Drug induced parkinsonism defined as patients fulfilling 1 or more of the following criteria: mean Simpson Angus Scale (SAS) score >0.33 in patients with baseline mean SAS ≤0.33; start of anticholinergic medication; or marked increase of total SAS ratings of ≥2 in patients with baseline positive rating for EPS.

Serious risk of bias: described as not blinded – downgraded 1 level

Serious risk of indirectness: 1) heterogeneous sample of diagnoses −28% of olanzapine arm and 26% of aripiprazole arm had schizophrenia spectrum disorders,(rest of the population were a mix of mood spectrum disorders (50% in olanzapine arm vs 41% in aripiprazole arm) and aggression spectrum disorder (22% in olanzapine arm vs 33% in aripiprazole arm) 2) Comorbidities – ADHD (44% of olanzapine arm vs 41% of aripiprazole arm), OCD (5% of olanzapine arm vs 5% of aripiprazole arm), SUD (22% of olanzapine arm vs 9% of aripiprazole arm) 3) Co-medications – psychostimulants (16% of olanzapine arm vs 21% of aripiprazole arm), antidepressants (17% of olanzapine arm vs 26% arm), mood stabiliser (43% of olanzapine arm vs 21% of aripiprazole arm) - downgraded 1 level

Single study analysis

Serious imprecision as the 95% CIs are wide and crosses over the default appreciable benefit (0.75) - downgraded 1 level

Dyskinesia defined as treatment emergent dyskinesia in patients without dyskinesia in a prior rating, or as an increase of ≥2 on the abnormal involuntary movement scale in patients with dyskinesia at baseline

Very serious imprecision as the 95% Cis are wide and crosses over both the default appreciable benefit and harm (0.75 and 1.25) – downgraded 2 levels

Akathisia defined as a rating of >1 on the BARNES Akathisia rating scale (BARS)

Reported as discontinuation due to extrapyramidal side effect

Serious risk of bias 1) blinding not described

Serious indirectness 1) heterogeneous sample of diagnoses 2) cocomintant medictions permitted but numbers not reported

Single study analysis

Serious imprecision as confidence interval cross the default appreciable harm (+0.5 standard deviations)

Serious imprecision as confidence interval crosses the default appreciable benefit

Drug induced parkinsonism defined as patients fulfilling 1 or more of the following criteria: mean Simpson Angus Scale (SAS) score >0.33 in patients with baseline mean SAS >0.33; start of anticholinergic medication; or marked increase of total SAS ratings of ≥2 in patients with baseline positive rating for EPS.

Serious risk of bias: described as not blinded – downgraded 1 level

Serious risk of bias: 1) heterogeneous sample of diagnoses −28% of olanzapine arm and 26% of aripiprazole arm had schizophrenia spectrum disorders,(rest of the population were a mix of mood spectrum disorders (50% in olanzapine arm vs 41% in aripiprazole arm) and aggression spectrum disorder (22% in olanzapine arm vs 33% in aripiprazole arm) 2) Comorbidities – ADHD (44% of olanzapine arm vs 41% of aripiprazole arm), OCD (5% of olanzapine arm vs 5% of aripiprazole arm), SUD (22% of olanzapine arm vs 9% of aripiprazole arm) 3) Co-medications – psychostimulants (16% of olanzapine arm vs 21% of aripiprazole arm), antidepressants (17% of olanzapine arm vs 26% arm), mood stabiliser (43% of olanzapine arm vs 21% of aripiprazole arm) - downgraded 1 level

Single study analysis

Very serious imprecision as the 95% CIs are wide and crosses over both the default appreciable benefit and harm (0.75 and 1.25) - downgraded 2 levels

Dyskinesia defined as treatment emergent dyskinesia in patients without dyskinesia in a prior rating, or as an increase of ≥2 on the abnormal involuntary movement scale in patients with dyskinesia at baseline

Akathisia defined as a rating of >1 on the BARNES Akathisia rating scale (BARS)

Serious risk of bias as allocation to treatment groups not described in detail, blinding not described.

