Abstract
Considerable recent attention has been given to the question of whether it is possible, or could become possible, to predict the initial onset of schizophrenia and related psychotic disorders. The question of whether it is possible to predict the imminent onset of an episode of psychotic relapse, however, has gained considerably less attention in recent times. This is despite the answer to the question remaining unclear and the widespread development of clinical programs that to some degree presume that it is possible and practical with current clinical strategies.
The capacity to predict the initial onset of a low prevalence disorder such as schizophrenia is dependent on the incidence of the disorder in the baseline population or population in which a detection program is implemented [1]. In contrast, the incidence of relapse in patients who have experienced one or more psychotic episodes is high and ranges between 50 and 80%% in the year following an episode of psychosis without treatment [2]. There is evidence that the presence of psychosis itself may have a detrimental effect on the long-term prognosis of the disorder. Evidence for this comes from two areas. Firstly, a number of studies have suggested a relationship between the duration of untreated psychosis (DUP) in patients with first-onset psychosis and prognosis. Patients with a longer duration of untreated symptoms have been shown to have a slower and less complete recovery with higher relapse rates and substantially higher health care costs over the first 3 years following the initial episode [3–6] although a number of recent studies have produced conflicting results [7–9]. The interpretation of these studies is complicated by their capacity to only demonstrate an association between DUP and prognosis rather than a causal relationship. It is possible that patients with slow onset insidious disorders have a poorer prognosis and these patients may exist ‘quietly’ in the community for months or years prior to coming to clinical attention or that increased DUP is associated with other ‘poor outcome’ clinical variables. Studies controlling for some suggested clinical factors appear to show a relationship between DUP and outcome [10], [11]. The study of Wyatt et al. also appears to deal with this complication [12]. In this study the original findings of May et al. published in 1976 [13] were reanalysed. In this, hospitalised psychotic patients were randomised to receive one of psychotherapy, antipsychotic medication, electroconvulsive therapy or a combination of psychotherapy and a biological treatment. The patients receiving a biological therapy were shown, in the original paper, to do significantly better that those receiving psychotherapy alone. The reanalysis of Wyatt looked at the post discharge outcome between the various groups. The patients receiving psychotherapy later received medication and achieved adequate clinical responses. However, over the subsequent 3 years these patients suffered greater relapse rates and were more frequently hospitalised despite no differences in their ongoing treatment. These patients were also reported to have slightly lower global assessment of functioning (GAF) scores 6 years following the original treatment. Wyatt et al. argued that the primary difference between the patient groups was the delay in somatic treatment in the psychotherapy group that is proposed to have contributed to poorer outcome [12].
There also appears to be an association between relapse and outcome for subsequent episodes of illness. In particular, Wiersma et al. have demonstrated that with each subsequent episode of relapse, there is a progressive increase in the proportion of patients categorised as no longer responding to treatment [2] and these results appear to be confirmed by the preliminary results of a second study [14].
If there is an association between the actual presence of psychotic symptoms and prognosis, it is crucial to predict and prevent the onset of the disorder, and once individuals have experienced a psychotic episode, to predict and prevent relapse. The most powerful means of preventing relapse is with the provision of maintenance antipsychotic medication. It is reasonable to assume however, that most patients will cease medication, for a variety of reasons, at one, if not many times during a life history of schizophrenia. This then raises several crucial questions: Can we adequately predict relapse? If so, how? And, how can this knowledge be best implemented in clinical programs to improve patient outcomes?
Can we predict relapse?
Prior to the 1980s, there was little systematic research in this area. Several early reports concentrated on the phenomenological description of the process of the prodrome of relapse. In particular, a temporal pattern was suggested in which patients experienced a series of stages prior to the onset of psychosis. These stages varied between studies, but in general it was suggested that patients would progress from general nonpsychotic symptoms through ‘sub threshold’ psychotic symptoms, to complete relapse [15–17]. It has been suggested that this sequence would be independent of the rapidity with which relapse would occur [15]. Although many of the subsequent studies have been based on the presumption that the emergence of psychosis would follow these nonspecific symptoms, it is not clear if this is actually the case. It has been suggested that when carefully assessed, low-grade psychotic symptoms may in fact be present during the time which is regarded as prodromal [18], [19] and that in fact the non-specific symptoms may be secondary to the patients difficulty in coping with these actual emerging psychotic symptoms [20].
