Extrapolation between measures of symptom severity and change: An examination of the PANSS and CGI

https://doi.org/10.1016/j.schres.2007.09.006Get rights and content

Abstract

Objective

Research based on pooling data from clinical trials suggests that it is possible to extrapolate from a CGI-S to a PANSS severity score and from the CGI-I to a PANSS percentage change score. This research has not been replicated nor examined in individual trials. This study aims to examine the feasibility of extrapolation from the CGI to the PANSS across and within 4 large clinical trials of antipsychotic medication.

Methods

Equipercentile linking is used to examine extrapolation (a) from CGI-S to PANSS severity ratings and (b) from CGI-I to PANSS percentage change (n = 2698). Linking is conducted at baseline and after 2, 4, 6 and 8 weeks of treatment from ITT clinical trial participants with schizophrenia.

Results

Across weeks 2, 4, 6 and 8, being considered ‘not ill’ according to the CGI-S corresponded to PANSS scores of 31–2. The relationship between the CGI-S and the PANSS followed an increasing trend, such that ‘very mild’ corresponded with 41–7, ‘mild’ corresponded with 55–62, ‘moderate’ corresponded with 71–7, ‘marked’ corresponded with 88–94, ‘severe’ corresponded with 105–110, and ‘extremely severe’ corresponded with 126–134. The relationship between CGI-I ratings and percentage change followed a linear trend, such that ‘very much improved’ corresponded to PANSS percentage change scores from 79 to 75, ‘much improved’ corresponded with 45 to 49, ‘minimally improved’ corresponded with 21 to 23, ‘unchanged’ corresponded with 2 to 3, ‘minimally worse’ corresponded with − 15 to − 20, ‘much worse’ corresponded with − 44 to − 51. Generally, within the trials the cut-off ranges identified overlapped within around 10 points of those found in the pooled analysis.

Conclusions

Despite trial heterogeneity, the results support the extrapolation from the CGI-I to PANSS percentage change. Extrapolation of the CGI-S to the PANSS is observed, except in the case of severe symptomatology which is rare. Collectively, the results support the extrapolation between the PANSS and CGI.

Introduction

Two of the most widely used measures of treatment efficacy in clinical trials of antipsychotic medications are the Positive and Negative Symptoms Scale (PANSS) and the Clinical Global Impression Scale (CGI). The PANSS aims to provide a comprehensive measure of symptomatology (Kay, 1990, Kay et al., 1987, Kay et al., 1989). It consists of 18 items from the Brief Psychiatric Rating Scale (Overall and Gorham, 1962) measuring positive symptoms, general psychopathology and affective symptoms, and 12 items from the psychopathology rating schedule (Singh and Kay, 1975). Together these form 7 items to measure positive symptoms, 7 items to measure negative symptoms and 16 items to measure general psychopathology. The PANSS items are frequently summed to provide a general index of symptom severity (Kay et al., 1987). The CGI (Guy, 1970) has two versions each consisting of one item on a seven point scale that is rated by symptom severity or change. Therefore, the CGI provides an overall index of symptom severity or change. Thus while the PANSS is more comprehensive, the CGI may be more practical to administer. Also, the CGI is easier to use and understand, although the PANSS may be preferred due to its strong psychometric properties of reliability and validity (Reviewed in Kay et al., 2000). The clinical implications of the PANSS, however, are sometimes not readily apparent (Leucht et al., 2006). For instance, it is difficult to provide a clinical judgment based on a PANSS total score of 50 or 70. Similarly, PANSS change scores range from 20% to 50% (e.g., Leucht et al., 2006). The use of the PANSS in this way assists in the definition of “response”, but what these cut-offs mean from a global “clinical” perspective is unclear. The CGI, however, is a frequently used instrument that appears to be more informative in this regard. It describes a patient's overall clinical state as a global impression by the rater. It therefore provides more readily understood clinical information than the PANSS which has more desirable psychometric features. This research is informative to clinicians since it explains the relationship between the PANSS (a widely used 30 item research instrument of symptom severity) with global ratings that are easily understood (i.e., the CGI).

This makes it desirable to extrapolate from the CGI to the PANSS. Past research based on pooling data from clinical trials indicates that it is possible to extrapolate from the CGI to the PANSS (Leucht et al., 2005b). For instance, past research over a 6 week period indicates that a CGI rating of mildly ill corresponds to a PANSS total of 58 at baseline. A minimal improvement according to the CGI corresponds to a PANSS mean percentage reduction of 19% after 1 week, a figure that increases during the course of a trial. The purpose of the current study is to examine the extrapolation between CGI and PANSS severity or change across and within individual clinical trials of antipsychotic medication over an 8 week period.

