Which Antipsychotic is the Best?

In the online edition of the Lancet of June 27th, 2013, Stefan Leucht from the Technical University of Munich and colleagues report on their recent meta-analysis entitled “Comparative efficacy and tolerability of 15 antipsychotic drugs in schizophrenia: a multiple-treatments meta-analysis.” (Leucht et al., Lancet 26.6.2013; Epub ahead of print). The authors conducted a so-called “multiple-treatments” meta-analysis, which is based not only on direct, but also indirect comparisons of the investigated compounds. The analysis was based on randomized, double-blind studies in patients with schizophrenia or related disorders (schizoaffective disorder, delusional disorder). Published studies were supplemented by data from the FDA and pharmaceutical companies. The authors finally analyzed 212 studies with data from 43049 patients.

Cipriani et al. (2012) describe the „Multiple Treatments Metaanlyse“ as follows: „Standard meta-analyses are an effective tool in evidence-based medicine, but one of their main drawbacks is that they can compare only two alternative treatments at a time. Moreover, if no trials exist which directly compare two interventions, it is not possible to estimate their relative efficacy. Multiple treatments meta-analyses use a meta-analytical technique that allows the incorporation of evidence from both direct and indirect comparisons from a network of trials of different interventions to estimate summary treatment effects as comprehensively and precisely as possible.“ (Cipriani et al.: What is a multiple treatments meta-analysis? Epidemiol Psychiatr Sci 2012; 21:151-153)

Primary outcome parameters in the present meta-analysis was efficacy, secondary outcome parameters were discontinuations for any reason, weight gain, extrapyramidal side effects, increase in prolactin, QTc interval, and sedation. Continuous variables were expressed as standardized mean differences (SMD), which is a measure of the effect size.

All antipsychotics were significantly more effective than placebo. The authors found the following SMDs for the efficacy (in descending order, in parentheses the 95% credibility intervals):

  • Clozapine 0.88 (0.73 – 1.03)
  • Amisulpride 0.66 (0.53 – 0.78)
  • Olanzapine 0.59 (0.53 – 0.65)
  • Risperidone 0.56 (0.50 – 0.63)
  • Paliperidone 0.50 (0.39 – 0.60)
  • Zotepine 0.49 (0.31 – 0.66)
  • Haloperidol 0.45 (0.39 – 0.51)
  • Quetiapine 0.44 (0.35 – 0.52)
  • Aripiprazole 0.43 (0.34 – 0.52)
  • Sertindole 0.39 (0.26 – 0.52)
  • Ziprasidone 0.39 (0.30 – 0.49)
  • Chlorpromazine 0.38 (0.23 – 0.54)
  • Asenapine 0.38 (0.25 – 0.51)
  • Lurasidone 0.33 (0.21 – 0.45)
  • Iloperidone 0.33 (0.22 – 0.43)

Also with regard to the frequency of treatment discontinuation all compounds – with the exception of zotepine – were significantly better than placebo (ie, discontinuations were less frequent). The authors calculated odds ratios between 0.43 (95% credibility interval 0.32 to 0.57) for amisulpride (with a value of 1 for placebo, ie, with amisulpride was the risk of discontinuation of therapy less than half as large) and 0.80 (0.71 to 0.90) for haloperidol.

Only three of the 15 studied antipsychotics caused weight gain that was not significantly greater than with placebo: haloperidol, ziprasidone, and lurasidon. The  ranking was as follows (with increasing SMDs for greater weight gain compared to placebo):

  • Haloperidol 0.09 (–0.00 – 0.17) not significant
  • Ziprasidone 0.10 (–0.02 – 0.22) not significant
  • Lurasidone 0.10 (–0.02 – 0.21) not significant
  • Aripiprazole 0.17 (0.05 – 0.28)
  • Amisulpride 0.20 (0.05 – 0.35)
  • Asenapine 0.23 (0.07 – 0.39)
  • Paliperidone 0.38 (0.27 – 0.48)
  • Risperidone 0.42 (0.33 – 0.50)
  • Quetiapine 0.43 (0.34 – 0.53)
  • Sertindole 0.53 (0.38 – 0.68)
  • Chlopromazine 0.55 (0.34 – 0.76)
  • Iloperidone 0.62 (0.49 – 0.74)
  • Clozapine 0.65 (0.31 – 0.99)
  • Zotepine 0.71 (0.47 – 0.96)
  • Olanzapine 0.74 (0.67 – 0.81)

