Placebo Response in Trials of Negative Symptoms in Schizophrenia: A Critical Reassessment of the Evidence

Abstract

Background

Summarizing evidence from clinical trials of patients with schizophrenia with predominant or prominent negative symptoms (NS), a prior meta-analysis reported a large placebo effect in negative symptoms (Cohen’s d = 2.909). Assuming that such an effect was clinically not plausible, we performed a critical re-assessment and an update of the previous results with newly available data from add-on and monotherapy studies.

Study Design

Random-effect meta/regression analysis of trials that focused on predominant or prominent NS; and adopted a double-blind, randomized, placebo-controlled design. The final pooled meta-analytic database, based on the available add-on and monotherapy studies combined, included 24 publications containing data on a total of 25 studies (21 add-on, 4 monotherapy).

Study Results

The pooled overall estimate for the placebo effect from the primary analysis for all included studies had a medium effect size, with a Cohen’s d value of 0.6444 (SE = 0.091). The estimates were similar in the add-on and monotherapy studies. Meta-regression indicated that the high placebo response was significantly associated with clinical trial characteristics, including the high ratio of patients assigned to active vs. placebo treatment and short trial duration.

Conclusions

These results represent a major downward correction for a current effect size estimate of the placebo response in the negative symptoms of schizophrenia. Our findings also pinpoint certain clinical trial characteristics, which may serve as important predictors of the placebo response. The knowledge of these factors can have important implications for drug development and trial design for new drugs for negative symptoms of schizophrenia.

Introduction

Despite major efforts to develop efficacious treatments, current antipsychotics (APs) have limited efficacy against the negative symptoms of schizophrenia. Negative symptoms (NS) of schizophrenia (eg, blunted affect, anhedonia, avolition, asociality and alogia) consist of two major classes with heterogeneous etiologies, namely the primary and secondary NS.¹ The primary NS are viewed as an intrinsic part of the schizophrenic illness, and are considered to be associated with specific disease course, and long-term morbidity.¹ By contrast, the secondary NS are associated with well-defined etiologies, eg, positive symptoms, comorbid depression, side effects of medications, or social deprivation.^1–4 The treatment of primary NS represents a major unmet need. They may be present before the onset of the first positive symptoms of schizophrenia and persist after the acute phase of the illness, and are associated with poor psychosocial functioning.^1,2,5 To approve a specifically targeted, domain-specific approach for the development of new drugs for the treatment of negative symptoms, regulators asked an important question: “Do negative symptoms respond differently to drug treatment than other schizophrenic symptoms?” ⁶

With accumulating evidence that available APs provide little benefit for NS, the domain-specific approach was endorsed by the US Food and Drug Administration (FDA) agency in the early 2000s.^6,7 Furthermore, a consensus between academic experts, regulators and industry representatives was also reached that clinical trials for NS need to develop their specific methodology requirements in order to implement this approach in practice.⁸ Specifically, trials of NS were expected to adopt specific sampling strategies (eg, selection of patients with stable and persistent prominent and predominant NS); inclusion criteria (eg, low severity in other symptom domains, such as positive, depressive, and extrapyramidal symptoms, and cognitive impairment to avoid pseudospecificity); sufficient duration to capture potential improvement; and validated measures of outcome (ie, specific measures of NS, applied as primary endpoint).⁸

Additionally, academic and regulatory experts reached consensus on the need of placebo control in clinical trials to ascertain the specificity of improvement in NS during treatment.⁷ In light of this, it is of interest to turn around the question posed by the regulators, and ask: Do negative symptoms respond to placebo in the same way as other symptoms of schizophrenia? If the placebo effect for NS is high in an adequately selected patient population (as in the case of other symptoms of schizophrenia^9–11), then it would be essential to find out whether certain clinical trial design or patient level prognostic factors exist that would reduce the placebo response. Identification of these factors could improve our ability to detect the effects of promising new agents for NS.

