- Email: [email protected]
Predicting clinical response in people at ultra-high risk of psychosis: a systematic and quantitative review
DRUDIS-1592; No of Pages 4 Drug Discovery Today Volume 00, Number 00 March 2015
PERSPECTIVE
feature Predicting clinical response in people at ultra-high risk of psychosis: a systematic and quantitative review Andrea Mechelli1, [email protected], Diana Prata2,3, Charles Kefford1 and Shitij Kapur1
Introduction Psychosis is common, severely disabling and has a major socioeconomic impact [1]. The onset of the illness is typically preceded by a prodromal phase that is characterised by the emergence of ‘attenuated’ psychotic symptoms; because only about one-third of people with these symptoms subsequently develop a psychotic disorder, this phase is described as ultra-high risk (UHR) or atrisk mental state (ARMS) of psychosis [2]. In the past two decades, with the increasing appreciation of the necessity for early intervention in psychosis worldwide [3], a number of treatments have been employed to delay or prevent the onset of psychosis in people at UHR of psychosis. These treatments include antipsychotic medication [4–6], nutritional supplements such as omega-3 fatty acids [7] and psychological treatment [8–13]. A recent meta-analysis
suggests that, based on evidence of very low to moderate quality, these interventions appear to be effective in delaying and even preventing the onset of psychosis [14]. However there is great individual variation in clinical response to treatment among people at UHR of psychosis. For example, although some respond well to antipsychotic medication and report minimal side-effects, others do not show any clinical benefits and suffer from life-threatening sideeffects such as weight gain, hyperlipidemia, movement disorders and agranulocytosis [15]. Furthermore, persisting with treatment in people who do not benefit from it could lead to a worsening of symptoms [16]. One way of addressing these challenges is to gain a better understanding of why some people at UHR of psychosis benefit from a certain intervention whereas others do not, and then use this
Features PERSPECTIVE
People at ultra-high risk (UHR) of psychosis have 30% chance of developing the illness within two years. A range of pharmacological and psychosocial interventions are now available but there is great individual variation in clinical response. Here we examine the evidence for clinically applicable predictors of clinical response in people at UHR of psychosis. We report that currently there are no reliable predictive markers that can be used to optimise treatment. We argue that there is an urgent need for a better understanding of why some people at UHR of psychosis benefit from a certain treatment whereas others do not. This information will help clinicians make more-effective treatment decisions, and improve long-term clinical outcomes in this population.
information to target delivery of intervention to the subgroup of patients most likely to benefit. In the present systematic and quantitative report, we therefore consider the current literature on predictors of clinical response to available treatments in this population (i.e. predictive markers).
Literature search and article evaluation We searched the PubMed database for extant literature published online up to 16th February 2015, using the following search syntax: (marker OR biomarker OR outcome OR response OR prognosis OR prediction OR predictor OR predict) AND (at-risk mental state OR ultra-high risk OR ultrahigh risk). This identified articles containing the terms ‘marker’, ‘biomarker’, ‘outcome’, ‘response’, ‘prognosis’, ‘prediction’, ‘predictor’ or ‘predict’ in conjunction with ‘at-risk mental state’,
1359-6446/ß 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). www.drugdiscoverytoday.com http://dx.doi.org/10.1016/j.drudis.2015.03.003
Please cite this article in press as: Mechelli, A. et al. Predicting clinical response in people at ultra-high risk of psychosis: a systematic and quantitative review, Drug Discov Today (2015), http://dx.doi.org/10.1016/j.drudis.2015.03.003
1
DRUDIS-1592; No of Pages 4 PERSPECTIVE
Drug Discovery Today Volume 00, Number 00 March 2015
TABLE 1
Criteria for quality of evidence and effect size to grade predictive biomarkersAdapted, with permission, from [17]. Quality of evidence An observation of a positive result (P < 0.05, corrected for multiple comparisons)
Approximate equivalence to drug effects
Score
–
0
In a study controlled for relevant extraneous variables (confounding, nuisance or effect modifiers) , that is, matched, restricted or adjusted for treatment, age, gender and (for genetic studies) ethnicity
Exploratory
1
In a study as above (grade 1), but with an explicit a priori intent to discover a precisely defined biomarker, that is, with a given measure or modality, cut-off and direction of effect of biomarker and response
Phase I drug trial
2
In a study as above (grade 2), but designed with adequate power informed by previous positive studies of the same biomarker, that is, replication in a larger cohort
Phase II drug trial
3
In at least two studies as above (grade 3)
Phase III drug trial
4
In an uncontrolled study a
Effect size
Features PERSPECTIVE
An observation of a statistically significant (positive) result (P < 0.05, corrected for multiple comparisons)
Approximate equivalence to drug effectsa
Score
With estimate from studies with quality of evidence 1
–
0
With marginal effect (OR < 1.3, SMD < 0.2 or r < 0.1)
–
1
With small effect size (OR = 1.3–1.5, SMD = 0.2–0.5 or r = 0.1–0.3)
Effect of psychiatric drugs on symptom improvement (median SMD = 0.41, mean SMD = 0.49)
2
With medium effect size (possibly rivalling the drug effect; OR = 1.5–2.0, SMD = 0.5–0.8 or r = 0.3–0.5)
Effect of antipsychotic drugs on acute symptom improvement (SMD = 0.51)
3
With large effect size (possibly exceeding the usual drug effect; OR > 2.0, SMD > 0.8 or r > 0.5)
Effect of antipsychotic drugs on relapse prevention (SMD = 0.92)
4
Abbreviations: OR, odds ratio; SMD, standard mean difference; r, correlation coefficient. a In genetic studies, this only needs to be applied in relation to the outcome variable, provided that genotypes are unknown (and thus naturally randomised) before subject inclusion, with the exception of ethnicity which affects linkage disequilibrium patterns.
