Published on 18 September 2026
Negative symptoms of schizophrenia: are 451 trials measuring the right symptom?
The essentials
This meta-analysis pools 451 randomized trials and 42,566 participants with schizophrenia spectrum disorders, each comparing an active intervention with a non-active comparator on the severity of negative symptoms. Its main contribution is not a drug but a threshold. Using the method of their own 2015 work, the authors re-estimate, across the 122 trials reporting the Clinical Global Impression-Severity scale, the standardized mean difference that corresponds to a one-point improvement on that scale: 0.457. Restricted to the 214 studies of high methodological quality (19,746 participants), the analysis finds 16 statistically significant subcategories, but only 6 cross this threshold of clinical relevance: antibiotics, integrated psychosocial interventions, antidepressants, physical activity, transcranial direct current stimulation and immunomodulators, most often as adjuncts to antipsychotics. GRADE certainty ranges from very low to high depending on the subcategory. Antipsychotics themselves do not cross the threshold in any combination. Two caveats govern any reading. This is a pairwise meta-analysis against non-active comparators, so there is no head-to-head comparison between interventions. And the authors themselves state that the work was not designed to separate primary from secondary negative symptoms.
Context
Negative symptoms (avolition, anhedonia, social withdrawal, blunted affect, alogia) are the main driver of functional disability in schizophrenia, and the area where the therapeutic toolkit is thinnest. In 2015, Fusar-Poli and colleagues pooled 168 placebo-controlled trials and concluded that no statistically significant effect reached the threshold for clinically significant improvement. Eleven years later, the same group returns with a corpus almost three times larger and the same tool, re-estimated: a clinical relevance threshold derived from the data themselves, rather than borrowed from conventional rules for interpreting effect sizes. The authors explicitly tie this method to their 2015 work, from which they take it. What is new here is the size of the corpus on which the threshold is recalculated.
The regulatory picture deserves precision, because a widespread belief holds that no medication is authorized for negative symptoms. France offers a checkable example, and there the belief is wrong. The summary of product characteristics of a French amisulpride 100 mg product, updated on January 7, 2026 and consulted in the French public medicines database on August 11, 2026, lists an indication for acute and chronic schizophrenic disorders with positive symptoms (such as delusions, hallucinations, thought disorders, hostility, suspicious behavior) and/or negative symptoms (deficit syndrome) such as blunted affect and emotional and social withdrawal, and it states that this includes patients in whom negative symptoms are predominant. The dosage specified for that situation is 50 to 300 mg per day. A marketing authorization whose wording explicitly covers the predominantly negative presentation therefore does exist.
What does not exist in that example is an authorization whose indication is negative symptoms in isolation, or authorizations for the add-on strategies studied here. Cariprazine is instructive. Its European indication is the treatment of schizophrenia in adults, with no mention of negative symptoms. In its opinion dated February 6, 2019, the Transparency Committee of the French National Authority for Health (Haute Autorité de santé) did not accept the claimed superiority over risperidone in predominant negative symptoms, citing the small size of the additional effect and methodological limitations of the trial, namely a risperidone dose that was not optimized and the choice of symptom rating scale. It rated the actual clinical benefit as moderate and the improvement in actual clinical benefit at level V. No antidepressant, no anti-inflammatory drug and no tetracycline holds a French indication for negative symptoms of schizophrenia, so in France every add-on strategy in this meta-analysis is off-label prescribing. The example is French, the structure of the problem is not: readers should check what their own regulator has authorized, and on what wording, before assuming either that an option is approved or that none is.
The core of the problem
Primary or secondary: what the measurement cannot tell apart
A falling negative symptom score does not say why it falls. It may fall because the patient is less delusional and therefore less withdrawn, because drug-induced extrapyramidal symptoms ease and facial expression returns, because comorbid depression improves, because sedation decreases, or because a primary deficit recedes. The first four mechanisms define secondary negative symptoms. Only the last one is the target that treatment claims to address. This has been the central question of the field for thirty years, and it is the question to put to any study that aggregates global scores.
