Optimal Doses of Specific Antipsychotics for Relapse Prevention in a Nationwide Cohort of Patients With Schizophrenia

Heidi Taipale; Antti Tanskanen; Jurjen J. Luykx; Marco Solmi; Stefan Leucht; Christoph U. Correll; Jari Tiihonen


Schizophr Bull. 2022;48(4):774-784. 

In This Article


To our knowledge, this is the first real-world study examining relapse risks for different dosages and formulations of maintenance therapy with antipsychotics in individuals with schizophrenia. Overall, our findings indicate that the best outcomes were associated with standard doses (0.9–<1.1 DDDs/day) in nine of the 15 antipsychotics, and with the next lower dose range (0.6–<0.9 DDDs/day) in 4 of the 15 antipsychotics. However, many low doses (<0.6 DDDs/day) and all high doses (≥1.6 DDDs/day) showed higher risk of relapse than standard dose. Consistent with prior meta-analyses of randomized controlled trials, lower[6,7] doses than these two (0.6–<0.9 and especially 0.9–<1.1) dose levels were found to be overall associated with poorer relapse outcomes. Two exceptions to these findings were detected, ie better outcomes for relatively low-dose oral perphenazine and for high-dose olanzapine LAI. Relative to all other antipsychotics, olanzapine LAI at DDD of 1.4–<1.6 (equivalent to 405 mg/4 weeks) was associated with the best outcomes in primary and secondary analyses.

The most striking finding in this study was that two widely used antipsychotics with strong D2-blockade, risperidone, and perphenazine, at >1.6 DDDs/day were associated with 34%–235% higher risk of rehospitalization compared with nonuse of antipsychotics. Concerning perphenazine, even the standard dose of 0.9–1.1 DDD/day, corresponding to 30 mg/day was associated with significantly higher risk of rehospitalization than nonuse of antipsychotics. It is possible that for some patients, the increased risk of relapse during high-dose treatment may be attributable to exceptionally severe symptoms or a recent increase in dose to control the symptoms. The within-subject approach controls much better for such "confounding by indication" than conventional between-subject analyses, but some residual confounding may exist. However, this explanation is unlikely, because this observation was rather specific for perphenazine and risperidone (oral and LAI formulations), while for weaker D2-antagonists (eg, olanzapine, clozapine, quetiapine, or aripriprazole) we did not find this association.[13] These findings could also be in part attributable to breakthrough psychosis[14–16] during ongoing use, or to rebound psychosis after abrupt discontinuation because of extrapyramidal or other side effects.[13] Leucht et al[6] found that in RCTs higher doses than one DDD are associated with more side effects. Thus, patients who receive high doses may not tolerate them, discontinue use, and relapse. Since the vast majority of the prescriptions in Finland are filled for 90 days, hospitalizations are labeled as having happened during the use of the medication in many cases where the medication was discontinued abruptly during this 90-day period. However, a similar dose-response relationship was observed for oral and LAI formulations, which implies that abrupt discontinuation may not be the only or major contributing factor. Another possible explanation for poorer outcome in the higher dose strata is dopamine supersensitivity,[17–21] as well as that the current reference DDD value[9] of 30 mg for perphenazine and 5 mg for risperidone may be too high. On the basis of our results, the appropriate standard dose as DDD would be 18 mg or less for perphenazine, and 3–4.5 mg for risperidone. Perphenazine is a very old antipsychotic and its DDD was defined at a time when higher doses were customary. Indeed, a similar trend was found for zuclopenthixol, for which 0.6 < 0.9 DDD were as good as 0.9–<1.1. A major aim of the WHO DDD classification system is to provide a method to study drug consumption, and it is not meant for recommendations for use or judgements about relative efficacy of drugs[9] – although it is possible that it might have unintentional implications related to these issues among prescribers. If there is a large gap between DDD of a drug vs the actual commonly used dosage, it might be reasonable to adjust the DDD value. Our results showed that the most frequently used dose category for risperidone was <0.6 DDD (less than 3 mg/day), and in our previous study on this nationwide cohort, the median dose for risperidone was 2.8 mg.[22] If these results were replicated in other countries, it would imply that the DDD for risperidone should be 3 mg instead of 5 mg.

The sensitivity analyses stratified based on age, schizoaffective diagnosis, and first-episode status confirmed the results of primary analysis. The results on high-dose perphenazine and risperidone were even more extreme in the first-episode cohort which is in line with guideline recommendations that first-episode patients need lower doses.[23] Certain risperidone and perphenazine doses below 0.9 DDD/day were associated with 21%–45% lower risk of re-hospitalization (P < .001) than the standard dose of 0.9–1.1 DDD/day. When the first 30 days were omitted from the beginning of all antipsychotic use and nonuse periods, these results remained the same (23%–45% difference), indicating that recent dose increase to standard dose due to putative worsening of the symptoms does not explain the findings. Additional head-to-head analyses comparing doses of each antipsychotic with its own standard dose as reference (including only those patients who had used standard dose), as well as head-to-head comparison with high-dose olanzapine as reference (including only those patients who had used high-dose olanzapine) were well in line with the primary analysis, indicating the robustness of the findings.

