 (0.08~0.70).
Among the different types of control conditions, when psychological treatment was compared to a no-treatment control condition, the pooled mean effect size in the 2 unpublished studies was g = 0.77 (-0.07~1.61) versus g = 1.01 (0.63~1.40) in the 5 published studies. Adding unpublished studies to published studies resulted in a 4% decrease in effect size point estimate to g = 0.97 (0.62~1.32). Comparing psychological treatments to treatment control conditions resulted in a pooled mean effect size of g = 0.11 (-0.12~0.35) for the 4 unpublished studies and g = 0.37 (0.25~0.48) for the 15 published studies. Adding unpublished to published studies resulted in a 29% decrease in effect size point estimate to g = 0.26 (0.02~0.51). For all the above-mentioned comparisons, psychological treatments remained significantly more efficacious than the control conditions after the unpublished data were added to the published data.
Among the different types of treatment control conditions (treatment-as-usual, pill-placebo, and non-specific [psychological placebo]), the largest differential between published and unpublished findings was apparent for comparisons of psychological treatment to pill-placebo (S1 Table). The pooled mean effect size of psychological treatment relative to pill-placebo control condition was g = -0.09 (-0.59~0.40) in 1 unpublished study and g = 0.34 (0.14~0.53) in 5 published studies. Adding unpublished studies to published studies resulted in a 45% decrease in effect size point estimate to g = 0.19 (-0.21~0.59).
Psychological treatment versus other psychological treatments.
Two unpublished studies involving a total of 164 participants were found comparing different types of psychological treatments. The pooled mean effect size in these 2 studies was g = -0.05 (CI95% -0.49~0.38), while the pooled mean effect size in the 12 published studies totaling 617 participants was g = 0.17 (-0.04~0.38). Adding the unpublished studies to the published studies resulted in a 24% decrease in effect size point estimate to g = 0.13 (-0.06~0.31), indicating no significant differences between the psychological treatments that the investigators expected, in our opinion, to be most efficacious, and those expected to be less efficacious. We conducted a sensitivity analysis for one study intended to minimize allegiance effects . Changing the psychological treatment ranks did not alter the pattern of results (S1 Table). We also conducted a sensitivity analysis for the one study that did not report the data required to analyze the results in a meta-analysis . Counting this study as unpublished instead of published did not alter the pattern of results (S1 Table).
Psychological treatment versus antidepressant medication.
When psychological treatment was compared to antidepressants, the pooled mean effect size was g = -0.21 (CI95% -0.53~0.11) for 3 unpublished studies (n = 231) and g = 0.01 (-0.13~0.16) for 15 published studies (n = 1675). Adding unpublished to published studies resulted in a pooled mean effect size of g = -0.05 (-0.25~0.15), indicating no significant difference between psychological treatments and antidepressants. This pattern of results was not altered when we included the estimates from the unpublished study (Gottlieb) for which the original data were not available (S1 Table).
Combined psychological treatment and antidepressants versus antidepressant monotherapy.
The pooled mean effect size of combined treatment in which psychological treatment was added to antidepressant medication versus medication monotherapy was g = 0.37 (CI95% -0.14~0.89) for 2 unpublished studies (n = 85) and g = 0.22 (-0.00~0.44) for 9 published studies (n = 812). Adding unpublished to published studies resulted in an effect size point estimate of g = 0.24 (0.04~0.45), indicating that combined treatment was superior to antidepressant medications alone. Including the estimates from one unpublished study for which the original data were not available (Gottlieb) did not alter this result pattern (S1 Table).
Reasons for non-publication
Of the 13 unpublished studies, only two were submitted for review; neither was accepted for publication. Six other studies were never submitted for review, although, in three instances, the investigators still hoped to do so. For the remaining five studies, it was unclear whether the authors tried to submit their findings for publication. Explanations that the investigators gave for not submitting manuscripts included that they did not think the findings were interesting enough to warrant publication, that they got distracted by other obligations or that they had practical problems.
