Effectiveness of Second-generation Antipsychotics: A Naturalistic, Randomized Comparison of Olanzapine, Quetiapine, Risperidone, and Ziprasidone

Erik Johnsen; Rune A Kroken; Tore Wentzel-Larsen; Hugo A Jørgensen


BMC Psychiatry. 2010;10:26 

In This Article


The study represents a naturalistic approach to the issue of effectiveness among first-choice SGAs and which of these should be preferred for a patient suffering from psychosis. About two-thirds were males, fifty-three percent represented first-time admittances, and 44% were antipsychotic drug-naïve. The mean PANSS total score at baseline was 74, range 51–110. The sample thus represents a heterogeneous group of patients with psychosis. The mean daily doses of the SGAs were in the lower end of the therapeutic range with large standard deviations, probably reflecting the relatively high proportion of drug-naïve patients who in general respond to lower doses of antipsychotic drugs.

Global Outcomes

The SGAs performed equally in the ITT analyses regarding times until discontinuation of the first offered antipsychotic drug, until discharge from index admission, and until readmission. Olanzapine-treated FCGs showed a significantly longer time to discontinuation compared with the ziprasidone-treated FCGs in the secondary analyses. Superior drug survival or better adherence for patients treated with olanzapine was also found in the systematic review on head-to-head effectiveness of SGAs, but only in chronic patients.[9,28–30] In one study on chronic patients who had discontinued perphenazine, both olanzapine and quetiapine groups had significantly longer time until treatment discontinuation than risperidone.[30] In the EUFEST study comparing haloperidol with SGAs in first-episode psychosis differences in all-cause discontinuation risk were lower with amisulpride, olanzapine, quetiapine, and ziprasidone, compared with haloperidol.[27] Because our sample consisted of both first-episode and chronically ill patients it seems reasonable that our results regarding drug survival was intermediate between those from chronic phase and first-episode studies. Alternatively the limited N in our study could represent a risk of a statistical type I error because of inadequate power, and we may accordingly have missed further differences among the groups.

Symptom Reduction

The outcomes for symptom reduction were unexpected. Quetiapine was consistently superior for all outcomes except reduction of PANSS negative symptoms and depressive symptoms according to CDSS. The mean CDSS baseline score was rather low, however. The results were similar for both RGs and FCGs, and their validity is further strengthened by the inherent consistency among outcomes on different rating scales, and that similar trends were found in supplemental analyses before and after 90 days. The latter analyses only revealed a few statistically significant differences between drugs, probably because of reduced statistical power in the supplemental analyses. To the authors' best knowledge, this is the first effectiveness study to show such differences among SGAs. In the systematic review on antipsychotic effectiveness the SGAs performed equally regarding their ability to alleviate symptoms of psychosis in all the acute phase studies including studies on first-episode patients, and in all but one chronic phase study.[8,28–40] The latter study found olanzapine to be superior to quetiapine in chronic schizophrenia patients that had previously discontinued an SGA because of intolerability.[29] In one study quetiapine performed better than risperidone on depression outcomes.[36] In the EUFEST study there were no differences between the treatment groups with regards to the PANSS and CDSS scores.[27] There were significant differences for the CGI and GAF scores, and amisulpride had the most favorable and haloperidol the least favorable outcomes in this regard. In the CUtLASS study comparisons between FGAs versus SGAs revealed no differences between the groups with regards to the PANSS, GAF, and CDSS scores.[41] In our study the quetiapine and ziprasidone treated RGs had higher percentages of antipsychotic drug naïve patients, defined as having no life-time exposure to antipsychotic drugs, at baseline compared to the other groups. Hypothetically, this could influence the results as the response to antipsychotics is usually better for first episode patient compared to chronic multi-episode patients. The differences between groups regarding fractions of antipsychotic drug naïve patients were not statistically significant, however, and additional sensitivity analyses revealed essentially the same results. We have not been able to find any differences in baseline demographic or clinical characteristics that could introduce a systematic bias to the results. In the secondary analyses based on FCGs the only significant difference among the drugs was a slightly higher PANSS positive score for the olanzapine group at baseline. As the outcome measure is reduction of PANSS positive score per day, the expected bias could actually be in favor of olanzapine as a higher baseline score has a higher potential for decrease. One could argue that given the naturalistic design with assessments not restricted to the time frame of actual use of the first SGA, the outcomes may not be related to that particular SGA but to subsequent medications. We have, however, demonstrated that about three-quarters of the patients did not change their original SGA, and that there were no differences among groups in the rate of antipsychotic medication changes or the choice of a new antipsychotic agent for those who did change. Furthermore, time until discontinuation was generally the same for all SGAs with the exception of olanzapine- versus ziprasidone-treated FCGs. Nor were there any differences in prescription rates of concomitant benzodiazepines, antidepressants, additional antipsychotics, or mood stabilizers.


