An umbrella review of adverse effects associated with antipsychotic medications: the need for complementary study designs

Antipsychotic medications are widely prescribed in psychotic illnesses and other mental disorders. Effectiveness is well-established, particularly for reducing symptoms such as delusions and hallucinations, but can be impacted by tolerability. Adverse effects are wide-ranging, and vary between antipsychotics, which is clinically important. This umbrella review aimed to comprehensively summarise the extent and quality of evidence for adverse effects associated with antipsychotic use in people with mental disorders. We included 32 meta-analyses of randomised trials and observational studies. The overall robustness of reported associations was considered in terms of review quality, heterogeneity, excess significance bias, and prediction intervals. Using this approach, endocrine and metabolic, movement-related, and sedation and sleep problems were the clinical domains with strongest evidence. The overall quality of included meta-analyses was low, and individual adverse effects were not typically examined in meta-analyses of both randomised trials and observational study designs. Future reviews should focus on adhering to methodological guidelines, consider the complementary strengths of different study designs, and integrate clinically relevant information on absolute rates and severity of adverse effects.


Introduction
Antipsychotics remain the drug class with the most robust evidence of effectiveness in psychotic disorders such as schizophrenia (Ceraso et al., 2022).They are also widely used in a range of other mental disorders, including in severe depression and bipolar disorder, and as off-licence treatments for symptom management in personality disorder, dementia, and some anxiety disorders.Altogether, nearly 800,000 individuals are prescribed these medications in England, (NHS Business Services Authority, 2020) and prevalence is estimated at 1.7% in the US, (Dennis et al., 2020) with increasing rates in children and adolescents (Radojčić et al., 2023).
In psychotic illness, first generation antipsychotics (also known as typical or conventional) exert their therapeutic effects most prominently in the reduction of positive symptoms such as delusions or hallucinations, through postsynaptic blockade of dopamine receptors.Second generation (atypical) antipsychotics are characterised by more weighting towards serotonergic modulation (Li et al., 2016).These profiles account for different adverse effect profiles between the two groups.
The clinical effectiveness of antipsychotics in symptom reduction is complicated by a range of potential adverse effects.The traditional division is that for first generation antipsychotics, dopamine receptor occupancy in nigrostriatal pathways is linked to musculo-skeletal problems, whereas the multi-receptor profile of second generation antipsychotics is associated with more metabolic adverse effects.A range of other potential effects, from cardiac effects to hyperprolactinaemia, may differ across individual antipsychotics.Factors such as anticholinergic and anti-histaminergic action have also been demonstrated to influence the adverse effect profiles of antipsychotics (Kaar et al., 2020).However, in contrast to the investigation of efficacy profiles, adverse effects of antipsychotics are not reported consistently enough to allow for a robust classification under the 'Dose, Time and Susceptibility' framework (Aronson and Ferner, 2003).As some adverse effects have serious health consequences, and these medications are widely prescribed, this needs clarification.
An effective way to address such a broad evidence gap is with an umbrella review, which comprehensively summarises and assesses the quality of systematic reviews (Fusar-Poli and Radua, 2018).This can provide new insights and a clearer overview of fields with a large or conflicting literature.Two relevant umbrella reviews on the adverse effects associated with antipsychotic use in adults have previously been undertaken.One reviewed the physical health effects of antipsychotic, antidepressant and mood stabilising medication (Correll et al., 2015).However, this was undertaken when umbrella review methodology was in its infancy, used unclear inclusion criteria, and did not include a quality assessment of included reviews.A second umbrella review of adverse effects of antipsychotics (Papola et al., 2019) included six systematic reviews of 58 observational studies, but its scope was limited to observational studies (not randomised controlled trials [RCTs]) of six life-threatening adverse events.Other umbrella reviews have considered pharmacological interventions more widely in mood disoders (Croatto et al., 2023) and child and adolescent populations, (Solmi et al., 2020) and focussed on metabolic adverse effects in children and adolescents treated with antipsychotics (Carnovale et al., 2023).
The current umbrella review sought to address evidence gaps and limitations in previous work by considering a broader range of all reported adverse effects associated with antipsychotic medication, and by including systematic reviews and meta-analyses of both randomised and non-randomised experimental study designs.The aims were to 1) map and summarise the extent and quality of evidence for adverse effects associated with antipsychotic use, 2) compare risk across adverse effect categories, and 3) identify the key gaps in existing literature for future work to target.