Quality of the included studies
We assessed the quality of the included studies (see Table 1) with regard to the four quality criteria. We were not able to retrieve quality ratings for three of the unpublished studies. Study quality was less than optimal; only 10 of the 53 (18.9%) studies were rated ‘low risk of bias’ on all four criteria. No significant difference was found in the proportion of such studies between the published (18.2%) and unpublished (22.2%) studies (Fisher’s Exact test, p = 1.00). Similarly, no significant difference was found between the published and unpublished studies with regard to the proportion of studies that reported intention-to-treat analyses (respectively, 63.6% and 55.6%, Fisher’s Exact test, p = .72) or that kept outcome assessors blind to treatment allocation (respectively, 56.8% and 55.6%, Fisher’s Exact test, p = 1.00). However, the number of studies that reported an adequate sequence generation evidenced a nonsignificant trend (38.6% and 77.8%, Fisher’s Exact test, p = .06) and the number of studies that reported that an independent party conducted the randomization was significantly smaller among the published studies (25.0%) than among the unpublished studies (77.8%; Fisher’s Exact test, p = .005). This pattern of findings did not change when we recoded the three unpublished studies with missing quality ratings as high risk of bias on all four criteria.
We also tested whether the mean number of participants per treatment condition differed between unpublished (31.9, SD = 22.2) and published (41.9, SD = 42.5) studies. Only about 75% as many patients were used per condition in the unpublished studies compared to the published studies, but this difference was not significant (t(54) = 0.79, p = .44).
We found clear indications of study publication bias in NIH-funded randomized clinical trials examining the efficacy of psychological treatment for major depressive disorder; 23.6% (13/55) of the grant-funded studies did not result in a publication. The binomial 95% confidence interval of this proportion (13%-37%) overlaps considerably with that of the 31.1% (23/74; binomial confidence interval 21%-43%) non-publication rate reported by Turner and colleagues for industry-funded trials of antidepressant medications .
When the unpublished findings were added to the published findings for comparisons of psychological treatments vis-à-vis control conditions (in aggregate), the effect size point estimate was reduced 0.13 standard deviations (from g = 0.52 to g = 0.39). Using data from the earlier study by Turner and colleagues  showed that the effect size for antidepressant medications should be adjusted downward by 0.07 standard deviations when unpublished findings were added those that had been published. Our findings are based on different sets of trials than Turner’s and we can have less confidence in our estimate for psychotherapy since it was based on a smaller sample. Our findings indicate that psychological treatment is efficacious and specific, but, as is the case for antidepressants, less than the published literature conveys.
Our findings are in line with previous studies that suggest the presence of publication bias in studies of psychotherapy for depression [11, 74]. However, in contrast to those studies, which relied on an inferred estimate based on statistical procedures, this study ascertains the extent of study publication bias through direct empirical assessment. By tracking a cohort of US NIH-funded grants, we were able to determine how many trials went unpublished. It is reassuring that all but one of the published studies in our sample reported data in a manner suitable for meta-analysis and that, for the unpublished studies, there was a general willingness to share data.
A major limitation of this study is that we were not able to ascertain the extent to which the published results might have been affected by outcome reporting bias, thus exaggerating their effects. Unfortunately, there is no equivalent to the FDA database that would allow us to do for psychological treatment studies what Turner and colleagues were able to do with antidepressant trials , namely to determine whether, and quantify the degree to which, the published results had been “spun”. We tried to obtain the original grants from the NIH, but only a handful of the earlier protocols (before 2000) had been retained. This contrasts with the success of Chan and colleagues in obtaining protocols from the Canadian Institutes of Health Research . If we had had similar access to the NIH protocols, we could have ascertained the degree to which the authors adhered to the a priori methods when generating the corresponding publications. We believe that the NIH should archive protocols and make them publicly available so that outcome reporting bias can be ruled out by researchers and discouraged among investigators.
However, even with full access to the protocols, we still would not know whether the results reported were obtained using the methods prespecified a priori in the protocols. Therefore, we cannot say to what extent our effect sizes (adjusted for study publication bias alone) represent the ‘true’ effect of psychological treatment. If we could have detected outcome reporting bias, the adjusted effect would probably have been lower than what we report after adjusting for study publication bias alone. In essence, we can never come to terms with the extent of publication bias in the literature unless we can correct for both study publication bias and outcome reporting bias. The field needs both a clinical trials registry and a data repository for psychological treatment trials similar to the one the FDA maintains for industry-funded pharmacotherapy trials.
