Systematic review

The systematic review identified a total of 11 studies meeting the inclusion criteria. Of these, 10 studies had a randomised design and 8 included a clear blinding method to reduce risk of bias.

A direct comparison of one type of antipsychotic with placebo was included in 6 of the studies. A further 4 studies compared multiple antipsychotics with each other and 1 study compared different doses of the same antipsychotic. In total, 8 different antipsychotics (paliperidone, ziprasidone, aripiprazole, quetiapine, risperidone, molindone, olanzapine and haloperidol) were identified within the treatment arms of the included studies.

Network meta-analysis

A total of 11 studies were included in the network meta-analysis with secondary outcomes reported less frequently as indicated below. For the one study comparing different doses of the same antipsychotic, the analysis reclassified the lower dose as a placebo arm. The one non-randomised study was also included within the final analysis as it was the only study with haloperidol in a treatment arm. Sensitivity analyses following removal of each of these two studies produced similar results to the final analysis.

From 11 studies, the primary outcome of change in total PANSS scores from baseline to 6 weeks compared to placebo indicated statistically significant improvements in symptoms with:

• Molindone (mean difference in total PANSS score [MD] = −13.51, 95% Credible Interval [CrI] = −24.12 to −1.45).
• Olanzapine (MD= −13.03, 95% CrI= −20.22 to −5.10).
• Risperidone (MD= −11.57, 95% CrI= −16.91 to −5.53).

The largest effect was found for haloperidol (MD= −15.55, 95% CrI= −35.37 to 4.11) however the result was not statistically significant. The remaining 4 antipsychotics also indicated statistically non-significant improvements in total PANSS score compared to placebo for aripiprazole (MD= −6.45, 95% CrI= −14.65 to 2.40), paliperidone (MD= −5.51, 95% CrI= −15.21 to 3.85), quetiapine (MD= −8.20, 95% CrI= −16.84 to 0.71) and ziprasidone (MD= −2.67, 95% CrI= −11.27 to 6.19).

Further analyses of pairwise treatment comparisons did not find any statistically significant differences in total PANSS scores between antipsychotics.

Using the Bayesian approach, haloperidol was ranked with the greatest probability (0.49) to be the best treatment for reducing total PANSS score, molindone (0.25) as the next, followed by the remaining antipsychotics all with a probability less than 0.13.

From 10 studies, the secondary outcome of change in positive PANSS scores from baseline to 6 weeks compared to placebo indicated statistically significant improvements in symptoms with:

• Haloperidol (mean difference in positive PANSS score [MD] = −8.48, 95% Credible Interval [CrI] = −15.60 to −1.28).
• Olanzapine (MD= −3.46, 95% CrI= −6.23 to −0.49).
• Risperidone (MD= −3.20, 95% CrI= −5.58 to −0.87).

The remaining 5 antipsychotics indicated statistically non-significant improvements in positive PANSS score compared to placebo for aripiprazole (MD= −2.22, 95% CrI= −5.73 to 1.42), molindone (MD= −3.44, 95% CrI= −7.64 to 0.91), paliperidone (MD= −1.66, 95% CrI= −5.60 to 2.29), quetiapine (MD= −2.45, 95% CrI= −6.11 to 1.26) and ziprasidone (MD= −1.32, 95% CrI= −4.95 to 2.46).

Further analyses of pairwise treatment comparisons did not find any statistically significant differences in positive PANSS scores between antipsychotics.

Using the Bayesian approach, haloperidol was ranked with the greatest probability (0.85) to be the best treatment for reducing positive PANSS score. Detailed statistical ranking probabilities for the remaining antipsychotics are not provided within the study.

From 10 studies, the secondary outcome of change in negative PANSS scores from baseline to 6 weeks compared to placebo indicated statistically non-significant improvements for all 8 antipsychotic treatments. Whilst the largest effects were found for haloperidol and molindone (both MD= −3.42), haloperidol had the largest amount of uncertainty (95% CrI= −11.55 to 4.65) of all the antipsychotics. Risperidone came closest to achieving statistical significance (MD= −2.87, 95% CrI= −5.32 to 0.08) and ziprasidone was closest to the line of no difference (MD= −0.38, 95% CrI= −4.51 to 3.92).

Further analyses of pairwise treatment comparisons did not find any statistically significant differences in negative PANSS scores between antipsychotics.

Using the Bayesian approach, haloperidol was ranked with the greatest probability (0.40) to be the best treatment for reducing negative PANSS score. Detailed statistical ranking probabilities for the remaining antipsychotics are not provided within the study.

From 10 studies, the secondary outcome of change in weight from baseline to 6 weeks compared to placebo indicated statistically significant weight gain with:

• Olanzapine (mean difference in weight [MD] = 3.96, 95% CrI= 2.43 to 5.44).
• Quetiapine (MD= 2.41, 95% CrI= 0.41 to 4.32).
• Risperidone (MD= 1.49, 95% CrI= 0.30 to 2.82).

Statistically non-significant weight gain compared to placebo was found for aripiprazole (MD= 0.90, 95% CrI= −0.97 to 2.74) and paliperidone (MD= 0.89, 95% CrI= −0.92 to 2.76).

Haloperidol (MD= −1.68, 95% CrI= −5.26 to 1.89), molindone (MD= −1.82, 95% CrI= −3.96 to 0.41) and ziprasidone (MD= −0.09, 95% CrI= −1.93 to 1.73) indicated statistically non-significant weight reduction compared to placebo.