The studies that have been conducted in this area are further complicated by a frequent failure to clearly distinguish what the authors of the study are actually trying to predict. In particular, many studies fail to distinguish the detection of the exacerbation of psychotic symptoms in patients who have incompletely responded to treatment, from the occurrence of relapse in patients who have achieved remission, at least of positive symptoms [20]. This distinction is of considerable clinical relevance as patients in true ‘remission’ are likely at some stage to stop medication, either supervised or otherwise, and are likely to have considerable psychosocial ‘gains’ that are at risk with relapse.
The first systematic studies of relapse prediction were published in the 1980s. Herz and Melville [21] carried out detailed interviews with 145 patients and families and reported that the majority of both groups detected nonpsychotic changes preceding relapse. There was a reasonable concordance between families and patients (two thirds of the time). The early signs most commonly reported were the non-specific signs of nervousness, increased tension and an alteration of sleep and appetite. Several similar studies have been published with the consistent report of the common occurrence of nonspecific early warning signs (EWS) [22–24].
Further development of knowledge in this field came from the interpretation of studies of medication treatment through the 1980s and early 1990s [25–27]. Most of these studies were not designed to study the prediction of relapse but important information may be drawn from them. Several of these studies were based around trying to establish the role of maintenance versus intermittent medication treatment strategies. In several studies, patients were allocated to continual treatment or treatment withdrawal with targeted intervention at the detection of EWS of relapse. In general, these studies have shown an advantage of maintenance treatment over intermittent treatment [28]. However, the overall rate of relapse in the targeted treatment groups were reduced compared to what would be expected with the natural history of the disorder although this has not been directly addressed in most studies [28]. In addition, relapse rates appear to be lower when intermittent medication treatment is compared to targeted nonpharmacological treatment. Gaebel et al. [25] studied a group of patients treated over two years with one of three conditions: (i) maintenance antipsychotic medication, (ii) gradual medication withdrawal and recommencement at the earliest sign of relapse, and (iii) medication withdrawal and (nonmedication) crisis intervention at the early signs of relapse. In the crisis intervention group, medication was only recommenced upon complete relapse (as defined by Brief Psychiatric Rating Scale (BPRS) change criteria). Prediction of relapse was based around six non-specific symptoms drawn from earlier research. Two interesting observations can be drawn from this study. Firstly, the sensitivity and the positive predictive value of the EWS were poor although the specificity of the predictive measures was high. Secondly, despite the poor predictive power of the measures, the clinical outcome of the intermittent medication group was superior to that of the crisis intervention group. Thus, although the predictive value of the non-specific EWS was low, there was clinical value in the early detection and intervention strategy applied.
The poor predictive value of non-specific measures is a considerable problem found in a number of other studies. The positive predictive value (PPV) of a measure is dependent on its sensitivity (true positive rate – in this case the percentage of episodes preceded by EWS), and its specificity (true negative rate – percentage of patients who do not relapse and do not demonstrate EWS) as well as the incidence rate of relapse in the population. As the incidence increases, so does the PPV for a measure with the same sensitivity and specificity. Thus the capacity to predict relapse in the clinical setting where relapse is relatively common, is likely to be a much more achievable goal than the prediction of the onset of a low prevalence disorder, such as schizophrenia, in the general population.