Methods
This umbrella review followed PRISMA harms reporting guidelines (Zorzela et al., 2016) and the PRIOR statement for reviews of healthcare interventions (Appendix 1) (Gates et al., 2022).The protocol was registered on PROSPERO (CRD42020223706).

Data sources
A systematic search was conducted of PubMed, Embase, PsycINFO, Scopus, CINAHL and the Cochrane Library of Systematic Reviews, from inception to 30th July 2023.The search strategy incorporated the names of individual antipsychotic medications and search strings comprising of database-specific indexing terms (e.g.MeSH terms) attached to adverse effect-related subheadings.Study design filters developed by the National Library of Medicine (National Library of Medicine, 2019) and British Medical Journal (BMJ Best Practice, n.d.) were incorporated in the search strategy to identify relevant studies on PubMed and Embase respectively.See Appendix 2 for the full search strategy for all databases.

Record screening and eligibility
We aimed to include systematic reviews with meta-analyses of RCTs or observational studies with a cohort, case-control, nested case-control or cross-sectional design, which 1) reported associations between adverse effects and any of the 32 antipsychotic medications of interest (see below), 2) examined antipsychotic prescription as either monotherapy or combination therapy at any dose, 3) investigated specified adverse effects associated with antipsychotic use in human populations of any age with any psychiatric illness or medical condition, 4) reported adverse effects with study-level data that allowed for the calculation of risks using odds ratios (ORs) or relative risks (RRs), and 5) measured adverse effects as primary or secondary outcome.
Several criteria were used to focus the range of reported adverse effects.First, adverse effects with data obtained from fewer than 3 primary studies were excluded due to the limitations of estimating effect size heterogeneity in these cases (von Hippel, 2015).Second, where appropriate, clinical diagnoses (e.g.extrapyramidal disorder) and their synonymous or associated symptoms (e.g.akathisia, dystonia) were considered as the same outcome.Third, to ensure clinical relevance of the adverse effects considered, observations that were not anchored in diagnoses, or were not of a clinically relevant threshold or defined in a reproducible manner (e.g.increased or decreased appetite, increased or decreased duration of sleep, without further specification) were excluded.We limited the range of cause-specific mortality outcomes to suicidal mortality, cardiac death and sudden death.
We aimed to examine antipsychotic use in predominantly community settings, as these likely represent more stable prescription patterns, whereas treatment in the acute phase may more likely be complicated by polypharmacy or underreporting of adverse effects due to symptom severity.Therefore, meta-analyses which specified they included only inpatient samples, and/or reported non-stratified data that included > 30% primary studies with institutionalised samples were excluded (threshold set a priori through discussion within research team to balance the number of studies included with representativeness of samples).Meta-analyses on surgical or palliative care patients were also excluded as these are distinct populations.
Where multiple meta-analyses of effect sizes for the same antipsychotic medication / adverse effect / study type / population combination were identified, only one of the meta-analyses was included.See Appendix 3 for the process for this selection.This eliminated the potential for there to be any overlap between primary studies (i.e.doublecounting) incorporated in pooled effect-sizes.

Data extraction
Full texts of potentially relevant records were retrieved and screened using a standard template.For each included systematic review and meta-analysis, data extracted included: design of primary studies; number of databases searched; year range of review; number of primary studies; study location; settings; number of participants; participant age; participant diagnoses; type(s) of antipsychotics; adverse effects; effect sizes with 95% confidence intervals; and for reviews of observational studies, the number of cases exposed to the adverse outcome and the number of cases without exposure to the adverse outcome.RC undertook data extraction, which was independently duplicated for a random 20% of included studies by LF and any discrepancies resolved by a third reviewer (DW).

Analysis
To account for high heterogeneity, wherever possible, summary effect estimates and corresponding 95% confidence intervals from included meta-analyses were re-estimated using the Dersimonian-Laird random-effects model.For 9 included meta-analyses (28%), insufficient primary study-level data was reported to allow this re-estimation, and effect sizes were reported directly.For each adverse event, associations with zero events in the exposed and non-exposed groups were excluded during re-estimation.If a meta-analysis reported multiple associations respectively for different types of antipsychotics, doses and/ or patient characteristics, these associations were pooled to calculate an overall summary effect estimate and heterogeneity statistics using the Dersimonian-Laird random-effects model.The respective effect sizes and information on individual antipsychotics, dose-response relationships, duration, and patient characteristics were discussed and investigated via subgroup analyses.
For observational studies, where adverse effect incidence is typically lower, ORs and RRs are approximately equivalent.For RCTs, incidence is typically higher, and so where meta-analyses reported effect sizes other than RRs, if sufficient information was reported this was used to re-estimate effect size as RRs.Conversion between ORs and RRs requires baseline prevalence data, and as this was not typically available, universal conversion was not undertaken.