‘ultra-high risk’ or ‘ultrahigh risk’ or related medical subject heading (MeSH) terms such as ‘biological markers’ or ‘psychotic disorders’ in any of the fields searchable in PubMed. We read the abstracts and/or full text of the articles yielded from this search with the aim of identifying those studies that had used a longitudinal design to examine predictors of clinical response to treatment. We then used a recently developed two-dimensional scale [17] to rate the clinical applicability of the predictive markers identified in these studies. This scale, which is reported in Table 1, was adapted from a biomarker rating system proposed by Lassere [18] and is based on quality of evidence (scored 1–4) and effect size (scored 1–4). Quality of evidence is assessed based on whether studies include a control group, test-specific hypotheses, are prospectively designed, appropriately powered and independently replicated; whereas effect size is 2
assessed based on odds ratios. Predictive markers that reach a sum score of 6 (out of 8) are considered clinically applicable (i.e. particularly worthy of clinical consideration).
Systematic and quantitative review observations Our literature search identified 314 articles published between 1990 and 2015. Careful examination of each article, however, revealed that only two studies had investigated predictors of clinical response to a given treatment and, as such, were relevant to the present article. In the first study, Morrison and colleagues carried out exploratory analyses of the impact of pretreatment metacognitive and sociotropic scores on clinical response to cognitive therapy [13]; these exploratory analyses were carried out in the treatment group but not the control group and did not detect any statistically significant
effects, resulting in a grade of 1 for quality of evidence and for effect size. According to our scoring system (Table 1) this yielded a sum score of 0 that was well below our threshold for clinical applicability. In the second study, Amminger and colleagues examined biological and clinical factors associated with clinical improvement following 12 weeks of treatment with long-chain omega-3 (v-3) polyunsaturated fatty acids (PUFAs) [19]. An initial analysis using standard univariate statistics revealed that higher levels of erythrocyte membrane a-linolenic acid (ALA; the parent fatty acid of the v-3 family) predicted subsequent functional improvement in the treatment group but not in the placebo group. A subsequent analysis using multivariate machine learning confirmed that baseline fatty acids allowed prediction of response to treatment in the v-3 PUFA group with high levels of sensitivity, specificity and accuracy. According to our
www.drugdiscoverytoday.com Please cite this article in press as: Mechelli, A. et al. Predicting clinical response in people at ultra-high risk of psychosis: a systematic and quantitative review, Drug Discov Today (2015), http://dx.doi.org/10.1016/j.drudis.2015.03.003
DRUDIS-1592; No of Pages 4 Drug Discovery Today Volume 00, Number 00 March 2015
scoring system (Table 1), quality of evidence was rated as 1 on the basis that there was no a priori hypotheses concerning the direction of the effect; and effect size was rated as 4 based on a standard mean difference equal to 1.1. This yielded a sum score of 4 that was still below our threshold for clinical applicability. The vast majority of the other articles were concerned with nonspecific predictors of clinical outcome irrespective of treatment (n = 90), reported crosssectional rather than longitudinal comparisons (n = 82), reviewed the existing literature without presenting new data (n = 64), tested the main effect of a treatment of interest without considering predictors of clinical response (n = 14) or focused on other clinical groups (n = 32). See Supplementary material available online for the full list of searched articles divided by category.
PERSPECTIVE
Studies of predictive markers Score
1 2 3 4
Score
Quality of evidence
Effect size
1
Controlled study
Marginal
2
+ A priori definition of marker
+
+ Adequate power
Small
3
Medium
4
+ 1 positive study of score 3
Large
Sum score
>6
Clinical applicability
Randomised controlled trials Risks & side-effects Ethical considerations Inconvenience
Differentiating between prognostic and predictive markers
Relevance of outcome
Clinical utility Drug Discovery Today
FIG. 1
From clinical applicability to clinical utility. Once longitudinal studies identify a predictive marker that is clinically applicable (using the two-dimensional scale in Table 1), further randomised clinical trials (of marker vs standard care) should be carried out so that additional factors such as cost of administration, potential risks and side effects, inconvenience and delays associated with testing can be balanced and a decision on its real-world clinical utility can then be made. Adapted, with permission, from [17].
effect on clinical outcome but no interactive effect with treatment; that is, the effect of treatment on individual subjects does not depend on this pre-treatment variable [22]. Following this methodological framework [22], future studies should be able to differentiate between predictive (i.e. treatment-specific) and prognostic (i.e. treatment-nonspecific) markers, in people at UHR of psychosis and those who have developed the illness.