These characteristics have been analysed in several studies of relapse prediction. Jolley et al. studied a group of patients receiving either maintenance medication or placebo over the course of 2 years and looked at the relationship between non-specific EWS and relapse [26], [29]. Sensitivity rates fell from 73%% in year 1 to 53%% in year 2 and the overall PPV was only 16%%. Several specific studies of relapse prediction have also been published. Birchwood et al. developed an early signs scale (ESS) that consisted of 34 items drawn from family and patient interviews [30]. The ESS was applied to 19 patients every 2 weeks at clinical review or by post. It appeared to have reasonable predictive value with a PPV over 70%% using either an increase in symptoms or a score above a cut off point of 30. However, the study had a small sample size, has not been replicated and used problematic definitions of relapse (readmission or incipient relapse based on clinical judgement). In addition, the study did not distinguish between patients in remission who experienced an episode of relapse versus those with ongoing symptoms who experienced an exacerbation in their clinical state. Malla and Norman also looked at the relationship between nonspecific EWS and the worsening of psychotic symptoms (not relapse from remission) in 55 outpatients [31]. Nonspecific symptoms showed a high degree of specificity but low sensitivity and poor predictive quality with either a dimensional or categorical analysis.
The most recent study utilising non-specific EWS is that of Jorgensen [32]. This study used the ESS and an eight-item derivative (the warning signals scale or WSS) to predict the subsequent development of delusions (not generalised relapse). The WSS was defined in the first phase of the study and its usefulness tested in a sample of 71 patients followed twice a week over 6 months. The WSS demonstrated reasonable sensitivity (77%%) and specificity (68%%). It is a quick and easy tool to implement with just eight yes or no items. The positive predictive value was reasonable (78%%) but the capacity to generalise these results to a broader definition of relapse is unclear.
Several attempts have been made to use other symptom clusters or phenomena to improve prediction. Tarrier et al. followed 56 patients post discharge [33]. An increase in Psychiatric Assessment Scale (PAS) scores (rather than non-specific symptoms) was related to the subsequent development of relapse that was defined as an increase in positive symptoms for at least a week. This resulted in reasonable sensitivity and specificity but the criteria for prediction and relapse appear to be too similar to be meaningfully applied, especially to subjects in remission.
One of the more innovative approaches has been to study the value of an individual relapse signature in the prediction of relapse. Marder et al. studied patients receiving low dose maintenance medication who were randomised to receive either antipsychotic medication or placebo at the earliest sign of relapse [34]. The use of a placebo group allows an accurate calculation of the relationship between EWS and relapse that is not affected by the initiation of treatment that may avert relapse. Three scales were utilised to predict relapse: (1) the anxiousdepressive subscale of the BPRS (2) a modified version of the ESS and (3) the Idiosyncratic Prodrome Scale (IPS). The IPS was developed to rate each patient's individual ‘relapse signature’. On the basis of each individual case history, a scale was developed including ratings of the three symptoms thought to occur most regularly prior to relapse. Both the BPRS subscale and the IPS predicted relapse but with specificity levels likely to result in low PPV. The value of the IPS was further studied with similar methods in a second study [35]. In this population, the sensitivity was as low as 37%% with a 43%% PPV. Interestingly, like the study of Jolley et al. [26], [29], there was a decrease in sensitivity in the medication treated group over time. This suggests the clinicians got better at predicting relapse that was then prevented by the implementation of treatment.
In general these studies fail to confirm that relapse can be predicted with the accuracy required for clinical implementation. As suggested, there are several major issues with the studies themselves that may contribute to this negative conclusion and the studies suggest factors that may produce more useful results. Firstly, the timeliness of monitoring appears important as suggested by Gaebel et al. where less frequent reviews produced more incorrect predictions [25]. Secondly the type of symptoms rated appears crucial. Most of the studies reviewed used non-specific symptoms, usually rated on structured rating scales with a cut-off defining the possibility of relapse. However, Gaebel et al. indicates that prediction is better when based upon a combination of rating scales and clinical judgement [25]. It would also appear to make sense that the addition of early or sub threshold psychotic symptoms to these non-specific symptoms should enhance predictive value and re-emergent psychotic symptoms are likely to be more acceptable to clinical staff in the adjustment/initiation of antipsychotic medication [36]. In addition, where utilised, the combination of carer and patient ratings appears of value (for example Birchwood et al. [30]) and there is evidence of consistency between these responses. Of clinical significance, the studies of Jolley et al. and Gaebel et al. both suggest that the predictive value of EWS appears independent of whether patients are currently receiving medication or not [25], [26]. This has importance in the clinical setting where implementation programs may be useful both in maintenance treatment and where patients have been withdrawn from therapy or chosen to cease treatment themselves.