Quality and credibility assessment
The methodological quality of included reviews was critically appraised using the AMSTAR 2 tool.Of the 16 items, 7 domains are regarded as critical (items 2, 4, 7, 9, 11, 13 and 15) (Shea et al., 2017).Each item was rated "yes, "partial yes" or "no".To summarise quality, an inadequate rating (i.e.scoring "no") on a critical domain was defined as a critical flaw, and on a non-critical domain defined as a non-critical weakness.Based on this, overall confidence in meta-analyses' findings were rated as high (no or one non-critical weakness), moderate (more than one non-critical weakness), low (one critical flaw with or without non-critical weakness) or critically low (more than one critical flaw with or without non-critical weaknesses) (Shea et al., 2017).Quality assessment was undertaken by RC, and checked independently for a random 20% of studies independently by LF, with any discrepancies resolved by a third reviewer (DW).
Two adaptations of the standard AMSTAR 2 criteria were made: 1) for item 7, the inclusion of a full list of references was not practical in most cases, (Hailes et al., 2019) and so this criterion was modified to state that a summary of excluded studies, in the form of PRISMA flowchart or equivalent description in the results section, would suffice; 2) for item 15, investigation of publication bias via graphical or statistical tests was not possible or appropriate for some reviews, and so this criterion was modified to state that specific attempts to identify publication bias would suffice.

Overall evidence consistency and robustness
We used an overall score for consistency and robustness developed for umbrella reviews (Hailes et al., 2019).Each identified adverse outcome was assigned a score on four criteria: between-study heterogeneity (< 50% = 1, 50-75% = 0.5, > 75% = 0), prediction intervals (rejects the null hypothesis = 1, no = 0), excess significance (no = 1, yes = 0), and AMSTAR 2 rating (high = 1, moderate = 0.5, low = 0, critically low = 0).See Appendix 5 for full statistical methods.The four quality scores were then summed to determine an aggregate overall rating within the range of 0-4.Composite scores of 3 or 4 (out of 4) indicate high overall consistency/robustness of evidence for the respective adverse outcome, with scores less than 1.5 indicating low consistency/robustness.

Deviations from protocol
Deviations from the pre-registered protocol are detailed in Appendix 4.

Characteristics of included studies
In total, of 6206 unique records identified, 895 full texts were scrutinized, and 32 meta-analyses were included (Appendix 6).The included reviews were published between 2007 and 2023 and reported on 38,661,668 participants and 1438 associations from primary studies.Eligible systematic reviews provided effect size data for 270 adverse effects (see Appendix 7 for all excluded adverse effects with reasons and Appendix 8 for eligible but excluded studies).Of the 166 associations for which data extraction was double checked, there were only 2 disagreements in effect sizes or confidence intervals, which were clarified.The weighted kappa of quality rating duplication was 0.89.
Eleven (34%) of the included articles were meta-analyses of observational studies, (Correll et al., 2022;Indrakusuma et al., 2022;Liu et al., 2021;Mortensen et al., 2020;Nosè et al., 2015;Papola et al., 2018;Seppala et al., 2018;Vancampfort et al., 2015;Wang et al., 2021;Yang et al., 2018;Zivkovic et al., 2019) reporting data on 15 adverse effects and 262 studies, with sample sizes ranging from 11,789 to over 28,000,000 (total N = 38,350,349), which compared exposed to non-exposed individuals.Primary studies were mostly case-control and cohort designs, with some self-controlled case series, cross-over or cross-sectional designs.Most reviews included non-specified or multiple diagnostic groups, and two were focused on individuals diagnosed with schizophrenia-spectrum or mood disorders.Table 2 presents the characteristics of included reviews of observational studies.

Quality assessment of included studies
Of the total group of 32 included reviews, 4 (13%) were rated high quality according to AMSTAR 2, 5 (16%) were moderate, 12 (38%) were low, and 11 (34%) were critically low.For the 7 AMSTAR 2 critical domains, the lowest ratings (i.e.proportion of studies scoring "yes" in each domain) were for: 44% of included reviews explicitly established a protocol before conducting the review, 50% accounted for risk of bias in primary studies when interpreting and discussing results, and 72% adequately investigated publication bias and discussed its impact on the results of the review (see Fig. 1 for AMSTAR 2 scores across reviews).