From clinical applicability to clinical utility Although we have used a two-dimensional scale to rate the clinical applicability of predictive biomarkers in people at UHR of psychosis, we note that clinical applicability per se does not assure real-world clinical utility. This is because the clinical utility of a predictive marker depends not only on the effect size but also on the clinical context: a test with a small effect size that would prevent a fatal side-effect is more useful than one with a large effect that would prevent a minor rash [17]. In addition, clinical utility is dependent on the ability of a marker to generate a ‘divergent prediction’: a test that is highly
accurate at predicting clinical response to a given treatment is not particularly useful if there are no alternative treatments [23]. A number of indices have been proposed in recent years with the aim of encapsulating the clinical utility of a marker in real-world clinical practice. For example, the number-needed-to-assess (NNA) refers to the number of individuals who need to be screened to result in one additional positive outcome [24]; this index was adapted from a well established measure of the impact of a new treatment known as number-needed-to-treat (NNT) [25]. Although these indices provide a better proxy of clinical utility than traditional statistical measures, it remains challenging to encapsulate the clinical utility of a marker into a simple and universal formula owing to the large number of factors such as cost of administration, potential risks and side effects, inconvenience and delays associated with testing (Fig. 1).
Concluding remarks Here, we show that, at present, a clinician has no way of predicting whether a person at UHR of psychosis will or will not benefit from a given
www.drugdiscoverytoday.com 3 Please cite this article in press as: Mechelli, A. et al. Predicting clinical response in people at ultra-high risk of psychosis: a systematic and quantitative review, Drug Discov Today (2015), http://dx.doi.org/10.1016/j.drudis.2015.03.003
Features PERSPECTIVE
Whereas previous reviews on clinical outcome in people at UHR of psychosis have focused on predictors of clinical outcome irrespective of treatment (i.e. prognostic markers) [20], here we examine the existing evidence for predictors of clinical response to available treatments (i.e. predictive markers). Our systematic and quantitative report indicates that at present we have no clinically applicable predictive markers that could be used to optimise treatment of this population. This conclusion is consistent with our previous systematic and quantitative review of predictors of treatment response in people with already established psychosis [17], where we found only one predictive biomarker meeting our criteria for clinical applicability (the HLADQB1 6672G>C SNP which predicts side-effects of clozapine administration). The current paucity of predictive markers, in UHR and established psychosis, might be explained by the methodological challenges of differentiating between predictors of clinical response to a given treatment of interest and nonspecific predictors of clinical outcome. However, although there is no doubt that designing a study that differentiates between these two types of markers requires careful consideration, this is by no means impossible based on lessons from other areas of medicine [21]. In particular, a pre-treatment variable can be considered a predictor of clinical response if it can be shown to have an interactive effect with treatment on clinical outcome; in other words, the effect of treatment on individual subjects must depend on this pre-treatment variable [22]. By contrast, a pre-treatment variable should be considered a nonspecific predictor of clinical outcome if it has a main
Lower number-needed-to-assess
Cost
DRUDIS-1592; No of Pages 4 PERSPECTIVE
treatment. There is therefore an urgent need for a better understanding of why some people at UHR of psychosis benefit from a certain treatment whereas others do not. This information will help clinicians make more-effective treatment decisions in their everyday clinical practice. For example, those people at UHR of psychosis who are likely to benefit could be offered psychological treatment, whereas those who are unlikely to benefit could be offered alternative interventions, such as fish oil. Such stratification of treatment would result in better long-term clinical outcome and quality of life among people at increased risk of developing psychosis.
Drug Discovery Today Volume 00, Number 00 March 2015
5
6
7
8
Acknowledgements This research was supported by a grant (ID99859) from the Medical Research Council (MRC) to A.M. and S.K. and (to D.P.) a National Institute for Health Research (NIHR) fellowship (PCSKBZA). The funders had no role in the design of the study, the analysis and interpretation of the data, the drafting of the manuscript or the decision to submit the article for publication.
9
10
11
Appendix A. Supplementary data
12
Supplementary material related to this article can be found, in the online version, at http://dx. doi.org/10.1016/j.drudis.2015.03.003.