Is it possible to produce a better predictive outcome? The studies reviewed suggest the best approach required the utilisation of clinical judgement, non-specific and specific symptoms, frequent assessments and the involvement of patients, clinicians and carers. Can this approach be supported? Two recent studies provide evidence for this. The first of these is a study of the use of diazepam for early relapse [37]. In this study, 53 patients were withdrawn from antipsychotic medication and carefully monitored. At the detection of EWS they were randomised to diazepam, fluphenazine or placebo. They received this treatment over the next 6 weeks or until ‘further exacerbation’ occurred as defined by either clinician judgement or rating scale (BPRS and CGI) scores. The program for the detection of EWS involved weekly review by an experienced clinician. Early warning signs were defined by one of clinician judgement, an increase on one or more of several BPRS items, the development of marked insomnia or family reports of changes sleep pattern or a CGI rating increase. Importantly, BPRS items rated included both non-specific (somatic concern, anxiety, hostility) and specific psychosis related (conceptual disorganisation, suspiciousness, hallucinatory behaviour) items. Additionally, assessments were not only frequent, but also timely. They were provided on a 24 h basis, 7 days per week when required, both for the detection of EWS and the detection of relapse once patients were randomised to treatment. Progression rates varied from 47%% in the diazepam group, 56%% for fluphenazine and 70%% in the placebo group. In the context of this discussion, there were two important results. Firstly, no patient suffered a relapse without prior detection of EWS (100%% sensitivity). Secondly, of the patients experiencing EWS who were randomised to placebo, 70%% progressed to meet relapse criteria (70%% specificity). The PPV of 70%% is likely to be an underestimation as non-specific aspects of the placebo condition are likely to have had some effect in averting progression to relapse. This study was designed to analyse the usefulness of diazepam therapy in early relapse, not the predictive value of the measures applied and the number of patients in the study arms was small. However, the study produced promising results and outlines a comprehensive methodology for relapse prediction. In addition, it clearly indicates that interventions can be applied to reduce the rate of progression to relapse. However, this is applied in a treatment intensive manner that is likely to consume considerable resources if applied to large numbers of patients.
The second and most comprehensive study to address the question of relapse prediction and intervention was published recently by Herz et al. [38]. In this study, 82 patients were randomised to either treatment as usual (TAU – biweekly individual therapy and medication review) or the program for relapse prevention (PRP). PRP was multifaceted and involved components of (a) patient and family education, specifically in regards to relapse prevention (b) active monitoring of early relapse signs by patients, families, clinicians and others (c) biweekly multifamily education groups (d) weekly supportive group or individual therapy (e) early treatment for EWS that included crisis problem solving, supportive therapy visits and increased medication (usually increased dose by ∼20%%). A shortened version of the ESS was administered prior to each individual therapy session. Prodrome of relapse was defined by the judgement of the treating team psychiatrist and could include an increase in dysphoria, mild psychotic symptoms or signals identified from previous episodes. Clinicians in the PRP program were provided prodrome recognition education prior to the study commencement.
The study produced several results that are worthy of comment. Firstly, patients in the PRP group experienced a significantly lower rate of relapse and hospitalisation. In regards to prediction, 35%% of relapse episodes (7 out of 20) experienced by patients in the TAU group were not preceded by the identification of prodrome. This was significantly higher than the figure for the PRP group (1 out of 24 (4%%)). The PRP program was associated with a higher rate of false positives (prodromes to stable outcomes) but a significantly higher prodrome to relapse rate. Due to the active intervention applied to all identified with prodrome, it is not possible to accurately calculate specificity rates. However, the authors conclude that they appeared to have had a reasonable rate of successful early intervention to false positives.
This study is of considerable significance as it is the most comprehensive attempt to implement and evaluate a multifactorial early detection and intervention program. It is not possible to determine whether individual aspects of the program contributed excessively to its successful results. It was of interest though that only 29%% of patients in the PRP group had family able to participate in the multifamily groups, and these patients were felt by the authors to do particularly well. It is possible that the capacity to enrol as higher proportion of patients with intact carer networks would produce even more compelling results.