Summary of adverse effects data
Included meta-analyses reported on 47 different adverse effects.In RCT meta-analyses, 38 adverse effects were reported, and in observational study meta-analyses 15 adverse effects were reported.Only 5 adverse effects (myocardial infarction, stroke, all-cause mortality, cardiac death, suicide mortality, and pneumonia) were examined in systematic reviews of both RCTs and observational studies (and for pneumonia the review of RCTs consisted of only a single trial).Table 3 summarises the effect size data across included reviews.

Circulatory system
One observational meta-analysis in people with any diagnosis reported associations with pulmonary embolism (OR 3.7, 95% CI 1.2-11.1)and venous thromboembolism (OR 1.6, 95% CI 1.4-1.8)(Liu et al., 2021) compared to those without exposure to antipsychotics.None of the four cardiovascular outcomes examined in RCTs (hypertension, hypotension, myocardial infarction and stroke) were found to

Table 3
Effect size data across adverse effect categories from included reviews of RCTs and observational studies.Cell shading corresponds to overall summary evidence score for consistency/robustness (red, < 1.5; yellow, 1.5-2.5;green, 3-4).*Insufficient data available.RCT, randomised controlled trial; OR, odds ratio; RR, risk ratio; CI, confidence interval; I 2 , heterogeneity of the effect size estimate with cut-offs for low, moderate and high of 25%, 50% and 75% respectively; PI, prediction interval; ESS, excess significance bias.Lower 95% CIs for effect sizes in bold do not cross 1.

Fig. 2.
Pooled effect sizes for adverse effects with a summary quality score of 1.5 or more from meta-analyses of RCTs.
be associated with antipsychotic use, though meta-analyses of observational studies reported associations for stroke and myocardial infarction.
Second generation antipsychotics used off-licence increased the risk of endocrine abnormalities (except diabetes) in RCTs (RR 2.2, 95% CI 1.3-3.8).Among pregnant women, antipsychotic exposure was associated with the risk of gestational diabetes mellitus in observational studies (RR 1.2, 95% CI 1.1-1.4).

Musculoskeletal and movement-related outcomes
Meta-analysis of RCTs of individuals with schizophrenia treated with paliperidone versus placebo demonstrated associations with hyperkinesia (RR 1.6, 95% CI 1.0-2.9)and hypertonia (RR 2.9, 95% CI 1.3-6.5)(Nussbaum and Stroup, 2008).The association with gait abnormalities in an RCT meta-analysis of individuals with dementia treated with second generation antipsychotics also remained significant after re-estimating the original effect size using a random-effects model (RR 2.7, 95% CI 1.7-4.5).None of these outcomes were examined in meta-analyses of observational studies.

Neuropsychiatric outcomes
Eleven different neuropsychiatric outcomes were considered across RCT meta-analyses, none of which were also examined in meta-analyses of observational studies.The strongest associations were with sedation (RR 2.5, 95% CI 2.1-2.9) and somnolence (RR 2.9, 95% CI 2.5-3.3).When stratified by age/population, larger RRs for sedation were generally found for older adults with dementia than adults without dementia.Across both populations, the largest individual effect sizes were Fig. 3. Pooled effect sizes for adverse effects with a summary quality score of 1.5 or more from meta-analyses of observational studies.
for quetiapine (RR 3.7, 95% CI 2.3-6.0 for older adults with dementia and RR 2.9, 95% CI 2.6-3.2 for adults), followed by olanzapine, aripiprazole and risperidone.Significant associations (with effect sizes ranging from 1.7 to 3.6) were also found for hypersomnia in individuals with treatment-resistant depression treated with olanzapine, and for insomnia and agitation in individuals with depression, bipolar depression or mixed affective states treated with aripiprazole or cariprazine.More equivocal effects were found in meta-analyses reporting anxiety, suicidal ideation, suicide attempts, depression, treatment-emergent mania, and seizures (effect sizes ranged from 0.7 to 1.3).

