Key Question 1. Disorder-Specific and Nonspecific Symptoms

The findings for symptom improvement are presented for each condition in Table 45. With the exception of studies examining schizophrenia, the evidence comparing FGAs with SGAs and antipsychotics within each class was limited. For most conditions, the majority of the findings focused on the comparison of SGA versus placebo. Comparisons and outcomes for which the evidence was graded as insufficient are not discussed.

A total of 11 studies on pervasive developmental disorders reported measures of symptom improvement. No significant difference was observed between FGAs and SGAs for autistic symptoms. SGAs were favored over placebo for autistic and obsessive-compulsive symptoms, but no difference was found on the CGI. The strength of evidence for these findings was low.

Eight studies reported the effect of antipsychotics on symptoms in ADHD and disruptive behavior disorders. SGAs were superior to placebo on various measures of behavior symptoms and on the CGI (moderate strength of evidence). There was no difference between SGAs and placebo for aggression or anxiety (low strength of evidence).

Eleven bipolar disorder studies reported symptom improvement. SGAs were favored over placebo on the CGI (moderate strength of evidence). Studies showed no significant difference for depression and a significant difference for mania favoring SGAs over placebo (low strength of evidence).

A total of 25 studies reported symptom improvement in patients with schizophrenia or schizophrenia-related psychosis. SGAs were favored over placebo on the CGI and for positive and negative symptoms (moderate strength of evidence). SGAs were significantly favored over FGAs on the CGI (low strength of evidence). No significant difference was found between clozapine and olanzapine or olanzapine and risperidone on the CGI or for positive and negative symptoms (low strength of evidence).

Five studies provided evidence on symptom improvement in Tourette syndrome. SGAs were favored over placebo for tics (moderate strength of evidence).

Four studies examined patients with behavioral issues, such as aggression. One study found greater improvement in autistic symptoms with risperidone than placebo (low strength of evidence).

Key Question 2. Adverse Events

The results for adverse events are summarized by drug comparison across all conditions in Table 46. Although many studies reported on the incidence of adverse events, the evidence on the persistence and reversibility of harms was very limited.

Twelve studies provided adverse events data for FGAs versus SGAs. For extrapyramidal symptoms, SGAs were significantly favored over haloperidol (low strength of evidence). Haloperidol was favored over olanzapine for body composition (low strength of evidence). All other adverse events were not significant (low strength of evidence) or had insufficient evidence.

For all comparisons of different FGAs or FGA with placebo for adverse events, there was insufficient evidence.

A total of 25 studies compared the adverse event profiles of various SGAs. Risperidone was favored over olanzapine for dyslipidemia (moderate strength of evidence). Olanzapine was favored over risperidone for prolactin-related events (moderate strength of evidence). Both quetiapine and risperidone were favored over olanzapine for body composition (moderate strength of evidence).

Adverse events were reported in 36 studies comparing SGAs with placebo. For nearly all outcomes and comparisons, the placebo group experienced significantly fewer adverse events than the group receiving SGAs. One exception was a significant effect in favor of aripiprazole for prolactin-related adverse events (moderate strength of evidence).

Key Question 3. Other Short- and Long-Term Outcomes

The findings for other short- and long-term outcomes are presented separately for each condition in Table 47. The evidence was rated as insufficient to draw conclusions for health-related quality of life, involvement with the legal system, and other patient-, parent-, or care provider–reported outcomes for all conditions. Comparisons and outcomes for which the evidence was graded as insufficient are not discussed.

Short- and long-term outcomes were reported in nine studies examining pervasive developmental disorders and in eight studies examining ADHD and disruptive behavior disorders. Medication adherence was not significantly different between SGAs and placebo for both conditions (low strength of evidence).

Eleven bipolar studies provided data for other outcomes. Medication adherence was significantly better for placebo than for SGAs (low strength of evidence). There was no significant difference between SGAs and placebo for suicide-related behaviors (moderate strength of evidence).

A total of 22 studies provided data on a variety of short- and long-term outcomes for patients with schizophrenia and related psychosis. The studies found no significant difference in medication adherence between FGAs and SGAs, olanzapine and risperidone, or SGAs and placebo (low strength of evidence). Similarly, SGAs and placebo did not differ in suicide-related behaviors (low strength of evidence).

Other outcomes were reported by four and two studies for Tourette syndrome and behavioral issues, respectively. The evidence was insufficient for all of the outcomes and comparisons examined in these studies.