Serious indirectness 1) Heterogeneous sample of diagnoses – schizophrenia spectrum (35% vs 31%), mood spectrum disorders (40% vs 22%), behavioural disorders (12% vs 27%), other diagnoses (14% vs 19%) 2) comedications – antidepressants (32% vs 9%), benzodiazapines (41% vs 25%), mood stabilisers (16% vs 12%) 3) substance use – tobacco (37% vs 36%), alcohol (28% vs 29%), cannabis (30% vs 32%)

Not estimable as number of events is 0

Serious imprecision as 95%CIs wide and cross the default appreciable harm (+1.25)

Carbon 2015: Reported as discontinuation due to extrapyramidal side effect

Noguera 2013: Reasons included adverse reaction (n=2 vs 1); insufficient response (n=5 vs 1); lost to follow up (n=1 vs 1); not available (n=0 vs 1)

Arango 2014: Reasons included lost to follow up (n=2 vs 7), symptom remission (n=5 vs 2), change of treatment (n=1 vs 4), intolerance (n=1 vs 0), nonadherence (n=0vs1), drug withdrawal (n=0 vs 1)

Serious risk of bias – blinding not described, subjects allowed to change treatment during the course of the study, unclear whether doses stated are medians or means as text says medians but table states means.

Serious indirectness 1) Heterogeneous sample of diagnoses at baseline (schizophrenia n=8); schizoaffective disorder (n=5), schizophreniform disorder (n=30), psychotic disorder not otherwise specified (n=42); major depressive disorder with psychotic symptoms (n=12) and bipolar disorder (n=13)

Serious risk of bias as blinding not described

Serious indirectness as 1) heterogenous sample of diagnoses 2) comedications permitted but numbers not reported

Single study analysis

Serious imprecision as confidence interval crosses the default appreciable harm (+0.5 standard deviations)

Allocation to treatment groups not described in detail – 51% were already receiving treatment, unclear whether these subjects carried on with same treatment or not. Study also not blinded.

Serious indirectness: 1) heterogeneous sample of diagnoses (schizophrenia type disorder – 39%; psychotic disorder NOS 38%; depressive disorder 12%; bipolar/manic episode with psychotic symptoms 11%) – breakdown by study arms not reported

Serious risk of bias as allocation to treatment groups not described in detail, blinding not described.

Serious indirectness 1) Heterogeneous sample of diagnoses – schizophrenia spectrum (35% vs 31%), mood spectrum disorders (40% vs 22%), behavioural disorders (12% vs 27%), other diagnoses (14% vs 19%) 2) comedications – antidepressants (32% vs 9%), benzodiazapines (41% vs 25%), mood stabilisers (16% vs 12%) 3) substance use – tobacco (37% vs 36%), alcohol (28% vs 29%), cannabis (30% vs 32%)

Serious risk of bias as allocation to treatment groups and blinding not described. Percentages and numbers reported in study do not match in all cases as denominators are not clearly reported – these have therefore been re-calculated by analyst based on N reported in figure 1 of study at different time points.

Serious indirectness as heterogeneous sample of diagnoses, comedications permitted and substance use.

Serious imprecision as 95%CI crosses the default appreciable benefit (−0.5)

Serious risk of bias as blinding not described, subjects allowed to change treatment during the course of the study, unclear whether doses stated are medians or means as text says medians but table states means.

Serious indirectness 1) Heterogeneous sample of diagnoses at baseline (schizophrenia n=8); schizoaffective disorder (n=5), schizophreniform disorder (n=30), psychotic disorder not otherwise specified (n=42); major depressive disorder with psychotic symptoms (n=12) and bipolar disorder (n=13)

Serious risk of bias as allocation to treatment groups not described, open label, inclusion and exclusion criteria not reported, concomitant medications used not reported.

Single study analysis

Serious risk of bias as blinding not described

Serious indirectness as 1) heterogenous sample of diagnoses 2) comedications permitted but numbers not reported

Allocation to treatment groups not described in detail – 51% were already receiving treatment, unclear whether these subjects carried on with same treatment or not. Study also not blinded.

