Schizophrenia, Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) and number needed to treat: how can CATIE inform clinicians?

Authors


Leslie Citrome, MD, MPH, Nathan S. Kline Institute for Psychiatric Research, 140 Old Orangeburg Road, Orangeburg, NY 10962, USA
Tel.: + 1 845 398 5595
Fax: + 1 845 398 5483
Email: citrome@nki.rfmh.org

Summary

The schizophrenia medication study conducted as part of the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) provided a large quantity of data. However, placing these data into a clinically meaningful context for the individual practitioner has been challenging. Effectiveness and safety outcome data were extracted from the three principal publications that documented the results of phases 1 and 2 of the CATIE schizophrenia study. Number needed to treat (NNT) and number needed to harm (NNH) were calculated from the categorical results, together with their confidence intervals. Olanzapine and clozapine demonstrated advantages over comparators in terms of all-cause discontinuation, largely driven by efficacy advantages. NNT for olanzapine compared with perphenazine, quetiapine, risperidone and ziprasidone ranged from 5.5 to 10.1 in phase 1. NNT for clozapine compared with risperidone or quetiapine was approximately 3 in phase 2. There were marked differences in association with weight gain and metabolic effects, with olanzapine demonstrating a NNH ranging from 12.4 to 17.7 in terms of discontinuation of treatment in phase 1 because of these effects. Results from phase 2 reflect phase 1 in this regard, and demonstrated an advantage for ziprasidone in terms of discontinuation because of weight gain or metabolic effects, with NNT ranging from 10.6 to 20.8. However, these notable differences in association with weight gain and metabolic effects did not seem to drive the differences in overall time to all-cause discontinuation. NNT and NNH can help place the wide array of CATIE results into clinical context, and permits quantification of the differences observed between the antipsychotics that were tested.

Introduction

Long-term, double-blind, randomised clinical trials, comparing second-generation antipsychotics for the treatment of schizophrenia have usually been conducted by the manufacturers of the antipsychotic medications, have usually included only two medications, and are not always published. Large-scale, independently conducted, multiple-medication trials are needed. The Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) have emerged as a rich source of information from which clinicians can gather some information about the relative strengths and weaknesses the second-generation antipsychotics have compared amongst each other, and also in comparison with a representative first-generation agent (1–3).

CATIE's design included three main phases. Phase 1 involved the randomisation of 1493 patients to receive one of five blinded antipsychotics for up to 18 months: perphenazine, olanzapine, quetiapine, risperidone or ziprasidone. Patients who discontinued phase 1 before 18 months were offered participation in phase 2, where 543 patients were randomised to one of five second-generation antipsychotics that they did not receive during phase 1. Those who prematurely discontinued phase 2 were offered open-label treatment with one or two antipsychotics. When enrolled, patients were made aware that these switches were possible. Nearly half of all patients who enrolled finished a full 18 months of follow-up.

Among the 1432 patients who received medication and who were included in the phase 1 analysis, 74% discontinued before 18 months. The antipsychotics differed in terms of time to all-cause discontinuation. In addition, olanzapine was statistically superior to both quetiapine and risperidone in pairwise comparisons. Of the 74% who discontinued phase 1, approximately half entered phase 2. Patients who had discontinued phase 1 because of lack of efficacy (24%) were encouraged to participate in phase 2 in a pathway that involved randomisation to open-label clozapine, or blinded olanzapine, quetiapine or risperidone. Patients who had discontinued because of intolerability (15%) were encouraged to participate in phase 2 in a pathway that involved randomisation to blinded ziprasidone, olanzapine, quetiapine or risperidone. Not all patients who discontinued phase 1 because of inefficacy chose to participate in the pathway involving clozapine. About half of those in the ziprasidone pathway had discontinued phase 1 because of inefficacy. The clozapine pathway results demonstrated superiority of clozapine in all-cause discontinuation to both risperidone and quetiapine. The ziprasidone pathway failed to demonstrate the superiority of ziprasidone for all-cause discontinuation; olanzapine and risperidone demonstrated a longer time to all-cause discontinuation compared with quetiapine and ziprasidone. Weight gain and metabolic effects with olanzapine in phase 2 mirrored that seen in phase 1. Ziprasidone had the most benign metabolic profile but there was a trend suggesting that overall risperidone was the best tolerated medication in both phases.