This study also has several additional limitations. (1) We were not able to obtain unpublished data for 2 of the 13 (15.4%) unpublished studies. In one case, the investigator was willing, but data collected in the 1980s were no longer available. However, in the other case, the investigator was reluctant to share her data, fearing that doing so would jeopardize her chances for independent publication. (2) Although unlikely, our search might have missed grant-funded randomized clinical trials that were begun but never published. We did find one instance in which we had excluded a grant that resulted in a publication that fell within our criteria. We suspect that this was a rare occurrence that came about only because the sample studied was available for inclusion in a separate acute treatment trial after it was excluded from the funded maintenance trial . (3) Our sample was restricted to studies funded by grants from the US government. Studies conducted in other countries or funded by other sources could show different characteristics and publication patterns. (4) No review protocol was registered for this study. When we started our review in 2009, registration of protocols for reviews was uncommon ; the international prospective register of systematic review protocols in health and social care (PROSPERO) was launched in February 2011, by which time our review was well underway. Nevertheless, review protocol registration was possible  and, in retrospect, would have increased the transparency of our study by allowing readers to compare our published methods with those originally planned. (5) The quality of the studies included in this meta-analysis was less than optimal. (6) Many of the analyses included only a small number of unpublished studies. (7) Combining different types of psychological treatments or different types of comparison conditions might have increased heterogeneity. (8) Given that we used depressive symptoms as the sole outcome for this review, it is possible that we might have missed studies that excluded such measures,but all published and unpublished studies did include such measures, such that no study was excluded from the analyses for this reason. (9) It is possible that we missed published articles; however, we think that is unlikely because we searched a comprehensive database of published randomized clinical trials of psychotherapy for depression, conducted a literature search for all publications by the grant’s principal investigator, then, when we could not locate a published study, contacted the investigators. (10) Our examination of the completeness of our grant search strategy, which involved checking for grant acknowledgements within all published studies listed in a database, could be criticized for introducing bias, as it could only result in the inclusion of additional published studies and not unpublished studies. However, excluding the one study that was added to the sample as a consequence of this examination from the analyses did not affect the general pattern of results. (11) Finally, some of the unpublished psychological treatment studies may still be published in the future. The investigators of two of the more recent trials that obtained relatively large effect sizes indicated that they still intended to submit their studies. If these were to be accepted that would lower the non-publication rate from 23.6% to 20%. It would not alter the “true” effect of psychotherapy based on all initiated trials; it would, however, widen the gap in effect size between the published and unpublished trials.
Most of the unpublished studies were not submitted for review. That is not to say that journals do not play a role in shaping author expectations, but the proximal cause of non-publication in our sample was more often non-submission by the authors rather than rejection by the journals. This is consistent with an earlier study of US NIH-funded clinical trials that found that findings were often not submitted when the investigators considered their results "not interesting" or when they “did not have enough time” . At the same time, we doubt that the authors were trying to suppress less than impressive findings. We would speculate that most research clinicians would benefit more from a publication than they would be adversely affected by null findings. Nonetheless, the bias against null findings is so ingrained in the field that we doubt that simple exhortations to authors or editors will do enough to change behavior. Instead, we join with others who recommend that funding agencies or journals should archive both original protocols and raw data from any funded randomized clinical trial [79–83].
In addition, we believe that a loophole should be closed in the NIH Public Access Policy (http://publicaccess.nih.gov/policy.htm). This policy requires that results from NIH-sponsored studies be made freely available to the public (via open access publication), but it applies only if the trial results are published in peer-reviewed journals. If an investigator chooses to leave NIH-sponsored study results unpublished, he or she is still free to do so. But we believe that an investigator who accepts public funds to conduct research has a fiduciary and ethical duty to provide a return on the public’s investment by making a good faith effort to publish. We therefore call upon the NIH to expand its Public Access Policy so that the public can have, full, not partial, access. The widespread belief that “negative” results are “not publishable” is no longer as true as it once was, because there are now journals committed to publishing study results without regard to their strength and direction, including Trials and PLOS ONE.