The authors answer it in their own limitations section, and explicitly: they state that the study was not specifically designed to disentangle primary from secondary negative symptoms. In practice, trials did not have to select patients with predominant or persistent negative symptoms to be included, and no mediation analysis is reported that would adjust the measured effect for concurrent changes in positive symptoms, depression or extrapyramidal effects. The only meta-regressions performed concern mean age, follow-up duration and baseline negative symptom severity. The authors also point to the conceptual overlap: several core negative symptom domains, particularly anhedonia, anergia and avolition, overlap substantially with depressive features, which complicates both clinical assessment and the interpretation of treatment effects. They place the primary versus secondary distinction among future research directions, not among the results.
| Question | What this work allows us to say | What it does not allow us to say |
|---|---|---|
| Effect on the negative symptom score | Yes, with an effect size, an interval and a GRADE rating for each subcategory | Nothing about the share attributable to the primary deficit |
| Adjunctive antidepressants | An effect reaching the clinical relevance threshold, moderate certainty | That the effect does not work through improvement of a depressive component |
| Antipsychotics | No category-by-duration combination reaches the threshold | That optimizing a poorly tolerated treatment is pointless: the reduction of secondary negative symptoms is not tested here |
| Ranking of interventions | Nothing: each intervention is compared with a non-active comparator, never with the others | That antibiotics “do better” than antipsychotics |
The practical consequence is simple. This work answers the question “does this intervention move the negative symptom scale compared with a non-active control?” It does not answer the question “does this intervention act on the primary deficit?” The two questions look alike and do not carry the same clinical value.
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| People with a schizophrenia spectrum disorder, with at least 80% of the sample diagnosed according to DSM or ICD with schizophrenia, schizophreniform disorder, schizoaffective disorder or delusional disorder. 451 trials, 42,566 participants, including 25,151 in active arms and 17,415 in control arms. No requirement for predominant or persistent negative symptoms at inclusion | |
| Intervention | |
| Five categories (antipsychotics, other pharmacological agents, brain stimulation, psychosocial interventions, lifestyle interventions), divided into 27 subcategories, 24 of which could be meta-analyzed in the high-quality subset. Most often as adjuncts to antipsychotic treatment | |
| Comparator | |
| Non-active comparators only: no treatment, placebo, sham stimulation, waiting list. No head-to-head comparison between active interventions | |
| Primary outcome | |
| Change in negative symptom severity on a validated scale (PANSS, SANS, BPRS, CAINS, BNSS), expressed as a standardized mean difference. Selection hierarchy when several scales were reported: PANSS negative subscale, then SANS, then BPRS, then BNSS, then CAINS | |
| Design | |
| Random-effects pairwise meta-analysis, PRISMA-compliant, protocol registered on PROSPERO under CRD42024613967. Search of PsycINFO and all databases of the Web of Science platform from inception to November 30, 2024, supplemented by hand-searching the reference lists of previous systematic reviews. Follow-up classified as short (under 7 weeks), middle (7 to 24 weeks) and long (over 24 weeks). Benjamini-Hochberg correction for multiple comparisons. Risk of bias assessed with the Cochrane tool version 1, certainty with GRADE · CEBM 1a | |
| Clinical relevance threshold | |
| Standardized mean difference of 0.457, the unstandardized coefficient of the regression of effect sizes on a one-point reduction in CGI-S, calculated across the 122 trials reporting that scale. Correlation between the two measures moderate to strong, r = 0.566 (95% CI 0.431 to 0.676), about one third of the variance explained. A close value, 0.453, when the calculation is restricted to the 44 high-quality trials. Complementary criterion: a 27% reduction in the negative score from baseline, corresponding to minimal improvement on the CGI-I according to the work of Leucht and colleagues |
Quality control
| Criterion | Status |
|---|---|
| Protocol registration | Solid |
