Stratified medicine in schizophrenia: how accurate would a test of drug response need to be to achieve cost-effective improvements in quality of life?

Significant efforts have been made to identify stratifiers for predicting antipsychotic treatment responses for patients with schizophrenia. However, little is known about how accurate such a stratifier needs to be to achieve cost-effective improvements in quality of life. To the best of our knowledge, only one previous study has assessed the cost-effectiveness of a stratifier for schizophrenia patients, which is a modelling study conducted by Perlis et al. [23]. The stratifier assessed was a genetic test that can identify individuals with greater likelihood of responding to second-line conventional antipsychotics, with a sensitivity of 95.9% and a specificity of 38.3%. Perlis et al. found that applying the test to treatment-naïve patients and using clozapine as a first-line antipsychotic for those predicted to respond to clozapine is more effective but also more expensive, comparing with no test and reversing clozapine as a third-line antipsychotic. However, the study conducted by Perlis et al. had some limitations: (1) it did not consider the health or cost impact of any adverse events of antipsychotics, which might have important impacts on the cost-effectives results; (2) PSA was not conducted to examine the impact of joint uncertainty of multiple parameters simultaneously; (3) this study was published in 2005, therefore, it has limited relevance to the current context, due to rapidly changing nature of treatments, health services, evidence base and unit costs. No previous studies have assessed the cost-effectiveness of a stratifier for schizophrenia patients who fail a first-line antipsychotic. In order to fill the gap, we used decision-analytic modelling to estimate the cost-effectiveness of a SMA for schizophrenia patients who failed a first-line antipsychotic, over a life-time time horizon from the health sector and social care perspective. Results indicate that the stratified medicine algorithm was more cost effective that treatment as usual, when both the sensitivity and specificity of the stratifier was set at 60%. One surprising finding of this study is, even when the sensitivity of the stratifier is set to 0%, as long as the specificity of the stratifier is no less than 11.50%, SMA is still more cost-effective than TAU. This was mainly because the proportion of patients who would respond to clozapine is much larger than the proportion of patients who would only respond to a second-line conventional antipsychotic. A 0% sensitivity means none of the AP2 responders are correctly identified as such—all of them are identified as AP2 non-responders and therefore be prescribed with clozapine. According to Agid’s study [13] and the CATIE trial [15], 16.67% of the target population are AP2 responders; and 71.16% of AP2 responders would also respond to clozapine. This means, only those AP2 responders who do not respond to clozapine (16.67%*(1–71.16%) = 4.81%) are adversely affected by the poor sensitivity of the stratifier, and even then, they are only on clozapine for 1 year in our model, before being switched to a non-clozapine antipsychotic, to which they would respond. For those AP2 responder who also respond to clozapine (16.67%*71.16% = 11.86%), our model shows that although they are slightly better-off with a conventional second-line antipsychotic comparing to clozapine, the overall magnitude of cost and QALY difference is very small. The specificity of the test, in this model, is its ability in correctly identifying AP2 non-responders, which includes clozapine responders and non-responders. According to Agid’s study [13], clozapine responders comprise 62.50% of the target population. A 11.50% specificity means, comparing to TAU, 7.19% (= 11.50%*62.50%) of the target population received the right treatment (clozapine) one cycle earlier, and therefore results in lower cost and higher QALY gains. To sum up, comparing to TAU, use of a stratifier with 0% sensitivity and 11.50% specificity results in worse outcomes for 4.81% patients (AP2 responders wrongly identified as clozapine responders, and who do not respond to clozapine) and better outcomes for 7.19% patients (clozapine responders correctly identified as AP2 non-responders). Therefore, after weighing up the clinical benefits, clinical harms and cost implications of the stratifier, SMA is considered to be more cost-effective than TAU.