The difficulty of demonstrating a significant effect against placebo in NS trials is highlighted by Krause et al’s meta-analysis,¹² which examined monotherapy trials of APs approved for schizophrenia by the U.S. FDA and the European Medicines Agency. This meta-analysis focused on patients with predominant or prominent NS, and found that amisulpride (AMI) was “the only antipsychotic that outperformed placebo in the treatment of predominant negative symptoms”.¹² It is also noteworthy that even for AMI, the evidence against placebo was equivocal, as all positive studies were conducted prior to the 2000s, before the above requirements for NS trials were stipulated (only one study was conducted after 2000, and it yielded negative results). The trials for the three other APs that were tested besides AMI (olanzapine, zotepine, and sulpride) took place after 2000. It is of note that Krause et al’s meta-analysis (which excluded the add-on trials) did not specifically examine the placebo effect, or address whether the failure of APs against placebo was due to a high placebo response.¹³

We found only one meta-analysis for the placebo effect in clinical trials of patients with schizophrenia with predominant or prominent NS. This meta-analysis by Fraguas et al¹⁴ covered data from an approx. 15 year period (from January 2002 to January 24, 2017). Unlike Krause et al’s study, Fraguas et al’s meta-analysis investigated placebo-controlled add-on trials (ie, monotherapy trials were not included). The test treatments in the included studies were pharmaceutical agents that have not received regulatory approval for the treatment of schizophrenia. Analyzing the data from 18 studies reported in 17 publications, the authors described a “large placebo response” for NS, with an estimated effect size of 2.909 in terms of Cohen’ d measure.

There may be several reasons why a placebo effect of the above magnitude is clinically not plausible in NS studies. First, the analyzed studies adopted the consensus criteria for NS trials which emerged during the early 2000s, including the methodologically stringent provisions for patient selection (eg, requirement for stable predominant or prominent NS, and low severity in other symptom domains). It can be expected that such a carefully selected target population, which presents with stable and enduring NS, is unlikely to evidence a large change in symptom severity over time. Second, it can be hypothesized that unspecific effects including psychological factors that may generally underlie placebo response, such as expectancy effects, optimism and cognitive reframing are modest in this mostly treatment resistant sample. Third, the effect size reported by Fraguas et al.¹⁴ would entail that, in most cases, the patients assigned to placebo would become almost completely asymptomatic with respect to NS within a trial’s time-frame, which is an unlikely clinical outcome.

In view of these reasons, our principal aim in the current study was to re-assess the placebo effect in randomized double-blind NS studies by focusing specifically on the improvement evidenced by patients in the placebo arms. Furthermore, we also wanted to examine whether the placebo response differs in the monotherapy and the add-on trials. Finally, we aimed to delineate the predictors of the placebo response in the NS studies. Our specific goal was to determine how certain clinical trial-design factors, some of which were reported in the previous meta-analysis as predictors of placebo response (number of arms, industry sponsorship), influence the placebo effect based on the updated estimates.

To accomplish these aims, we re-assessed the effect size estimates which were computed for the individual studies by Fraguas et al.¹⁴ We implemented adjustments for the estimates due to problems with the original effect size calculation (eg, using standard error, instead of SD, for the computation of Cohen’s d). Furthermore, we updated the analysis data set with new data by extending the study inclusion period to the past 20 years. We also broadened the scope of the analyses by including placebo-controlled monotherapy studies besides the add-on trials.

Methods

Selection of Publications

We searched for studies that met the following criteria: they (1) focused on predominant or prominent NS; (2) adopted a double-blind, randomized placebo-controlled design; and (3) administered the test treatments either in add-on or monotherapy trials. Similar to Fraguas et al., we selected trials that were published starting from 2002, since as stated by these authors “predominant/prominent negative symptoms only became a target of investigation in the early 2000s”.¹⁴ The end of the selection period for the current investigation lasted until January 24, 2022, which extended the selection period of the prior meta-analysis (which lasted until January 24, 2017) by 5 years. Methodological details of the search strategy are provided in supplementary material, part 1.