The results of phases 1 and 2 can be quantified in terms of the clinically intuitive concepts of number needed to treat (NNT) and number needed to harm (NNH) (4,5). NNT (NNH) can describe how many patients one needs to treat on drug A vs. drug B to see one extra success (or adverse event), and is calculated by taking the reciprocal of attributable risk (6). For example, the use of corticosteroids (vs. no treatment) prior to delivery of a premature infant will reduce the risk of fetal respiratory distress syndrome, with an NNT of 11 (95% confidence interval 8–16) (7,8). This means that one case of fetal respiratory distress syndrome is prevented for every 11 premature births where the intervention was provided. A small NNT means the relative advantage of one drug over another is big because you need to treat fewer patients to see one additional case of response or benefit than if you had been using the alternative. NNT can be calculated for relevant CATIE outcomes such as remaining on the randomised medication, avoidance of hospitalisation, as well as discontinuation because of adverse events. NNH can be calculated for safety outcomes such as new prescriptions for antidiabetic agents. A large NNH means the difference is small for the adverse event. The objective of this paper was to present the NNT and NNH, together with their relevant confidence intervals, for the range of outcomes already contained in the published CATIE phase 1 and 2 papers (1–3), and discuss how this can inform the clinician on potential antipsychotic choice.

Methods

Data were extracted from the principal results of CATIE phases 1 and 2, as reported in their respective journal articles (1–3). Attributable risk was calculated by subtracting the rate (frequency) of an event seen with drug A from the rate observed with drug B. The intention-to-treat populations reported in the original papers were used for the effectiveness outcomes (discontinuation from randomised drug). All randomised patients were used for the safety outcomes. For example, all-cause discontinuation on olanzapine in phase 1 was observed at a rate of 210/330 (0.636) (number of patients on olanzapine discontinuing early divided by the number of randomised patients receiving olanzapine), and that for perphenazine was 192/257 (0.747). Attributable risk in this case was 0.111. The number of people that the intervention has to be given in order to avoid the outcome (NNT) is calculated by taking the reciprocal of the attributable risk, in this case dividing 1 by 0.111, resulting in an NNT of 9.03. A 95% confidence interval was calculated for each NNT by determining the upper and lower bounds of the corresponding attributable risk, using the formula

image

where z = 1.96; this amount is added or subtracted from the attributable risk, and the reciprocal is then calculated to determine the upper or lower bound for the NNT (9). In addition, 98.33% (z = 2.3932) and 98.75% (z = 2.4977) confidence intervals were also calculated, as appropriate, for the effectiveness outcomes in order to be able to assess the results based on Hochberg correction for multiple pairwise comparisons, where the original papers used thresholds of p = 0.0167 when three pairwise comparisons were performed and p = 0.0125 when four pairwise comparisons were performed. However, unlike the principal analysis of phase 1 in the original paper (1), all randomised patients are included when comparing phase 1 discontinuation rates with ziprasidone. Because the report of the ziprasidone pathway of phase 2 (3) only examined effectiveness outcomes of the patients assigned to olanzapine, quetiapine or risperidone after the availability of ziprasidone, data from patients in the study before ziprasidone were available could not be included in the present NNT effectiveness analysis for phase 2. However, data from these patients are included in the safety analysis. Because the p values for the safety outcomes were presented for descriptive purposes only (1–3), a 95% confidence interval was selected for the purposes of this paper for safety outcome values. Because NNT (NNH) lends itself best to dichotomous outcomes, the outcomes selected for this analysis did not include changes in rating scales or laboratory measures. For the dichotomous safety outcomes that reached a p value of less than 0.05, the NNH was calculated in a similar way as for the NNT for effectiveness outcomes, namely by determining the attributable risk by subtracting the rate of an event seen with drug A from the rate observed with drug B. For the phase 2 ziprasidone pathway safety outcomes, percentages without the actual Ns were published (3). These Ns were later made available to us for the purposes of this paper.

