Fourteen reports of 13 RCTs (1,664 patients) were included, in addition to 17 open prospective studies (480 patients) 11 retrospective studies (1,446 patients), 12 case reports and 5 reports of switching studies.
Methodological flaws included studies that lacked adequate statistical power to detect differences between the treatments, the inclusion of treatment intolerant patients, and the variable definition of refractoriness among the studies.
Randomised double-blind comparisons (14 reports). Clozapine was effective in clearly defined treatment resistance. Clozapine was of greater efficacy than chlorpromazine and, in less well-defined resistance, haloperidol. The evidence relating to risperidone and olanzapine was inconclusive.
Studies with a refractory score of 5 (2 RCTs): clozapine was associated with a significantly greater response rate (30%) than chlorpromazine (4%) in one RCT with 268 patients (P<0.01); the other RCT (18 patients) compared risperidone and haloperidol, but changes in BRPS and clinical outcomes were not stated clearly.
Studies with a refractory score of 4 (2 RCTs): risperidone was associated with a significantly increased number of patients with improved BPRS scores than haloperidol (1 RCT, 67 patients); there was no significant difference in the response rates between olanzapine and chlorpromazine in the other RCT (84 patients).
In other reports, refractoriness was much less well-defined. Trials with a refractory score of 1 or 2 (5 RCTs): 2 RCTs with 60 patients (one involving 21 children and adolescents) showed clozapine to be superior to haloperidol.
Trials with a refractory score of 0 (4 RCTs): clozapine was more effective than haloperidol (1 RCT, 423 patients) and olanzapine was more effective than haloperidol (1 RCT, 526 patients). There was no difference between clozapine and either risperidone (1 RCT, 86 patients) and zotepine (1 RCT, 26 patients).
Prospective open studies (17 studies, including 2 controlled studies).
Clozapine appeared overall to be more effective than risperidone in 2 controlled trials (121 patients). Neither study was adequately powered to detect important differences. All of the other studies were uncontrolled and the findings differed considerably.