Computation of Effect Sizes

Similar to Fraguas et al, we used the Cohen’s d measure to characterize the placebo effect size for NS. The input parameters for the computation were the mean change of the primary measure of efficacy at the end of treatment in the placebo arm relative to the pretreatment baseline (Chg_PBO); and the within-group standard deviation of the change (SD_{Chg_PBO_within}). Cohen’s d was computed by dividing the mean change by its SD (ie, Cohen’s d = Chg_PBO/ SD_{Chg_PBO_within}). From 7 of the included publications the above statistics of change (ie, Chg_PBO, SD_{Chg_PBO_within}) could not be retrieved. Nonetheless, these publications provided the mean and the within-group SD at baseline and at the end of treatment; we used these for the effect size calculation. Specifically, Chg_PBO in these studies was computed as raw score difference between the baseline and endpoint (ie, Mean_{PBO_bas}, Mean_{PBO_end}). For the computation of the within-group SD (SD_{Chg_PBO_within}), we used the formula for correlated means provided by Cohen,¹⁵ since pairs of observations from the same subjects (ie, baseline/endpoint) are expected to be correlated. Accordingly, following Cohen,¹⁵ we applied the formula below:

where VAR_{PBO_bas} and VAR_{PBO_end} represent, respectively, the within-group variance of the efficacy measure at baseline and endpoint, and r denotes the correlation between baseline and endpoint. Since the r-value was not reported in the respective studies, we estimated Cohen’s d for various r-values. First, the r-value of .25, which represents a modest association between baseline and endpoint from the same subjects, was adopted for our primary analysis as a conservative assumption. Then, in sensitivity analyses we investigated whether the effect size estimates would remain similar under various alternative assumptions, including the one of higher correlation (r = .5) or no correlation (r = 0).

Moderator Variables for the Meta-Regression Analyses

Besides determining the overall effect size for placebo, we conducted meta-regression analyses to delineate potential predictors of placebo response. First, we investigated whether the placebo effect increased with time in NS trials of schizophrenia (similarly to previous trials of broad-spectrum AP agents which indicated increasing placebo response over the years in the overall symptom severity^11,16). For this analysis, we used publication date as a moderator variable in the meta-regression. Furthermore, we also examined whether industry sponsorship influenced the placebo response. Since industry sponsorship is typically associated with a large sample size, the total study sample size was also introduced as covariate to investigate whether industry sponsorship has a significant contribution after adjusting for this variable.

For further meta-regression analyses, we classified the potential predictor variables into two sets, and examined them in both univariate and multivariate analyses. One of the sets comprised clinical trial design variables, and included trial type (add-on/monotherapy); study duration (which due to the skewed distribution was classified as short-[≤8 weeks] or longer-term[>8 weeks]); sample size in the placebo arm; the ratio of the subjects in the test vs. placebo arm; and study site (used as a dichotomous variable [single center/multi-center], since the number of study sites had skewed distribution). In this analysis, we also investigated the interaction between the active vs. placebo ratio and sample size in the placebo arm to examine whether an association between the active vs. placebo ratio and placebo response varies with the number of patients in the placebo sample. The other predictor variable set comprised basic demographic and clinical characteristics including age; illness duration; gender proportion (female%); and baseline severity of NS, expressed as a fraction (%) within the theoretical severity range of the scale. Further information on the statistical analyses is provided in supplementary material, part 2 (Technical Details of the Statistical Analyses).

Results

Included Studies

Detailed description of the results of the automatic search and the manual selection process for the inclusion of publications for this meta-analysis is provided in supplementary material, part 3. Overall, the final pooled meta-analytic database, from the add-on and monotherapy studies jointly, includes 25 studies reported in 24 publications. Supplementary material, part 4 shows the list of publications included in the meta-analysis.

Overall, the current investigation extended the former dataset used in Fraguas et al’s meta-analysis from 18 studies (reported in 17 publications) to 25 studies (reported in 25 publications), by adding the 4 monotherapy studies,^17–20 and 3 further add-on studies.^21–23 We also note that for one add-on study included in the former meta-analysis²⁴ an updated version of the clinical data became available²⁵; we used this data in our analyses for the updated estimates. It should also be noted that for a monotherapy study (¹⁹) in the former meta-analysis the placebo effect size was determined for comparative purposes with the add-on studies. The effect size, however, was erroneously computed as Cohen’s d = 3.255 since SE instead of SD was used in the calculation; the corrected value is 0.366.