This re-examination of the CATIE data was carried out based on the published outcomes (1–3). The study itself was approved by the institutional review board at each study site. The CATIE Executive Committee reviewed the initial proposal for this secondary analysis, but aside from one of the authors (TSS), this analysis was carried out independently of the CATIE investigators.

Results

Table 1 provides the NNT for the CATIE phase 1 effectiveness outcomes of all-cause discontinuation, and separately for discontinuation because of lack of efficacy, intolerability, patient decision or other reasons. Table 2 presents the NNT (or more appropriately termed NNH) for specific dichotomous safety outcomes in phase 1 that reached a p-value of less than 0.05 for differences comparing all treatment groups, namely hospitalisation for exacerbation of schizophrenia, insomnia, urinary hesitancy/dry mouth/constipation, incontinence/nocturia, weight gain greater than 7% from baseline, discontinuation of treatment because of weight gain or metabolic effects, discontinuation of treatment because of extrapyramidal effects, prolonged corrected QT interval, use of antidepressants, use of sedatives/hypnotics, use of anxiolytics and use of anticholinergics. Tables 3 and 4 provide the analogous information for phase 2 for the available data as published. Olanzapine was selected as the principal reference medication for phase 1 comparisons because of its superiority in that phase. Likewise clozapine was selected as the principal reference medication for phase 2 comparisons involving the clozapine pathway. Ziprasidone was selected for the principal reference medication for phase 2 comparisons involving the ziprasidone pathway, in line with the original hypothesis stated in that paper (3). Positive numbers indicate an advantage for the reference drug, and in general can be interpreted as an NNT; negative numbers represent a disadvantage for that agent, and in general can be interpreted as an NNH. The smaller the absolute magnitude of the NNT (NNH), the more substantial the difference between the drugs being compared. Confidence intervals are presented as suggested by Altman (10). Narrower confidence intervals denote greater precision for the estimate being provided. When confidence intervals include both positive and negative numbers, this denotes a lack of statistical significance, and is further emphasised by including infinity in the range of possibilities (an NNT of infinity means no difference between the treatments). However, the NNT calculations here are not intended for hypothesis testing, but rather to place the results in a more clinically meaningful context. The reader is referred back to the original publications for the results of the statistical analysis designed to test the a priori hypotheses (1–3).

Table 1.   Phase 1 effectiveness outcomes – number needed to treat (NNT) and confidence intervals (CI) for discontinuation on randomised medication*
OutcomeNNT95% CI98.33% or 98.75% CI†
  1. OLZ, olanzapine; PER, perphenazine; QUE, quetiapine; RIS, risperidone; ZIP, ziprasidone. *Intention-to-treat population. †98.33% CI listed for comparisons with PER, QUE and RIS; 98.75% CI listed for comparisons with ZIP.

OLZ (n = 330) vs. PER (n = 257)
 All-cause9.05.4 to 27.45.0 to 50.0
 Lack of efficacy9.35.8 to 23.85.3 to 36.2
 Intolerability−31.0−10.7 to ∞ to 34.6−9.4 to ∞ to 23.6
 Patient decision15.8−109.6 to ∞ to 7.4−39.8 to ∞ to 6.6
 Other reasons−36.0−15.7 to ∞ to 124.0−14.0 to ∞ to 62.6
OLZ vs. QUE (n = 329)
 All-cause5.54.0 to 8.73.8 to 10.0
 Lack of efficacy7.45.1 to 13.84.8 to 17.0
 Intolerability−25.7−10.4 to ∞ to 55.2−9.2 to ∞ to 32.5
 Patient decision10.56.1 to 37.85.6 to 88.2
 Other reasons−112.2−21.8 to ∞ to 35.8−18.5 to ∞ to 27.7
OLZ vs. RIS (n = 333)
 All-cause10.15.9 to 34.45.4 to 73.6
 Lack of efficacy7.85.3 to 15.04.9 to 18.8
 Intolerability−11.7−7.2 to −30.7−6.6 to −48.1
 Patient decision14.9−2314.9 to ∞ to 7.4−65.3 to ∞ to 6.7
 Other reasons−104.1−21.6 to ∞ to 37.0−18.4 to ∞ to 28.4
OLZ vs. ZIP (n = 183)
 All-cause6.44.3 to 12.93.9 to 17.8
 Lack of efficacy10.56.0 to 44.85.3 to 420.5
 Intolerability−28.7−9.8 to ∞ to 31.1−8.3 to ∞ to 19.8
 Patient decision9.35.2 to 39.74.7 to 399.3
 Other reasons−83.2−18.3 to ∞ to 32.8−15.1 to ∞ to 23.7
Table 2.   Phase 1 safety outcomes*– number needed to harm (NNH) and confidence intervals (CI)
OutcomeNNH95% CI
  1. OLZ, olanzapine; PER, perphenazine; QUE, quetiapine; RIS, risperidone; ZIP, ziprasidone. *All randomised patients.