| FindingPROSPERO CRD42024613967, PRISMA compliance stated | |
| Clinical relevance threshold | Solid |
| FindingDerived from the data rather than borrowed from interpretive conventions, and stable when recalculated on high-quality trials only, 0.453 versus 0.457. The method is not new: the authors take it from their 2015 work. What is new is the re-estimation on a corpus of 122 trials. The threshold is estimated on the whole corpus, all interventions combined, then applied to the high-quality subset | |
| Restriction of the main analysis | Solid |
| FindingThe clinically meaningful conclusions rest on the 214 studies of high methodological quality, 19,746 participants, and not on the whole corpus | |
| Sources searched | Caveat |
| FindingPsycINFO and all databases of the Web of Science platform, listed in the supplementary material, supplemented by hand-searching the reference lists of previous systematic reviews. Embase and the Cochrane trials register are not named, and no trial registry appears in the described strategy | |
| Grey literature and language | Caveat |
| FindingResults limited to peer-reviewed journal articles, grey literature excluded. Non-English articles excluded a priori, the authors citing resource costs and methodological work concluding that the resulting bias is small or absent | |
| Risk of bias assessment | Caveat |
| FindingCochrane tool version 1, not RoB 2. For the psychosocial, lifestyle and brain stimulation categories, the blinding criterion was not applied to the overall risk of bias judgment because of design constraints. That criterion nevertheless feeds into the GRADE rating | |
| Measurement scales | Caveat |
| FindingThe PANSS negative subscale and the SANS account for 96.25% of the data. The second-generation scales, BNSS and CAINS, come last in the selection hierarchy, and the authors acknowledge that domain-specific scales were rarely used | |
| Heterogeneity | Caveat |
| FindingThe authors themselves name it as the main limitation of the work (“Heterogeneity was the main limitation.”). It is high in several of the clinically meaningful subcategories | |
| Publication bias | Caveat |
| FindingThe authors note that improvement in control groups varies widely between subcategories, and they raise the possible presence of publication bias. Funnel plots and Egger tests for the 24 subcategories are in the supplementary material. The test for antibiotics is not significant, but with three trials its confidence interval runs from −31.6 to 29.9: it proves nothing, in either direction | |
| Internal consistency of the figures | Caveat |
| FindingThree values in the main text do not match the corresponding supplementary tables: the sample size for transcranial direct current stimulation, 725 in the text versus 318 plus 307, that is 625, in the table, and the heterogeneity of that same subcategory and of antidepressants, 32.98% and 58.90% in the text versus 76.95% and 88.36% in the table. The other four clinically meaningful subcategories match exactly, including confidence interval bounds | |
| Sample overlap | Caveat |
| FindingDespite efforts to control for duplicates, the authors write that partial overlap between samples cannot be fully ruled out | |
| Missing treatments | Caveat |
| FindingClozapine and electroconvulsive therapy could not be meta-analyzed for lack of enough trials against a non-active comparator: they are usually tested against active treatment | |
| Funding and conflicts of interest | Solid |
| FindingPublic and institutional funding: the NextGenerationEU program of the Italian Ministry of University and Research, MNESYS project, for four authors, and grants from the UK National Institute for Health and Care Research and Medical Research Council for a fifth. Publication fees covered by the University of Pavia. No industry funding of the study is declared. Three authors declare conflicts of interest: Boehringer Ingelheim and Teva scientific committees, Janssen research funding and consulting fees from ten companies for the first; employment at HMNC Holding for the second; fees or consulting for Angelini, AbbVie, Boehringer Ingelheim, Lundbeck and Otsuka for the third. The other authors are not mentioned in this section | |
Results
The values below are expressed as magnitudes: they are reductions in negative symptom severity compared with the non-active comparator, in the subset of studies of high methodological quality. They come from the full text and the supplementary tables, consulted on the publisher’s website. The p values reported are those remaining after Benjamini-Hochberg correction.
