Psychiatry and Clinical Neurosciences (2001), 55, 587–593
Regular Article
Schizophrenic outpatient perceptions of psychiatric treatment and psychotic symptomatology:An investigation using structural equation modeling NAOKI HAYASHI, md, phd,1,3 MITSURU YAMASHINA, md,2 HISAKO TAGUCHI, md,3 NAOKO ISHIGE, md4 AND YOSHITO IGARASHI, md3,5 1
Department of Schizophrenia Research,Tokyo Institute of Psychiatry, 2Department of Psychiatry, School of Medicine, Juntendo University, 3Department of Psychiatry,Tokyo Metropolitan Matsuzawa Hospital, 4Tokyo Metropolitan Center for Mental Health, 5Department of Mental Health Services,Tokyo Institute of Psychiatry, Tokyo, Japan
Abstract
Schizophrenic patient perceptions of treatment have clinical value and deserve detailed psychiatric investigation. The present study sought a model indicating statistically estimated cause–effect relationships of perceptions and psychotic symptomatology of outpatients with schizophrenia by applying a method of structural equation modeling. The perceptions included in this model were patient satisfaction with treatment, perceptions of their treating psychiatrists, and patient-role perception. Scores of Positive and Negative Syndrome Scale and poor insight measures were added to the model as possible influential factors. The constructed model revealed that the poor insight exerted a major influence on the patient-role perception that had small effects on the reliable therapist perception and the satisfaction. It was also shown that satisfaction was chiefly determined by the reliable therapist perception that was formed in the treatment relationship, rather independently of the other construct. These findings were valuable in terms of their implications for understanding the makeup of the perceptions and the strategy for interventions to improve them.
Key words
insight into illness, patient attitude, patient perception, patient satisfaction, patient–physician relationship.
INTRODUCTION Perceptions of the patients are increasingly seen as essential in evaluating psychiatric treatment.1,2 Among others, patient satisfaction with treatment can be used as an index to assess the quality of treatment.1 Their perceptions of therapist and treatment would also be relevant. In addition, these patient appraisals have taken on some clinical significance. They are closely related to patient motivation to receive treatment and cooperation with therapists, and can form an important part of treatment
Correspondence address: Naoki Hayashi, Tokyo Institute of Psychiatry, 2-1-8 Kamikitazawa, Setagaya-ku, Tokyo 156-8585, Japan. Email:
[email protected] Received 22 December 2000; revised 8 June 2001; accepted 18 June 2001.
compliance.3–5 These perceptions deserve detailed investigation. There are supposedly several factors that have some influence on perceptions. For instance, distressing psychotic symptoms possibly lowers patient satisfaction with treatment. Moreover, they may affect the sense of reality, and produce distortions of perceptions. Among them, poor insight into illness, which many patients with schizophrenia suffer,6 could impair perceptions of treatment and appropriate patient attitude and treatment relationship. The aim of the present study was to obtain answers to a series of questions about the relationships of patient perceptions and psychotic symptoms: which psychotic symptoms affect patient perceptions, to what extent does it influence them, is it possible to postulate causal relationship among the perceptions, etc. For that purpose, we used the methodology of
N. Hayashi et al.
588
structural equation modeling, which has the potential to render a statistical model of the causal structure.7,8 While a cross-sectional study cannot establish causal directionality, the analysis permits us to postulate a credible view on the cause–effect relationships from data of that kind. The model would also be a useful guide to seek more favorable patient–therapist cooperation and to improve the quality of treatment.