Descriptive Data for the Included Studies

Table 1 provides a detailed characterization of the individual studies included in the current meta-analysis. The description of the NS measures and the computational details for the severity of NS at baseline are shown in supplementary material, part 5.

Regarding clinical trial characteristics, the majority of studies (18 of 25; 72.0%) were multicenter trials with an add-on design (21 of 25; 84%) and >8 weeks of treatment duration (19 of 25; 76.0%). In 10 of 25 (40.0%) studies, an unbalanced randomization scheme was applied, which assigned twice (in 28% of the studies) or three times (in 12% of the studies) as many patients to the test treatment as to placebo. The average sample size in the placebo arms was 63.7 (SD = 62.1); however, as indicated by the median, only half of the studies had a sample size of ≥34 patients in the placebo arms. The proportion of industry sponsorship was 44.0% (11 of the 25 PBO arms). However, it is of note that while the rate of industry sponsorship was 33.3% in the add-on studies, all monotherapy trials were industry sponsored. Detailed data on the trial features and the basic demographical and clinical characteristics of the study samples are shown in supplementary material, part 6.

Meta-Analyses

Figure 1 shows the Forest-plot of the effect sizes from the placebo arms of the individual studies and the pooled estimate from the primary analysis. We used a random-effect model to derive the effect size estimate since the Likelihood Ratio Test (LRT) based on the comparison of nested fixed and random-effect models using van Houwelingen et al’s approach²⁶ indicated significant heterogeneity in the effect size estimates across the included studies (LRT = 58.2, df = 1, P < .0001).

The pooled estimate was computed for all studies and for the add-on and the monotherapy studies, separately. As the plot indicates, all but two of the individual studies had a numerical improvement in the placebo arm, reaching statistical significance in the majority of cases. The pooled overall effect size estimate from the primary analysis for all included studies in terms of Cohen’s d was 0.6444 (SE = 0.091). The pooled estimates were similar for the add-on (Cohen’s d = 0.5898, SE = 0.077) and monotherapy studies (Cohen’s d = 0.6993, SE = 0.167).

Table 2 provides further details for the effect size estimates, including the results of the sensitivity analyses. These analyses were conducted since not all studies reported the basic summary statistics for the mean change from baseline to endpoint. As shown by the Table, the pooled estimates from the sensitivity analyses showed only little variation. Specifically, considering all included studies, the lowest and the highest pooled effect size was 0.6367 (SE = 0.092) and 0.6566 (SE = 0.091), respectively. For the add-on studies, the lowest and highest estimate from the sensitivity analysis was 0.5742 (SE = 0.078) and 0.6142 (SE = 0.077), respectively. We note that for the monotherapy trials, the sensitivity estimates did not vary as there were no missing data in the respective reports. The analyses did not indicate the presence of publication bias on the basis of the funnel plots using Begg & Mazumdar’s rank correlation approach²⁷ and the Egger test²⁸ (P > .1 in both analyses).

Together, our results indicate that the effect size estimate of the placebo response in NS is approximately one-fifth of the estimate reported by the previous meta-analysis (ie, Cohen’s d = 2.909).¹⁴Supplementary material, part 7 details the reasons why the approximately five-fold inflation of the effect size occurred in the previous meta-analysis.

Meta-Regressions

Table 3 displays the meta-regression findings for the putative predictors of placebo response of NS based on all trials (add-on/monotherapy combined). Regarding trial characteristics, the univariate analyses showed that the ratio of active vs. placebo treated patients and the sample size had a significant positive association with the placebo effect size. With respect to patient demographics, illness duration approached statistical significance (P = .061) with an inverse relationship with the placebo response: patients with longer illness duration tended to show lower placebo response. We also found that industry sponsorship and larger trial size were associated with greater response to placebo, while publication date had no association.