| Subcategory | Published result |
|---|---|
| Antibiotics (minocycline, D-cycloserine) | SMD 0.95 (95% CI 0.18 to 1.71); p = 0.026; 3 trials, 158 participants; I² = 76.95%; GRADE moderate |
| PEB readingThe largest figure rests on the narrowest base Three trials, an interval whose lower bound almost touches zero, 77% heterogeneity, and a p value that only just survives correction for multiple comparisons. The moderate GRADE rating is surprising given these indicators, and it has an explanation: the only downgrade applied is for inconsistency. The authors themselves recall that the largest minocycline trial, BeneMin, found no benefit, and they specify that it was excluded from the analysis for lack of raw follow-up data | |
| Integrated psychosocial interventions | SMD 0.93 (95% CI 0.53 to 1.33); p < 0.001; 7 trials, 468 participants; I² = 72.78%; GRADE very low |
| PEB readingThe only subcategory also crossing the 27% threshold An improvement of 28.61% from baseline (95% CI 12.11 to 45.11). But certainty is very low, blinding impossible, and the intervention is composite: cognitive behavioral therapy, cognitive remediation, motivational interviewing and family therapy combined, and therefore impossible to separate from one another | |
| Antidepressants (SSRIs, duloxetine, buspirone, reboxetine) | SMD 0.76 (95% CI 0.33 to 1.19); p = 0.002; 14 trials, 834 participants; I² = 58.90% in the text, 88.36% in the supplementary table; GRADE moderate |
| PEB readingThe most transferable result And the most exposed to the primary versus secondary confusion. It is also the strategy that the 2021 European guidance retains as an option. The question of how much of the effect is depressive is not settled by this work | |
| Physical activity | SMD 0.68 (95% CI 0.39 to 0.96); p < 0.001; 8 trials, 434 participants; I² = 50.18%; GRADE very low |
| PEB readingClear effect, very low certainty The rating reflects the impossibility of blinding and the fragility of the designs, not the absence of a signal. The benefit-risk balance remains favorable for other reasons, cardiometabolic ones in particular | |
| Transcranial direct current stimulation | SMD 0.52 (95% CI 0.17 to 0.86); p = 0.007; 14 trials, 725 participants according to the main text and 625 according to the supplementary table; I² = 32.98% in the text, 76.95% in the table; GRADE low |
| PEB readingThe least well documented of the six results 14 trials and a credible sham stimulation as comparator, but the lower bound of the interval falls well below the 0.457 threshold, and two of the published figures, sample size and heterogeneity, do not match between text and tables. This is direct current, not alternating current | |
| Immunomodulators (aspirin, celecoxib, fingolimod, methotrexate) | SMD 0.47 (95% CI 0.26 to 0.67); p < 0.001; 7 trials, 379 participants; I² = 0.00%; GRADE high |
| PEB readingThe highest certainty The smallest effect and the lowest heterogeneity. The point estimate exceeds the threshold by thirteen thousandths, and almost half of the confidence interval lies below it. This is the most consistent result of the work, and it plays out exactly at the edge of clinical relevance | |
| Antipsychotics | No category-by-duration combination reaches the 0.457 threshold. The authors describe the effects of second-generation antipsychotics as statistically significant but subclinical, with high GRADE certainty, one of only three in the entire work |
| PEB readingNot to be read as inefficacy The comparator is non-active, the measure is a global score, and the contribution of antipsychotics to reducing secondary negative symptoms is not isolated | |
Of the 11 category-by-duration combinations that could be meta-analyzed, two cross the threshold, both at middle follow-up, that is between 7 and 24 weeks: lifestyle interventions (7 trials, 403 participants, SMD 0.72, 95% CI 0.41 to 1.03, I² = 52.36%) and other pharmacological agents (89 trials, 6,243 participants, SMD 0.52, 95% CI 0.37 to 0.66, I² = 86.76%). That last heterogeneity value, close to 87%, indicates that the category lumps together things that do not have the same effect, which is precisely why the subcategories exist.