Serious indirectness: 1) heterogeneous sample of diagnoses (schizophrenia type disorder – 39%; psychotic disorder NOS 38%; depressive disorder 12%; bipolar/manic episode with psychotic symptoms 11%) – breakdown by study arms not reported

Serious imprecision as 95%CIs wide and crosses the default appreciable harm (+0.5 standard deviations)

Serious risk of bias – blinding not described, subjects allowed to change treatment during the course of the study, unclear whether doses stated are medians or means as text says medians but table states means.

Serious indirectness 1) Heterogeneous sample of diagnoses at baseline (schizophrenia n=8); schizoaffective disorder (n=5), schizophreniform disorder (n=30), psychotic disorder not otherwise specified (n=42); major depressive disorder with psychotic symptoms (n=12) and bipolar disorder (n=13)

Serious risk of bias as allocation to treatment groups not described in detail, blinding not described.

Serious indirectness 1) Heterogeneous sample of diagnoses – schizophrenia spectrum (35% vs 31%), mood spectrum disorders (40% vs 22%), behavioural disorders (12% vs 27%), other diagnoses (14% vs 19%) 2) comedications – antidepressants (32% vs 9%), benzodiazapines (41% vs 25%), mood stabilisers (16% vs 12%) 3) substance use – tobacco (37% vs 36%), alcohol (28% vs 29%), cannabis (30% vs 32%)

Very serious imprecision as 95%CIs wide and crosses both the default appreciable benefit and harm (−0.5 and +0.5 standard deviations)

Serious imprecision as 95% CIs wide and crosses the defaultappreciable benefit (−0.5)

Drug induced parkinsonism defined as patients fulfilling 1 or more of the following criteria: mean Simpson Angus Scale (SAS) score >0.33 in patients with baseline mean SAS ≤0.33; start of anticholinergic medication; or marked increase of total SAS ratings of ≥2 in patients with baseline positive rating for EPS.

Serious risk of bias: described as not blinded – downgraded 1 level

Serious risk of bias: 1) heterogeneous sample of diagnoses −28% of olanzapine arm and 26% of aripiprazole arm had schizophrenia spectrum disorders,(rest of the population were a mix of mood spectrum disorders (50% in olanzapine arm vs 41% in aripiprazole arm) and aggression spectrum disorder (22% in olanzapine arm vs 33% in aripiprazole arm) 2) Comorbidities – ADHD (44% of olanzapine arm vs 41% of aripiprazole arm), OCD (5% of olanzapine arm vs 5% of aripiprazole arm), SUD (22% of olanzapine arm vs 9% of aripiprazole arm) 3) Co-medications – psychostimulants (16% of olanzapine arm vs 21% of aripiprazole arm), antidepressants (17% of olanzapine arm vs 26% arm), mood stabiliser (43% of olanzapine arm vs 21% of aripiprazole arm) - downgraded 1 level

Single study analysis

Very serious imprecision as the 95% CIs are wide and crosses over both the default appreciable benefit and harm (0.75 and 1.25) - downgraded 2 levels

Dyskinesia defined as treatment emergent dyskinesia in patients without dyskinesia in a prior rating, or as an increase of ≥2 on the abnormal involuntary movement scale in patients with dyskinesia at baseline

Akathisia defined as a rating of >1 on the BARNES Akathisia rating scale (BARS)

Carbon 2015 - reported as discontinuation due to extrapyramidal side effect

Castrofornilles 2008 – discontinued due to side effects, details not reported

Noguera 2013 – reasons included adverse reaction (n=2 vs 5); insufficient response (n=5 vs 11), lost to follow up (n=1 vs 4); other (n=0 vs 1); not available (n=0 vs 1).

Arango 2014 – reasons included lost to follow up (n=2 vs 17), symptom remission (n=5 vs 2), change of treatment (n=1 vs 8), intolerance (n=1 vs 0), drug withdrawal (n=0 vs 4), nonadherence (n=0 vs 2), lack of efficacy (n=0 vs 1), other reasons (n=0 vs 5). Dropouts 3 to 6 months were similar reasons (not due to adverse effects).

Allocation to treatment groups not described in detail – 51% were already receiving treatment, unclear whether these subjects carried on with same treatment or not. Study also not blinded.