We also performed multivariate analyses to understand the specific contribution of the variables in the various sets of predictors (labeled as set 1 through 3 in table 3) in determining the placebo response. The results in table 3 show that in the set of trial characteristics included, the ratio of active vs. placebo treated patients and the effect of trial duration reached significance in the multivariate analysis. Shorter trial duration and greater active vs. placebo ratio were associated with larger placebo response. There was no significant interaction between the active vs. placebo ratio and the number of patients in the placebo group (t = −0.59, P = .571). The joint additive association of placebo response with the ratio of active vs. placebo ratio and trial duration is presented in figure 2, which shows that smaller placebo effects are associated with a combination of longer trial duration (>8 weeks) and a balanced active vs. placebo ratio (ie, 1:1 assignment). We also found that in the set of “patient characteristics” variables included in table 3 (set 2), illness duration remained marginally significant after adjusting for the other variables in the set. Our third analysis showed that, after adjusting for each other’s contribution, in the multivariate analysis neither trial size (ie, total N) nor industry sponsorship was able to serve as an independent predictor, as both of these variables lost significance.

Discussion

This study provides an update and critical reassessment of a previous meta-analysis of the placebo effect in negative symptoms in schizophrenia.¹⁴ Two research questions were evaluated: 1) What is the magnitude of the placebo response (the effect size) in patients with predominant or prominent NS? 2) What are the predictors of placebo response in these studies?

Effect Size Estimation

With respect to the first question, our results indicated a medium effect size (Cohen’s d = 0.6444). It is noteworthy that the results showed no difference between the add-on and monotherapy trials; they were numerically almost identical. Overall, our findings represent a major downward correction of the estimate from the previous meta-analysis, which reported approximately five times higher effect size. We found that much of the discrepancy between the prior and current estimates is due to the incorrect computation of the effect size in the previous study. With this in mind, we would like to note that in terms of widely accepted benchmarks (which describe Cohen’s d values of 0.2, 0.5, and 0.8, as small, medium, and large, respectively (Cohen¹⁵)), the placebo effect size estimate (0.6444) that we found for NS can be viewed as being at the higher end of the medium effect size range.

The finding that the placebo effect size estimate is substantially smaller than the cited estimate is consistent with empirical data indicating that negative symptoms, especially the primary negative symptoms demonstrate marked stability during the entire disease course.^{1,2,4,13,29,30} Stable disease course and chronicity have been shown to be associated with lower placebo response rates across various psychiatric illnesses including depression,³¹ while fluctuating course is a predictor of high placebo response, as many episodes tend to be time limited.³² With respect to NS studies in schizophrenia, the expert consensus opinion for patient selection criteria stated that “Negative symptoms should be stable and persistent. If negative symptoms fluctuate during a trial, this may increase the proportion of subjects who improve on the control condition.” ⁸ Collectively, our results that show only a medium effect size for placebo in the NS studies can be viewed as a corroboration that the consensus recommendations are reaching their goals clinical trials’ practice. Nonetheless, we should also bear in mind that, despite the recommendations, a substantial variation in the study samples and in the assessment and choice of rating scales for NS can still be present in the randomized clinical trials that can introduce heterogeneity and bias, which adds to the complexity of interpreting changes in negative symptoms.^29,33

The question can be raised whether the placebo effect size for NS is comparable to the placebo effect size for other symptoms of schizophrenia. Since comparative meta-analytic summaries of the placebo response across various symptoms dimensions of schizophrenia are lacking, this question is currently difficult to address directly. With respect to the placebo effect for overall symptom severity, a large meta-analysis by Agid et al.,¹¹ which used the same effect size measure as we used in the current meta-analysis, yielded a Cohen’s d value of 0.33. This value is somewhat lower than the placebo effect size we found for the negative symptoms (ie, Cohen’s d = 0.6444). However, two important caveats have to be taken into consideration in order to interpret the lower effect size in Agid et al’s study. First, similar to other studies which used raw score based measures to investigate placebo response (eg,^10,34), Agid at al’s meta-analysis included many studies where a single-blind initial placebo run-in period was used to exclude placebo responders to increase assay-sensitivity. Thus, patients who were likely to evidence a large response to placebo were not included in the double-blind phase of these trials. Second, the fact that the overall symptom severity measures also included negative symptoms precludes a clear conclusion regarding the placebo effect size in NS vs. the rest of symptoms of schizophrenia, which highlights the need for further studies.