Other outcomes
Other outcomes examined in RCT meta-analyses are summarised in Table 3. Off-label antipsychotic use was significantly associated with fatigue (RR 2.2, 95% CI 1.6-2.9).Olanzapine-fluoxetine in combination versus fluoxetine-placebo was significantly associated with peripheral oedema (reported as a pooled RR of 10).Second generation antipsychotic monotherapy and adjunctive therapy to antidepressants in major depressive disorder were significantly associated with blurred vision (RR 3.1, 95% CI 1.7-5.5).Adjunctive aripiprazole was associated with the highest risk of blurred vision (RR 4.1, 95% CI 1.7-9.8),while no significant effect was found for other antipsychotics.There was an increased risk of urinary incontinence among individuals treated with antipsychotics for psychosis associated with dementia compared to placebo (RR 3.9, 95% CI 1.7-9.1).Meta-analysis of observational studies found a significant OR for pneumonia associated with antipsychotic use in patients with all age groups (OR 1.8, 95% CI 1.6-2.1);this yielded a larger effect size for pneumonia than the re-estimated RCT effect size (OR 0.3, 95% CI 0.0-5.2).The RCT data for pneumonia was extracted from a single trial with a sample of only 517 participants.Therefore, this effect size was excluded from summary assessments.Meta-analysis of observational studies found no association between antipsychotic use and breast cancer (OR 1.1, 95% CI 0.9-1.2).

Subgroup analyses
Most summary effect estimates of outcomes reported in metaanalyses of observational studies had high heterogeneity.Where separate effect estimates were reported in included meta-analyses, these were examined as possible sources of heterogeneity.This was possible for classes of antipsychotics, individual antipsychotics, age, diagnosis, and observational study designs.No statistically significant differences between subgroups were found.Due to the numerous variations in pharmacological and patient variables in RCTs, no subgroup analyses were possible.

Discussion
This umbrella review systematically assessed the current state of research evidence for adverse effects associated with antipsychotic medication.We have synthesized evidence from 32 meta-analyses, which reported on over 1399 associations examined in almost 40 million participants.Umbrella review methodology was employed, including thorough assessment of the quality, consistency, and robustness of meta-analytic findings.We discuss overall conclusions on the validity and scope of the current evidence base, implications for clinical practice, and the direction of future research.
First, the overall quality of included meta-analyses was low, with 23 out of 32 included reviews rated as low or critically low according to a quality rating tool for umbrella reviews, AMSTAR 2. The weakest areas of methodology were the justification of included study designs, a priori establishment of review protocols, and sufficient consideration of risk of bias and heterogeneity.Meta-analyses without pre-registered protocols are especially prone to selective reporting (Ayorinde et al., 2020).Included meta-analyses that did account for risk of bias in primary studies used the Cochrane Risk of Bias tool, Newcastle-Ottawa Scale, and SIGN (Scottish Intercollegiate Guidelines Network) checklist.A few included meta-analyses reported high risk of selective reporting within the primary studies they included.Other sources of potential bias were inadequate matching of participant variables, funding from pharmaceutical companies, and high non-response rate.
Second, meta-analyses of RCTs and observational studies did not typically consider the same adverse effects.These different study designs are prone to specific biases, and a more complete evidence base requires information from both RCT and observational data.For example, the majority of effects examined in meta-analyses of observational studies were found to be associated with antipsychotic use, but most effect sizes had high heterogeneity, and adjustment for confounding within analyses was variable.These findings are likely therefore to be confounded.Case-control studies are particularly liable to potential selection bias, due to the difficulties of identifying a well-matched control group.Observational studies, especially of a cross-sectional design, also typically offer less strong inference of causality.Trial evidence is also prone to selection bias due to narrow inclusion criteria and restrictive recruitment processes.This can limit ecological validity or generalisability, as those entering RCTs are often less unwell than the real-world clinical population to which their findings would apply (Kennedy--Martin et al., 2015).For example, people with comorbidities or elevated suicidal risks are typically excluded from antipsychotic trials.Similarly, trial evidence is focused on short term outcomes (Blonde et al., 2018).Included reviews varied widely in this regard, and the trial period was 12 weeks or shorter in eleven included meta-analyses.While low heterogeneity was found among effect sizes extracted from RCTs, limited representativeness could underestimate heterogeneity in reported outcomes (Kravitz et al., 2004).Furthermore, adverse effects are often the secondary outcome in RCT meta-analyses, meaning their reporting may be less comprehensive compared to the primary outcome (i.e.efficacy).
Third, the differences that we found between observational and RCT meta-analyses suggest a clear role for observational studies in evidence synthesis.The clearest examples were for rarer and more severe outcomes, including mortality, MI, and stroke where the RCT evidence had wide confidence intervals.Complementing these were observational studies, where estimates were more precise owing to larger sample sizes, although the validity of conclusions may be affected by selection bias.In addition to the importance of complementarity, there were adverse effects where there was no RCT meta-analysis, including for fractures, metabolic syndrome, pulmonary emboli and venous thromboemboli, but for which observational studies provided evidence.These are important outcomes to prevent, and in the absence of RCT evidence, findings from observational designs could inform assessment and prevention.At the same time, well-powered feasible RCTs need funding.