13
Features PERSPECTIVE
References 1 Reininghaus, U. et al. (2014) Mortality in schizophrenia and other psychoses: a 10-year follow-up of the SOP first-episode cohort. Schizophr. Bull. http://dx.doi.org/ 10.1093/schbul/sbu138 2 Fusar-Poli, P. et al. (2013) The psychosis high-risk state: a comprehensive state-of-the-art review. JAMA Psychiatry 70, 107–120 3 McGorry, P.D. et al. (2008) Early intervention in psychosis: concepts, evidence and future directions. World Psychiatry 7, 148–156 4 McGlashan, T.H. et al. (2003) The PRIME North America randomized double-blind clinical trial of olanzapine
4
14
15
16
versus placebo in patients at risk of being prodromally symptomatic for psychosis. Schizophr. Res. 61, 7–18 Phillips, L.J. et al. (2009) Randomized controlled trial of interventions for young people at ultra-high risk of psychosis: study design and baseline characteristics. Aust. N. Z. J. Psychiatry 43, 818–829 Ruhrmann, S. et al. (2007) Acute effects of treatment for prodromal symptoms for people putatively in a late initial prodromal state of psychosis. Br. J. Psychiatry Suppl. 51, s88–s95 Amminger, G.P. et al. (2010) Long-chain omega-3 fatty acids for indicated prevention of psychotic disorders: a randomized, placebo-controlled trial. Arch. Gen. Psychiatry 67, 146–154 Bechdolf, A. et al. (2007) Randomized controlled multicentre trial of cognitive behaviour therapy in the early initial prodromal state: effects on social adjustment post treatment. Early Interv. Psychiatry 1, 71–78 Addington, J. et al. (2011) A randomized controlled trial of cognitive behavioral therapy for individuals at clinical high risk of psychosis. Schizophr. Res. 125, 54–61 van der Gaag, M. et al. (2012) Cognitive behavioral therapy for subjects at ultrahigh risk for developing psychosis: a randomized controlled clinical trial. Schizophr. Bull. 38, 1180–1188 Morrison, A.P. et al. (2012) Early detection and intervention evaluation for people at risk of psychosis: multisite randomised controlled trial. BMJ 344, e2233 Morrison, A.P. et al. (2004) Cognitive therapy for the prevention of psychosis in people at ultra-high risk: randomised controlled trial. Br. J. Psychiatry 185, 291–297 Morrison, A.P. et al. (2007) Three-year follow-up of a randomized controlled trial of cognitive therapy for the prevention of psychosis in people at ultrahigh risk. Schizophr. Bull. 33, 682–687 Stafford, M.R. et al. (2013) Early interventions to prevent psychosis: systematic review and meta-analysis. BMJ 346, f185 McGorry, P.D. et al. (2013) Randomized controlled trial of interventions for young people at ultra-high risk of psychosis: twelve-month outcome. Clin. Psychiatry 74, 349–356 Dunn, G. et al. (2012) Effective elements of cognitive behaviour therapy for psychosis: results of a novel type of subgroup analysis based on principal stratification. Psychol. Med. 42, 1057–1068
17 Prata, D. et al. (2014) Clinically meaningful biomarkers for psychosis: a systematic and quantitative review. Neurosci. Biobehav. Rev. 45, 134–141 18 Lassere, M.N. (2008) The Biomarker-Surrogacy Evaluation Schema: a review of the biomarker-surrogate literature and a proposal for a criterion-based, quantitative, multidimensional hierarchical levels of evidence schema for evaluating the status of biomarkers as surrogate endpoints. Stat. Methods Med. Res. 17, 303–340 19 Amminger, G.P. et al. (2015) Predictors of treatment response in young people at ultra-high risk for psychosis who received long-chain omega-3 fatty acids. Transl. Psychiatry 5, e495 20 Jung, W.H. et al. (2012) Gray matter volumetric abnormalities associated with the onset of psychosis. Front. Psychiatry 3, 101 21 Arteaga, C.L. et al. (2012) Treatment of HER2-positive breast cancer: current status and future perspectives. Nat. Rev. Clin. Oncol. 9, 16–32 22 Kraemer, H.C. et al. (2002) Mediators and moderators of treatment effects in randomized clinical trials. Arch. Gen. Psychiatry 59, 877–883 23 Perlis, R.H. (2011) Translating biomarkers to clinical practice. Mol. Psychiatry 16, 1076–1087 24 Uher, R. et al. (2012) Biomarkers predicting treatment outcome in depression: what is clinically significant? Pharmacogenomics 13, 233–240 25 Laupacis, A. et al. (1988) An assessment of clinically useful measures of the consequences of treatment. N. Engl. J. Med. 318, 1728–1733
Andrea Mechelli1,*, Diana Prata2, 3, Charles Kefford1, Shitij Kapur1 1 Department of Psychosis Studies, Institute of Psychiatry, Psychology & Neuroscience, King’s College London, 16 De Crespigny Park, SE5 8AF, UK 2 Centre for Neuroimaging Sciences, Institute of Psychiatry, Psychology & Neuroscience, King’s College London, 16 De Crespigny Park, SE5 8AF, UK 3 Instituto de Medicina Molecular, Faculdade de Medicina da Universidade de Lisboa, Av. Professor Egas Moniz, 1649-028 Lisboa, Portugal
*Corresponding author: Department of Psychosis Studies, Institute of Psychiatry, Psychology & Neuroscience, King’s College London, 16 De Crespigny Park, SE5 8AF, UK.