SUBJECTS AND METHODS Patients and psychiatrists One hundred and eighteen outpatients with schizophrenia (68 males and 50 females) and their treating psychiatrists (seven males and two females) participated in the present study. The subject patients were currently treated at Tokyo Metropolitan Matsuzawa Hospital, Tokyo, Japan, a large psychiatric center in central Tokyo for regional psychiatric services and treatment of difficult patients. The patients were clinically in a stable condition without mental retardation, organic brain diseases or severe physical diseases. Diagnosis was made on the basis of examining case records according to DSM-IV criteria. We obtained written informed consent from the subjects to include them in this study. The patients comprised mostly of long-term users of psychiatric services. Mean age (SD) at investigation and mean age at onset were 41.4 (11.4) and 25.1 (7.9) years, respectively. They had been hospitalized as inpatients a mean (SD) of 3.4 (2.7) times. Mean inpatient days in the past 3 years was 131 (221) days. On average they were taking 924 (794) mg of chlorpromazine equivalent neuroleptic dose. Mean duration of successive outpatient treatment was 1427 (1879) days, and that of the current treatment relationship, 993 (1066) days. Sixty-two (51.8%) of the subjects had a history of full-time regular work. Only 33 patients (12 males, 21 females) had ever married. The mean duration of patient full-time education was 12.7 (2.3) years. The mean age of the psychiatrists who were in charge of the patients was 34.8 (5.5) years. They had clinical experience of a mean of 9.3 (5.4) years of primarily inpatient treatment for psychotic illnesses. The treatment dealt with in the present study setting is included in the concept of supportive psychotherapy or clinical management.9 The psychiatrists, who took part in the treatment with the patient as a therapist, provided psychological support, helped the patient cope with the manifestations and consequences of illness, facilitated adherence to a medication regimen, and gave practical advice and guidance when neces-
sary. They had an interview with their patients at 1–4 week intervals.
Assessment of the patients Scale of clinical symptoms Current clinical symptoms of the patients were rated according to the Positive and Negative Syndrome Scale (PANSS).10 The scale has been widely used for evaluating psychotic symptoms, consisting of 30 symptom items, rated on a 7-point scale, and three subscales. With respect to the reliability of PANSS, our research group reported previously that the rating attained a satisfactory level of interrater reliability.11 Positive and Negative Syndrome Scale positive and negative subscale scores were used in the model construction.
Measures of poor insight For the purpose of assessing insight into illness, we firstly used PANSS item G12 (the 12th item of general psychopathology subscale), lack of judgment and insight. This item includes examination of patient recognition of psychiatric illness, symptoms, and need for treatment. However, PANSS item G12 is a single item scale, and rather weak in assessing insight since it is alleged to be a multidimensional construct.6 Therefore, we added three general items from Amador et al.’s insight scale.12 They are 5-point scales on current recognition of illness, effects of medication, and social consequences of illness, which are supposed to represent some aspects of the insight concept. Our previous study5,11 showed that scores of the three items and PANSS item G12 have reliability and validity at a certain level.
Measure of the patient-role perception To assess the patient-role perception of the subjects, we used Awareness of Being a Patient Scale (ABPS),5 a 25-item, 4-point self-report scale that has been developed for measuring schizophrenic patients’ psychological attitude towards the treatment situation. The scale is divided into two subscales: recognition of the need for treatment (subscale 1) and acceptance of the treatment situation (subscale 2), which are in accordance with two components of the sickrole fulfillment on the side of the patients in the Parsons’ concept.13 In the present study, the scores of the two ABPS subscales were used. A high ABPS score indicated a more appropriate patient-role perception.
Patient perceptions and psychotic symptoms
The patient perception postulated in ABPS is closely related to insight into illness. However, there is an essential difference. The scale does not address the precise understanding of the illness and symptoms, which is the core component of insight. It is a patient perspective that ABPS measures, while insight is assessed by whether the patient recognizes the one postulated from the side of psychiatry. It is also suggested that the makeup differs from that of impaired insight in being markedly influenced by psychosocial factors, while the latter is virtually included in psychotic symptomatology.5,14,15
Reliable therapist perception scale The method for evaluating perceptions that patients possess in the treatment relationship through the Semantic Differential technique16 was applied. We previously reported that composite scales consisting of the estimations of treating psychiatrists by patients had validity in characterizing the treatment relationship.17 The scale used in the present study was that of a reliable therapist perception of patients, which has been ascertained to reflect patient–therapist cooperation,17 and reported to have value in predicting the favorable outcome of a psychoeducation program.18
Satisfaction measure A newly devised self-report scale was used to measure patient satisfaction with treatment. The scale is a 7-point Lickert scale composed of six items, which each patient rates on the degree of satisfaction with treatment. The items were on (i) the psychiatrist’s explanation for neuroleptic treatment, (ii) the explanation of the illness, (iii) advice and guidance, (iv) the level of understanding of the patient, (v) the clinical competence of the psychiatrist, and (vi) the matching of the treatment. A high score on the satisfaction scale indicated that the patient was more satisfied with the treatment.