In conclusion, study publication bias is present in studies funded by US government grants to examine the efficacy of psychological treatment for depression, and this leads to an overestimation of its effects. The reason why is open to debate, but the fact that it occurs is not. Our findings indicate that psychological treatment is efficacious and specific, but, as is the case for antidepressants, less than the published literature conveys. Clinicians, guidelines developers, and decision makers should be aware of overestimated effects of the predominant treatments for major depressive disorder.
S1 Table. Meta-analyses of studies examining the effect of psychological treatment for depression (Additional and sensitivity analyses).
* p < .05; ** p < .01; Italic numbers indicate a non-significant trend (p < .10). a difference between the effect size point estimates of the published studies only and the published + unpublished studies, calculated using 5 decimals. b 1 degree of freedom. c Sensitivity analysis changing psychotherapy ranks in the published study, which was intended to minimize allegiance effects . d Sensitivity analysis counting  that does not report effect size data as unpublished. e Sensitivity analysis including the unpublished Gottlieb study, for which the effect size was estimated to be zero because of missing data. Note: NSt = number of studies; PT = psychological treatment; Qbetw. = Q between value.
Conceived and designed the experiments: ED SDH CLHB EHT. Performed the experiments: ED SDH. Analyzed the data: ED EHT. Contributed reagents/materials/analysis tools: SDH PC EHT. Wrote the paper: ED SDH CLHB PC EHT.
References marked with an asterisk indicate studies included in the meta-analysis.
- 1. Dickersin K. The existence of publication bias and risk factors for its occurrence. JAMA. 1990;263: 1385–1389. pmid:2406472
- 2. Rosenthal R. The “file drawer problem” and tolerance for null results. Psychol Bull. 1979;86: 638–641.
- 3. Kessler RC, Berglund P, Demler O, Jin R, Koretz D, Merikangas KR, et al. The epidemiology of major depressive disorder: Results from the National Comorbidity Survey Replication (NCS-R). JAMA. 2003;289: 3095–3105. pmid:12813115
- 4. Kessler RC. The costs of depression. Psychiatr Clin North Am. 2012;35: 1–14. pmid:22370487
- 5. Kupfer DJ, Frank E, Phillips ML. Major depressive disorder: new clinical neurobiological, and treatment perspectives. Lancet. 2012;379: 1045–1055. pmid:22189047
- 6. Mathers CD, Loncar D. Projections of global mortality and burden of disease from 2002 to 2030. PLoS Med. 2006;3: e442. pmid:17132052
- 7. American Psychiatric Association. Practice guideline for the treatment of patients with major depressive disorder. 3rd ed. Arlington (VA): American Psychiatric Association; 2010.
- 8. National Institute for Health and Clinical Excellence. Depression: The treatment and management of depression in adults (update). 2009. Available: http://guidance.nice.org.uk/CG90/.
- 9. Marcus SC, Olfson M. National trends in the treatment of depression from 1998 to 2007. Arch Gen Psychiatry. 2010;67: 1265–1273. pmid:21135326
- 10. Turner EH, Matthews AA, Linardatos E, Tell RA, Rosenthal R. Selective publication of antidepressant trials and its influence on apparent efficacy. NEJM. 2008;358: 252–260. pmid:18199864
- 11. Cuijpers P, Smit F, Bohlmeijer E, Hollon SD, Andersson G. Efficacy of cognitive-behavioural therapy and other psychological treatments for adult depression: meta-analytic study of publication bias. Br J Psychiatry. 2010;196: 173–178. pmid:20194536
- 12. Duval S, Tweedie R. A nonparametric ‘trim and fill’ method of accounting for publication bias in meta-analysis. J Am Stat Assoc. 2000;95: 89–98.
- 13. Duval S, Tweedie R. Trim and fill: A simple funnel-plot-based method of testing and adjusting for publication bias in meta-analysis. Biometrics. 2000;56: 455–463. pmid:10877304
- 14. Sterne JA, Egger M, Smith GD. Systematic reviews in health care: investigating and dealing with publication and other biases in meta-analysis. BMJ. 2001; 323: 101–105. pmid:11451790
- 15. Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. Introduction to meta-analysis. UK: Wiley; 2009.