www.drugdiscoverytoday.com Please cite this article in press as: Mechelli, A. et al. Predicting clinical response in people at ultra-high risk of psychosis: a systematic and quantitative review, Drug Discov Today (2015), http://dx.doi.org/10.1016/j.drudis.2015.03.003
PERSPECTIVE
feature Predicting clinical response in people at ultra-high risk of psychosis: a systematic and quantitative review Andrea Mechelli1, [email protected], Diana Prata2,3, Charles Kefford1 and Shitij Kapur1
Introduction Psychosis is common, severely disabling and has a major socioeconomic impact [1]. The onset of the illness is typically preceded by a prodromal phase that is characterised by the emergence of ‘attenuated’ psychotic symptoms; because only about one-third of people with these symptoms subsequently develop a psychotic disorder, this phase is described as ultra-high risk (UHR) or atrisk mental state (ARMS) of psychosis [2]. In the past two decades, with the increasing appreciation of the necessity for early intervention in psychosis worldwide [3], a number of treatments have been employed to delay or prevent the onset of psychosis in people at UHR of psychosis. These treatments include antipsychotic medication [4–6], nutritional supplements such as omega-3 fatty acids [7] and psychological treatment [8–13]. A recent meta-analysis
suggests that, based on evidence of very low to moderate quality, these interventions appear to be effective in delaying and even preventing the onset of psychosis [14]. However there is great individual variation in clinical response to treatment among people at UHR of psychosis. For example, although some respond well to antipsychotic medication and report minimal side-effects, others do not show any clinical benefits and suffer from life-threatening sideeffects such as weight gain, hyperlipidemia, movement disorders and agranulocytosis [15]. Furthermore, persisting with treatment in people who do not benefit from it could lead to a worsening of symptoms [16]. One way of addressing these challenges is to gain a better understanding of why some people at UHR of psychosis benefit from a certain intervention whereas others do not, and then use this
Features PERSPECTIVE
People at ultra-high risk (UHR) of psychosis have 30% chance of developing the illness within two years. A range of pharmacological and psychosocial interventions are now available but there is great individual variation in clinical response. Here we examine the evidence for clinically applicable predictors of clinical response in people at UHR of psychosis. We report that currently there are no reliable predictive markers that can be used to optimise treatment. We argue that there is an urgent need for a better understanding of why some people at UHR of psychosis benefit from a certain treatment whereas others do not. This information will help clinicians make more-effective treatment decisions, and improve long-term clinical outcomes in this population.
information to target delivery of intervention to the subgroup of patients most likely to benefit. In the present systematic and quantitative report, we therefore consider the current literature on predictors of clinical response to available treatments in this population (i.e. predictive markers).
Literature search and article evaluation We searched the PubMed database for extant literature published online up to 16th February 2015, using the following search syntax: (marker OR biomarker OR outcome OR response OR prognosis OR prediction OR predictor OR predict) AND (at-risk mental state OR ultra-high risk OR ultrahigh risk). This identified articles containing the terms ‘marker’, ‘biomarker’, ‘outcome’, ‘response’, ‘prognosis’, ‘prediction’, ‘predictor’ or ‘predict’ in conjunction with ‘at-risk mental state’,
1359-6446/ß 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). www.drugdiscoverytoday.com http://dx.doi.org/10.1016/j.drudis.2015.03.003
Please cite this article in press as: Mechelli, A. et al. Predicting clinical response in people at ultra-high risk of psychosis: a systematic and quantitative review, Drug Discov Today (2015), http://dx.doi.org/10.1016/j.drudis.2015.03.003
1
DRUDIS-1592; No of Pages 4 PERSPECTIVE
Drug Discovery Today Volume 00, Number 00 March 2015
TABLE 1
Criteria for quality of evidence and effect size to grade predictive biomarkersAdapted, with permission, from [17]. Quality of evidence An observation of a positive result (P < 0.05, corrected for multiple comparisons)
Approximate equivalence to drug effects
Score
–
0
In a study controlled for relevant extraneous variables (confounding, nuisance or effect modifiers) , that is, matched, restricted or adjusted for treatment, age, gender and (for genetic studies) ethnicity
Exploratory
1
In a study as above (grade 1), but with an explicit a priori intent to discover a precisely defined biomarker, that is, with a given measure or modality, cut-off and direction of effect of biomarker and response
Phase I drug trial
2
In a study as above (grade 2), but designed with adequate power informed by previous positive studies of the same biomarker, that is, replication in a larger cohort
Phase II drug trial
3
In at least two studies as above (grade 3)
Phase III drug trial
4
In an uncontrolled study a
Effect size
Features PERSPECTIVE
An observation of a statistically significant (positive) result (P < 0.05, corrected for multiple comparisons)
Approximate equivalence to drug effectsa
Score
With estimate from studies with quality of evidence 1
–
0
With marginal effect (OR < 1.3, SMD < 0.2 or r < 0.1)
–
1
With small effect size (OR = 1.3–1.5, SMD = 0.2–0.5 or r = 0.1–0.3)
Effect of psychiatric drugs on symptom improvement (median SMD = 0.41, mean SMD = 0.49)
2
With medium effect size (possibly rivalling the drug effect; OR = 1.5–2.0, SMD = 0.5–0.8 or r = 0.3–0.5)
Effect of antipsychotic drugs on acute symptom improvement (SMD = 0.51)
3
With large effect size (possibly exceeding the usual drug effect; OR > 2.0, SMD > 0.8 or r > 0.5)
Effect of antipsychotic drugs on relapse prevention (SMD = 0.92)
4
Abbreviations: OR, odds ratio; SMD, standard mean difference; r, correlation coefficient. a In genetic studies, this only needs to be applied in relation to the outcome variable, provided that genotypes are unknown (and thus naturally randomised) before subject inclusion, with the exception of ethnicity which affects linkage disequilibrium patterns.