Data analysis The major feature of the present study was the use of structural equation modeling. This technique provides a method for estimating causal relationship among constructs in question as a path analysis of an advanced kind. In multiple indicator models6 based on this technique, latent constructs are hypothesized to determine its component measured variables, and are connected to one another with paths that assume the relationships. The latent variables (its component variables are indicated in parentheses) used to con-
589
struct the model were the psychotic symptom construct (PANSS positive and negative subscale scores), the poor insight construct (scores of the four poor insight measures), the patient-role perception construct (ABPS subscale 1 and 2 scores), the reliable therapist perception construct (reliable therapist perception scale score), and the patient satisfaction construct (patient satisfaction scale score). With regard to the last two constructs that had only single indicators, error variances of their indicators were fixed at the quantity of one minus Cronbach’s alpha coefficient multiplied by the variance of each indicator, according to the method applied by Vandenberg and Scarpello.19 The next stage of model construction was to draw paths between the constructs. The starting presupposition was that the first two symptom constructs, the psychotic symptom and poor insight constructs were to be located more upstream in the cause–effect model than the latter patient perception constructs. Psychotic symptomatology was of a type given to patients, while particular perceptions are formed in somewhat active judgmental processes. That psychotic symptoms are causal factors for the perceptions is clearly more likely than the reversal of these causes and effects because they usually antedate their formation, and must influence them. The next presupposition was that when there was a significant correlation between component variables of different constructs, some particular relationship could be postulated (i.e., a path could be put between the constructs on the basis of correlation analyses of their measured variables). The models were encoded into multiple regression equations: constructs or indicators directed by a straight arrow are dependent variables, and directing ones, independent variables, and a bi-directional arrow indicates a correlation. By means of calculating goodness of fit indexes of all possible models constructed with maximum likelihood estimations, the most fitting model was determined. Typically, a non-significant c2 statistic was the criterion for the goodness of fit of a model. However, in the present study we used a comparative fit index (CFI) greater than or equal to 0.9 as an adequacy index of model fit. This goodness of fit index was recommended by Bentler8 to be independent of the number of sample cases, while a substantial sample size can produce large c2 even for good fitting models. For performing structural equation modeling, we used EQS for Windows 5.7a, and for other analyses, the SPSS 10.0 statistical package (SPSS Inc., Chicago, IL, USA). A significance level of 0.05 was applied. Two-tailed probability was used in the correlation analysis.
N. Hayashi et al.
590
RESULTS Basic statistics and correlation analysis of the studied variables Table 1 shows the means, SD and ranges of the variables used in the model. In Table 1, PANSS item G12 was chosen as a representitive of poor insight measures. Means (SD) of Amodor’s scales on current recognition of illness, effects of medication, and social consequences of illness were 2.6 (0.7), 2.4 (0.7) and 2.6 (0.8), respectively. Pearson’s correlation coefficients of the PANSS G12 score with these scale scores were 0.737, 0.714 and 0.690, respectively. These correlations indicated that these measures were efficient, representing the domains within the clinical construct of poor insight. Internal consistency of the perception scales was satisfactory; Cronbach’s alpha coefficients for ABPS1, ABPS2, the reliable therapist perception scale and the satisfaction scale were 0.801, 0.754, 0.834 and 0.894, respectively. The results of the correlation analysis are also presented in Table 2. The three variables of psychotic symptoms, PANSS positive, negative and item G12 scores had significant positive correlations with one another, and negative ones with patient perceptions. The most salient correlation between symptoms and perceptions was found between poor insight and ABPS subscale 1. Reliable therapist perception and
patient satisfaction were strongly correlated, and had rather weak correlations with the ABPS subscales. On the basis of the two presuppositions mentioned before, we constructed structural equation models by drawing paths among the constructs. We tested the possible models, and determined the model with the highest CFI. Table 1. Demographic and clinical characteristics of the subjects Total sample (n = 118; 68 males, 50 females) Age at investigation (years)* Chronicity (years) Age at onset (years) Neuroleptic daily dose (mg)† Education (years) Inpatient days in last 3 years Duration of current therapy relationship (days) Lifetime hospitalizations GAS score PANSS positive scores PANSS negative scores PANSS general psychopathology
41.4 (11.4), 20–66 16.5 (10.3), 1–50 25.1 (7.9), 12–55 924 (794), 50–3800 12.7 (2.3), 6–16 131 (221), 0–1060 1091 (1173), 27–3689 3.4 (2.7), 0–12 53.7 (7.9), 31–73 16.0 (5.4), 7–30 18.2 (4.8), 9–32 35.0 (7.3), 19–54
* Lines below this indicate means (SD), range; † chlorpromazine equivalents.
