Methods	Allocation: random, no further details. Blindness: double, double-dummy protocol. Duration: 29 weeks. Design: parallel. Location: multicentre.
Participants	Diagnosis: (DSM-IV) schizophrenia (n = 93) or schizoaffective disorder (n = 14), treatment resistance, moderate severity score on BPRS or SANS. N = 107. Sex: 84 M, 23 F. Age: 18-60 years (mean = 42 years). History: duration ill n.i., age at onset mean clozapine = 23 years, mean risperidone = 22 years. Setting: in- and outpatient.
Interventions	Clozapine: flexible dose. Allowed dose range: 500-800 mg/day. Mean dose: 456.7 mg/day. N= 53. Risperidone: flexible dose. Allowed dose range: 6-16 mg/day. Mean dose: 6.8 mg/day. N = 54
Outcomes	Leaving the study early: any reason, adverse events, inefficacy. Global State: CGI. General Mental State: BPRS total score, SANS. Cognitive functioning: Spatial working memory performance, social skill and problem solving assessment. Adverse effects: TESS, laboratory (hematology). Unable to use: SANS (modified version).
Notes
Risk of bias
Bias	Authors’ judgement	Support for judgement
Adequate sequence generation?	Unclear risk	Random, no further details.
Allocation concealment?	Unclear risk	No further details.
Blinding? subjective outcomes	Unclear risk	Double, double-dummy protocol. Whether blinding was successful has not been examined, but the compounds differ quite substantially in side effects. This can be a problem for blinding
Blinding? objective outcomes	Low risk	Objective outcomes such as laboratory measures or death are unlikely to have been much affected by problems of blinding
Incomplete outcome data addressed? All outcomes	High risk	The overall attrition was high 63.9%. The analysis was based on mixed effect models, but it is unclear whether any statistical method can account for such a high dropout rate
Free of selective reporting?	High risk	Outcome reporting was incomplete. Quote: “…adverse events data have been reported elsewhere (unpublished study of N.R. Schooler et al.)
Free of other bias?	Low risk	Review authors did not find other sources of bias. Probably yes