Serious indirectness: 1) heterogeneous sample of diagnoses (schizophrenia type disorder – 39%; psychotic disorder NOS 38%; depressive disorder 12%; bipolar/manic episode with psychotic symptoms 11%) – breakdown by study arms not reported

Serious risk of bias – blinding not described, subjects allowed to change treatment during the course of the study, unclear whether doses stated are medians or means as text says medians but table states means.

Serious indirectness 1) Heterogeneous sample of diagnoses at baseline (schizophrenia n=8); schizoaffective disorder (n=5), schizophreniform disorder (n=30), psychotic disorder not otherwise specified (n=42); major depressive disorder with psychotic symptoms (n=12) and bipolar disorder (n=13)

Serious risk of bias as allocation to treatment groups not described in detail, blinding not described.

Serious indirectness 1) Heterogeneous sample of diagnoses – schizophrenia spectrum (35% vs 31%), mood spectrum disorders (40% vs 22%), behavioural disorders (12% vs 27%), other diagnoses (14% vs 19%) 2) comedications – antidepressants (32% vs 9%), benzodiazapines (41% vs 25%), mood stabilisers (16% vs 12%) 3) substance use – tobacco (37% vs 36%), alcohol (28% vs 29%), cannabis (30% vs 32%)

Serious imprecision as 95% CIs wide and cross the default appreciable harm (1.25)

Serious risk of bias, allocation concealment not described, co-morbidities not reported – downgraded 1 level

Serious indirectness 1) heterogeneous sample of diagnoses (schizophrenia spectrum 31% in the olanzapine arm vs 68% in the risperidone arm; affective disorders 69% in the olanzapine arm vs 32% in the risperidone arm) 2) Concomitant medications (benztropine, amantadine, propranolol, lorazepam, initial antidepressant, mood stabiliser) – 88% of olanzapine arm vs 89% of risperidone arm

Single study analysis

Very serious imprecision as the 95% CIs are wide and crosses over both the default appreciable benefit and harm (−0.5 and +0.5 standard deviations) - downgraded 2 levels

Serious risk of bias – randomisation method and allocation concealment not described, open label study. Inclusion criteria not reported, small sample size.

Serious indirectness – use of concomitant medications (100% vs 69%), comorbidities (33% vs 54%)

Serious risk of bias as randomisation and allocation concealment not described. 285 subjects not enrolled – reasons not provided.

Serious indirectness – use of concomitant medication with anticholinergic agents, propranolol, benzodiazepines, mood stabilisers, antidepressants were guided by algorithms.

Serious imprecision as 95% CIs wide and cross over the default appreciable harm (+0.5 standard deviations) – downgraded 1 level

Unclear how akathisia was assessed

Serious risk of bias, allocation concealment not described, co-morbidities not reported – downgraded 1 level

Serious indirectness 1) heterogeneous sample of diagnoses (schizophrenia spectrum 31% in the olanzapine arm vs 68% in the risperidone arm; affective disorders 69% in the olanzapine arm vs 32% in the risperidone arm) 2) Concomitant medications (benztropine, amantadine, propranolol, lorazepam, initial antidepressant, mood stabiliser) – 88% of olanzapine arm vs 89% of risperidone arm

Very serious imprecision as the 95% CIs are wide and crosses over both the default appreciable benefit and harm (0.75 and 1.25) - downgraded 2 levels

Serious risk of bias – randomisation method and allocation concealment not described, open label study. Inclusion criteria not reported, small sample size.

Serious indirectness – use of concomitant medications (100% vs 69%), comorbidities (33% vs 54%)

Serious imprecision as 95% CIs wide and crosses over the default appreciable benefit (0.75)

Sikich 2004 - reasons for dropout: insufficient response (n=2 vs 2); EPS (n=0 vs 1); weight gain (n=0 vs 3); other side effects (n=0 vs 1); noncompliance (n=0 vs 2); moved to remote site (n=0 vs 1)

Mozes 2006 – reasons for dropout: contact lost (1 vs 0); lack of improvement (0 vs 3); severe hyperprolactinemia (0 vs 1)

Sikich 2008: Serious risk of bias as randomisation and allocation concealment not described. 285 subjects not enrolled – reasons not provided.