‘ultra-high risk’ or ‘ultrahigh risk’ or related medical subject heading (MeSH) terms such as ‘biological markers’ or ‘psychotic disorders’ in any of the fields searchable in PubMed. We read the abstracts and/or full text of the articles yielded from this search with the aim of identifying those studies that had used a longitudinal design to examine predictors of clinical response to treatment. We then used a recently developed two-dimensional scale [17] to rate the clinical applicability of the predictive markers identified in these studies. This scale, which is reported in Table 1, was adapted from a biomarker rating system proposed by Lassere [18] and is based on quality of evidence (scored 1–4) and effect size (scored 1–4). Quality of evidence is assessed based on whether studies include a control group, test-specific hypotheses, are prospectively designed, appropriately powered and independently replicated; whereas effect size is 2
assessed based on odds ratios. Predictive markers that reach a sum score of 6 (out of 8) are considered clinically applicable (i.e. particularly worthy of clinical consideration).
Systematic and quantitative review observations Our literature search identified 314 articles published between 1990 and 2015. Careful examination of each article, however, revealed that only two studies had investigated predictors of clinical response to a given treatment and, as such, were relevant to the present article. In the first study, Morrison and colleagues carried out exploratory analyses of the impact of pretreatment metacognitive and sociotropic scores on clinical response to cognitive therapy [13]; these exploratory analyses were carried out in the treatment group but not the control group and did not detect any statistically significant
effects, resulting in a grade of 1 for quality of evidence and for effect size. According to our scoring system (Table 1) this yielded a sum score of 0 that was well below our threshold for clinical applicability. In the second study, Amminger and colleagues examined biological and clinical factors associated with clinical improvement following 12 weeks of treatment with long-chain omega-3 (v-3) polyunsaturated fatty acids (PUFAs) [19]. An initial analysis using standard univariate statistics revealed that higher levels of erythrocyte membrane a-linolenic acid (ALA; the parent fatty acid of the v-3 family) predicted subsequent functional improvement in the treatment group but not in the placebo group. A subsequent analysis using multivariate machine learning confirmed that baseline fatty acids allowed prediction of response to treatment in the v-3 PUFA group with high levels of sensitivity, specificity and accuracy. According to our
www.drugdiscoverytoday.com Please cite this article in press as: Mechelli, A. et al. Predicting clinical response in people at ultra-high risk of psychosis: a systematic and quantitative review, Drug Discov Today (2015), http://dx.doi.org/10.1016/j.drudis.2015.03.003
DRUDIS-1592; No of Pages 4 Drug Discovery Today Volume 00, Number 00 March 2015
scoring system (Table 1), quality of evidence was rated as 1 on the basis that there was no a priori hypotheses concerning the direction of the effect; and effect size was rated as 4 based on a standard mean difference equal to 1.1. This yielded a sum score of 4 that was still below our threshold for clinical applicability. The vast majority of the other articles were concerned with nonspecific predictors of clinical outcome irrespective of treatment (n = 90), reported crosssectional rather than longitudinal comparisons (n = 82), reviewed the existing literature without presenting new data (n = 64), tested the main effect of a treatment of interest without considering predictors of clinical response (n = 14) or focused on other clinical groups (n = 32). See Supplementary material available online for the full list of searched articles divided by category.
PERSPECTIVE
Studies of predictive markers Score
1 2 3 4
Score
Quality of evidence
Effect size
1
Controlled study
Marginal
2
+ A priori definition of marker
+
+ Adequate power
Small
3
Medium
4
+ 1 positive study of score 3
Large
Sum score
>6
Clinical applicability
Randomised controlled trials Risks & side-effects Ethical considerations Inconvenience
Differentiating between prognostic and predictive markers
Relevance of outcome
Clinical utility Drug Discovery Today
FIG. 1
From clinical applicability to clinical utility. Once longitudinal studies identify a predictive marker that is clinically applicable (using the two-dimensional scale in Table 1), further randomised clinical trials (of marker vs standard care) should be carried out so that additional factors such as cost of administration, potential risks and side effects, inconvenience and delays associated with testing can be balanced and a decision on its real-world clinical utility can then be made. Adapted, with permission, from [17].
effect on clinical outcome but no interactive effect with treatment; that is, the effect of treatment on individual subjects does not depend on this pre-treatment variable [22]. Following this methodological framework [22], future studies should be able to differentiate between predictive (i.e. treatment-specific) and prognostic (i.e. treatment-nonspecific) markers, in people at UHR of psychosis and those who have developed the illness.