Analyses of pairwise treatment comparisons found statistically significant weight reduction for ziprasidone (MD= −4.06, 95% CrI= −6.31 to −1.73), risperidone (MD= −2.47, 95% CrI= −3.80 to −1.12) and paliperidone (MD= −3.07, 95% CrI= −5.34 to −0.79) compared to olanzapine.

Further analyses of pairwise treatment comparisons found statistically significant weight gain was associated with:

• Risperidone (MD= 3.3, 95% CrI= 1.33 to 5.26), quetiapine (MD= 4.23, 95% CrI= 1.30 to 7.17) and olanzapine (MD= 5.79, 95% CrI= 3.82 to 7.76) compared with molindone.
• Quetiapine (MD= 4.05, 95% CrI= 0.13 to 8.12) and olanzapine (MD= 5.63, 95% CrI= 2.15 to 9.09) compared with haloperidol.
• Olanzapine (MD= 3.08, 95% CrI= 0.71 to 5.45) compared aripiprazole.

Using the Bayesian approach, haloperidol was ranked with the greatest probability (0.50) to be the best treatment for weight change, followed by molindone (0.45) and the remaining antipsychotics all with a probability less than 0.03.

The secondary outcome of all-cause discontinuation was reported in 7 studies and included 5 of the antipsychotics in the treatment arms with placebo. Risperidone indicated statistically significant reduced odds of discontinuation compared to placebo (Odds ratio [OR] = 0.48, 95% CrI= 0.25 to 0.84). Statistically non-significant reduced odds of discontinuation compared to placebo was found for haloperidol (OR= 0.19, 95% CrI= 0.01 to 2.42), olanzapine (OR= 0.50, 95% CrI= 0.22 to 1.24), quetiapine (OR= 0.43, 95% CrI= 0.18 to 1.06) and ziprasidone (OR= 0.58, 95% CrI= 0.26 to 1.32).

Further analyses of pairwise treatment comparisons did not find any statistically significant differences in odds of all-cause discontinuation between antipsychotics.

Using the Bayesian approach, haloperidol was ranked with the greatest probability (0.67) to be the best for continuing treatment, followed by quetiapine (0.16) and the remaining antipsychotics all with a probability less than 0.08.

The secondary outcome of discontinuation due to adverse events was reported in 9 studies and included 7 of the antipsychotics in the treatment arms with placebo. Statistically non-significant increased odds of discontinuation due to adverse events were found for all 7 antipsychotics compared with placebo. Due to the paucity of data in the included studies for this outcome, no further analyses were conducted or reported.

The results suggest that whilst all the antipsychotic treatments indicated improved symptom efficacy compared with placebo, no significant differences were found when compared with one another. Although haloperidol seems to be ranked with the greatest probability of being the most effective treatment, the variation in certainty across results should be noted. A considerable amount of variation was also found for the effects of treatments on weight and rates of discontinuation.

Strengths

The target population in the study is directly relevant to the population in NICE guideline CG155. The study clearly defines the population as children and adolescents under the age of 18 who experience early-onset schizophrenia. The results of this study add to an otherwise limited evidence base as most studies on antipsychotics are conducted in an adult population.

A methodological strength of the study is the inclusion of an extensive range of antipsychotic medication and placebo as comparators. NICE guideline CG155 recommends the use of oral antipsychotics however does not specify particular medications. The comparison of multiple first and second generation antipsychotics in this study offers data on their relative efficacy for this population. Also, the study adequately reported the network of evidence and the inclusion of treatment ranking probabilities to provide further useful information on efficacy.

The primary and secondary outcomes in the study are directly relevant to NICE guideline CG155. The outcomes are good indicators of the efficacy of the included antipsychotic medications and the PANSS scale being a commonly used measure of symptoms in this population. The reporting of weight changes associated with medications gives further strength and relevance to the study as this is one of the most concerning side-effects for this population.

Further methodological strengths consist of the use of an adequate search strategy to identify relevant studies and the use of an appropriate risk of bias tool to assess the quality of the included randomised controlled trials.

To give the network meta-analysis further strength, the study has adequately justified the inclusion of a non-randomised design. The network meta-analysis also accounted for the pooling of data from different doses and the reclassification of a low dose to placebo. Appropriate sensitivity analyses were conducted to test the effect on results when these were excluded.

Limitations

The network meta-analysis includes a number of limitations in its methodology. Although a risk of bias tool was used for RCTs, the study does not specify the tool used for the non-randomised study. Also, the quality assessment of several included studies identified bias from lack of blinding, selective reporting, differences in baseline characteristics and failure to explain differences in drop-out rates.

Variations in baseline gender and mean age characteristics of the included studies indicated heterogeneity. However, a statistical test was not reported to confirm this or to account for the differences. These methodological limitations negatively impact upon the strength of the results and reduce the certainty of the treatment effects.

The authors also highlighted uncertainty in the treatment effects due to the limited number of included studies and small sample sizes. Further uncertainty in the network was created through the number of indirect comparisons and many comparisons being informed by single trials. The authors acknowledged that substantial responses for placebo have been previously reported. The effect of placebo alone was not investigated here which may reduce the reliability of the network meta-analysis results.