Another major finding was that relatively high dose of 1.4–1.6 DDD/day (equaling to 405 mg once in 4 weeks) of olanzapine LAI was associated with substantially (>50%) lower risk of re-hospitalization than any dose of any other antipsychotics. This result of our primary analysis was confirmed in between-individual analyses, indicating the robustness of the finding. In addition, in the first-episode cohort, aHR was 0.05 (0.01–0.16) for 1.1–<1.4 DDD/day, and 0.07 (0.03–0.19) for 1.4–<1.6 DDD/day for olanzapine LAI. This finding suggests that olanzapine LAI used at these doses is relatively more effective in reducing rehospitalizations than other antipsychotics and, even more pronounced, among first-episode patients. Moreover, the only randomized controlled dose-finding study of olanzapine LAI also suggested that at the highest dose examined, 300 mg biweekly, a plateau of the dose-response curve had not been reached yet.[6,24] However, the superior effectiveness of olanzapine LAI for relapse prevention must be balanced with the well-known adverse effects of this medication, including postinjection syndrome and metabolic adverse effects.[25–28] Moreover, as indicated before,[22,29–31] overall, LAIs were among the most effective treatments, and among oral compounds clozapine and olanzapine were most effective in standard doses. Antipsychotic polypharmacy is rather common and has recently become more acceptable option[32] but, due to enormous complexity of DDD analyses on a large number of specific antipsychotic combinations, we did not investigate this issue.

Commonly used antipsychotics were used with very different dose ranges in this real-world cohort of patients with schizophrenia. Oral olanzapine was commonly used in high dose (>1.6 DDDs/day), whereas risperidone and quetiapine were used in low dose (<0.6 DDDs/day). Many antipsychotics were relatively rarely used as 1 DDD/day (corresponding to 0.9–1.1 category in our study). This was found for most commonly used oral antipsychotics olanzapine which was more often used with higher dose, and for risperidone and quetiapine which were used with low dose more often than with standard dose. This demonstrates that general assumption of 1 DDD/day use applied in some previous studies is not a valid assumption for all patients in modeling antipsychotic use from register-based data, and that modeling has to be based on more sophisticated methods such as PRE2DUP. As DDDs have been developed as a tool for drug utilization research, for example, to monitor drug consumption over time, there is a reluctance to chance them.[33] However, in extreme cases such as perphenazine, they should be changed. It is important to note that our estimates on the used doses are not based on what was prescribed but on what the patients actually had available them based on what and when they picked up the antipsychotic medications from pharmacies.

Results of this study need to be interpreted within its limitations. First, our results are based on a Finnish nationwide cohort including all patients hospitalized with schizophrenia diagnosis. Thus, generalizability is not an issue concerning Finland – and probably also other higher-income countries with similar healthcare system – but the results may not apply to middle- and low-income countries without full reimbursement of medication costs for patients with schizophrenia. Second, our analyses were based on antipsychotic prescriptions dispensed from pharmacies, and it was not possible to assess how much of the dispensed medication had been actually used. However, blood level analyses have shown that our drug use modeling method determines the actual medication use rather accurately.[11] Third, the DDD methodology simply derives the DDDs of LAIs from the average recommended oral doses divided by the dosing interval, which may not always be appropriate as it has been shown eg for paliperidone, which was not included here.[34] Moreover, current DDDs may be mainly suitable for average persons, ie, weighting 70 kg. Fourth, we only focused on the 15 most commonly used antipsychotics in Finland. Therefore, data are lacking in this study on other antipsychotics that may be of interest. Amisulpride was not included in the analysis because it does not have marketing approval in Finland. Fifth, since observational studies are prone to selection bias, we used within-individual analysis to eliminate selection biases related to patient characteristics, such as sex, genetics, and initial severity of the illness. The duration of the illness, the temporal order of treatments, as well as concomitant use of antidepressants and benzodiazepines were adjusted for. However, we cannot exclude residual confounding in this nonrandomized study design. Because within-individual analysis includes only those patients with an outcome and variation in the exposure (not the same treatment all of the time), we also conducted between-individual analysis in which all patients were included. The results of those two complementary analyses were highly consistent with each other, strengthening our findings. Sixth, since our database does not include data on the reasons for drug discontinuation, it was not possible to include those in the analyses. Seventh, despite clinical relevance of antipsychotic polypharmacy, the highly complex analyses of DDD on a large number of specific antipsychotic combinations were beyond the scope of the study. Future analyses should focus on this issue specifically. Nevertheless, despite these limitations, our analyses provide relevant information on the clinically most effective doses of most antipsychotics for the maintenance treatment of schizophrenia occurring in real-world treatment settings.