From clinical applicability to clinical utility Although we have used a two-dimensional scale to rate the clinical applicability of predictive biomarkers in people at UHR of psychosis, we note that clinical applicability per se does not assure real-world clinical utility. This is because the clinical utility of a predictive marker depends not only on the effect size but also on the clinical context: a test with a small effect size that would prevent a fatal side-effect is more useful than one with a large effect that would prevent a minor rash [17]. In addition, clinical utility is dependent on the ability of a marker to generate a ‘divergent prediction’: a test that is highly
accurate at predicting clinical response to a given treatment is not particularly useful if there are no alternative treatments [23]. A number of indices have been proposed in recent years with the aim of encapsulating the clinical utility of a marker in real-world clinical practice. For example, the number-needed-to-assess (NNA) refers to the number of individuals who need to be screened to result in one additional positive outcome [24]; this index was adapted from a well established measure of the impact of a new treatment known as number-needed-to-treat (NNT) [25]. Although these indices provide a better proxy of clinical utility than traditional statistical measures, it remains challenging to encapsulate the clinical utility of a marker into a simple and universal formula owing to the large number of factors such as cost of administration, potential risks and side effects, inconvenience and delays associated with testing (Fig. 1).
Concluding remarks Here, we show that, at present, a clinician has no way of predicting whether a person at UHR of psychosis will or will not benefit from a given
www.drugdiscoverytoday.com 3 Please cite this article in press as: Mechelli, A. et al. Predicting clinical response in people at ultra-high risk of psychosis: a systematic and quantitative review, Drug Discov Today (2015), http://dx.doi.org/10.1016/j.drudis.2015.03.003
Features PERSPECTIVE
Whereas previous reviews on clinical outcome in people at UHR of psychosis have focused on predictors of clinical outcome irrespective of treatment (i.e. prognostic markers) [20], here we examine the existing evidence for predictors of clinical response to available treatments (i.e. predictive markers). Our systematic and quantitative report indicates that at present we have no clinically applicable predictive markers that could be used to optimise treatment of this population. This conclusion is consistent with our previous systematic and quantitative review of predictors of treatment response in people with already established psychosis [17], where we found only one predictive biomarker meeting our criteria for clinical applicability (the HLADQB1 6672G>C SNP which predicts side-effects of clozapine administration). The current paucity of predictive markers, in UHR and established psychosis, might be explained by the methodological challenges of differentiating between predictors of clinical response to a given treatment of interest and nonspecific predictors of clinical outcome. However, although there is no doubt that designing a study that differentiates between these two types of markers requires careful consideration, this is by no means impossible based on lessons from other areas of medicine [21]. In particular, a pre-treatment variable can be considered a predictor of clinical response if it can be shown to have an interactive effect with treatment on clinical outcome; in other words, the effect of treatment on individual subjects must depend on this pre-treatment variable [22]. By contrast, a pre-treatment variable should be considered a nonspecific predictor of clinical outcome if it has a main
Lower number-needed-to-assess
Cost
DRUDIS-1592; No of Pages 4 PERSPECTIVE
treatment. There is therefore an urgent need for a better understanding of why some people at UHR of psychosis benefit from a certain treatment whereas others do not. This information will help clinicians make more-effective treatment decisions in their everyday clinical practice. For example, those people at UHR of psychosis who are likely to benefit could be offered psychological treatment, whereas those who are unlikely to benefit could be offered alternative interventions, such as fish oil. Such stratification of treatment would result in better long-term clinical outcome and quality of life among people at increased risk of developing psychosis.
Drug Discovery Today Volume 00, Number 00 March 2015
5
6
7
8
Acknowledgements This research was supported by a grant (ID99859) from the Medical Research Council (MRC) to A.M. and S.K. and (to D.P.) a National Institute for Health Research (NIHR) fellowship (PCSKBZA). The funders had no role in the design of the study, the analysis and interpretation of the data, the drafting of the manuscript or the decision to submit the article for publication.
9
10
11
Appendix A. Supplementary data
12
Supplementary material related to this article can be found, in the online version, at http://dx. doi.org/10.1016/j.drudis.2015.03.003.