With regard to extrapyramidal side effects (EPS), it is the classical antipsychotics and substances with a higher affinity for the D2 receptor, which are significantly worse than placebo. Interestingly, clozapine was the only drug in this analysis, which caused even less EPS than placebo. With haloperidol the risk of EPS is almost five times higher than with placebo. The risk was in the following order (with ascending odds ratio versus placebo):

  • Clozapine 0.3 (0.12 – 0.62) significantly lower than placebo
  • Sertindole 0.81 (0.47 – 1.3) not significant
  • Olanzapine 1.00 (0.73 – 1.33) not significant
  • Quetiapine 1.01 (0.68 – 1.44) not significant
  • Aripiprazole 1.20 (0.73 – 1.85) not significant
  • Iloperidone 1.58 (0.55 – 3.65) not significant
  • Amisulpride 1.60 (0.88 – 2.65) not significant
  • Ziprasidone 1.61 (1.05 – 2.37)
  • Asenapine 1.66 (0.85 – 2.93) not significant
  • Paliperidone 1.81 (1.17 – 2.69)
  • Risperidone 2.09 (1.54 – 2.78)
  • Lurasidone 2.46 (1.55 – 3.72)
  • Chlorpromazine 2.65 (1.33 – 4.76)
  • Zotepine 3.01 (1.38 – 5.77)
  • Haloperidol 4.76 (3.70 – 6.04)

As expected, an increase in prolactin is the least likely to occur with aripiprazole compared to placebo. This compound even tended to reduce prolactin (not significant). For amisulpride, clozapine and zotepine there were not sufficient data from the available studies, although it is known that under amisulpride the strongest increases in prolactin are observed, while clozapine usually does not increase prolactin. The results were in the following sequence (in ascending odds ratio versus placebo):

  • Aripiprazole –0.22 (–0.46 – 0.03) not significant
  • Quetiapine –0.05 (–0.23 – 0.13) not significant
  • Asenapine 0.12 (–0.12 – 0.37) not significant
  • Olanzapine 0.14 (+0.00 – 0.28)
  • Chlorpromazine 0.16 (–0.48 – 0.8) not significant
  • Iloperidone 0.21 (–0.09 – 0.51) not significant
  • Ziprasidone 0.25 (0.01 – 0.49)
  • Lurasidone 0.34 (0.11 – 0.57)
  • Sertindole 0.45 (0.16 – 0.74)
  • Haloperidol 0.70 (0.56 – 0.85)
  • Risperidone 1.23 (1.06 – 1.40)
  • Paliperidone 1.30 (1.08 – 1.51)
  • Amisulpride NA
  • Clozapine NA
  • Zotepine NA

Almost all investigated antipsychotics prolong the QTc interval of the ECG. The extension  is most pronounced with sertindole. Amisulpride leads to a greater prolongation of the QTc interval than ziprasidone, which is not generally known. For clozapine, chlorpromazine and zotepine the authors had insufficient data available. The results were in the following sequence (in ascending odds ratio versus placebo):

  • Lurasidone –0.10 (–0.21 – 0.01) not significant
  • Aripirazole 0.01 (–0.13 – 0.15) not significant
  • Paliperidone 0.05 (–0.18 – 0.26) not significant
  • Haloperidol 0.11 (0.03 – 0.19)
  • Quetiapine 0.17 (0.06 – 0.29)
  • Olanzapine 0.22 (0.11 – 0.31)
  • Risperidone 0.25 (0.15 – 0.36)
  • Asenapine 0.30 (–0.04 – 0.65) not significant
  • Iloperidone 0.34 (0.22 – 0.46)
  • Ziprasidone 0.41 (0.31 – 0.51)
  • Amisulpride 0.66 (0.39 – 0.91)
  • Sertindole 0.90 (0.76 – 1.02)
  • Clozapine NA
  • Chlorpromazine NA
  • Zotepine NA