Critical appraisal
| Domain | Judgment |
|---|---|
| Primary versus secondary distinction | Major limitation |
| FindingAcknowledged by the authors. No selection on predominant or persistent negative symptoms, no reported mediation analysis on positive symptoms, depression or extrapyramidal effects. This is the limitation that caps the clinical reach of the whole work | |
| Scale generation | Caveat |
| FindingThe PANSS negative subscale and the SANS make up 96.25% of the data. These first-generation instruments mix diminished expression with avolition and include items sensitive to drug-induced slowing. The BNSS and the CAINS, built to separate the two factors, come last in the selection hierarchy. A sub-analysis of the 21 studies, 4.78% of the corpus, that used both scales finds similar effect sizes, 0.412 versus 0.373, which is reassuring about agreement between these two instruments, not about their construct validity | |
| Architecture of the synthesis | Caveat |
| FindingPairwise meta-analysis, not network. Each intervention is compared with a non-active comparator, in different populations, durations and settings. Sorting effect sizes in descending order produces an apparent ranking that has no comparative value | |
| Antibiotics result | Fragile |
| FindingThree trials, 158 participants split 79 and 79, I² at 77%, interval from 0.18 to 1.71, p at 0.026 after correction. The subcategory also groups two distinct mechanisms, minocycline and D-cycloserine, a partial agonist at the glycine site of the NMDA receptor. There is nothing here on which to base a practice | |
| Mismatch between GRADE and apparent fragility | Major limitation |
| FindingAntibiotics receive moderate certainty with three trials and 77% heterogeneity, while physical activity receives very low certainty with eight trials and 50% heterogeneity. The supplementary material gives the exact mechanics: the imprecision domain is downgraded only if the confidence interval crosses both the relevance threshold and zero, and by two levels if it crosses the threshold in both directions. Because the lower bound for antibiotics stays at 0.18, above zero, no downgrade for imprecision is applied despite three trials and 158 participants. Sample size never enters the calculation. The same document sets that threshold at 0.472 in this domain, a value that appears nowhere in the main text, and transcranial stimulation is downgraded for inconsistency even though the heterogeneity the text attributes to it, 32.98%, is below the 50% cut-off the authors set for themselves. A reader who takes GRADE labels as an overall reliability ranking will be misled | |
| Time horizon | Caveat |
| FindingThe two clinically meaningful combinations are at middle follow-up, between 7 and 24 weeks. Negative symptoms are, by definition, a chronic problem. The subcategories are not stratified by duration: their estimates pool all time horizons, which makes it impossible to know whether the effect of the six persists beyond 24 weeks. The only two long follow-up combinations that could be analyzed, other pharmacological agents and psychosocial interventions, do not cross the threshold | |
| Missing treatments | Caveat |
| FindingClozapine and electroconvulsive therapy absent from the analysis for methodological reasons. Their absence from the tables is not a judgment on their efficacy | |
| Consistency with earlier literature | Solid |
| FindingThe 2015 meta-analysis of 168 trials concluded that no effect reached the clinical relevance threshold, and it is the source of the CGI-S anchoring method used here. The corpus has nearly tripled, and six subcategories now cross the re-estimated threshold. The progress is real, and it concerns add-on interventions, not baseline treatment | |
It is worth being precise about what these data establish. What is demonstrated: in high-quality trials against non-active comparators, six add-on categories produce a reduction in the negative symptom score that exceeds a CGI-S-derived threshold, with certainty levels ranging from very low to high. What is suggested: that add-on strategies (antidepressant, exercise, integrated rehabilitation) have a place that antipsychotic treatment alone does not cover. What remains expert opinion: that the neuroinflammatory hypothesis finds support here. The consistency of the immunomodulators, with zero heterogeneity and high certainty, makes that reading attractive, but seven trials and 379 participants are not enough to establish it, especially for an effect with almost half of its confidence interval below the relevance threshold.