- 16. Ross JS, Tse T, Zarin DA, Xu H, Zhou L, Krumholz HM. Publication of NIH funded trials registered in ClinicalTrials.gov: Cross sectional analysis. BMJ. 2012;344: d7292. pmid:22214755
- 17. Spitzer RL, Endicott J, Robins E. Research diagnostic criteria: Rationale and reliability. Arch Gen Psychiatry. 1978;35: 773–785. pmid:655775
- 18. Feighner JP, Robins E, Guze SB, Woodruff RA, Winokur G, Munoz R. Diagnostic criteria for use in psychiatric research. Arch Gen Psychiatry. 1972;26: 57–63. pmid:5009428
- 19. Cuijpers P, van Straten A, Smit F, Andersson G. Is psychotherapy for depression equally effective in younger and older adults? A meta-regression analysis. Int Psychogeriatr. 2009;21: 16–24. pmid:19040783
- 20. Cuijpers P, van Straten A, Warmerdam L, Andersson G. Psychological treatment of depression: A meta-analytic database of randomized studies. BMC Psychiatry. 2008;8: 36. pmid:18485191
- 21. Borenstein M, Hedges L, Higgins J, Rothstein H. Comprehensive Meta-analysis Version 2. Englewood, NJ: Biostat; 2005.
- 22. Hamilton M. A rating scale for depression. J Neurol Neurosurg Psychiatry. 1960;23: 56–62. pmid:14399272
- 23. Beck AT, Ward CH, Mendelson M, Mock J, Erbaugh J. An inventory for measuring depression. Arch Gen Psychiatry. 1961;4: 561–571. pmid:13688369
- 24. Ioannidis JPA, Patsopoulos NA, Evengelou E. Uncertainty in heterogeneity estimates in meta-analyses. BMJ. 2007;335: 914–916. pmid:17974687
- 25. Orsini N, Higgins J, Bottai M, Buchan I. Heterogi: Stata module to quantify heterogeneity in a meta-analysis; 2005.
- 26. StataCorp. Stata Statistical Software: Release 11. College Station, TX: StataCorp LP; 2009.
- 27. Wampold BE, Mondin GW, Moody M, Stich F, Benson K, Ahn H. A meta-analysis of outcome studies comparing bona fide psychotherapies: empirically, “all must have prizes”. Psychol Bull. 1997;122: 203–215.
- 28. * Reynolds CF, Miller MD, Pasternak RE, Frank E, Perel JM, Cornes C, et al. Treatment of bereavement-related major depressive episodes in later life: a controlled study of acute and continuation treatment with nortriptyline and interpersonal psychotherapy. Am J Psychiatry. 1999;156: 202–208. pmid:9989555
- 29. * Areán PA, Raue P, Mackin RS, Kanellopoulos D, McCulloch C, Alexopoulos GS. Problem-solving therapy and supportive therapy in older adults with major depression and executive dysfunction. Am J Psychiatry. 2010;167: 1391–1398. pmid:20516155
- 30. * Barber JP, Barrett MS, Gallop R, Rynn MA, Rickels K. Short-term dynamic psychotherapy versus pharmacotherapy for major depressive disorder: a randomized, placebo-controlled trial. J Clin Psychiatry. 2012;73(1): 66–73. pmid:22152401
- 31. * Rush AJ, Beck AT, Kovacs M, Hollon SD. Comparative efficacy of cognitive therapy and pharmacotherapy in the treatment of depressed outpatients. Cognit Ther Res. 1977;1: 17–37.
- 32. * Beutler LE, Engle D, Mohr D, Daldrup RJ, Bergan J, Meredith K, et al. Predictors of differential response to cognitive, experiential, and self-directed psychotherapeutic procedures. J Consult Clin Psychol. 1991;59: 333–340. pmid:2030196
- 33. * Covi L, Lipman RS. Cognitive behavioral group psychotherapy combined with imipramine in major depression. Psychopharmacol Bull. 1987;23: 173–176. pmid:3602315
- 34. * Denton WH, Wittenborn AK, Golden RN. Augmenting antidepressant medication treatment of depressed women with emotionally focused therapy for couples: a randomized pilot study. J Marital Fam Ther. 2012;38(s1): 23–38.