Limitations
Despite the strengths of an umbrella review for summarising such a large literature, one limitation is that the defined parameters of the umbrella review (in terms of population, intervention, comparator and outcome) may not precisely match included meta-analyses.To address this, primary studies could be identified that examine relevant adverse outcomes but which are not included in the original meta-analyses.Such an undertaking was however outside the scope of the current umbrella review.
Regarding the quality assessments, AMSTAR 2 was used to assist in identifying high quality reviews, and the number of critical flaws and non-critical weaknesses was used to guide a summary of the confidence in the findings of each meta-analysis.This accepted a "partial yes" score as a positive dichotomous rating, which tends towards nonconservatism.Importantly, however, AMSTAR 2 was one aspect of how the robustness of the underlying review evidence was considered, and we tested other markers of this (including heterogeneity).The overall low quality of evidence suggests caution is warranted on the clinical interpretation of the findings.As there are many ways which quality and robustness of the evidence can be considered, and the way that the data quality is skewed, we chose to present more granular details from the underlying reviews rather than undertaking subgroup or sensitivity analyses.
In this review, we have considered the nature of the evidence for the association between antipsychotics and adverse effects in observational studies and trials.The issue of causality, and the biological underpinnings and causal pathways, is another important aspect of understanding which is not directly addressed by the current synthesis.This requires integration of other types of evidence, including experimental and imaging studies, and restricting findings to only those with these other designs and RCT evidence in support (Kaar et al., 2020).

Clinical recommendations
Study quality information was combined with information on prediction intervals, heterogeneity, and excess significance bias.Of the 47 adverse effects, the overall score for consistency and robustness calculated in this manner was rated as either moderate or high for 32 (68%) adverse effects.When effect sizes are considered in this context, three groups of adverse outcomes present with the most consistent evidence: endocrine and metabolic effects, movement-related outcomes (e.g.hypertonia), and sedation and sleep-related outcomes.Therefore, we would suggest these areas would be particularly relevant for clinical decision-making and discussions with patients about treatment (Achtyes et al., 2018).The review findings allow clinicians to consider the robustness and quality of the research evidence in such discussions, which will help in assisting patients to make the most informed decisions.For those adverse effects with the clearest evidence, discussion of the relative magnitude of the effect sizes summarised in this review can further add to decision-making about treatment options.Part of this discussion will also be information about patient preferences about what adverse effects are most problematic for any particular individual, and further research should gather patient views.For clinical application, it is important to note that some of the adverse effects discussed were only considered in specific described study populations (e.g.falls and fractures in studies of older adults).
A key clinical question not well covered by the included metaanalyses is absolute rates, or the magnitude of risk of adverse outcomes, given that the emphasis is on reporting of risks compared to control populations.In clinical discussion, this information would ideally be set in the context of the severity of the potential adverse outcome (e.g.mild, moderate or severe), individual preferences, and the comparative benefits on risk of relapse and other medication benefits.A less severe side effect may be more acceptable if the risk is similar to that of relapse prevention.At the same time, communicating benefits and risks will also rely on evidence synthesis of relative and absolute risks of adverse effects.This is particularly the case for communicating the risk of the most severe adverse effects, such as increased risk of sudden death which was reported by one meta-analysis of observational studies.

Implications for future research
This umbrella review has identified limitations to the quality of meta-analyses in the area of antipsychotic adverse effects, and future reviews should endeavour to address methodological shortfalls highlighted.Further, choice of study design and need for triangulation between study types should be considered in future reviews.
There was a clear lack of overlap between the adverse effects considered by reviews of observational and randomised studies, and further work should consider these gaps.However, both types of study design have limitations, and so other approaches should also be pursued.For example, confounding by indication and the difficulties of identifying matched control groups in observational studies can be addressed, to some extent, by using within-individual pharamco-epidemiological designs.This design investigates outcome rates in periods where the same individual is exposed or not exposed to medication, thereby controlling for time-invariant confounders.