Predictors of Placebo Response (Meta-Regression)

Besides the estimation of the placebo effect for NS, we conducted meta-regression analyses to identify the predictors of the placebo response. Regarding trial characteristics, we found that add-on and monotherapy trials were similar in the magnitude of the placebo effect. Furthermore, with respect to trial characteristics, one of the most robust predictors of placebo response was the active vs. placebo ratio, with congruent results from univariate and multivariate analyses. Specifically, we found that the higher the number of patients assigned to the active group(s) relative to placebo, the greater the placebo response. This finding for the negative symptoms is similar to most other findings reported for the placebo effect for the overall symptom severity in schizophrenia (eg,^11,35). Thus, our results can be viewed as an extension of the prior results to the negative symptoms of schizophrenia.

Additionally, we note that a consistent picture regarding the predictive role of the active vs. placebo ratio emerges when we compare the findings from schizophrenia trials to the findings from other psychiatric disorders, especially major depression where the role of the active vs. placebo ratio is well-documented.^36,37 This predictor, therefore, can be conceived of as a general methodological factor across various disease conditions, whose effect is mediated by unspecific factors rather than by the symptoms of the disorder. We think that one of the most likely candidates for such an unspecific predictor can be the expectancy effect, which can enhance treatment effects via expectations from either the patient or the physician through the anticipation that the current treatment is efficacious. Specifically, with higher active vs. placebo ratio a higher proportion of patients receive active treatment, which can increase the overall expectations and thereby results in higher placebo response, which we found in this study.

Study duration also reached significance in the multivariate analyses as a predictor of placebo response. Specifically, we found that the shorter the trial, the larger the placebo response. This finding is also consistent with results from former antipsychotic trials in schizophrenia (see, eg, Leucht et al’s meta-analytic summary¹⁰). It also highlights that longer trial duration is more suitable to reduce placebo effect, and to adequately capture changes in negative symptoms which are not associated with placebo response. Furthermore, it supports the conclusion of the expert panel that, while briefer duration of treatment of negative symptoms is acceptable for proof-of-concept trials, “pharmacological clinical trials need to be no less than 3 months but optimally 6 month or longer not including a prerandomization phase for the purpose of stabilization”.⁸

We also examined whether industry sponsorship is a predictor of placebo response in NS studies. Because industry sponsorship is often accompanied by larger sample size, we investigated these two variables (industry sponsorship, sample size) both in univariate and multivariate analyses. In the univariate analyses, consistent with former reports we found that both variables were associated with larger placebo response. As to the explanation of the trial size effect, we agree with Leucht et al’s conclusion that “… in studies with large sample size more patients who benefit from placebo are recruited compared to smaller trials where patients can be more carefully selected.” ¹⁰ We also concur with Leucht et al that while industry-sponsoring tends to increase placebo response, “industry sponsorship probably is a composite of multiple factors”,¹⁰ including the co-occurrence of larger sample size with the industry sponsored studies. Indeed, when these variables were examined jointly in our multivariate analysis, neither of them could provide a significant independent contribution.

In terms of patient characteristics, our analyses yielded only one marginally significant finding which occurred both in the univariate and the multivariate analysis. Specifically, the results indicated that longer illness duration was associated with lower placebo response independent of the effect of age, while age, in itself, was not associated with the placebo response as indicated by the multivariate analysis (P > .1). These findings for the NS, while only approached statistical significance, are congruent with the results of several meta-analyses which focused on the overall symptom severity, and showed a negative association with the illness duration.^10,11 It is important to note that the inverse relationship between illness duration and placebo response is not specific to schizophrenia; chronicity is associated with lower placebo response rates in other psychiatric illnesses as well (eg, depression³¹). An explanation for the association in schizophrenia was put forward by Agid et al who stated that patients who have “shorter duration of illness may be more susceptible to placebo response because of heightened expectations based on treatment earlier in the course of illness.“ ¹¹ Nonetheless, it is important to note that with respect to NS, the current study identified only a potential signal that illness duration can serve as a predictor of placebo response; this requires confirmation.