Key Question 4. Subpopulations

A total of 36 studies compared outcomes across various patient subpopulations. Sex and age were examined most frequently. Overall, few studies identified differences in the results across subpopulations. Few associations between the patient or clinical variables and outcomes were supported by more than one study. Studies frequently had discordant conclusions in whether there was a significant association between subpopulations and outcomes and the direction of this association. Two studies reported low intelligence to be associated with a lower treatment response rate. Six studies examined the association between sex and increases in prolactin levels during treatment; the results were discordant across the studies.

Applicability

The majority of the studies were small to moderate-sized RCTs that examined the efficacy of two or more intervention groups. Study populations generally excluded patients with additional diagnoses of psychiatric or behavioral conditions other than the condition of interest, as well as additional comorbidities such as mental retardation, psychosis, or substance abuse. Patients with a history of various adverse events, including tardive dyskinesia, suicide-related behaviors, neuroleptic malignant syndrome, or abnormal lab values, were often excluded.

Additional restrictions that were commonly applied were use of adjunctive medications (e.g., mood stabilizers or antidepressants) and previous unresponsiveness to the study medication. In addition, several studies excluded patients who did not meet minimum response criteria or were nonadherent during the run-in period prior to the double-blind treatment phase. Because patients in clinical practice often have multiple diagnoses and undergo cotreatment with several drugs, these restrictions reduce the applicability of this body of evidence. Exclusion of patients with comorbidities or a history of various adverse events may have overestimated the estimates of the efficacy and underestimated the harm of antipsychotics.

The majority of the studies in this review excluded young adults; therefore, the results may have limited applicability to this population. Young adults were included in approximately 25 percent of studies of schizophrenia, despite the natural history of schizophrenia which typically has its peak onset during these years. Although this population would be included in studies of adults, there are numerous unique issues associated with patients between the ages of 19 and 24, particularly because patients frequently lose access to services once they become legal adults at age 18. For conditions other than schizophrenia, young adults were rarely included.

Applicability may also be limited due to monitoring practices to ensure treatment adherence throughout the trials. In typical practice settings, it is likely that will patients have lower rates of medication adherence and therefore less symptom improvement.

Another factor that restricts the applicability of the studies is the short duration of followup. In particular, the median study duration of 8 weeks is insufficient to assess some long-term efficacy outcomes, such as school performance, maturation, emotional development, legal system interactions, or detect some adverse events, such as development of diabetes and cardiovascular disease-related events. Adverse events were likely underestimated due to the short followup period.

Limitations of the Existing Evidence

The strength of the evidence was low or insufficient for the majority of outcomes across the various drug comparisons and conditions. These low grades were driven by a high risk of bias within individual studies and a lack of consistency and precision among studies.

Although the majority of studies providing data for this report were RCTs, nearly all of the trials had a high risk of bias as assessed using an empirically derived tool developed by the Cochrane Collaboration.³³ Approximately half of the RCTs were rated as having adequately generated the allocation sequence, concealed allocation, and blinded study investigators and participants. Measures employed by the study investigators to ensure that the allocation sequence was truly random and that allocation occurred without foreknowledge of treatment assignments was often unclear in the trials. These features can always be conducted in trials and should be routinely employed in order to avoid selection bias.

Inadequate blinding is another important limitation of this body of evidence, as lack of blinding can lead to exaggerated treatment effects. Many of the outcomes, such as improvement on symptom scales, were assessed by physicians; therefore, blinding should extend beyond patients to include all outcome assessors. Half of the trials had incomplete outcome data due to loss to followup and inadequate handling of missing data in the reporting and analyses, which may exaggerate treatment effects.

The most problematic trial feature was the source of funding: nearly 80 percent of the trials were funded by industry. Empirical evidence has shown that pharmaceutical company sponsorship greatly increases the chances of pro-industry findings. ^34,124 Transparency in reporting the nature and extent of industry involvement in the design, conduct, and analysis of studies may help readers better evaluate the likelihood of industry bias in trials. We also encourage nonpharmaceutical organizations, such as government and advocacy groups, to become increasingly involved in funding future studies in this field in order to determine whether the results of industry trials can be replicated.

A total of 17 cohort studies were included in the review. Because these studies did not employ randomization, they are particularly vulnerable to selection bias and a lack of comparability between intervention groups. Moreover, the majority of the studies did not control for important potential confounders (e.g., age, disease severity) in their design or analyses.