- 35. * DeRubeis RJ, Hollon SD, Amsterdam JD, Shelton RC, Young PR, Salomon RM, et al. Cognitive therapy vs medications in the treatment of moderate to severe depression. Arch Gen Psychiatry. 2005;62: 409–416. pmid:15809408
- 36. * DiMascio A, Weissman MM, Prusoff BA, Neu C, Zwilling M, Klerman GL. Differential symptom reduction by drugs and psychotherapy in acute depression. Arch Gen Psychiatry. 1979;36: 1450–1456. pmid:518245
- 37. * Weissman MM, Prusoff BA, Dimascio A, Neu C, Goklaney M, Klerman GL. The efficacy of drugs and psychotherapy in the treatment of acute depressive episodes. Am J Psychiatry. 1979;136: 555–558. pmid:371421
- 38. * Frank E, Cassano GB, Rucci P, Thompson WK, Kraemer HC, Fagiolini A. Predictors and moderators of time to remission of major depression with interpersonal psychotherapy and SSRI pharmacotherapy. Psychol Med. 2011;41(1): 151–162. pmid:20380782
- 39. * Wright AGC, Hallquist MN, Swartz HA, Frank E, Cyranowski JM. Treating co-occurring depression and anxiety: modeling the dynamics of psychopathology and psychotherapy using the time-varying effect model. J Consult Clin Psychol. 2014;82: 839–853. pmid:24041230
- 40. * Lustman PJ, Griffith LS, Freedland KE, Kissel SS, Clouse RE. Cognitive behavior therapy for depression in type 2 diabetes mellitus. A randomized, controlled trial. Ann Intern Med. 1998;129: 613–621. pmid:9786808
- 41. * Clarkin JF, Glick ID, Haas GL, Spencer JH, Lewis AB, Peyser J, et al. A randomized clinical trial of inpatient family intervention. V. Results for affective disorders. J Affect Disord. 1990;18: 17–28. pmid:2136866
- 42. * Glick ID, Clarkin JF, Haas GL, Spencer JH. Clinical significance of inpatient family intervention: conclusions for a clinical trial. Hosp Community Psychiatry. 1993;44: 869–873. pmid:8225301
- 43. * Goldman RN, Greenberg LS, Angus L. The effects of adding emotion-focused interventions to the client-centered relationship conditions in the treatment of depression. Psychother Res. 2006;16: 537–549.
- 44. * Hersen M, Bellack AS, Himmelhoch JM, Thase ME. Effects of social skills training, amitriptyline and psychotherapy in unipolar depressed women. Behav Ther. 1984;15: 21–40.
- 45. * Hollon SD, DeRubeis RJ, Evans MD, Wiemer MJ, Garvey MJ, Grove WM, et al. Cognitive therapy and pharmacotherapy for depression: singly and in combination. Arch Gen Psychiatry. 1992;49: 774–781. pmid:1417429
- 46. * Hollon SD, DeRubeis RJ, Fawcett J, Amsterdam JD, Shelton RC, Zajecka J, et al. Effect of cognitive therapy with antidepressant medications vs antidepressants alone on the rate of recovery in major depressive disorder. JAMA Psychiatry. 2014;71(10): 1157–1164. pmid:25142196
- 47. * Elkin I, Shea MT, Watkins JT, Imber SD, Sotsky SM, Collins JF, et al. National Institute of Mental Health Treatment of Depression Collaborative Research Program. General effectiveness of treatments. Arch Gen Psychiatry. 1989;46: 971–982. pmid:2684085
- 48. * Jacobson NS, Dobson K, Fruzzetti AE, Schmaling KB, Salusky S. Marital therapy as a treatment for depression. J Consult Clin Psychol. 1991;59: 547–557. pmid:1918559
- 49. * Dimidjian S, Hollon SD, Dobson KS, Schmaling KB, Kohlenberg RJ, Addis ME, et al. Randomized trial of behavioral activation, cognitive therapy, and antidepressant medication in the acute treatment of adults with major depression. J Consult Clin Psychol. 2006;74: 658–670. pmid:16881773
- 50. * Jacobson NS, Dobson KS, Truax PA, Addis ME, Koerner K, Gollan JK, et al. A component analysis of cognitive-behavioral treatment for depression. J Consult Clin Psychol. 1996;64: 295–304. pmid:8871414
- 51. * Jarrett RB, Schaffer M, McIntire D, Witt-Browder A, Kraft D, Risser RC. Treatment of atypical depression with cognitive therapy or phenelzine: a double-blind, placebo-controlled trial. Arch Gen Psychiatry. 1999;56: 431–437. pmid:10232298
- 52. * Waters SJ, Strauman TJ, McKee DC, Campbell LC, Shelby RA, Dixon KE, et al. Self-system therapy for distress associated with persistent low back pain: a randomized clinical trial. Psychother Res. 2015; 16: 1–12.