The outcomes for tolerability were generally the same across groups. This is consistent with the findings of other effectiveness studies in which the SGAs performed equally on most tolerability outcomes.[9,28–40] The most consistent difference between the SGAs across studies in the systematic review where related to weight gain and adverse influence on cholesterol and triglyceride levels.[8] In the EUFEST study there were only differences between haloperidol, amisulpride, olanzapine, quetiapine, and ziprasidone with regards to akathisia, parkinsonism, weight gain from baseline, and hyperprolactinemia.[27] Whereas there were no significant differences between the drugs with regards to sexual dysfunction, prevalence of overweight, weight gain > 7% from baseline, proportions with hyperglycemia, hypercholesterolemia, low HDL concentration, high LDL concentration, triglyceride concentration, or change from baseline of these metabolic variables, change from baseline of fasting insulin, or proportion with QTc interval prolongation.[27] In the CutLASS study no significant differences were found between FGA and SGA groups.[41] The results may suggest that clear-cut side-effect profiles from premarketing RCTs are less distinct in a naturalistic setting where samples are more heterogeneous and concomitant medications less restricted. Our secondary outcomes on metabolic effects were somewhat surprising as the olanzapine-treated FCG was the only group that had a reduction of triglycerides, and even though patients in all FCGs gained weight and BMI, olanzapine-treated patients did so to a lesser degree than those in the other groups. One explanation may be that there is a high awareness among clinicians of olanzapine-associated metabolic adverse effects and that patients at risk of massive weight gain were identified very early and changed to another antipsychotic agent. Obviously, interactions with concomitant psychotropic drugs may also confuse the picture. The finding of equality among FCGs regarding neurological side effects must be interpreted bearing in mind that there was a significant difference among FCGs in the use of anti-cholinergic drugs, with the risperidone-treated patients having the highest rate of anti-cholinergic prescriptions. The finding of equality among FCGs regarding prolactin elevation should also be interpreted in light of the significantly higher baseline prolactin level in risperidone-treated patients compared with patients in both quetiapine-and ziprasidone-treated groups. The outcomes for autonomic side effects should be interpreted with caution because of their borderline internal reliability.


Some limitations to the study need to be discussed. The randomization was open to both the treating clinician and the patient. Systematic utilization differences among the SGAs before the start of the study could theoretically have introduced bias if some of the SGAs under investigation were associated with more prior experience among the RGs. The direction of such theoretical bias is hard to predict, as both negative and positive prior experiences could influence the attitude towards the SGAs under investigation. We registered prior life-time use of antipsychotic drugs as a "yes" or "no" variable. The registration of antipsychotic drug use in the 12 months prior to study inclusion was limited to whether or not antipsychotics were used, and if so, which antipsychotic drug was used. Even though more detailed information on the duration of this treatment would have added value to the paper; that was not the primary target of the study. There were, however, no substantial differences between the RGs regarding proportions with life-time exposure to antipsychotic drugs or with regards to the agents used in the 12 months prior to inclusion. Furthermore, there were no substantial differences between the randomization groups regarding the proportion who accepted the SGA listed as 1. Theoretically, the open design could also introduce bias if some of the SGAs were more popular among the clinicians or patients. However, we have no clear indications of any trends in the choices of SGAs or later changes.

Even though the exclusion criteria were limited compared with those in RCTs for efficacy, the sample represents only about 30% of those assessed for eligibility which could be a source of selection bias. Others have found the proportion of patients included in clinical trials to be in the range of 7%-14% of those initially assessed.[42–44] At least at the lower end of this range, the inference of trial results to the whole population under investigation can be questioned.

There was a high attrition rate in the study, which could be a source of bias if attrition was significantly different in the treatment groups. However, we found no significant difference in total attrition between treatment groups, and total attrition was not significantly related to baseline characteristics. Also, the primary analyses were ITT analyses based on the randomization groups.

For some of the outcome measures, the sample size may have been too small to detect actual differences among the drugs, resulting in type II errors. This may particularly be true for the survival outcomes. In addition the high rate of dropouts through follow-up led to much censoring resulting in less statistical power. The symptom outcomes and tolerability outcomes are less vulnerable because of the statistical method used.

The naturalistic design aspires to mimic clinical practice in which the antipsychotic treatment is initiated before the diagnosis for a particular patient is specified although for some of the SGAs this represents off-label use. Patients were included consecutively because of psychosis per se, and diagnostic evaluations were performed later by the treating clinicians. Accordingly, the sample is heterogeneous with respect to diagnoses and duration of the psychotic disorder among others, and there were insufficient statistical power to perform secondary analyses in subgroups which limits the inference of trial results to selected sub-populations suffering from psychosis.

Finally, the CDSS was used to assess symptoms of depression. The CDSS is primarily developed to assess depression in patients with schizophrenia, and may not be the optimal tool in assessing depressive features in other diagnostic categories.