Table 2. Means, standard deviations (SD), and Pearson’s correlation coefficients of patient perceptions and psychotic symptoms Mean (SD) Range PANSS P PANSS N PANSS G12 ABPS1 ABPS2 RTP PS
16.0 (5.4) 7–30 18.2 (4.8) 9–32 3.3 (1.0) 1–6 7.1 (7.4) - 3–30 21.2 (6.4) 11–39 1.2 (8.6) - 14–25 14.8 (6.3) 6–31
PANSS P
PANSS N
G12
ABPS1
1.000
0.357***
0.473***
- 0.170
- 0.182*
- 0.183*
- 0.287**
1.000
0.456***
- 0.211*
- 0.234*
- 0.256**
- 0.319**
1.000
- 0.499***
- 0.211*
- 0.195*
- 0.230*
1.000
ABPS2
RTP
PS
0.378***
0.156
0.348***
1.000
0.190*
0.290**
1.000
0.676*** 1.000
* P < 0.05; ** P < 0.01; *** P < 0.001. PANSS P, PANSS positive subscale score; PANSS N, PANSS negative subscale score; PANSS G12, score of lack of judgment and insight; ABPS1, score of ABPS subscale 1, recognition of need for treatment; ABPS2, score of ABPS2, acceptance of treatment situation; RTP, reliable therapist perception scale score; PS, patient satisfaction scale score.
Patient perceptions and psychotic symptoms
The constructed structural equation model The constructed model and its standardized parameters are shown in Fig. 1. Five constructs are connected with unidirectional or bi-directional paths. Unidirectional paths between the constructs represent statistically assumed causal relationships, and a bidirectional one, a correlation between the connected constructs. The values put on unidirectional arrows of the figure are standardized regression coefficients of regression equations that composed the model, and the one on the bi-directional arrow is a correlation coefficient. All values presented in Fig. 1 are statistically significant by the Z-test. The values represent the extent of the influence of the directing variable on the directed variable. This model attained an acceptable goodness of fit with a CFI of 0.958 (c2 = 52.7, d.f. = 32, P = 0.0012; GFI = 0.924; AGFI = 0.869; RMSEA = 0.075). Figure 1 shows that the psychotic symptom construct and poor insight construct are strongly linked, and that patient perceptions were mutually connected with arrows. Reliable therapist perception had an especially strong effect on patient satisfaction. Patient-role perception had some effects on patient satisfaction and reliable therapist perception. Taking a bridging position between psychotic symptomatology
591
and patient perceptions was the relationship between poor insight and patient-role perception. Although Fig. 1 did not indicate that psychotic symptomatology directly influenced the two downstream patient perceptions, reliable therapist perception and patient satisfaction, poor insight appeared to have some indirect effects on them. Standardized indirect effects of poor insight were –0.16 for reliable therapist perception and –0.30 for patient satisfaction.
DISCUSSION From the constructed model that provided statistically estimated causal relationship among the patient perceptions and symptoms, certain clues for possible determinants of patient perceptions are indicated. Influencing factors of each perception will be discussed together with related findings of previous studies. First, patient satisfaction, as the model of the present study indicated, depended chiefly on the extent to which patients saw their therapist as reliable, and to a lesser degree, on the awareness of the patient-role. In line with the first finding, previous studies have stressed the significance of personal relations with treatment staff. The overview of Corrigan20 revealed that a number of studies supported the
Figure 1. The model of causal relationships of psychotic symptomatology and patient perceptions. F, factors (constructs); X, indicators (measured variables); D, disturbance variables; E, error variances: X1, PANSS positive subscale score; X2, PANSS negative subscale score; X3, score of PANSS G12, lack of judgment and insight; X4, current recognition of illness; X5, current recognition of effects of medication; X6, current recognition of social consequences of illness; X7, score of ABPS Subscale 1, recognition of need for treatment; X8, score of ABPS2, acceptance of treatment situation; X9, reliable therapist perception scale score; X10, patient satisfaction scale score.