OLZ (n = 336) vs. PER (n = 261)
 Hospitalisation22.7−85.9 to ∞ to 10.0
 Number of hospitalisations per total person-year of exposure4.63.2 to 7.8
 Insomnia11.26.4 to 43.0
 Urinary hesitancy, dry mouth, constipation−59.8−11.9 to ∞ to 19.7
 Incontinence, nocturia−32.7−16.5 to −2425.9
 Weight gain >7%−5.6−4.1 to −8.7
 Discontinuation of treatment because of weight gain or metabolic effects−12.4−8.8 to −21.2
 Discontinuation of treatment because of extrapyramidal effects16.510.2 to 43.4
 Prolonged corrected QT interval86.0−227.6 to ∞ to 36.2
 Use of antidepressants−85.0−15.9 to ∞ to 25.4
 Use of hypnotics or sedatives44.2−48.2 to ∞ to 15.1
 Use of anxiolytics18.89.4 to 1902.1
 Use of anticholinergics39.7−48.3 to ∞ to 14.1
OLZ vs. QUE (n = 337)
 Hospitalisation11.37.0 to 29.4
 Number of hospitalisations per total person-year of exposure2.72.2 to 3.5
 Insomnia13.57.2 to 113.5
 Urinary hesitancy, dry mouth, constipation6.04.1 to 10.8
 Incontinence, nocturia−110.4−24.0 to ∞ to 42.4
 Weight gain >7%−7.2−4.9 to −13.6
 Discontinuation of treatment because of weight gain or metabolic effects−17.6−10.7 to −50.2
 Discontinuation of treatment because of extrapyramidal effects170.5−54.0 to ∞ to 33.1
 Prolonged corrected QT interval35.719.9 to 168.9
 Use of antidepressants−27.8−12.3 to ∞ to 105.2
 Use of hypnotics or sedatives−41.8−17.3 to ∞ to 99.7
 Use of anxiolytics22.6−267.8 to ∞ to 10.8
 Use of anticholinergics−23.9−13.2 to −126.7
OLZ vs. RIS (n = 341)
 Hospitalisation27.4−69.8 to ∞ to 11.5
 Number of hospitalisations per total person-year of exposure6.24.1 to 13.0
 Insomnia12.67.1 to 51.6
 Urinary hesitancy, dry mouth, constipation89.2−18.8 to ∞ to 13.2
 Incontinence, nocturia50.6−59.1 to ∞ to 17.7
 Weight gain >7%−6.3−4.5 to −10.5
 Discontinuation of treatment because of weight gain or metabolic effects−13.4−9.2 to −24.6
 Discontinuation of treatment because of extrapyramidal effects118.4−61.0 to ∞ to 30.0
 Prolonged corrected QT interval31.118.0 to 114.8
 Use of antidepressants25.4−79.0 to ∞ to 11.0
 Use of hypnotics or sedatives35.2−81.0 to ∞ to 14.5
 Use of anxiolytics221.6−25.3 to ∞ to 20.6
 Use of anticholinergics51.4−44.8 to ∞ to 16.3
OLZ vs. ZIP (n = 185)
 Hospitalisation15.37.7 to 1813.2
 Number of hospitalisations per total person-year of exposure3.62.6 to 5.8
 Insomnia7.24.6 to 16.2
 Urinary hesitancy, dry mouth, constipation−28.5−9.2 to ∞ to 26.2
 Incontinence, nocturia13.17.7 to 45.3
 Weight gain >7%−4.4−3.4 to −6.3
 Discontinuation of treatment because of weight gain or metabolic effects−16.7−10.0 to −50.7
 Discontinuation of treatment because of extrapyramidal effects71.3−55.8 to ∞ to 21.7
 Prolonged corrected QT interval74.0−196.5 to ∞ to 31.1
 Use of antidepressants46.5−25.4 to ∞ to 12.1
 Use of hypnotics or sedatives37.9−43.7 to ∞ to 13.2
 Use of anxiolytics18.6−170.6 to ∞ to 8.8
 Use of anticholinergics786.8−21.7 to ∞ to 20.6
Table 3.   Phase 2 effectiveness outcomes – number needed to treat (NNT) and confidence intervals (CI) for discontinuation on randomised medication*
OutcomeNNT95% CI98.33% CI
  1. CLO, clozapine; OLZ, olanzapine; QUE, quetiapine; RIS, risperidone; ZIP, ziprasidone. *Intention-to-treat population.