Lack of consistency and precision of results across studies also contributed to the low strength of evidence grades for the majority of outcomes. Consistency was often unknown due to the few studies comparing the same interventions. In addition, lack of consistency is also attributable to the various scales and surrogate measures that were used to assess efficacy outcomes and adverse events. Precision was often poor due to the small sample sizes in many of the studies, which may have resulted in insufficient power to detect differences between groups. Both consistency and precision may have been affected by variations in the clinical populations assessed across the studies, such as the number, type, and severity of comorbidities, patient age, sex ratio, and dose of antipsychotics.

Currently, there is little consensus regarding which outcomes and measures are important to evaluate in studies in this field. Across the 81 studies included in this review, more than 60 outcome scales were reported. Although some outcomes and scales were assessed fairly consistently for some conditions, such as the Young Mania Rating Scale (YMRS) for bipolar disorder and the Positive and Negative Syndrome Scale (PANSS) for schizophrenia, there was great diversity in the scales used in studies for the other conditions. Further, response and remission were based on different outcome measures and criteria across studies. This heterogeneity makes comparisons across studies and interventions challenging. Professional associations in the field of child psychiatry should take a lead role in initiating discussion and consensus on which measures should be included in studies and what degree of change in these measures would be considered a clinically important finding. This information is necessary for designing future research (e.g., planning for adequate sample sizes) and interpreting the findings.

Further, few studies described the use of standardized scales to systematically assess the incidence of adverse events. As such, it is unclear whether studies simply evaluated side effects through passive elicitation by patients or open-ended questions. Studies also rarely reported definitions for what constituted various adverse events (e.g., criteria for metabolic syndrome) and whether adverse events were self-reported or assessed by clinicians. Varied time points at which outcomes were assessed also contributed to inconsistency.

The studies rarely evaluated key outcomes that are important to patients. These include health-related quality of life, social and occupational functioning, involvement with the legal system, and patient-, parent-, or care provider–reported outcomes. The duration of followup was brief in the majority of studies (median of 8 weeks; IQR, 6 to 12 weeks); therefore, long-term data are needed to assess the impact of antipsychotic use on these outcomes. Similarly, longer followup is needed to determine the association of acute adverse effects, such as antipsychotic-induced weight gain, on downstream effects such as diabetes, dyslipidemia, hypertension, and cardiovascular morbidity. Few studies reported on the persistence or reversibility of harms observed during the study period. Although many of the studies included open-label extension phases to assess efficacy or harm data, the majority failed to provide comparative data, precluding evaluation of effects between groups. Providing long-term comparative data for studies comparing an active treatment with a placebo may not be feasible. As such, observational studies are needed to provide data on patients using different antipsychotics over the course of several years to determine the comparative benefits and risks associated with these drugs.

Few discernable trends were noted for the effects of subpopulations on treatment outcomes due to heterogeneity in the patient factors and outcomes examined. The results of subgroup and regression analyses were often poorly described in the studies (e.g., few studies reported whether an association was significant), limiting the conclusions that could be drawn. Information on patient subpopulations that are associated with positive and negative outcomes is crucial for informing clinical practice; therefore, future studies assessing the impact of important subgroups (e.g., sex, age) on key outcomes (e.g., prolactin levels, weight gain) are needed.

This comparative effectiveness review has several limitations. Only English-language studies were eligible for inclusion in the review; therefore, it is possible that relevant studies published in other languages may have affected the review findings. We based our assessments of methodological quality on study publications and did not contact authors to verify the methods used. Some studies may have been adequately conducted, but the methods were poorly reported. The scope of this report was limited to the direct comparison of various antipsychotics and the comparison of antipsychotics with placebo. As such, evidence on the use of other drug classes (e.g., anticonvulsants) that are frequently used in the treatment of these patient populations is not considered in this report.

This report presents a synthesis of the available evidence on the effectiveness and safety of antipsychotics in the pediatric and young adult populations. However, we do not make clinical recommendations on the use of these medications, as this is the purview of the user group. We trust that the evidence presented in this report will be helpful in the further development of clinical practice guidelines in this field.

