Whether early improvement to antipsychotic drugs (at least 20% PANSS total score reduction at two weeks) can be used as a test for subsequent response and remission in acute schizophrenia.

We will carry out a comprehensive literature search of electronic databases (EMBASE, MEDLINE, PubMed, BIOSIS, CINAHL, PsycINFO, Cochrane Library), with terms combining antipsychotic drugs, schizophrenia and prediction of response: [(schizophrenia MeSH or schizo*) AND (antipsychotic agents MeSH or antipsychoti* or neurolept*) AND (early N2 (improvement or early non-respon* or respon* or prediction*)]. We will additionally contact study authors and pharmaceutical companies for further trials. Search methods are being developed with the help of a search specialist, Samantha Roberts, to ensure the highest standards in our search methods.

We will include the studies irrespective of the blinding procedure, setting (inpatient and outpatient) and design (e.g. RCTs and case series). There will be no language restriction to avoid the problem of language bias (Egger et al. 1997).

We will include all studies that examined the identification of responders to an antipsychotic at follow-up by the degree of improvement of schizophrenic symptoms at two weeks.

We will include the studies irrespective of the blinding procedure, setting (inpatient and outpatient) and design (e.g. RCTs and case series). The minimum follow-up to assess response will be four weeks, whereas the maximum will be 12 weeks. There will be no language restriction to avoid the problem of language bias (Egger et al. 1997). The treatment will be any antipsychotic that is marketed in at least one country. According to all major treatment guidelines there is no major difference in the efficacy between the available antipsychotics (except for clozapine which will thus be excluded). Therefore, pooling the results of studies that used different antipsychotic drugs is justified (Lehman et al. 2004, Falkai et al. 2005, Tandon et al. 2010). There will be no restriction in antipsychotic dose as long as doses were in the range of a recent international consensus document (Gardner et al. 2010). The antipsychotic has to be given orally (oral tablets, oral liquid), because intramuscular formulations are either given for short-term sedation or as depots for long-term relapse prevention. We will only include studies on monotherapy with one antipsychotic, because the review question is whether patients who have not improved at two weeks treatment with one drug should be switched . Nevertheless, other comedication than antipsychotic drugs such as benzodiazepines, antidepressants or anticholinergics will be allowed. The justification is that these concomitant medications are given in clinical routine, as well, enhancing generalisability. 

Acute exacerbation of schizophrenia or schizophrenia(-like) psychoses (mainly schizophreniform and schizoaffective disorders), irrespective of the diagnostic criteria used.

We will include people with an acute exacerbation of schizophrenia (no restriction on age, setting, gender, ethnicity) or schizophrenia(-like) psychoses (mainly schizophreniform and schizoaffective disorders), irrespective of the diagnostic criteria used. There is no clear evidence that the schizophrenia-like psychoses are caused by fundamentally different disease processes or require different treatment approaches (Carpenter and Buchanan 1994). We will include studies regardless of diagnostic criteria used such as the International Classification of Diseases (ICD-10) or Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) since in clinical routine these criteria are not meticulously used either. Therefore, the use of our broad inclusion criteria will enhance generalisability and representativeness. This strategy is also generally used by the Cochrane Schizophrenia Group.

Index test

The index test being evaluated in this review is early improvement, defined primarily as at least 20% reduction of the Positive and Negative Syndrome Scale (PANSS, Kay et al. 1994) from baseline score to two weeks. We have shown that this cutoff means minimal improvement according to the Clinical Global Impression (Guy 1974) of the raters (Leucht et al. 2005 and Levine et al. 2008). Other cutoff points ranging from at least over 0% to at least 50 % PANSS reduction will also be examined.

Reference standards  

Response at follow-up (reference tests):

In our context, the reference standard will be subsequent response to antipsychotics, primarily defined as at least 50% reduction of the PANSS/BPRS. In two independent studies, we have shown that this definition is clinically meaningful, equalling to “much improved” from the perspective of the raters using the equipercentile linking method (Leucht et al. 2005 and Levine et al. 2008). Secondary definitions of subsequent response will be at least 20% reduction of PANSS/BPRS and remission as defined by Andreasen et al. (2005) and vanOs et al. (2006). As mentioned already, at least 20% reduction of the Positive and Negative Syndrome Scale (PANSS, Kay et al. 1994) means minimal improvement according to the Clinical Global Impression (Guy 1974) of the raters (Leucht et al. 2005 and Levine et al. 2008). Regarding the definition by Andreasen, criteria for remission are fulfilled if eight core items of the PANSS are not rated higher than mildly present.

Separate (stratified) analyses will be carried out for these three different reference standards, and/or data will be clearly presented separately. The exact definition of a positive outcome of the reference standard in every study will be also clearly described. If possible, above data will be also combined.

Primary outcomes

The outcome in a diagnostic test review can be defined as the sensitivity and specificity of the test. In our review, this is the sensitivity and specificity of the cutoff 'at least 20% PANSS improvement at two weeks' as a definition of early improvement. We will extract the necessary numbers for the two-by-two tables to calculate these parameters together with their respective 95% confidence interval for each individual study. Then, an average summary value of sensitivity and specificity will be computed. We will primarily attempt to calculate these numbers based on the cutoff 'at least 20% PANSS reduction at two weeks' and, for that reason, we will send a request for further information to the authors.

Secondary outcomes

Different cutoffs, ranging for at least above 0% to 50% reduction, will also be examined.