Clozapine pathway
CLO (n = 45) vs. OLZ (n = 17)
 All-cause6.6−9.1 to ∞ to 2.4−6.0 to ∞ to 2.1
 Lack of efficacy4.1−312.7 to ∞ to 2.0−17.4 to ∞ to 1.8
 Intolerability−19.1−5.1 to ∞ to 10.8−4.4 to ∞ to 8.0
 Patient decision10.6−6.6 to ∞ to 2.9−4.8 to ∞ to 2.5
 Other reasons−7.5−4.3 to −29.4−3.9 to −82.9
CLO vs. RIS (n = 14)
 All-cause3.31.9 to 14.81.7 to 62.2
 Lack of efficacy3.21.7 to 23.6−54.5 to ∞ to 1.5
 Intolerability−9.0−4.9 to −51.8−4.5 to ∞ to 993.2
 Patient decision6.4−8.4 to ∞ to 2.3−5.5 to ∞ to 2.0
 Other reasons−16.2−4.4 to ∞ to 9.5−3.8 to ∞ to 7.0
CLO vs. QUE (n = 14)
 All-cause2.71.8 to 5.71.6 to 7.6
 Lack of efficacy3.21.7, 23.6−54.5 to ∞ to 1.5
 Intolerability9.7−7.7 to ∞ to 3.0−5.5 to ∞ to 2.6
 Patient decision11.7−5.6 to ∞ to 2.9−4.2 to ∞ to 2.4
 Other reasons−7.5−4.3 to −29.4−3.9 to −82.9
Ziprasidone pathway
ZIP (n = 135) vs. OLZ (n = 66)
 All-cause−9.6−4.2 to ∞ to 33.0−3.7 to ∞ to 16.7
 Lack of efficacy−11.9−4.7 to ∞ to 22.8−4.2 to ∞ to 13.9
 Intolerability17.8−17.8 to ∞ to 5.9−12.3 to ∞ to 5.2
 Patient decision−14.4−5.2 to ∞ to 18.2−4.5 to ∞ to 12.1
 Other reasons−148.5−17.0 to ∞ to 22.0−14.2 to ∞ to 17.5
ZIP vs. RIS (n = 69)
 All-cause−7.5−3.8 to ∞ to 915.9−3.4 to ∞ to 32.6
 Lack of efficacy−19.9−5.6 to ∞ to 12.6−4.8 to ∞ to 9.2
 Intolerability−25.4−7.6 to ∞ to 18.9−6.6 to ∞ to 13.6
 Patient decision−15.6−5.3 to ∞ to 16.9−4.7 to ∞ to 11.6
 Other reasons47.8−23.4 to ∞ to 11.8−17.6 to ∞ to 10.1
ZIP vs. QUE (n = 63)
 All-cause14.1−22.8 to ∞ to 5.4−14.4 to ∞ to 4.7
 Lack of efficacy26.2−9.7 to ∞ to 5.6−7.4 to ∞ to 4.8
 Intolerability29.5−13.0 to ∞ to 6.9−9.9 to ∞ to 5.9
 Patient decision−16.9−5.4 to ∞ to 14.6−4.6 to ∞ to 10.4
 Other reasons17.2−47.7 to ∞ to 7.3−26.0 to ∞ to 6.5
Table 4.   Phase 2 safety outcomes*– number needed to harm (NNH) and confidence intervals (CI)
OutcomeNNH95% CI
  1. CLO, clozapine; OLZ, olanzapine; QUE, quetiapine; RIS, risperidone. *All randomised patients.