Suggestions
The prediction of relapse appears an achievable goal for a substantial proportion of patients with schizophrenia. Successful prediction appears most likely to require a combination of factors built into an integrated program. These factors would appear to include:
(a) The measurement of non-specific and psychosis related symptoms. Particularly useful non-specific symptoms include mood symptoms (dysphoria, tension, anxiety) and ‘vegetative’ symptoms (altered sleep and appetite). Specific symptoms include ‘sub threshold’ or transient psychotic symptoms and may also include symptoms specific to the individual patient as determined by the review of previous episodes.
(b) The inclusion of carers, patients and clinicians in the monitoring process. The necessity for engagement of families in the recovery process from a psychotic episode is well recognised. It appears clear that family psycho-education should include discussions of early signs of relapse, at risk periods and possible early intervention strategies. The provision of this education and support on an ongoing basis appears to be superior to single education sessions. Multi-family groups appear to be an appropriate and resource-effective format for these programs.
(c) The use of clinical judgement in tandem with structured assessments. Tools such as the WSS or ESS may be usefully adapted to clinical practice and act as an aid for less experienced clinicians. However, these should not be used as a substitute for clinical training, education and supervision. The therapeutic relationship built between an individual clinician and patient appears a very important factor in the recognition of early relapse and the strengthening of longitudinal relationships should be encouraged.
(d) Assessments need to be frequent and timely. The frequency of monitoring of patients in the maintenance phase of treatment is often gradually reduced in clinical settings and falls below the frequency levels applied in the programs described (even the frequency of contact in the TAU group of Herz et al. was described as being more intense than local usual practice). This appears a natural response when patients are stable and not requiring intensive treatment. It is also a practical response to the needs of busy services. Programs such as that described by Herz et al. [38] utilise experienced clinicians and are likely to be service intensive. Can these principles be adapted to other clinical settings that are less intensive or service demanding? Several possibilities exist. Firstly, many of the principles described are already part of many standard community based treatment programs. These programs may only need to be changed slightly to improve their capacity to detect prodromal relapse symptoms. Secondly, a variety of shared care and primary care programs for the management of individuals with schizophrenia are under development or investigation. The education of family physicians in the principles of detection of EWS, the provision of brief structured assessments and the provision of on-call specialist services for further assessment would all seem to be useful and realistic. The implementation of an ongoing relationship between specialist services and family physicians involving training, supervision and secondary consultation would appear to offer substantial advantages over once-off education sessions. Finally, where as the widespread/ongoing application of these principles seems ideal, it may not always be clinically practical, for example, for a patient who has been in remission for a period of time and has returned to full-time employment. Under these circumstances, a targeted approach may be useful where ‘high-risk’ periods could be identified in advance, perhaps with specification in a management plan, and more intensive monitoring initiated intermittently. Appropriate times would include during the adjustment of medication dosage, during particularly stressful times or during substance misuse or withdrawal.
(e) It would appear useful prior to the implementation of an individual monitoring program, to clearly distinguish whether the patient is relatively stable but with persisting psychotic symptoms that have not fully responded to treatment or whether the patient actually is in remission. In the former case it is reasonable to assume that deterioration is likely to present with a worsening of the chronic symptoms, whereas for the patient in remission the more detailed considerations described above apply. This distinction could be documented as part of a management plan that describes the relevant EWS.
Conclusions
Evidence is progressively accumulating that the prediction of episodes of psychotic relapse is a realistic goal. Most importantly, evidence indicates that interventions based upon programs of early detection can reduce rates of illness relapse. This will reduce suffering and the secondary morbidity associated with schizophrenia. A reduction in relapse rates is likely to lower costs, especially associated with hospitalisation, and this may substantially offset the costs associated with monitoring programs. Further research is required to establish the usefulness of monitoring programs in general psychiatry and family practice settings, to clarify the predictive value of brief assessment tools and to establish the cost benefits of detection and intervention programs.