- 53. * Hayden T, Perantie DC, Nix BD, Barnes LD, Mostello DJ, Holcomb WL, et al. Treating prepartum depression to improve infant developmental outcomes: a study of diabetes in pregnancy. J Clin Psychol Med Settings. 2012;19: 285–292. pmid:22526914
- 54. * Lynch TR, Morse JQ, Mendelson T, Robins CJ. Dialectical behavior therapy for depressed older adults: a randomized pilot study. Am J Ger Psychiatry. 2003;11: 33–45.
- 55. * Manber R, Edinger JD, Gress JL, San Pedro-Salcedo MG, Kuo TF, Kalista T. Cognitive behavioral therapy for insomnia enhances depression outcome in patients with comorbid major depressive disorder and insomnia. Sleep. 2008;31: 489–495. pmid:18457236
- 56. * Miller IW, Norman WH, Keitner GI. Cognitive-behavioural treatment of depressed inpatients. Behav Ther. 1989;20: 25–47.
- 57. * Miranda J, Chung JY, Green BL, Krupnick J, Siddique J, Revicki DA, et al. Treating depression in predominantly low-income young minority women: a randomized controlled trial. JAMA. 2003;290: 57–65. pmid:12837712
- 58. * Mohr DC, Boudewyn AC, Goodkin DE, Bostrom A, Epstein L. Comparative outcomes for individual cognitive-behavior therapy, supportive-expressive group psychotherapy, and sertraline for the treatment of depression in multiple sclerosis. J Consult Clin Psychol. 2001;69: 942–949. pmid:11777121
- 59. * Murphy GE, Simons AD, Wetzel RD, Lustman PJ. Cognitive therapy and pharmacotherapy. Singly and together in the treatment of depression. Arch Gen Psychiatry. 1984;41: 33–41. pmid:6691783
- 60. * Murphy GE, Carney RM, Knesevich MA, Wetsel RD, Whitworth P. Cognitive behavior therapy, relaxation training and tricyclic antidepressant medication in the treatment of depression. Psychol Rep. 1995;77: 403–420. pmid:8559866
- 61. * O'Hara MW, Stuart S. Gorman LL, Wenzel A. Efficacy of interpersonal therapy for postpartum depression. Arch Gen Psychiatry. 2000;57: 1039–1045. pmid:11074869
- 62. * Rohan RJ, Roecklein KA, Tierney Lindsey K, Johnson LG, Lippy RD, Lacy TJ, et al. A randomized controlled trial of cognitive-behavioral therapy, light therapy and their combination for seasonal affective disorder. J Consult Clin Psychol. 2007;75: 489–500. pmid:17563165
- 63. * Schulberg HC, Block MR, Madonia MJ, Scott P, Rodriguez E, Imber SD, et al. Treating major depression in primary care practice. Eight-month clinical outcomes. Arch Gen Psychiatry. 1996;53: 913–919. pmid:8857868
- 64. * Linde JA, Simon GE, Ludman EJ, Ichikawa LE, Opersalski BH. A randomized controlled trial of behavioral weight loss treatment versus combined weight loss/depression treatment among women with comorbid obesity and depression. Ann Behav Med. 2011;41: 119–130. pmid:20878292
- 65. * Spinelli MG, Endicott J. Controlled clinical trial of interpersonal psychotherapy versus parenting education program for depressed pregnant women. Am J Psychiatry. 2003;160: 555–562. pmid:12611838
- 66. * Swartz HA, Frank E, Zuckoff A, Cyranowski JM, Houck PR, Cheng Y, et al. Brief interpersonal psychotherapy for depressed mothers whose children are receiving psychiatric treatment. Am J Psychiatry. 2008;165: 1155–1162. pmid:18558645
- 67. * Talbot NL, Chaudron LH, Ward EA, Duberstein PR, Conwell Y, O’Hara MW, et al. A randomized effectiveness trial of interpersonal psychotherapy for depressed woman with sexual abuse histories. Psychiatr Serv. 2011;62: 374–380. pmid:21459988
- 68. * Strachowski D, Khaylis A, Conrad A, Neri E, Spiegel D, Taylor CB. The effects of cognitive behaviour therapy on depression in older adults with cardiovascular risk. Depress Anxiety. 2008;25: E1–E10.