Clozapine pathway
CLO (n = 49) vs. OLZ (n = 19)
 Insomnia8.5−17.9 to ∞ to 3.4
 Urinary hesitancy, dry mouth, constipation−4.9−3.2 to −11.0
 Sialorrhoea−4.5−2.4 to −32.5
CLO vs. RIS (n = 16)
 Insomnia3.72.0 to 26.4
 Urinary hesitancy, dry mouth, constipation−7.1−3.3 to ∞ to 45.2
 Sialorrhoea−5.0−2.4 to ∞ to 142.0
CLO vs. QUE (n = 15)
 Insomnia10.8−11.3 to ∞ to 3.7
 Urinary hesitancy, dry mouth, constipation3.8−71.6 to ∞ to 1.8
 Sialorrhoea−3.1−2.2 to −5.1
Ziprasidone pathway
ZIP (n = 137) vs. OLZ (n = 108)
 Hospitalisation−20.2−7.4 to ∞ to 27.8
 Insomnia−5.4−3.5 to −11.9
 Sex drive, sexual arousal, sexual orgasm74.7−12.6 to ∞ to 9.4
 Orthostatic faintness26.6−47.7 to ∞ to 10.4
 Skin rash−39.6−14.7 to ∞ to 57.5
 Discontinuation of treatment because of weight gain or metabolic effects12.07.4 to 32.0
 Weight gain >7%5.53.7 to 10.8
ZIP vs. RIS (n = 104)
 Hospitalisation−148.4−10.1 to ∞ to 11.6
 Insomnia−12.0−5.1 to ∞ to 34.3
 Sex drive, sexual arousal, sexual orgasm7.44.2 to 34.2
 Orthostatic faintness47.2−29.8 to ∞ to 13.2
 Skin rash72.0−23.5 to ∞ to 14.2
 Discontinuation of treatment because of weight gain or metabolic effects20.811.2 to 143.6
 Weight gain >7%15.6−136.4 to ∞ to 7.4
ZIP vs. QUE (n = 95)
 Hospitalisation25.4−16.2 to ∞ to 7.1
 Insomnia−6.4−3.8 to –20.4
 Sex drive, sexual arousal, sexual orgasm−20.8−7.4 to ∞ to 26.1
 Orthostatic faintness11.16.1 to 62.5
 Skin rash24.7−39.8 to ∞ to 9.4
 Discontinuation of treatment because of weight gain or metabolic effects10.66.5 to 27.9
 Weight gain >7%13.3−923.4 to ∞ to 6.6

In the NNT analyses for the phase 1 effectiveness outcomes, olanzapine demonstrated advantages over perphenazine, quetiapine, risperidone and ziprasidone with regard to all-cause discontinuation with relatively small and thus meaningful NNT magnitudes ranging from 5.5 to 10.1. This advantage of olanzapine over all the other drugs was seen also for discontinuation because of lack of efficacy (NNT ranges 7.4–10.5). Risperidone had a substantial advantage over olanzapine with regard to discontinuation because of intolerability (NNH 11.7). In phase 2, clozapine showed substantial advantages compared with risperidone and quetiapine with NNTs of 3.3 and 2.7, respectively, for all-cause discontinuation. Although discontinuation specifically because of lack of efficacy also resulted in NNTs around 3 when comparing clozapine with risperidone or quetiapine, these estimates were imprecise as indicated by wide confidence intervals. Comparison of clozapine and olanzapine resulted in an NNT of 6.7 for all-cause discontinuation and 4.1 for discontinuation because of lack of efficacy, but with broad confidence intervals that included infinity. Comparisons of ziprasidone in phase 2 resulted in NNTs of magnitude −9.6 to 14.1 with broad confidence intervals for effectiveness as measured by all-cause discontinuation of medication. Other effectiveness measures of ziprasidone in phase 2 also had broad confidence intervals.