Level of evidence
PEB assessment: high confidence that immunomodulators have an effect of small magnitude, moderate confidence in the effect of adjunctive antidepressants, low to very low confidence for physical activity, transcranial direct current stimulation and integrated psychosocial interventions, low confidence in the antibiotics result despite its moderate rating, and complete uncertainty about how much of any of these effects reflects the primary deficit. Re-estimating the clinical relevance threshold on a corpus of this size is the contribution that will outlast the individual results. What limits the reach of the work is not the rigor of the synthesis but the nature of the material: first-generation global scales, short trials, non-active comparators and a causal question left open by design.
The colleague test
What an experienced colleague would say after a two-minute summary of this study, between two appointments.
“I’m keeping the 0.457 threshold. It’s useful and I’ll use it elsewhere. Six options above it, fine, but every one of them is something you add on, and I still can’t tell whether I treated a deficit or a depression that looks like one. Antibiotics on three trials and 158 patients, I won’t mention to anyone. An add-on antidepressant and exercise, on the other hand, I can start tomorrow.”
In practice: faced with any study on negative symptoms, the first question is not “how much did the score fall?” but “what, within the score, actually fell?” The same question works for depression, ADHD and cognitive disorders.
What you can do with this on Monday morning. This work does not overturn prescribing, but it gives structure to the approach when a patient is fading.
- Look for secondary causes before adding anything: residual positive symptoms, drug-induced extrapyramidal symptoms or akinesia, sedation, comorbid depression, the effects of an understimulating environment, substance use. The 2021 guidance from the European Psychiatric Association explicitly recommends optimizing antipsychotic treatment to avoid secondary negative symptoms related to adverse effects and positive symptoms.
- Document the starting point with an instrument, ideally a second-generation one, BNSS or CAINS, which separates diminished expression from avolition. It is the only way to know, three months later, what has changed. The PANSS negative subscale remains acceptable and remains the reference in the literature.
- An add-on antidepressant is a reasonable and documented option, supported here with moderate certainty. The European guidance retains it at grade B, but in a specific order: it comes in if negative symptoms do not improve after antipsychotic treatment has been optimized, and it should be stopped if it brings nothing for negative symptoms or depression, to avoid polypharmacy. In France this is off-label for that purpose, and it is worth checking the status in your own jurisdiction; in either case, record the rationale in the chart, set an outcome criterion and a review date. An improvement that works through the depressive component is still an improvement for the patient, but it should not be presented as an effect on the primary deficit.
- Prescribe physical activity without waiting for high-certainty evidence on this target. GRADE certainty is very low here, but the measured effect is clear, tolerability is excellent and the cardiometabolic benefit is established elsewhere, in a population with reduced life expectancy.
- Do not prescribe minocycline, celecoxib or aspirin on the basis of these data. The antibiotics result rests on three trials and 158 participants, and the largest minocycline trial in recent-onset psychosis is negative, without having been included in the analysis. The immunomodulator result is consistent but sits exactly at the edge of clinical relevance, on seven trials.
- Refer toward what structures the day. The subcategory with the best result in relative improvement from baseline is not a single technique but an integrated program combining cognitive behavioral therapy, cognitive remediation, motivational interviewing and family therapy, often in groups and over a long period. Taken separately, cognitive remediation and social skills training do not cross the threshold in this work. Certainty remains very low.
- Take the threshold with you. A standardized mean difference of 0.457 corresponds to one CGI-S point. This benchmark was built on negative symptom trials in schizophrenia and applies to them first, but the approach, anchoring an effect size to a scale clinicians use every day, carries over to other fields.
- The course of action is set out in the NICE decision tree for depression in adults.
Frequently asked questions
Does this study show that antibiotics work better than antipsychotics for negative symptoms?
No, and that reading is methodologically impossible. This is a pairwise meta-analysis: each intervention is compared with a non-active comparator, never with another intervention. Setting side by side effect sizes from different populations, durations and comparators is not a comparison. On top of that, the antibiotics result rests on three trials and 158 participants, with 77% heterogeneity.
Is there an approved medication specifically for negative symptoms?
