The heterogeneity of schizophrenia in disease states

Abstract

Previous presentation: Some of the contents of this paper have been previously presented at the 16th Annual Meeting of the International Society for Technology Assessment in Health Care June 20, 2000 in the Hague, Netherlands and at the 21st Annual Meeting of the Society for Medical Decision Making as a poster on October 3, 1999 in Reno, NV. Background: Studies of schizophrenia treatment often oversimplify the array of health outcomes among patients. Our objective was to derive a set of disease states for schizophrenia using the Positive and Negative Symptom Assessment Scale (PANSS) that captured the heterogeneity of symptom responses. Methods: Using data from a 1-year clinical trial that collected PANSS scores and costs on schizophrenic patients (N=663), we conducted a k-means cluster analyses on PANSS scores for items in five factor domains. Results of the cluster analysis were compared with a conceptual framework of disease states developed by an expert panel. Final disease states were defined by combining our conceptual framework with the empirical results. We tested its utility by examining the influence of disease state on treatment costs and prognosis. Results: Analyses led to an eight-state framework with varying levels of positive, negative, and cognitive impairment. The extent of hostile/aggressive symptoms and mood disorders correlated with severity of disease states. Direct treatment costs for schizophrenia vary significantly across disease states (F=27.47, df=7, p<0.0001), and disease state at baseline was among the most important predictors of treatment outcomes. Conclusion: The disease states we describe offer a useful paradigm for understanding the links between symptom profiles and outcomes.

Introduction

The ability to understand how effectively interventions reduce morbidity from mental illness is central to improving the quality of psychiatric care. However, the complex nature of the symptoms of many psychiatric disorders makes defining clinically intuitive outcomes measures often difficult. A researcher faces a constant tension between descriptive realism and simplicity. A case in point is the treatment of schizophrenia, in which patients may manifest a diversity of symptoms. How can one compare, for example, the movement from a florid phase of positive symptoms, marked by hallucinations and delusions, to one marked by hostility and suicidal ideation? Has the patient's situation improved?

Both the Positive and Negative Symptom Assessment Scale (PANSS) (Kay et al., 1987) and the Brief Psychiatric Rating Scale (BPRS) (Overall and Gorham, 1962), have enhanced a researcher's ability to capture the wide array of observed schizophrenic symptoms. Recent factor analyses of data from these instruments have further refined our understanding that the disease may consist of several distinct syndromes (Peralta et al., 1994), which may respond differentially to treatment. Despite these advances in measurement, treatment outcomes continue to be reported in summary measures, such as hospitalizations or percentage changes in mean PANSS scores, which simplify the array of treatment responses among schizophrenic patients. Treatments may change the profile of a patient's symptoms in addition to the severity of specific types of symptoms.

The objective of this study was to derive a clinically intuitive and empirically justifiable set of disease states for schizophrenia using data from the PANSS that captures the heterogeneity of symptoms in schizophrenia. The intent was to bridge the gap between common, but simplistic, outcome measures for schizophrenia treatment and the detail provided in the PANSS. We sought to provide measures of health outcomes that:

• reflect the underlying disease process;

• are generalizable across treatment settings and time;

• are generic with respect to type of interventions being used; and

• can be applied consistently across the wide spectrum of persons suffering from schizophrenia.

Our major goals were to define a set of disease states that are suitable for the study of preferences for health outcomes in schizophrenia and that also describe differential resource use, so that they could be used for analyzing the cost-effectiveness of various treatment interventions for schizophrenia.

Disease states are a means of classifying patients according to their clinical profiles. Defining diseases in terms of discrete, mutually exclusive states can be helpful for several reasons. They can be used to understand disease progression (as in cancer staging), to measure the impact of interventions, and to elicit judgments about how different manifestations of the disease can affect quality of life. Moreover, disease states can provide a framework for clinical decision making. Placing (“diagnosing”) a patient in a particular disease state, for example, may be a trigger for applying more aggressive treatment methods or an entirely different treatment approach. To the extent that the effectiveness of interventions varies by disease state, they can serve as a basis for the development of practice guidelines and, ultimately, measuring the quality of patient care. For cost-effectiveness analysis, the costs of being in different disease states over time can be combined with state-specific patient or societal preferences to compare cost per quality-adjusted life year for various interventions. In essence, the use of disease states can help to distill the complexity of a disease into consistently defined and easily understood measures.

A common definition of disease states for schizophrenia, which has been used as a basis for developing practice guidelines, divides this illness into acute, stabilization, and stable phases (American Psychiatric Association, 1997). While treatment goals and approaches may vary across these phases, these states are difficult to define. As the American Psychiatric Association notes, boundary lines between the phases are neither absolute, nor clear (American Psychiatric Association, 1997). Moreover, these phases are often relative to an individual patient's condition, and, thus, cannot be applied consistently across the wide spectrum of persons suffering from schizophrenia.

Lacking a clear definition of a psychotic episode, most economic studies of schizophrenia have used hospitalizations as a proxy for psychotic relapse Bond et al., 1995, Glazer and Ereshefsky, 1996, Langley-Hawthorne, 1997. While easy to measure, this proxy misses clinically significant exacerbations of psychotic symptoms for patients who are not admitted. Also, the hospitalization proxy requires a dramatic change in patient functioning and does not encourage intervention when patient functioning is poor, although stable. Although other definitions of relapse emphasize elevation of positive symptoms as measured by the BPRS or PANSS (Schooler et al., 1997), the focus on positive symptoms ignores the potentially differential effect of treatments on other schizophrenic symptoms.

Revicki et al. (1996) expands these measures by defining five disease states that describe manifestation of positive or negative symptoms, levels of functioning, and whether the patient is in an inpatient or an outpatient setting. Notably, the states they define are neither exhaustive nor mutually exclusive, nor are they clinically defined. Also, these states are not generalizable across different practice styles and over time, as they reflect explicit clinical decisions regarding hospitalization.

In recent years, cluster analysis has been used to define disease states empirically for some psychiatric disorders, such as depression (Sugar et al., 1998) and schizophrenia Chouinard and Albright, 1997, Lee et al., 2000, Lenert et al., 2000. Cluster analysis is a statistical technique used to find naturally occurring subgroups in data (Kaufman and Rousseeuw, 1990). Chouinard and Albright (1997) conducted a cluster analysis of endpoint PANSS scores on 135 patients with chronic schizophrenia. In their analysis, five clusters were identified, but small sample size enabled the evaluation of only three of those, which corresponded to mild, moderate and severe symptoms based on mean PANSS scores. The Chouinard et al. analysis, while interesting, is lacking in detail about both methods and description of final disease states. These features limit the replication of their results and the utility of their disease states for future outcomes measurement. In addition, the small sample size for their analysis considerably restricted the possible range of disease states that might be considered.

Lee et al. (2000), and subsequently Lenert et al. (2000), conducted analyses that broke symptoms of schizophrenia into four dimensions (thought disorder; withdrawal/retardation; hostility/suspiciousness; and anxiety/depression) of effects prior to clustering. To study associations between these dimensions, these investigators transformed BPRS items using factor weights from a published meta-analysis and then clustered the factor-weighted scores. The resulting cluster analysis identified a model with six discrete states, with differing levels of symptoms in each of the four dimensions. One limitation of these studies was the adoption of a model based on the BPRS scale, which contains less information on schizophrenic symptoms than the PANSS, and may not fully capture the range of symptoms experienced by patients. We sought to improve upon previous research by using the PANSS to derive a set of disease states for schizophrenia that captured the heterogeneity of symptom responses.