Data sources

Existing Cochrane reviews were updated with relevant randomised controlled trials (RCTs) found from comprehensive literature searches. Search strategies focused on retrieving RCTs of atypical antipsychotic drugs and non-randomised studies of rare or long-term adverse events. In addition to extensive database searching, ongoing trial registers were searched and the reference lists of retrieved papers scanned.

A systematic review of cost-effectiveness was undertaken using the same sources. In addition, an economic model was constructed using data from the systematic review of clinical effectiveness.

Inclusion criteria

Data extraction

Two reviewers undertook data extraction independently, any discrepancies being discussed and resolved with reference to the original papers and, if necessary, a third reviewer.

Individuals who left studies early were considered to have had a negative outcome, except in the case of death. The impact of including studies with high attrition rates (25-50%) was analysed in a sensitivity analysis. For studies with greater than 50% attrition, all data were excluded other than the outcome 'leaving the study early'.

A validity assessment for RCTs was undertaken using the following criteria: adequacy of randomisation; adequacy of blinding; comparability of groups at baseline; attrition rate; adequacy of description of withdrawals; adequacy of intention-to-treat data analysis; appropriate dose of comparator drug; adequate washout period.

A validity assessment for non-randomised studies was performed using appropriate CRD checklists.

The validity of existing reviews was summarised using criteria for inclusion in the DARE database.

Data synthesis

For binary outcomes, the pooled relative risk and its 95% confidence interval were calculated for all included RCTs; a fixed-effects model was used.

To investigate the possibility of heterogeneity, a chi-squared test was used, together with visual inspection of graphs. A significance level of < 0.10 was interpreted as evidence of heterogeneity. The studies responsible for the heterogeneity were summated and presented separately, and the possible reasons for the heterogeneity explored.

Data from all included studies were entered, if possible, into a funnel plot (trial effect against trial size) in an attempt to investigate the likelihood of overt publication bias. If possible, reviewers entered data in such a way that the area to the left of the line of no effect in the resulting graph indicated a favourable outcome for an atypical antipsychotic drug.

For each non-randomised study, the main results and aspects of study design were summarised.

Literature search

A total of 171 RCTs were included, of which 28 comprised wholly or partly commercial-in-confidence data from drug manufacturers.

Additional safety data were found in 52 non-randomised studies, of which seven (all relating to sertindole) were commercial-in-confidence. In addition to 31 published economic evaluations, six commercial-in-confidence evaluations were submitted.

Validity

Evidence for the effectiveness of new atypical antipsychotic drugs compared with older drugs was, in general, of poor quality, based on short-term trials and difficult to generalise to the whole population with schizophrenia. Evidence for the effectiveness of new atypical antipsychotic drugs compared with each other was limited, as was evidence for their cost-effectiveness in the UK compared with each other and with older drugs. Thus the conclusions are based on limited evidence and should be treated with caution.

There was no evidence for the effectiveness of atypical versus typical antipsychotic drugs for individuals with concurrent substance abuse problems or comorbid mental illness, such as depression. There are few implications for those with related disorders such as schizoaffective and schizophreniform disorders, other than that ziprasidone, risperidone or olanzapine may be effective.

Effectiveness/safety

Cost-effectiveness

Amisulpride was more effective than haloperidol and, if ziprasidone remains unlicensed, represents the most cost-effective atypical antipsychotic drug.

Clozapine was more cost-effective than haloperidol and appeared from the model to be cost-effective compared with other atypical antipsychotic drugs; however, the cost of weekly blood monitoring was not included and the total cost figure is likely to be significantly higher in practice.

Olanzapine was the cheapest atypical antipsychotic drug but may be less effective than the others (not statistically significant). Some side-effects, such as weight gain associated with olanzapine treatment, were not included in the estimation of quality-adjusted life-years (QALYs), hence the effectiveness of olanzapine may have been overestimated.

Quetiapine was not more cost-effective than haloperidol and, compared with other atypical drugs, it was not a cost-effective treatment option (differences are not, however, statistically significant).

Risperidone had the highest costs after amisulpride but was also associated with higher QALYs than other atypical antipsychotic drugs. It did not appear to be superior and was dominated by ziprasidone.

Sertindole was dominated by chlorpromazine (apart from in final-line therapy for which it has better outcomes). Its costs and QALYs were both lower than those for haloperidol. Sertindole did not seem to be superior to other atypical drugs and is dominated (higher costs and lower number of QALYs) by ziprasidone.

Zotepine was cheaper but less effective than haloperidol. It did not appear to be superior to other atypical antipsychotic drugs.

Cost-effectiveness

Given the uncertainty about the validity of the clinical data for typical antipsychotic drugs and what is an acceptable cost/QALY, it was not possible to reach any definite conclusions as to whether the additional costs and benefits represent value for money.