Not for negative symptoms in isolation, but an authorization can cover them by name. France gives a concrete example. The summary of product characteristics of amisulpride, in its version updated on January 7, 2026, lists an indication for acute and chronic schizophrenic disorders with positive and/or negative symptoms, and states that this includes patients in whom negative symptoms are predominant, with a dosage of 50 to 300 mg per day for that situation. The product is also marketed and reimbursed there, but authorization, marketing and reimbursement are three separate questions, and the clinically relevant one is the first. By contrast, none of the add-on strategies assessed in this meta-analysis (antidepressants, anti-inflammatory drugs, tetracyclines, transcranial stimulation) holds a French indication for negative symptoms. The example is French, the structure of the problem is not: check what your own regulator has authorized, and in which wording, before assuming an option is or is not approved.
Why does the distinction between primary and secondary negative symptoms change everything?
Because it decides the treatment. Social withdrawal secondary to persecutory delusions, to antipsychotic-induced akinesia or to depression calls for optimizing the current treatment, not for an extra drug. The primary deficit is what remains once all those causes have been treated, and that is where the toolkit is empty. A global scale does not tell the two apart, and the authors write that their work was not designed to do so.
How much is the 0.457 threshold worth?
It is the most solid contribution of the work, provided it is not credited with a novelty it does not have: the method comes from the same group’s 2015 meta-analysis and is reapplied here to a corpus three times larger. Instead of using the usual conventions for interpreting effect sizes, the authors regressed standardized mean differences on the reduction in CGI-S score in the 122 trials reporting that scale, and kept the value corresponding to one point of improvement. The link between the two measures is real but partial, r = 0.566, about one third of the variance. Recalculated on high-quality trials only, the threshold barely moves, 0.453. Two limits remain: the CGI-S is a clinician’s global impression, not a measure of functioning reported by the patient or their family, and the threshold describes an average group shift, not the minimal important difference for an individual patient.
Why do immunomodulators get high certainty while physical activity gets very low certainty?
GRADE certainty depends on risk of bias, consistency across studies, precision and directness of the evidence. Immunomodulator trials are double-blind and placebo-controlled, and perfectly consistent with one another, with zero heterogeneity. Physical activity trials cannot be blinded, which here earns them a two-level downgrade for risk of bias and structurally penalizes their rating. The number of participants does not enter the calculation as the authors defined it: their precision criterion is triggered only if the confidence interval crosses both the relevance threshold and zero. That is why three trials are enough for antibiotics to keep moderate certainty. Very low certainty does not mean the effect is absent; it means the estimate could be substantially different if better trials existed.
What should we make of the absence of clozapine and electroconvulsive therapy?
They could not be analyzed for lack of enough trials against a non-active comparator, since these treatments are usually compared with active treatment. Their absence from the tables is a consequence of the inclusion criterion, not a result. It illustrates a general limitation of placebo-controlled syntheses: the best-established treatments are the least represented.
Annotated bibliography
Damiani S, Stefanelli R, Fortea L, D’Imperio A, Calò M, Casarini F, Crippa A, Esposito CM, Leggi R, Orlandi M, Patron S, Peviani A, Piccolo A, Provenzani U, Santilli F, Spallarossa C, Papanastasiou E, Cella M, Patel R, Solmi M, Galderisi S, Leucht S, Stahl D, Radua J, Fusar-Poli P (2026). Interventions for negative symptoms in schizophrenia: efficacy and clinical interpretability in a meta-analysis of 451 randomized controlled trials. Molecular Psychiatry, 31(8), 4259-4269. DOI 10.1038/s41380-026-03543-1 · PMID 41872518. Source study analyzed here. Received September 7, 2025, accepted March 10, 2026, published online March 23, 2026. PROSPERO protocol CRD42024613967, search of PsycINFO and all databases of the Web of Science platform from inception to November 30, 2024, Creative Commons Attribution 4.0 license. Full text, Table 1 of GRADE ratings and three supplementary files consulted on the publisher’s website: GRADE downgrading methods, results tables by subcategory for the three pre-post correlations tested, funnel plots and Egger tests. Italian public funding (NextGenerationEU program and MNESYS project) and UK funding (National Institute for Health and Care Research and Medical Research Council), publication fees covered by the University of Pavia. Conflicts of interest declared by three of the twenty-five authors, two with the pharmaceutical industry and one involving employment at a biotechnology company. No erratum or correction attached at the time of verification.