Two reviewers will independently extract data from all selected trials. We will extract the diagnostic two-by-two table (true positive, false positive, true negative, and false negative index test results) from the publications with respective 95% confidence intervals, or if not available, reconstruct the two-by-two table using information on relevant parameters (sensitivity, specificity or predictive values). Eligible studies for which the diagnostic two-by-two table could not be reconstructed wil be presented in the review, but not included in the quantitative analyses. Disagreements will be resolved by discussion and, if necessary, by consulting a third review author. If consensus can still not be obtained, we will contact the study authors to resolve the dilemma.

In addition to the values needed for the two-by-two tables, our standardised extraction sheets will include characteristics of participants and study methods, index tests and reference standards.

Methodological quality will be assessed by two independent authors using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) instrument (Whiting 2003; Whiting 2004 and 2006). QUADAS consists of 14 items that refer to internal validity (for example, blind assessment of index and reference test, or avoidance of verification bias). We have added one item regarding the prior or posterior to the start of the study establishment of cut-off values. We will classify each item as 'yes' (adequately addressed), 'no' (inadequately addressed), or 'unclear' according to the criteria listed below. Disagreements will be resolved by discussion and, if necessary, by consulting a third review author. We will not exclude studies based on this assessment, but we will evaluate the overall quality of the available data.

The two key and commonly reported parameters of diagnostic test accuracy are sensitivity and specificity. Because a trade-off may exist between these two parameters, they should be analysed jointly. Sensitivities and specificities for each index test with 95% confidence intervals will be presented in forest plots. In addition, a scatterplot of study-specific estimates of sensitivity and specificity will be used to display data in Receiver Operating Characteristic (ROC) space in Review Manager.    

For the meta-analysis of diagnostic accuracy measures, two models are available: the bivariate model (Reitsma 2005) and the mathematically equivalent hierarchical summary ROC (HSROC) model (Rutter 2001). These models both take into account within study variation and between study variation. Although both models will give the same results when no covariates are added, their focus may be slightly different: the bivariate ROC model focuses on estimating a summary estimate of sensitivity and specificity (and thus focuses at one operating point), whereas the HSROC model focuses on the summary ROC curve as a whole (and thus not on one operating point but on the global accuracy over a range of operating points). For policy making decisions and calculations of costs (financial, psychological and societal), summary sensitivity and specificity are more useful. Thus, we will estimate the diagnostic accuracy for early identification of patients who will probably respond or not to antipsychotic treatment by analysing the results of the included studies separately for each predefined cutoff point.On the other hand, diagnostic tests may report different cut-off values for early improvement. If this is indeed the case, and the quantitative tests are all done at different cut-off values, we will use the HSROC model.

We will summarise the findings of the review in a 'Summary of results' table. This table will include, if possible, a summary estimate of either the diagnostic odds ratio (DOR) or another description of the HSROC curve (when using the HSROC model) or sensitivity and specificity (when using the bivariate model) for relevant subgroups of studies. We will furthermore provide an explanation of these results plus any potential impact for practice, as well as a comment on the quality of the evidence we provide. The presentation of this summary table will make diagnostic information more accessible to healthcare providers and other end users. As predictive values are particularly relevant for policy makers, we will calculate these measures from the pooled estimates obtained from the models. The exact mode of calculation will depend on whether there are a limited number of operating points per test, or not.

Investigations of heterogeneity  

Heterogeneity will be investigated initially through visual examination of forest plots of sensitivities and specificities and through visual examination of the ROC plot of the raw data. It will be formally assessed by examining differences in diagnostic accuracy between subgroups of studies or, if possible, patients. Although both meta-analytic models can be extended to include a covariate into the models, again their interpretation will vary. The HSROC model allows a statistical assessment of the evidence for an association between the position and shape of the SROC curve and potential sources of heterogeneity, while the bivariate model allows a statistical assessment of the evidence for an association between the summary sensitivity and specificity and potential sources of heterogeneity.

Several factors (next to variability in the positivity threshold) may contribute to heterogeneity in diagnostic performance across studies. Sources of heterogeneity that we will assess will be based on data extracted such as age, sex, psychiatric history, baseline severity, treatment and study design. Where there are sufficient data, we will include these features as covariates in the models.

Sensitivity analyses  

Furthermore, in order to prove that the findings from our systematic review are not dependent on arbitrary or unclear decisions, we will perfrom a sensitivity analysis. The meta-analysis will be repeated, substituting alternative decisions of ranges of values for decisions that were arbitrary or unclear. For example, if the eligibility of some studies in the meta-analysis is dubious because they do not contain full details, sensitivity analysis may involve undertaking the meta-analysis twice: first including all studies, and second, only including those that are definitely known to be eligible. We define a priori that we will conduct a sensitivity analysis excluding trials with low quality according to our assessment with the QUADAS instrument (Whiting et al. 2003). Again, if other sensitivity analyses should be made, these will be clearly described as post-hoc.

Myrto Samara

Trogerstrasse 7, 81675, Munich

samaramyrto@gmail.com

TU-München

http://www.psykl.med.tum.de/

Dr Myrto Samara, TU-München
Dr Claudia Leucht, TU-München
Professor Stefan Leucht, TU-München

Mrs Samantha Roberts, University of Nottingham

03 September 2012

15 March 2013

German Ministry of Education and Research

None known

English

Germany

Subject indexing assigned by CRD

Diagnostic Tests, Routine; Humans; Probability; Schizophrenia; Schizophrenic Psychology; Time Factors

05 September 2012

05 September 2012