Future Research

The following general recommendations for future research are based on the preceding discussion regarding the limitations of the current evidence:

• Studies examining long-term efficacy and, particularly, the safety of antipsychotics over the course of several years are needed. Future research should evaluate long-term developmental outcomes, such as growth, maturation, and cognitive and emotional development.
• Future studies should evaluate outcomes that are important to patients and parents, including health-related quality of life, school performance, and involvement with the legal system.
• Studies examining the impact of key patient subpopulations on important outcomes are needed to inform clinical practice. In particular, subgroup analyses examining young adults would be helpful in guiding clinical decisions due to the unique issues associated with this population.
• Future research should seek to minimize risk of bias by blinding study participants and outcome assessors, adequately concealing allocation, and handling and reporting missing data appropriately.
• Consensus on outcomes and outcome measures is needed to ensure consistency and comparability across future studies. Moreover, consensus on minimal clinically important differences is needed to guide study design and interpretation of results.
• Study authors should explicitly disclose sources of funding and the nature and extent of industry involvement in the design, conduct, supply of materials, analysis of outcomes, and reporting of studies.
• Large-scale effectiveness studies that use inclusive patient-selection criteria and closely match typical clinical practice are needed to achieve greater applicability of results.

This review identified several areas for which the evidence is sparse and which are priorities for future research. More than one third of the studies included in this review examined an FGA or SGA with only placebo and no active comparator. Based on the findings of this report, antipsychotics can be considered to have superior symptom improvement and inferior adverse event profiles than placebo. Evidence was insufficient to evaluate the comparative effectiveness of various antipsychotic agents. Future studies should be composed of two or more active treatment groups in order to address this evidence gap.

One of the greatest priorities for future research is the systematic evaluation of adverse events. The evidence for the safety of antipsychotics in the pediatric and young adult populations is sparse, resulting in much uncertainty and controversy regarding the use of these drugs in children. The dearth of safety data is of particular concern given that prescription rates of psychotropic medications among children and adolescents have nearly reached adult proportions, and adverse events may have greater long-term impact in children.³⁷ Although a high proportion of the studies included in this review reported some adverse events, few studies systematically evaluated harms using standardized measures. The use of standardized pediatric side-effect scales such as the Safety Monitoring Uniform Report Form¹²⁵ or a simplified version of this scale¹²⁶ have been recommended for all pediatric psychopharmacologic studies. Guidelines regarding which adverse events should be routinely monitored in studies evaluating the use of antipsychotics in children are available.³⁷ Cohort studies that examine the long-term comparative safety of drugs are also needed. Correll et al.¹¹² provides a good example of a well designed and conducted cohort study in this field.

Although numerous studies have investigated the use of antipsychotics for on-label conditions, such as schizophrenia, evidence for off-label use in treating behavior symptoms associated with conditions such as disruptive behavior disorders or Tourette syndrome is sparse. Future research efforts should examine whether these agents are effective in treating disorder-specific symptoms and how these agents compare with other available interventions.

Trials and cohort studies should be designed and conducted to minimize risk of bias where at all possible. Authors may find tools such as the CONSORT¹²⁷ and STROBE¹²⁸ statements helpful in designing and reporting on randomized controlled trials and cohort studies, respectively.

Conclusions

The efficacy and safety of FGAs and SGAs have been studied in children, adolescents, and young adults (ages≤24 years) for the following conditions: pervasive developmental disorders, ADHD and disruptive behavior disorders, bipolar disorder, schizophrenia, Tourette syndrome, and behavioral problems. Overall, data for head-to-head comparisons (FGAs vs. SGAs, FGAs vs. FGAs, and SGAs vs. SGAs) were generally of low strength of evidence; therefore, few conclusions regarding the relative efficacy and safety of antipsychotics could be drawn. However, evidence consistently showed that SGAs resulted in greater symptom improvement and a poorer safety profile than placebo. Evidence was sparse for several patient important outcomes, such as health-related quality of life, involvement with the legal system, and school performance. Few studies reported long-term data.

Treatment benefit and risks were examined most frequently for schizophrenia. Fewer studies examined other conditions. In addition, no eligible studies examining obsessive-compulsive disorder, post-traumatic stress disorder, or anorexia nervosa. Young adults were rarely examined, particularly for conditions other than schizophrenia. Additional research is needed to assess the treatment efficacy, and particularly, the safety of antipsychotics in these populations. In addition, studies that examine outcomes separately for subpopulations are needed to better predict the effects of treatment for individual patients in clinical practice.

Future research should incorporate design elements to minimize bias in treatment effects including adequate allocation concealment, adequate blinding of patients and outcome assessors, comparability of study groups, and appropriate handling and reporting of missing data. Consensus is needed on outcomes and outcome measures, as well as minimum clinically important differences. Standardized assessment of adverse events, and analysis and comprehensive reporting of subpopulations in future studies will allow for more accurate interpretation of findings across studies as well as greater understanding with respect to the applicability of the findings.