Fusar-Poli P, Papanastasiou E, Stahl D, Rocchetti M, Carpenter W, Shergill S, McGuire P (2015). Treatments of negative symptoms in schizophrenia: meta-analysis of 168 randomized placebo-controlled trials. Schizophrenia Bulletin, 41(4), 892-899. DOI 10.1093/schbul/sbu170 · PMID 25528757. The essential point of comparison, signed by part of the same group. It concluded that although statistically significant effects existed, none reached the threshold for clinically significant improvement. The reach of the 2026 work is measured against that conclusion.
Galderisi S, Kaiser S, Bitter I, Nordentoft M, Mucci A, Sabé M, Giordano GM, Nielsen MØ, Glenthøj LB, Pezzella P, Falkai P, Dollfus S, Gaebel W (2021). EPA guidance on treatment of negative symptoms in schizophrenia. European Psychiatry, 64(1), e21. DOI 10.1192/j.eurpsy.2021.13 · PMID 33726883. Source of the practical recommendations cited in the box: optimizing antipsychotic treatment to prevent secondary negative symptoms, an add-on antidepressant as an option, social skills training, cognitive remediation when cognition is impaired, physical exercise. A learned society guidance document, hence an expert synthesis with levels of evidence, not a systematic review with meta-analysis. Silvana Galderisi, first author of this guidance, is also a co-author of the meta-analysis analyzed here: the two sources are not independent of each other.
Deakin B, Suckling J, Barnes TRE, et al. (2018). The benefit of minocycline on negative symptoms of schizophrenia in patients with recent-onset psychosis (BeneMin): a randomised, double-blind, placebo-controlled trial. The Lancet Psychiatry, 5(11), 885-894. DOI 10.1016/S2215-0366(18)30345-6 · PMID 30322824. The largest controlled trial of minocycline, cited by the authors themselves. Negative conclusion: minocycline had no effect on negative symptoms or on biomarkers of inflammation or neurodegeneration, and the results do not support its use as an add-on within two to five years of illness onset. The authors of the meta-analysis state that they could not include it, for lack of raw follow-up data. To be set systematically against the pooled result from three trials.
Haute Autorité de santé (French National Authority for Health). REAGILA (cariprazine), avis de la Commission de la transparence du 6 février 2019. has-sante.fr. Opinion no. 3 of the Transparency Committee, examined on September 5, 2018 and dated February 6, 2019 after a hearing of the manufacturer. Actual clinical benefit rated moderate, improvement in actual clinical benefit at level V. The marketing authorization indication recalled there is the treatment of schizophrenia in adults, with no mention of negative symptoms, wording identical to section 4.1 of the summary of product characteristics published by the European Medicines Agency. The claimed superiority over risperidone in predominant negative symptoms was not accepted by the Committee. Consulted on August 11, 2026. Document in French.
Agence nationale de sécurité du médicament et des produits de santé (ANSM, French National Agency for Medicines and Health Products Safety). AMISULPRIDE VIATRIS 100 mg, comprimé sécable, résumé des caractéristiques du produit, mise à jour du 7 janvier 2026. French public medicines database (base de données publique des médicaments). Source of the French regulatory example cited in the context section and in the frequently asked questions. A marketed product, reimbursed at 65% in France. Section 4.1 explicitly includes patients with predominant negative symptoms, section 4.2 sets the corresponding dosage at 50 to 300 mg per day. Consulted on August 11, 2026. Document in French. Marketing authorization, actual marketing and reimbursement are three distinct notions, and they vary from one country to another: they happen to be combined here, which is not the case for all the products cited in this analysis.