According to a recent systematic review [14], there are only two trials which examined patients’ response to a second-line antipsychotic after failing a first-line antipsychotic: Agid et al. [13] and the OPTiMiSE trial [14]. Both studies found that for patients who failed a first-line conventional antipsychotic, their response to a second-line conventional antipsychotic is poor—in fact, the OPTiMiSE trial found that for patients who did not achieve remission with a first-line antipsychotic within 4 weeks, it might be better for them to continue their current antipsychotic, rather than being switched to a second-line conventional antipsychotic. The findings of these two studies, combined with the result of our model (a stratifier with very low sensitivity and specificity is still considered to be cost-effective) raises the question of whether a stratifier is necessary for patients who failed a first-line antipsychotic or not—using clozapine as a second-line treatment without any stratifier might turn out to be more cost-effective, compared to use of a stratifier with low sensitivity and/or specificity. Although a few studies have argued the potential benefits of using clozapine as a second-line or even first-line treatment [24‐26], there is currently a lack of high-quality evidence about clozapine’s efficacy for schizophrenia who failed a first-line antipsychotic. Further research is urgently needed to establish clozapine’s clinical efficacy and safety profile for patients as a second-line treatment. With publication of such results, the input data of the present analysis can be updated to shed light on whether clozapine should be used as a second-line antipsychotic.

To our knowledge, this is the first study to explore the pharmacoeconomic implications of using a stratified medicine approach in schizophrenia patients who failed a first-line antipsychotic. A major strength of this study was the collective use of the state-of-the-art methodology and most up-to-date data to evaluate the cost-effectiveness of a hypothetical stratifier. Decision analytic modelling provides a framework to quantify and synthesize the clinical benefits, clinical harms and cost implications of the stratifier. Comprehensive sensitivity analyses have been conducted to test the uncertainty of the result. Therefore, commissioners and clinical researchers can be reassured that the conclusion of this paper is robust to large changes or errors in the input parameters.

There are a number of limitations of the economic model presented here, the majority of which derive from limitations in the evidence base used to populate the model. For example: (1) patients’ responses to second-line antipsychotics were calculated based on the data for 60 schizophrenia patients included in the Agid et al. [13], as there is a lack of larger trials which specifically assessed patients’ responses to a second-line antipsychotic after failing a first-line antipsychotic. There is uncertainty about the generalizability of the results reported by Agid et al. due to its small sample size; (2) patients’ probability of developing adverse events were obtained from the network meta-analyses reported in the NICE schizophrenia guideline [2]. It was noted that the population included in those network meta-analyses were ‘general’ schizophrenia patients in remission, some of whom might have been on antipsychotics for many years. It is unknow whether those patients would have different probability of developing adverse events, compared with our target population. However, in this study, the limitation related to parameter uncertainty has been partially mitigated by extensive sensitivity analyses. Secondly, discontinuation of antipsychotics due to intolerable adverse events was only modelled for clozapine, but not for other conventional antipsychotics. This simplification is against the use of clozapine, and thus against the use of stratifier, as the key benefit of using the stratifier is that clozapine responders who are correctly identified as AP2 non-responders can initiate clozapine sooner than the current practice. The base case results show that even when this simplification was used, SMA still dominates TAU, which demonstrates the robustness of our base case conclusion. Thirdly, the cost and health impacts of adverse events of antipsychotics have not been fully captured. Other than the four adverse events modelled in this study (i.e. weight gain, diabetes, acute EPS and neutropenia), there are other adverse events which might impact on patients’ QALY and cost outcomes, such as tardive dyskinesia, sexual dysfunction, increase in prolactin levels, and agranulocytosis. Even for those adverse events which were considered in the model, the impacts of complications of those adverse events were not fully captured: for example, the impact of complications of diabetes (e.g. myocardial infarction and stroke etc.) on mortality were not considered in the model. Fourthly, besides the health and social care costs that were considered in this analysis, SMA is likely to have an impact on the wider societal costs (such as costs borne to the criminal justice system and productivity losses for schizophrenia patients and their carers). When wider societal costs were considered, the gap between treatment cost of remitted patients and relapse patients is likely to widen [27], and thus making SMA more cost-effective.