Conclusion
We have synthesized the meta-analytic evidence from RCTs and observational studies of adverse effects of antipsychotics, and appraised the robustness of reported associations in multiple ways.Endocrine and metabolic, movement-related, and sedation/sleep-related adverse effects were clinical domains with the strongest evidence for their association with antipsychotics.Overall, however, the quality of metaanalytic evidence was low and a number of key evidence gaps remain.Future reviews should focus on adhering to methodological guidelines, consider triangulating across different study designs, and integrate information on absolute rates and relative risks to aid clinical discussion and collaborative decision-making.

CRediT authorship contribution statement
RC completed data screening, data extraction, statistical analysis, and writing (original draft and editing).DW reviewed data extraction, and led on data visualisation, writing (original draft and editing).SF led on conceptualisation, supervision, and writing (review and editing).LF reviewed data extraction, and writing (review and editing).EO and AC writing (review and editing).All authors contributed to methodology and interpretation of results.RC is the guarantor of this review.

Declaration of Competing Interest
None.

Rationale 3
Describe the rationale for conducting the overview of reviews in the context of existing knowledge.

3-4
Objectives 4 Provide an explicit statement of the objective(s) or question(s) addressed by the overview of reviews.

Eligibility criteria 5a
Specify the inclusion and exclusion criteria for the overview of reviews.If supplemental primary studies were included, this should be stated, with a rationale.

5-7 5b
Specify the definition of 'systematic review' as used in the inclusion criteria for the overview of reviews.

5
Information sources 6 Specify all databases, registers, websites, organizations, reference lists, and other sources searched or consulted to identify systematic reviews and supplemental primary studies (if included).
Specify the date when each source was last searched or consulted.

5
Search strategy 7 Present the full search strategies for all databases, registers and websites, such that they could be reproduced.Describe any search filters and limits applied.

Appendix 2
Selection process 8a Describe the methods used to decide whether a systematic review or supplemental primary study (if included) met the inclusion criteria of the overview of reviews.11a Describe the methods used to assess risk of bias or methodological quality of the included systematic reviews.

8-9
11b Describe the methods used to collect data on (from the systematic reviews) and/or assess the risk of bias of the primary studies included in the systematic reviews.Provide a justification for instances where flawed, incomplete, or missing assessments are identified but not re-assessed.

7-8
11c Describe the methods used to assess the risk of bias of supplemental primary studies (if included).

n/a
Synthesis methods 12a Describe the methods used to summarize or synthesize results and provide a rationale for the choice(s).

7-8
12b Describe any methods used to explore possible causes of heterogeneity among results.

8
12c Describe any sensitivity analyses conducted to assess the robustness of the synthesized results.

18
Reporting bias assessment 13 Describe the methods used to collect data on (from the systematic reviews) and/or assess the risk of bias due to missing results in a summary or synthesis (arising from reporting biases at the levels of the systematic reviews, primary studies, and supplemental primary studies, if included).

7-8
Certainty assessment 14 Describe the methods used to collect data on (from the systematic reviews) and/or assess certainty (or confidence) in the body of evidence for an outcome.9

RESULTS
Systematic review and supplemental primary study selection 15a Describe the results of the search and selection process, including the number of records screened, assessed for eligibility, and included in the overview of reviews, ideally with a flow diagram.
10, Appendix 6 15b Provide a list of studies that might appear to meet the inclusion criteria, but were excluded, with the main reason for exclusion.
Appendix 8 Characteristics of systematic reviews and supplemental primary studies 16 Cite each included systematic review and supplemental primary study (if included) and present its characteristics.
Tables 1 and  2 Primary study overlap 17 Describe the extent of primary study overlap across the included systematic reviews.Risk of bias in systematic reviews, primary studies, and supplemental primary studies 18a Present assessments of risk of bias or methodological quality for each included systematic review.
11, Figure 1 18b Present assessments (collected from systematic reviews or assessed anew) of the risk of bias of the primary studies included in the systematic reviews.
11, Figure 1 18c Present assessments of the risk of bias of supplemental primary studies (if included).
n/a Summary or synthesis of results 19a For all outcomes, summarize the evidence from the systematic reviews and supplemental primary studies (if included).If metaanalyses were done, present for each the summary estimate and its precision and measures of statistical heterogeneity.If comparing groups, describe the direction of the effect.
Table 3 19b If meta-analyses were done, present results of all investigations of possible causes of heterogeneity.

15
19c If meta-analyses were done, present results of all sensitivity analyses conducted to assess the robustness of synthesized results.