N. Hayashi et al.
592
importance of personal relations in favorable subjective treatment evaluation for both inpatients and outpatients. Bene-Kociemba et al. contended that patients with schizophrenia tended to be satisfied in proportion to the degree to which they felt they were understood and helped.21 Elbeck and Fecteau also reported that patient satisfaction was mainly determined by the quality of the personal relations they had with their treatment staff.22 Regarding the influence of the patient-role perception, in the study of Svensson and Hansson, the willingness of patients to participate in treatment was shown to improve patient satisfaction.23 That finding, as well as that of the present study, suggested that patient attitude determines patient satisfaction to some extent. Clinically, the importance of forming an adequate treatment relationship and fostering patient-role taking in the psychiatric situation needs to be stressed to improve patient satisfaction. Second, reliable therapist perception was supposed to reflect the quality of the treatment relationship, and to be central to patient–therapist cooperation and an affective or relational component of the therapeutic alliance,17 the formation of which is usually a delicate task, and is highlighted as crucial in the treatment.9 Another clinical significance of the therapist perception was that it was predictive of a smooth acceptance of an appropriate attitude in the treatment situation as our previous study showed.18 The present findings indicated that the patient-role perception had some influence on the reliable therapist perception. Some studies have demonstrated an inverse relationship between therapeutic alliance and severity of psychotic symptoms.24,25 However, in the present study, the relationship was indirect, and was mediated by the patient-role perception and insight into illness. Since the measures used in the present study could explain only a small proportion of the variance of the therapist perception scale, there are supposedly other relevant determinants such as psychiatrist characteristics or patient–psychiatrist matching, which future studies need to take into consideration. Third, the patient-role perception was strongly affected by impairment of insight. The efficiency of insight concept such as having values on outcome and treatment compliance has been repeatedly reported.6 It may well have a strong influence on patient perceptions. However, the present study suggested that the direct effect of poor insight was solely on the patientrole perception assessed by ABPS, and that other psychotic symptoms, both positive and negative ones, did not have a direct impact on the patient perceptions. This perception is also assumed to be a facilitating
factor for treatment compliance independent of insight as shown in our previous study.5 As suggested in the constructed cause–effect model of the present study, its improvement can enhance patient satisfaction and reliable therapist perception. Both insight and patient-role perception are considered susceptible to treatment. A number of studies have demonstrated that psychiatric treatment ameliorated poor insight into illness.19,26–28 The patient-role perception was also supposed to be a feasible focus for psychoeducational intervention.19 It is considered that both insight and patient-role perception could be important levers for treatment to improve patient perception. In conclusion, the findings of the present study indicated particular interrelationships among the perceptions, which are seen as indices of the quality of treatment and of some clinical significance. The model that illustrated the cause–effect relationships among the perceptions and symptomatology was valuable for indicating their possible makeup. The model also suggested that the perceptions or poor insight situated upstream in the model would exert influence on the downstream perceptions, and therefore, improvement of the former would lead to that of the latter. In considering an intervention plan to improve the perceptions, the model would provide useful guidelines. There were several limitations of the present study. First, the model was only a statistical estimation of the causal relationships, and not one obtained in a robust study design. We must be cautious of the interpretation of the findings. Then there are issues of sample selection. It could bias the results that this study did not employ systematic sampling. Also, the clinical efficiency of the perception scales needs further validation. In addition, there may be other perceptions of clinical relevance not included in this investigation. Further studies are required to be directed to the following issues. The clinical features of the perceptions such as their susceptibility to treatment and their role taken in treatment compliance, are to be further clarified. For that purpose, longitudinal investigations addressing the present model hypotheses are indispensable. These efforts are especially contributory to improving the quality of treatment for schizophrenic patients and their treatment compliance.
REFERENCES 1. Coursey RD, Keller AB, Farrell EW. Individual psychotherapy and persons with serious mental illness: The clients’ perspective. Schizophr. Bull. 1995; 21: 283–301.