The safety analysis reveals consistent patterns in phase 1 and phase 2 with regard to the disadvantage for olanzapine regarding weight gain and metabolic effects. In phase 1, statistically significant disadvantage for olanzapine for weight gain greater than 7% was observed in comparisons with perphenazine, quetiapine, risperidone and ziprasidone, with NNT (or more appropriately termed NNH) ranging from 4.4 to 7.2. Discontinuation of treatment because of weight gain or metabolic effects resulted in NNH ranging from 12.4 to 17.7. It is unknown what the discontinuation rates were for weight gain separate from the discontinuation rates due to metabolic effects as these two somewhat different outcomes were combined when the original data were collected. In phase 2, a comparison of ziprasidone with olanzapine, risperidone and quetiapine revealed a statistically significant advantage for ziprasidone in terms of discontinuation of treatment because of weight gain or metabolic effects, with NNT ranging from 10.6 to 20.8. Comparisons for weight gain greater than 7% in phase 2 revealed an advantage for ziprasidone when compared with olanzapine, with an NNT of 5.5.

Discussion

The concept of NNT allows the clinician to estimate a medication's potential relevant effect. Examining the magnitudes of NNT and NNH, the clinician can start to make risk–benefit decisions tailored to the individual patient's needs or preferences. Olanzapine's superiority to other first-line second-generation antipsychotics in terms of overall effectiveness needs to be tempered by its proclivity to be associated with weight gain and metabolic problems. The NNH data point out that these adverse events do not occur with every patient, and can be expected every 12–18 patients in terms of discontinuation from treatment because of this, or every four to seven patients in terms of weight gain greater than 7%. On the other hand, olanzapine's efficacy advantage can be apparent every seven to 11 patients. Olanzapine's efficacy profile is consistent with the safety data regarding hospitalisation required because of exacerbation in phase 1, particularly when computing the effect as measured by number of hospitalisations per total person-year of exposure, where the NNT (NNH) ranged from 2.7 to 6.2 in favour of olanzapine compared with the other antipsychotics.

Phase 2 data need to be interpreted with caution. Patients who discontinued phase 1 because of inefficacy did not necessarily enter the clozapine pathway for phase 2. About equal numbers of patients who entered the phase 2 ziprasidone pathway discontinued phase 1 because of tolerability or efficacy reasons. Once randomised, patients in the clozapine pathway knew if they were getting clozapine or a blinded medication, and may have adjusted their attitudes towards continuation accordingly based on expectations of benefit, requirement of titration to a suitable dose, or degree of clinical contact (weekly rather than monthly). Moreover relatively small numbers of patients enrolled in the clozapine pathway of phase 2, resulting in few patients assigned to each blinded treatment arm. Small sample sizes reduce the statistical power available to resolve any differences should they exist.

The analysis presented is substantially different from the treatment discontinuation analysis reported in the original papers, where treatment groups were compared using Cox proportional hazard regression models, and where a number of adjustments were made (for example, whether or not the patient had an exacerbation of schizophrenia in the preceding 3 months or whether the patient had tardive dyskinesia at study baseline). Moreover, in the phase 1 analysis originally presented by Lieberman et al. (1), comparisons involving the perphenazine group were limited to patients without tardive dyskinesia, and comparisons involving the ziprasidone group were limited to the cohort of patients who underwent randomisation after ziprasidone was added. Thus, although the Cox-model pairwise comparisons for phase 1 did not reveal a statistically significant difference for all-cause discontinuation for olanzapine vs. ziprasidone (p value of 0.027 did not meet the threshold of 0.013) (1), the NNT calculated here was statistically significant. The opposite occurred when the NNT was calculated for phase 2 – comparisons of ziprasidone vs. risperidone or olanzapine did not result in statistically significant NNTs for all-cause discontinuation, however Cox-model pairwise comparisons did demonstrate the superiority of olanzapine or risperidone to ziprasidone (3).