13
Features PERSPECTIVE
References 1 Reininghaus, U. et al. (2014) Mortality in schizophrenia and other psychoses: a 10-year follow-up of the SOP first-episode cohort. Schizophr. Bull. http://dx.doi.org/ 10.1093/schbul/sbu138 2 Fusar-Poli, P. et al. (2013) The psychosis high-risk state: a comprehensive state-of-the-art review. JAMA Psychiatry 70, 107–120 3 McGorry, P.D. et al. (2008) Early intervention in psychosis: concepts, evidence and future directions. World Psychiatry 7, 148–156 4 McGlashan, T.H. et al. (2003) The PRIME North America randomized double-blind clinical trial of olanzapine
4
14
15
16
versus placebo in patients at risk of being prodromally symptomatic for psychosis. Schizophr. Res. 61, 7–18 Phillips, L.J. et al. (2009) Randomized controlled trial of interventions for young people at ultra-high risk of psychosis: study design and baseline characteristics. Aust. N. Z. J. Psychiatry 43, 818–829 Ruhrmann, S. et al. (2007) Acute effects of treatment for prodromal symptoms for people putatively in a late initial prodromal state of psychosis. Br. J. Psychiatry Suppl. 51, s88–s95 Amminger, G.P. et al. (2010) Long-chain omega-3 fatty acids for indicated prevention of psychotic disorders: a randomized, placebo-controlled trial. Arch. Gen. Psychiatry 67, 146–154 Bechdolf, A. et al. (2007) Randomized controlled multicentre trial of cognitive behaviour therapy in the early initial prodromal state: effects on social adjustment post treatment. Early Interv. Psychiatry 1, 71–78 Addington, J. et al. (2011) A randomized controlled trial of cognitive behavioral therapy for individuals at clinical high risk of psychosis. Schizophr. Res. 125, 54–61 van der Gaag, M. et al. (2012) Cognitive behavioral therapy for subjects at ultrahigh risk for developing psychosis: a randomized controlled clinical trial. Schizophr. Bull. 38, 1180–1188 Morrison, A.P. et al. (2012) Early detection and intervention evaluation for people at risk of psychosis: multisite randomised controlled trial. BMJ 344, e2233 Morrison, A.P. et al. (2004) Cognitive therapy for the prevention of psychosis in people at ultra-high risk: randomised controlled trial. Br. J. Psychiatry 185, 291–297 Morrison, A.P. et al. (2007) Three-year follow-up of a randomized controlled trial of cognitive therapy for the prevention of psychosis in people at ultrahigh risk. Schizophr. Bull. 33, 682–687 Stafford, M.R. et al. (2013) Early interventions to prevent psychosis: systematic review and meta-analysis. BMJ 346, f185 McGorry, P.D. et al. (2013) Randomized controlled trial of interventions for young people at ultra-high risk of psychosis: twelve-month outcome. Clin. Psychiatry 74, 349–356 Dunn, G. et al. (2012) Effective elements of cognitive behaviour therapy for psychosis: results of a novel type of subgroup analysis based on principal stratification. Psychol. Med. 42, 1057–1068
17 Prata, D. et al. (2014) Clinically meaningful biomarkers for psychosis: a systematic and quantitative review. Neurosci. Biobehav. Rev. 45, 134–141 18 Lassere, M.N. (2008) The Biomarker-Surrogacy Evaluation Schema: a review of the biomarker-surrogate literature and a proposal for a criterion-based, quantitative, multidimensional hierarchical levels of evidence schema for evaluating the status of biomarkers as surrogate endpoints. Stat. Methods Med. Res. 17, 303–340 19 Amminger, G.P. et al. (2015) Predictors of treatment response in young people at ultra-high risk for psychosis who received long-chain omega-3 fatty acids. Transl. Psychiatry 5, e495 20 Jung, W.H. et al. (2012) Gray matter volumetric abnormalities associated with the onset of psychosis. Front. Psychiatry 3, 101 21 Arteaga, C.L. et al. (2012) Treatment of HER2-positive breast cancer: current status and future perspectives. Nat. Rev. Clin. Oncol. 9, 16–32 22 Kraemer, H.C. et al. (2002) Mediators and moderators of treatment effects in randomized clinical trials. Arch. Gen. Psychiatry 59, 877–883 23 Perlis, R.H. (2011) Translating biomarkers to clinical practice. Mol. Psychiatry 16, 1076–1087 24 Uher, R. et al. (2012) Biomarkers predicting treatment outcome in depression: what is clinically significant? Pharmacogenomics 13, 233–240 25 Laupacis, A. et al. (1988) An assessment of clinically useful measures of the consequences of treatment. N. Engl. J. Med. 318, 1728–1733
Andrea Mechelli1,*, Diana Prata2, 3, Charles Kefford1, Shitij Kapur1 1 Department of Psychosis Studies, Institute of Psychiatry, Psychology & Neuroscience, King’s College London, 16 De Crespigny Park, SE5 8AF, UK 2 Centre for Neuroimaging Sciences, Institute of Psychiatry, Psychology & Neuroscience, King’s College London, 16 De Crespigny Park, SE5 8AF, UK 3 Instituto de Medicina Molecular, Faculdade de Medicina da Universidade de Lisboa, Av. Professor Egas Moniz, 1649-028 Lisboa, Portugal
*Corresponding author: Department of Psychosis Studies, Institute of Psychiatry, Psychology & Neuroscience, King’s College London, 16 De Crespigny Park, SE5 8AF, UK.
www.drugdiscoverytoday.com Please cite this article in press as: Mechelli, A. et al. Predicting clinical response in people at ultra-high risk of psychosis: a systematic and quantitative review, Drug Discov Today (2015), http://dx.doi.org/10.1016/j.drudis.2015.03.003