18
Reporting biases 20 Present assessments (collected from systematic reviews and/or assessed anew) of the risk of bias due to missing primary studies, analyses, or results in a summary or synthesis (arising from reporting biases at the levels of the systematic reviews, primary studies, and supplemental primary studies, if included) for each summary or synthesis assessed.

Discussion
22a Summarize the main findings, including any discrepancies in findings across the included systematic reviews and supplemental primary studies (if included).

16
22b Provide a general interpretation of the results in the context of other evidence.

16-17
22c Discuss any limitations of the evidence from systematic reviews, their primary studies, and supplemental primary studies (if included) included in the overview of reviews.Discuss any limitations of the overview of reviews methods used.

18
22d Discuss implications for practice, policy, and future research (both systematic reviews and primary research).Consider the relevance of the findings to the end users of the overview of reviews, e.g., healthcare providers, policymakers, patients, among others.

Registration and protocol
23a Provide registration information for the overview of reviews, including register name and registration number, or state that the overview of reviews was not registered.

5
23b Indicate where the overview of reviews protocol can be accessed, or state that a protocol was not prepared.Report which of the following are available, where they can be found, and under which conditions they may be accessed: template data collection forms; data collected from included systematic reviews and supplemental primary studies; analytic code; any other materials used in the overview of reviews. .

Appendix 2
Search #4(haloperidol or pimozide or prochlorperazine or trifluoperazine or benperidol or zuclopenthixol or acepromazine or droperidol or loxapine or triflupromazine or perphenazine or thioridazine or molindone or thiothixene or aripiprazole or clozapine or olanzapine or quetiapine or risperidone or amisulpride or flupentixol or levomepromazine or pericyazine or promazine or sulpiride or chlorpromazine or fluphenazine or paliperidone or ziprasidone or asenapine or cariprazine or lurasidone):ti,ab.
Approach to overlapping reviews.
For each adverse effect association reported in reviews of RCTs on the same antipsychotic medication and population, the meta-analysis with the highest quality was selected when multiple meta-analyses reporting effect sizes for the same outcome were identified, to avoid data duplication.When there was a partial overlap or no overlap in the antipsychotic medication and population of these meta-analyses, the largest meta-analysis with the largest number of trials was selected.In circumstances where more than one meta-analysis included the same number of trials, the meta-analysis with more participants was included.Given the more inclusive inclusion criteria of meta-analyses of observational studies, the most recent meta-analysis was selected.When one or more meta-analyses were published in the same year, the meta-analysis with the largest number of studies was selected.

523c
Describe and explain any amendments to information provided at registration or in the protocol.Indicate the stage of the overview of reviews at which amendments were made.financial or non-financial support for the overview of reviews, and the role of the funders or sponsors in the overview

Table 1
Characteristics of included RCT meta-analyses.

Table 1
(continued ) (continued on next page) R.T.S.Chow et al.

Table 2
Characteristics of all included meta-analyses of observational studies.

Table 3
(continued ) *Information unavailable; a Derived from pooling multiple associations on review level when no sufficient primary study data were available (corresponding to e.g.individual antipsychotics, doses, diagnoses), with 95% PIs calculated from review-level heterogeneity statistics.Adverse effect categories adapted from Common Terminology Criteria for Adverse Events (CCTAE) v5.0.CI, confidence interval; I 2 , heterogeneity of the effect size estimate (for meta-analyses which reported the heterogeneity of their primary studies with the Cochran's Q statistic, Q was converted into I 2 ); PI, prediction interval; MAs, meta-analyses.R.T.S. Chow et al.

Table 4
Adverse effects for which summary score for consistency/robustness of effect size data was rated as moderate or high.
paliperidone or ziprasidone or asenapine or cariprazine or lurasidone) OR AB ( haloperidol or pimozide or prochlorperazine or trifluoperazine or benperidol or zuclopenthixol or acepromazine or droperidol or loxapine or triflupromazine or perphenazine or thioridazine or molindone or thiothixene or aripiprazole or clozapine or olanzapine or quetiapine or risperidone or amisulpride or flupentixol or levomepromazine or pericyazine or promazine or sulpiride or chlorpromazine or fluphenazine or paliperidone or ziprasidone or asenapine or cariprazine or lurasidone))) AND ((TI systematic review OR TI meta analysis OR AB systematic review OR AB meta analysis) OR (PT systematic review OR PT meta analysis)).Search strategy for Cochrane Library of Systematic Reviews.

Table A1 (
continued ) (continued on next page) R.T.S.Chow et al.