All antipsychotics also lead to a more or less pronounced sedation, which was not significantly larger with four of the investigated substances than with placebo. The three tricyclic compounds chlorpromazine, zotepine and clozapine elicit the strongest sedation. The results were in the following sequence (in ascending odds ratio versus placebo):

  • Amisulpride 1.42 (0.72 – 2.51) not significant
  • Paliperidone 1.40 (0.85 – 2.19) not significant
  • Sertindole 1.53 (0.82 – 2.62) not significant
  • Iloperidone 1.71 (0.63 – 3.77) not significant
  • Aripiprazole 1.84 (1.05 – 3.05)
  • Lurasidone 2.45 (1.76 – 3.35)
  • Haloperidol 2.76 (2.04 – 3.66)
  • Asenapine 3.28 (1.37 – 6.69)
  • Olanzapine 3.34 (2.46 – 4.50)
  • Quetiapine 3.76 (2.68 – 5.19)
  • Ziprasidone 3.80 (2.58 – 5.42)
  • Chlorpromazine 7.56 (4.78 – 11.53)
  • Zotepine 8.15 (3.91 – 15.33)
  • Clozapine 8.82 (4.72 – 15.06)

The study provides important parameters especially with regard to the tolerability of the 15 most important available antipsychotics. But important questions remain. Thus, haloperidol continues to appear overdosed in most studies, although the authors emphasize that they have considered only studies in which the doses were given based on an international consensus. Here, however, haloperidol 10 mg and olanzapine 20 mg were considered equipotent, and that seems highly questionable. It is also noticeable that newer agents (such as iloperidone and lurasidone) were less efficacious than older ones. It is somewhat likely that the larger placebo response in the studies of recent years has influenced the results. Ultimately, the ideal antipsychotic for each patient will still have to be individually selected.

 

This post is also available in: German

One thought on “Which Antipsychotic is the Best?

  1. I found this page following a link from another blog. I wanted to pose a simple foundational question about the nature of these studies. I had an academic background in the foundations of mathematics, so am well attuned to thinking about how decision processes rest upon assumptions.

    Clearly you do not decide how to fix a car with gearbox trouble by looking at the colour of the paint, and doing a randomised controlled trial of potential replacement gearboxes for ‘cars with red paint’. Indeed it is not enough to do a randomised controlled trial of potential replacement gearboxes for ‘Ferrari cars with red paint’ either. Indeed as any engineer knows, the colour of paint has no effect upon the correct choice of replacement gearbox, but the make _and_ model of car does.

    Sometimes we can fix a broken machine with a replacement part without needing to know the ‘make and model’.

    When deciding how to treat a patient, the symptoms which lead to diagnoses such as schizophrenia are only part of the picture, and the diagnostic label itself effectively disregards much of the detail in the original symptom picture.

    What is the empirical basis for the assumption that things beyond the symptoms in the psychiatric diagnostic manual _can be safely ignored in all cases_, or indeed _can be safely ignored in most cases_?

    What is the empirical basis for the assumption that the diagnostic label of ‘schizophrenia’ is sufficiently specific or the outcome of a clinical trial such as the one cited above to be meaningful in the clinical scenario where a clinician has a single patient in front of them?

    The result of these trials is essentially the result of a massive statistical averaging process. Nearly all available information on the patients involved is disregarded by such statistical procedures. If something of major importance is disregarded by these statistical procedures, how will you know? On what basis do you consider it safe to make decisions primarily upon the results of such trials?

    Moreover, faced with the clinical scenario of a single consultant with a single patient, how significant is the heavily averaged and general information resulting from trials such as the above?

    For a medical example of the kind of over-generalisation I am describing, consider a hypothetical controlled trial for cancer treatment in which the type of cancer was totally ignored. How useful would the results be?

    All mathematical abstraction, including statistical methods, is a game of selective ignorance. The idea is to ignore what you can safely ignore, so as to aid the reasoning process. If you ignore something you cannot safely ignore, you reasoning process will be faulty, as will all which rely upon it. Are medical researchers using such abstractions as diagnostic labels and statistical studies safely? And if so, where is the rigorous justification that they are?

Please leave a comment!