- 69. * Gallagher DE, Thompson LW. Treatment of major depressive disorder in older adult outpatients with brief psychotherapies. Psychotherapy (Chic). 1982;19: 482–490.
- 70. * Thompson LW, Gallagher D, Breckenridge JS. Comparative effectiveness of psychotherapies for depressed elders. J Consult Clin Psychol. 1987;55: 385–390. pmid:3597953
- 71. * Thompson LW, Coon DW, Gallagher-Thompson D, Sommer BR, Koin D. Comparison of desipramine and cognitive/behavioral therapy in the treatment of elderly outpatients with mild-to-moderate depression. Am J Ger Psychiatry. 2001;9: 225–240.
- 72. * Foley SH, Rounsaville BJ, Weissman MM, Sholomskas D, Chevron E. Individual versus conjoint interpersonal psychotherapy for depressed patients with marital disputes. International Journal of Family Psychiatry. 1989;10: 29–42.
- 73. * Wright JH, Wright AS, Albano AM, Basco MR, Goldsmith LJ, Raffield T, et al. Computer-assisted cognitive therapy for depression: Maintaining efficacy while reducing therapist time. Am J Psychiatry. 2005;162: 1158–1164. pmid:15930065
- 74. Flint J, Cuijpers P, Horder J, Koole SL, Munafò MR. Is there an excess of significant findings in published studies of psychotherapy for depression? Psych Med. 2014;25: 1–8 [Epub ahead of print]
- 75. Chan A-W, Krleža-Jerić K, Schmid I, Altman DG. Outcome reporting bias in randomized trials funded by the Canadian Institutes of Health Research. CMAJ. 2004;171: 735–740. pmid:15451835
- 76. Reynolds CF, Frank E, Perel JM, Imber SD, Cornes C, Miller MD, et al. Nortriptyline and interpersonal psychotherapy as maintenance therapies for recurrent major depression. A randomized controlled trial in patients older than 59 years. JAMA. 1999;281: 39–45. pmid:9892449
- 77. Liberati A, Altman DG, Tetzlaff J, Mulrow C, Gøtzsche P, Ioannidis JP, et al. The PRISMA statement for reporting systematic reviews and meta-analyses of studies that evaluate health care interventions: Explanation and elaboration. PLoS Med. 2009;6: e1000100. pmid:19621070
- 78. Dickersin K, Min YI. NIH clinical trials and publication bias. Online J Curr Clin Trials. 1993;2: Doc No 50.
- 79. Chan A-W, Upshur R, Singh JA, Ghersi D, Chapuis F, Altman DG. Research protocols: waiving confidentiality for the greater good. BMJ. 2006;332: 1086–1089. pmid:16675819
- 80. Vickers AJ. Making raw data more widely available. New initiative by the Wellcome Trust sets out some guiding principles. BMJ. 2011;342: d2323. pmid:21543405
- 81. Alsheikh-Ali AA, Qureshi W, Al-Mallah MH, Ioannidis JPA. Public availability of published research data in high-impact journals. PLoS ONE. 2011;6: e24357. pmid:21915316
- 82. Smith R, Roberts I. Patient safety requires a new way to publish clinical trials. PLoS Clin Trials. 2006;1: e6. pmid:16871328
- 83. Perneger TV. Sharing raw data: another of Francis Galton’s ideas. BMJ. 2011;342: d3035. pmid:21586466