The absolute magnitudes of the NNT that are reported here are not dissimilar to other examples in the psychiatric literature, where for example a comparison of antipsychotic vs. placebo for the treatment of schizophrenia results in a range of NNT of 2–5 for the outcome of a 40% reduction in the Brief Psychiatric Rating Scale score or a rating of ‘much improved’ on Clinical Global Impression, of family intervention vs. usual care for patients with schizophrenia which nets an NNT of 7 for the outcome of relapse, antidepressant compared with placebo for the treatment of major depression which nets an NNT of 3 for the outcome of a 50% reduction in the Hamilton Rating Scale for Depression, and antidepressants compared with placebo for the treatment of bulimia nervosa which results in an NNT of 9 for remission (4). Kraemer and Kupfer (6) put forth that an NNT of 2.3, 3.6 and 8.9 correspond to a Cohen's d of 0.8, 0.5 and 0.2, respectively, representing effect sizes that are ‘large’, ‘medium’ and ‘small’. The more devastating the outcome, the more importance placed on differences, thus explaining why avoidance of stroke in a patient with a previous thromboembolic event demands the use of agents such as warfarin (vs. no treatment), despite a relatively high NNT of 18 (11).

The manufacturers of the first-line second-generation antipsychotics have spent a great deal of effort in attempting to differentiate their products from each other. One source of debate has been adverse events such as the development of new-onset diabetes mellitus (12). From the CATIE data one can calculate the NNH for new prescriptions of antidiabetic agents during phase 1. Although not statistically significant, the NNH comparing olanzapine with the other antipsychotics ranged from approximately 60 to 82 for this outcome. This is consistent with the NNH that can be calculated from the incidence data presented by Leslie and Rosenheck (13) comparing second-generation antipsychotics with first-generation antipsychotics, where NNH was 48 for clozapine, 125 for quetiapine, 159 for olanzapine and 2000 for risperidone (5). NNH for the prescription of a cholestatin also yielded high NNH, however there was a statistically significant difference when comparing olanzapine and ziprasidone (NNH for olanzapine was 29.6, with a 95% confidence interval of 16.5–139.0). This information about metabolic risk needs to be balanced by the therapeutic needs of the patient (14), and consideration of efficacy differences, which in the groups evaluated in the CATIE trial appear larger in magnitude than the differences seen for tolerability and safety. Nonetheless, when treating individual patients, management of risk, by monitoring and rapid intervention when required, is essential to good clinical care (15).

Limitations of this NNT/NNH analysis include the absence of information about relative efficacy on continuous measures such as the Positive and Negative Syndrome Scale, or on safety outcomes such as plasma levels of prolactin, cholesterol or triglycerides, all of which showed differences among the agents in the CATIE study reports (1–3). Also missing are data on categorical changes for the safety outcomes just mentioned, for example the number of patients for whom prolactin or cholesterol levels changed from normal to abnormal.

Conclusions

We re-examined CATIE data using the concepts of NNT and NNH to place the results into clinical perspective. All-cause discontinuation is an integration of efficacy, tolerability and patient preference effects. An efficacy advantage for olanzapine drove the overall low (and thus advantageous) NNT observed for olanzapine compared with the other antipsychotics in terms of all-cause discontinuation, particularly in phase 1. The NNH for discontinuation because of weight gain or metabolic effects for olanzapine, although statistically significant, did not reach the same level of magnitude as the NNT for efficacy. Data from phase 2 were largely consistent with phase 1, with important advantages noted for clozapine in terms of all-cause discontinuation in the clozapine pathway, and less marked advantages of olanzapine and risperidone in the ziprasidone pathway. In both phase 1 and 2, ziprasidone presented with the most favourable metabolic profile while risperidone appeared to have the best tolerability overall. Relative importance of these differences among the second-generation antipsychotics will vary from patient to patient, but the NNT and NNH calculated from the group data can help the clinician quantify this risk–benefit decision.

Acknowledgements

Leslie Citrome, MD, MPH, is a consultant for, has received honoraria from, or has conducted clinical research supported by the following: Abbott Laboratories, AstraZeneca Pharmaceuticals, Barr Laboratories, Bristol-Myers Squibb, Eli Lilly and Company, Janssen Pharmaceuticals, Jazz Pharmaceuticals, and Pfizer Inc.

T. Scott Stroup, MD, MPH, is a consultant for, has received honoraria from, or has conducted clinical research supported by the following: Bristol-Myers Squibb, Eli Lilly and Company, Janssen Pharmaceuticals, and Pfizer Inc.

Ancillary