We found no change in the placebo response with calendar time, which is noteworthy in the context of several reports on an increasing placebo response over the years.^9,11,16 It should be noted, however, that in some of the studies the time effect was not present³⁵ and that the meta-analyses that reported increasing placebo response over time typically span longer than the two decades¹¹ period, which we focused on in the current meta-analysis. One also needs to consider that the calendar time effect on placebo response is closely related to changes in trial design factors, eg, increase in the number of subjects and sites, and change in treatment settings over time.¹¹ These factors remained relatively stable during the period that we examined, since the required trial design features were stipulated in the early 2000s for the NS studies.

Our study has several limitations. First, the small number of monotherapy trials did not make possible a detailed analysis of these trials. Nonetheless, the pooled effect size for these trials did not show a considerable deviation from the pooled effect size of the add-on trials, indicating that the lack of significant difference was not due the lack of statistical power, but to the rather small difference between the two types of studies. Second, some of the included studies did not report the difference scores from baseline to endpoint, but instead reported the raw scores and standard deviations at the baseline and the end point, separately. Thus, the effect size measures from these studies were estimated on the basis of various assumptions on the correlations between the baseline and endpoint scores. We conducted sensitivity analyses to examine the impact of these assumptions, and found that the results did not exhibit a substantial change as a function of the various assumptions. Furthermore, we note that of some of the information on patient characteristics was missing from certain studies. However, since the proportion of missing data was rather small, we think that our regression estimates are robust with respect to the missing data.

Finally, it should also be noted that, due to several potential biases that occur in the clinical trial environment, the medium placebo effect size that we found for NS can still be an overestimate of the “true” placebo effect. For example, it is noteworthy that despite the strict inclusion criteria for the NS studies, and the low level of symptom fluctuations in patients with primary NS, the regression to the mean effect can still be operant since patients with high baseline severity are enrolled in these studies, who can spontaneously improve over time. It is also conceivable that some of the clinical trials might include patients with transient (secondary) negative symptoms besides their primary negative symptoms (or even patients without primary negative symptoms). Furthermore, because the severity of NS is required to exceed a certain threshold, baseline inflation can also occur due to clinician bias to enroll subjects in the study, which may lead to increased placebo response.³⁸ Study participants could also evidence the clinical trial (Hawthorne) effect, which can be mediated through various factors.^39,40 These can include the impact of receiving more attention and being better observed in a clinical study, slightly better care during the trial, and selection bias from the physicians to work with more adherent patients.

Overall, our update and critical reassessment of the previous meta-analysis of the placebo effect on negative symptoms in schizophrenia revealed a medium effect (Cohen’s d = 0.6444) for the placebo response in the negative symptoms of schizophrenia. This represents a major downward adjustment compared to the previous report. In terms of meta-regression results, the current study uncovered that several study design factors have a profound impact on the magnitude of the placebo response. For example, a longer-term trial with a balanced active placebo ratio is expected to produce an effect size of 0.389 (SE = 0.0619); by contrast a short-term trial with unbalanced active vs. placebo ratio is associated with a placebo effect size of 1.162 (SE = 0.1483). The knowledge of these factors has important implications for drug developers: it can help reducing the placebo response with a suitable study design, thereby increasing the assay-sensitivity to detect the effects of promising new agents for negative symptoms.

Acknowledgments

IB has been an advisory board member/consultant/lecturer in the last 5 years for ACADIA, Angelini, Gedeon Richter, Hikma Pharmaceuticals, Janssen/Janssen Cilag, Lundbeck, Medichem Pharmaceuticals, Inc. by Unilab, Inc. PC and BK declare no conflicts of interest in relation to the subject of this study. This study is based on aggregated data published in the literature. The individual studies included in the meta-analytic database were conducted in accordance with the ethical standards of the responsible committees (institutional and national) for medical research involving human subjects and with the Helsinki Declaration of 1975, and its applicable revisions at the time of the investigation. Informed consent was obtained from patients included in the original studies.

Funding

The study was supported by the Hungarian Brain Research Program (2017-1.2.1-NKP-2017-0002).

References