Sikich 2008: Serious indirectness – use of concomitant medication with anticholinergic agents, propranolol, benzodiazepines, mood stabilisers, antidepressants were guided by algorithms.

Serious imprecision as 95% CIs wide and crosses over the default appreciable benefit (1.25)

¹³

Serious indirectness, comorbid ADHD, ODD, CD (23% vs 33%), comorbid anxiety disorders (8% vs 50%), concomitant medications including valproate sodium, clomipramine hydrochloride, guanfacine hydrochloride, lorazepam, diphenhydramine hydrochloride (92% vs 58%)

Single study analysis

Very serious imprecision as the 95% CIs are wide and crosses over both the default appreciable benefit and harm (−0.5 and +0.5 standard deviations) - downgraded 2 levels

Serious risk of bias as allocation concealment is not described

Serious indirectness as concomitant medications received (lithium, depakoate, mood stabilisers, antidepressants, stimulants, naltrexone – numbers by arm not reported)

Serious imprecision as confidence interval crosses the default appreciable benefit (−0.5)

Serious imprecision as confidence interval crosses the default appreciable harm (+0.5 standard deviations)

Serious indirectness, comorbid ADHD, ODD, CD (23% vs 33%), comorbid anxiety disorders (8% vs 50%), concomitant medications including valproate sodium, clomipramine hydrochloride, guanfacine hydrochloride, lorazepam, diphenhydramine hydrochloride (92% vs 58%)

Single study analysis

Serious imprecision as the 95% CIs are wide and crosses the default appreciable harm (1.25) – downgraded 1 level

Serious risk of bias as allocation concealment is not described

Serious indirectness as concomitant medications received (lithium, depakoate, mood stabilisers, antidepressants, stimulants, naltrexone – numbers by arm not reported)

Very serious imprecision as the 95% CIs are wide and crosses over both the default appreciable benefit and harm (0.75 and 1.25) - downgraded 2 levels

Serious imprecision as the 95% CIs are wide and crosses the default appreciable benefit (0.75) – downgraded 1 level

Shaw 2006: 8/10 subjects in the olanzapine arm for who follow up data were obtained at 2 years dropped out and changed to clozapine due to treatment resistance. 1/10 subjects in the clozapine arm discontinued due to extreme weight gain.

Kumra 2008: reasons for dropout included nonresponse (6 vs 0), treatment emergent neutropenia (1 vs0), excessive weight gain (0 vs 2); treatment non-response (0 vs 1); withdrawal of consent (0 vs 1)

As measured with the UKU scale

Very serious risk of bias 1) allocation concealment not described 2) open label (neither participants/researchers blinded to treatment) 3) poor reporting of data: percentages and denominator for number of events reported in study do not match up, therefore number randomised has been used as the denominator

Serious indirectness 1) heterogeneous sample of diagnosis – schizophrenia (35% vs 33%); bipolar disorder (19% vs 33%); psychosis NOS (33% vs 25%), schizoaffective disorder (25% vs 25%), schizophreniform disorder (17% vs 25%); major depressive disorder (25% vs 25%) 2) Concomitant medications including anticholinergics, beta-blockers, benzodiazepines, antidepressants, antiepileptics, lithium, analgesics, iron compounds, cough medications, all patients were also prescribed risperidone flexible dose 3–5 days prior to randomisation.

Single study analysis

Serious imprecision as the 95% CIs are wide and crosses the default appreciable harm (1.25) – downgraded 1 level

Very serious imprecision as the 95% CIs are wide and crosses over both the default appreciable benefit and harm (0.75 and 1.25) - downgraded 2 levels

Reasons for drop out included loss to follow up (n=4 vs 3); symptom remission (n=1 vs 0); poor response (n=2 vs 0); poor compliance (n=0 vs 1);change to treatment (n=2 vs 0); adverse events (n=0 vs 0)

Serious risk of bias as concomitant medications permitted

Serious indirectness as study includes subjects up to 40 years of age, mean age of 24 vs 22.6 years in intervention and comparator arm respectively

Single study analysis