Patient perceptions and psychotic symptoms
2. Gilbert S, Ugelstad E. Patients’ own contributions to long-term supportive psychotherapy in schizophrenic disorders. Br. J. Psychiatry 1994; 23 (Suppl.): 84–88. 3. Soskis DA, Bowers MB. The schizophrenic experience: A follow-up study of attitude and posthospital adjustment. J. Nerv. Ment. Dis. 1969; 149: 443–449. 4. Buchanan A. A two-year prospective study of treatment compliance in patients with schizophrenia. Psychol. Med. 1992; 22: 787–797. 5. Hayashi N, Yamashina M, Igarashi Y. Awareness of being a patient and its relevance to insight into illness of patients with schizophrenia. Compr. Psychiatry 1999; 40: 377–385. 6. Amador XF, Strauss DH, Yale SA, Gorman JM. Awareness of illness in schizophrenia. Schizophr. Bull. 1991; 17: 113–131. 7. Bollen KA. Structural Equations with Latent Variables. John Wiley & Sons, New York, 1989. 8. Bentler PM. EQS Structural Equations Program Manual. Multivariate Software Inc, Enrico, 1995. 9. American Psychiatric Association. Practice guideline for the treatment of patients with schizophrenia. Am. J. Psychiatry. 1997; 154 (Suppl.): 1–63. 10. Kay SR, Opler LA, Fiszbein A. Positive and Negative Syndrome Scale. MultiHealth Systems, Toronto, 1991. 11. Igarashi Y, Hayashi N, Yamashina M et al. Interrater reliability of the Japanese version of the Positive and Negative Syndrome Scale (PANSS) and the appraisal of its training effect. Psychiatry Clin. Neurosci. 1998; 52: 467–470. 12. Amador XF, Strauss DH, Yale SA, Flaum M, Endicott J, Gorman JM. Assessment of insight in psychosis. Am. J. Psychiatry 1993; 150: 873–879. 13. Parsons T. Social Structure and Personality. Free Press, New York, 1964. 14. Kemp R, Lambert TCJ. Insight in schizophrenia and its relationship to psychopathlogy. Schizophr. Res. 1995; 18: 21–28. 15. Cuesta MJ, Peralta V, Zarzuela A. Reappraising insight in psychosis. Br. J. Psychiatry 2000; 177: 233–240. 16. Osgood CE, Suci GJ, Tannenbaum PH. The Measurement of Meaning. University of Illinois Press, Urbana, 1957.
593
17. Hayashi N, Yamashina M, Ishige N et al. Perceptions of schizophrenic patients and their therapists: Application of the semantic differential technique to evaluate the treatment relationship. Compr. Psychiatry 2000; 41: 197–205. 18. Hayashi N, Yamashina M, Igarashi Y, Kazamatsuri H. Improvement of patient attitude toward treatment among inpatients with schizophrenia and its related factors: Controlled study of a psychological approach. Compr. Psychiatry 2001; 42: 240–246. 19. Vandenberg RJ, Scarpello V. The matching model. An examination of the processes underlying realistic job previews. J. Appl. Psychol. 1990; 75: 60–67. 20. Corrigan PW. Consumer satisfaction with institutional and community care. Community Mental Health J. 1990; 26: 151–164. 21. Bene-Kociemba A, Cotton PG, Fortgang RC. Assessing patient satisfaction with state hospital and aftercare service. Am. J. Psychiatry 1982; 139: 660–662. 22. Elbeck M, Fecteau G. Improving the validity of measures of patient satisfaction with psychiatric care and treatment. Hospital Community Psychiatry 1990; 41: 998–1001. 23. Svensson B, Hansson L. Patient satisfaction with inpatient psychiatric care. The influence of personality traits, diagnosis and perceived coercion. Acta. Psychiatr. Scand. 1994; 90: 379–384. 24. Frank AF, Gunderson JG. The role of therapeutic alliance in the treatment of schizophrenia. Relationship to course and outcome. Arch. Gen. Psychiatry 1990; 47: 228–236. 25. Marder SR, Mebane A, Chien CP, Winsdale WJ, Swann E, van Putten T. A comparison of patients who refuse and consent to neuroleptic treatment. Am. J. Psychiatry 1983; 140: 470–472. 26. McEvoy JP, Freter S, Everett G. Insight and clinical outcome of schizophrenic patients. J. Nerv. Ment. Dis. 1989; 177: 48–51. 27. Carroll A, Fattah S, Clyde Z, Coffey I, Owens DGC, Johnstone EC. Correlates of insight and insight change in schizophrenia. Schizophr. Res. 1999; 35: 247–253. 28. Kemp R, Kirov G, Everitt B, Hayward P, David A. Randomized controlled trial of compliance therapy: 18month follow-up. Br. J. Psychiatry 1998; 172: 413–419.