Section snippets

Methods

PANSS and CGI data on 2698 persons were extracted and examined from four randomized controlled clinical trials of antipsychotic medication used to treat schizophrenia. These included: INT-2, n = 1362 (Peuskens, 1995) which compared risperidone to haloperidol; INT-3, n = 520 (Marder and Meibach, 1994) which compared risperidone, haloperidol and placebo; USA-121, n = 283 (Kane et al., 2003) which compared long acting injectable risperidone to placebo; and INT-35, n = 533 (Schooler et al., 2005) which

CGI–PANSS correlations

Spearman correlation coefficients between CGI-S and the PANSS were r = .61 (n = 2621) at baseline, r = .68 at week 2 (n = 2432), r = .71 at week 4 (n = 2227), week 6 r = .72 (n = 2001), and r = .73 at week 8 (n = 1810). This pattern indicated that the correlation between CGI-S and the PANSS increased with time and repeated testing. The correlations between PANSS percentage change and the CGI were on weeks 2 through 8: .61, .67, .67 and .68, respectively (all correlations, p < .001).

Linking the CGI-severity score and the PANSS total score

Total PANSS and CGI-severity score

Discussion

In contrast to previous studies using equipercentile linking (Leucht et al., 2006, Leucht et al., 2005a, Leucht et al., 2005b) the current study examines the correspondence between the PANSS and CGI within and across clinical trials. Despite seemingly heterogeneous trials a general trend emerges providing robust estimates linking the CGI and the PANSS both by severity and change. The results differ to past report (Leucht et al., 2005b) in one main respect. The upper range of symptom severity

Conclusions

In conclusion, the current study indicates that it is possible to retrieve more comprehensive PANSS total and change ratings from the more general and shorter CGI. The PANSS and CGI correspond but are not interchangeable since the PANSS allows the assessment of specific symptoms (e.g., positive and negative). Thus as the results show the PANSS and CGI show a high degree of correspondence, however, since they differ in meaning, there is added benefit of using both measures. Furthermore, our

Role of funding source

No funding was provided for this research.

Contributors

Authors Levine and Rabinowitz wrote up the main part of this article. Authors Levine and Rabinowitz interpreted the results and Levine drafted the manuscript. Rabinowitz edited and commented on the manuscript and approved it for submission. The above was done in total conjunction with our colleagues in Germany, Engel, Leucht and Etschel, all of whom contributed substantially and analytically to the body of this work. Levine submitted this research.

Conflict of interest

The authors have no conflict of interest relating to this article.

Acknowledgement

The authors wish to acknowledge Johnson & Johnson Pharmaceutical Research and Development, Beerse, Belgium, for providing the data for this work.

References (19)

  • LeuchtS. et al.

    What does the PANSS mean?

    Schizophr. Res.

    (2005)
  • BernsteinI.H. et al.

    Factoring items and factoring scales are different: spurious evidence for multidimensionality due to item categorization

    Psychol. Bull.

    (1989)
  • GuyW.B.

    Manual for the ECDEU Assessment Battery

    (1970)
  • JensenA.R. et al.

    Equating the advanced and standard forms of the Raven progressive matrices

    Educ. Psychol. Meas.

    (1988)
  • KaneJ.M. et al.

    Long-acting injectable risperidone: efficacy and safety of the first long-acting atypical antipsychotic

    Am. J. Psychiatry

    (2003)
  • KayS.R.

    Positive–negative symptom assessment in schizophrenia: psychometric issues and scale comparison

    Psychiatr. Q.

    (1990)
  • KayS.R. et al.

    The positive and negative syndrome scale (PANSS) for schizophrenia

    Schizophr. Bull.

    (1987)
  • KayS.R. et al.

    The Positive and Negative Syndrome Scale (PANSS) Manual

    (2000)
  • KayS.R. et al.

    The Positive and Negative Syndrome Scale (PANSS): rationale and standardisation

    Br. J. Psychiatry Suppl.

    (1989)
There are more references available in the full text version of this article.

Cited by (105)

  • ABCB1, ABCG2 and CYP2D6 polymorphism effects on disposition and response to long-acting risperidone

    2021, Progress in Neuro-Psychopharmacology and Biological Psychiatry
  • Association between discontinuation due to withdrawal of consent and use of long-acting injectable antipsychotics: A meta-analysis of randomized trials for schizophrenia

    2021, Journal of Psychiatric Research
    Citation Excerpt :

    According to a recent study of prescription data, the DDD method was more applicable in clinical practice and drug utilization research than was the chlorpromazine equivalent, which is a traditional method that has been used for decades to indicate the dose of antipsychotics (Lin et al., 2018). We defined the studies including patients in the maintenance phase as follows: (1) studies reported to include only clinically stable patients, (2) mean Clinical Global Impressions-Severity of Illness (CGI-S) (Guy and Bonato, 1970) score <4 at randomization (because most acute schizophrenia trials enroll patients with CGI ≥4) (Supplementary Table 1), and (3) mean Positive and Negative Syndrome Scale (PANSS) (Kay et al., 1987) total score <75 at randomization (because the range of PANSS total scores for a CGI-S score of 4 was calculated as 71–77) (Levine et al., 2008). More of the studies included in our meta-analysis used CGI-S instead of PANSS for baseline symptom assessment (Supplementary Table 1).

View all citing articles on Scopus
View full text