The main findings of the present study are that in a consecutive sample of patients acutely admitted to hospital with symptoms of psychosis more than two-thirds had hallucinations at admittance whereas only one-third was hallucinating at discharge/6 weeks. The mean PANSS hallucination score also declined significantly. A substantial reduction of frequency and severity of hallucinations can thus be expected in about half the cases during the first month after hospital admittance. Moreover, differential anti-hallucinatory effectiveness was found in the primary intention-to-treat analyses among the drugs since ziprasidone and quetiapine were both superior to risperidone in reducing the hallucination score, with olanzapine in an intermediate position, when the patients were followed for up to 2 years. The finding is somewhat unexpected as olanzapine and risperidone generally are found to be slightly superior to some of the other SGAs in recent meta-analyses of overall antipsychotic efficacy [21, 22]. The samples of most other randomized controlled antipsychotic drug trials are however highly selected . Our sample is more heterogeneous as reflected by the diagnostic diversity and relatively low mean PANSS scores with substantial inter-individual variation, which may explain the apparent discrepancy of the findings between our study and others. For comparison, Sommer et al.  recently published outcomes for hallucinations from the EUFEST study comparing the effectiveness of haloperidol, olanzapine, amisulpride, quetiapine, and ziprasidone in patients with first-episode schizophrenia. In their total sample of 498 patients, 73% had scores of at least 3 on the PANSS hallucinatory behavior item, which is a comparable proportion to the 68% in our study. The EUFEST subgroup with a score of at least 3 on the PANSS hallucinations item had a mean PANSS hallucinations score reduction from 4.4 to 2.5 during the first four weeks of treatment, and a further reduction to 1.5 after 6 months . Only 8% had hallucinations after 12 months. The corresponding reduction in our study from baseline to maximally 6 weeks was from 3.6 to 2.0 points. One might expect that symptom reduction would be slower in our more heterogeneous sample as response to antipsychotics is generally best in first-episode psychosis patients, with increasing duration of psychotic symptoms with each relapse [25, 26]. At least in the acute phase, no apparent differences in overall reduction of hallucinations were evident between the EUFEST sample  and our sample. With regards to comparisons between the different antipsychotic drugs, haloperidol had the least steep reduction of hallucinations for the groups included in the EUFEST sample, and the difference was largest compared to olanzapine, although not statistically significant after adjusting for multiple comparisons . The trend for superiority of olanzapine in the EUFEST sample is, according to the previous discussion, as expected, whereas the finding of superiority for ziprasidone and quetiapine in our sample is more surprising. Based on visual inspection of the steepness of the growth curves for the individual drugs in the EUFEST, ziprasidone and quetiapine seem to be in intermediate positions among the drugs with regards to mean decrease in hallucination severity . This discrepancy in results may indicate that our findings regarding ziprasidone and quetiapine should be interpreted with caution until replicated. The inferiority of haloperidol in its effect on hallucinations is intriguing in relation to our finding of risperidone as having the least steep reduction for the hallucination score. All currently available antipsychotic drugs antagonize dopaminergic transmission at the dopamine type 2 (D2) receptor, while SGAs are characterized by a lower affinity for the D2 receptor than the first-generation drugs, combined with a relatively stronger serotonergic antagonism at the 5HT2A receptor [27, 28]. Pharmacologically, risperidone is the SGA that resembles haloperidol closest with regards to D2 receptor affinity, both drugs being very potent D2 receptor antagonists [29, 30]. Antipsychotic drug D2 receptor occupancy of 65-70% has been proposed as the optimal therapeutic window for antipsychotic efficacy . However, accumulating evidence indicates a more complex interplay with the dopaminergic system and that additional mechanisms beyond D2 antagonism are at work mediating the antipsychotic drug effects. It was recently demonstrated in a meta-analysis that D2 receptor occupancy accounts for less than 20% of the clinical response variance . Both quetiapine and the SGA prototype – clozapine – have low affinities for the D2 receptor and do not reach receptor occupancies corresponding to the therapeutic window at clinical doses [31, 32]. The inferiority of haloperidol and risperidone found in the EUFEST sample and our sample, respectively, could thus be interpreted as additional support that potent D2 receptor antagonism is not the only mediator of anti-hallucinatory drug effects. With regards to alternative biological substrates of drug effects, a striking feature is the pharmacological heterogeneity of antipsychotics [30, 33]. Moreover, there is emerging evidence from both clinical and preclinical studies suggesting differential effects among the antipsychotic drugs on non-dopaminergic, non-serotonergic drug targets suggesting that differential drug effectiveness is to be expected in clinical samples with schizophrenia [34–40]. The general lack of finding robust differences for antipsychotics may also be attributed to methodological issues, as pointed out by Leucht and collaborators . An advantage of the present study is the use of the pragmatic, randomized design, emphasizing more representative samples and treatment settings than in traditional RCTs of efficacy, among others . A different aspect is that the vast majority of studies report sum scores. However, idiosyncratic symptom profiles from different sub-samples may level each other out in the data of the collected sample. Another advantage of the present study is the specific focus on hallucinations. To the best of our knowledge, clinical antipsychotic drug trials very rarely study drug effects on single psychotic symptoms, and the current results should encourage an increase in more symptom-based studies, which could be extended to other domains, such as gene-screening studies. The present study focuses exclusively on data obtained during the period of actual use of the drugs studied, which should increase confidence in the findings of differential effectiveness. The results are further strengthened by the serum level measurements which revealed serum drug levels within the reference ranges for all the comparators.
Some limitations should be mentioned. The pragmatic design allows for broader inclusion criteria and fewer exclusion criteria than for traditional RCTs of efficacy, but still only 30% of those assessed for eligibility were included in the study. Although this proportion is at the higher end compared to many antipsychotic drug trials, which may include as little as 10% of the population under investigation , there is a risk of selection bias. The direction of the influence of any selection bias related to our results is hard to predict. Another concern is the high attrition rate in our study. Attrition is a major problem in all antipsychotic drug studies and can exceed 40% in studies of only 4 to 10 weeks’ duration . Total attrition was, however, not related to any baseline clinical or demographical characteristics and there were no differences among the randomization groups with regards to mean duration of treatment. The randomization was open to both the attending clinicians and the patients in order to mimic usual clinical practice, but this could theoretically have introduced bias if there were systematic utilization differences among the drugs before the start of the study and some of the SGAs under investigation were associated with more prior experience than the others or were more popular among the clinicians or patients. 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. There were no substantial differences between the randomization groups with regards to the agents used in the 12 months prior to inclusion or in the proportion who accepted the first SGA on the list. About half the sample had life-time exposure to antipsychotic drugs at study inclusion but noncompliance is a common problem in this patient group and a frequent cause of relapse , and most likely only a minority had used antipsychotic drugs according to their prescriptions in the last period of time before admittance. Serum drug levels were not measured at admittance, so the exact figures cannot be verified. The secondary analyses based on the actually chosen drugs failed to find statistically significant differences among the groups.
Our primary analyses were, however, intention-to-treat analyses based on the randomization groups. The secondary analyses do not take advantage of the randomization and are accordingly vulnerable to confounding factors which could have biased the results. There were indeed statistically significant differences among the first-choice groups on the PANSS positive subscale score.