Design
The data for this study came from a randomized controlled trial on the effectiveness of cognitive-behavioral group training (CBT) for patients with UPS [22]. Patients with UPS were randomized either to CBT or to a waiting list after they had completed the baseline measurement (T0). The second measurement (T1) was made directly after the training (13 weeks), or, for those on the waiting list, after the same period. The maintenance of the effect of the group training was investigated in a non-randomized one-year follow-up. To this end, patients who had been randomized to the waiting list and had waited started the training after their second measurement (T1). Patients who had attended the training directly after randomization or after the waiting period were followed-up three months after the end of treatment (T2), and again one year later (T3).
The study was approved by the Erasmus Medical Research Ethics Committee, and registered in the Dutch Trial Register (NTR 1609) [23]. A detailed description of the protocol has been published earlier in this journal [24].
Participants
Patients were recruited between February 2005 and September 2008 in general practices, in outpatient clinics at general hospitals, and by Riagg Rijnmond, a secondary community mental-health service for the greater Rotterdam area in the Netherlands. General practitioners and specialists were asked to refer patients aged between 18 and 65 years whose physical symptoms, according to clinical judgment, could not be fully explained on the basis of a known medical condition. Patients were included if they signed the informed consent and if their UPS fulfilled the DSM-IV criteria for an undifferentiated somatoform disorder or a chronic pain disorder.
We chose UPS classified with DSM-IV as ‘undifferentiated somatoform disorder’ or as ‘chronic pain disorder’, as these disorders were given clinical relevance by their high prevalence – in general practices, they are the most prevalent of all somatoform disorders [25] – and as they could be selected by valid and reliable instruments. ‘Undifferentiated somatoform disorder’ and ‘chronic pain disorder’ are non-overlapping disorders because of criterion E in the DSM-IV criteria for ‘undifferentiated somatoform disorder’, which states that the disorder can be assigned only if the symptoms are not better accounted for by another mental disorder such as another somatoform disorder.
To verify whether UPS fulfilled all DSM-IV criteria for either ‘undifferentiated somatoform disorder’ or ‘chronic pain disorder’, we used the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I) [26], a semi-structured validated and reliable interview for making the major DSM-IV Axis I diagnoses.
Patients were excluded if they did not provide informed consent, or if poor language skills or handicaps such as cognitive impairment prevented them from understanding the CBT group training.
CBT group training
The intervention, a CBT group training called ‘Coping with the consequences of unexplained physical symptoms’, is a weekly two-hour manual-based [27] training that is held over a 13-week period. It uses the following CBT techniques: psychoeducation, response prevention, pacing activity, graded activity, graded exercise, problem-solving, breathing and relaxation exercise, cognitive intervention using the Ellis’ ABC worksheet, and relapse prevention. Its aim is to improve health-related quality of life. A more detailed description of this CBT has been published elsewhere [24, 28].
CBT outcome measurement
In the randomized controlled trial, the primary outcome was the summary scales of the 36-item Medical Outcomes Study Short-Form General Health Survey (SF-36) [29]: Physical Component Summary’ (PCS) and ‘Mental Component Summary’ (MCS). In the present study, the PCS was chosen as outcome measurement, because patients reported the quality of life in the physical domain as most burdensome. The group training significantly improved quality of life in the physical domain, and this positive effect was maintained during the entire one-year follow-up period [30].
The PCS summarizes functional health and well-being in the physical domain over the past four weeks. This summary is transformed into T-scores with a mean of 50 and standard deviation of 10. A higher summary score indicates a better quality of life. The CBT outcome score was the difference between the baseline PCS score and the following: post-treatment PCS scores (immediate outcome); three-month follow-up PCS scores (short-term outcome); and one-year follow-up PCS scores (long-term outcome). A higher CBT outcome score indicates more improvement of quality of life in the physical domain.
Predictors
Psychological symptoms
Psychological symptoms were measured using the revised 90-item Symptom Checklist (SCL-90-R). This is a validated and reliable self-report questionnaire with 90 questions and five fixed-response alternatives (Likert-type format: Not at all; Somewhat; Moderately; Very much; Absolutely) for evaluating a broad range of psychological symptoms, including anxiety and depression, over the past week [31]. The responses are summed up in the ‘Global severity index’. A higher score on this index indicates more severe psychological symptoms or a higher number of psychological symptoms.
Personality-disorder characteristics
Personality-disorder characteristics were measured using the Vragenlijst Kenmerken van Persoonlijkheid (VKP), a Dutch self-report questionnaire based on the International Personality Disorder Examination [32]. The VKP is a validated and reliable self-report questionnaire with 197 questions and three fixed-response alternatives (true; ?; false) for assessing the presence of DSM-IV axis II criteria of personality disorders over the past five years. ‘Personality-disorder characteristics’ were calculated by summing DSM-IV axis II criteria, to which was responded with “true”. A higher sum score indicates a higher number of DSM-IV axis II criteria confirmed.
Presence of DSM-IV axis I disorders in the past (‘psychiatric history’)
The presence per patient of DSM-IV axis I disorders, both currently and over their lifetime, was measured using the Structured Clinical Interview for DSM-IV axis I disorders (SCID-I) [26]. This is a semi-structured validated interview for classifying the major DSM-IV axis I disorders. The presence of these disorders in the past (‘psychiatric history’) was calculated by summing disorders in lifetime that were not currently present, and splitting the sum score into two categories (no DSM-IV axis I in the past, 0; or one or more DSM-IV axis I disorders in the past that were not currently present, 1).
Health-related quality of life in the mental domain (‘mental component summary’)
Health-related quality of life in the mental domain was measured using the ‘Mental Component Summary’ (MCS) of the SF-36 [29]. The MCS summarizes functional health and well-being in the mental domain over the past four weeks. This summary is transformed into T-scores with a mean of 50 and standard deviation of 10. A higher MCS-score indicates a better health-related quality of life in the mental domain.
Control variables
Pretreatment PCS scores, age, gender, marital status and employment status were used as control variables.
Statistical analyses
Required sample size
The randomized controlled trial on the effectiveness of the CBT group training resulted in a group of 162 patients. To verify whether this fixed number of patients was also sufficient for the present study, we applied a power analysis to calculate the sample size required for the present study [33, 34].
For this power analysis, the anticipated effect size of the set predictors was set at f2=0.15 [35]. We decided that the set should at least have this medium effect, because the predictors would exclude patients from treatment that had an exceptionally small risk of adverse events [36], and, also, because the selection and allocation assessment needed for this exclusion would raise costs. The desired statistical power level was set at 0.80 and the alpha at 0.05; both by convention [35]. The number of predictors was four, while the number of control variables was five. The predictors were selected on the basis of assumptions practiced in clinical practice. The control variables were chosen on the basis of findings of other studies that indicated the potential relevance of these variables for CBT outcome. By selecting predictors used in clinical practice and by choosing control variables indicated by studies as potentially relevant, we reduced the number of predictors and control variables, and prevented ‘fishing’.
A power analysis with these parameters led to a minimum required sample size of 113 patients [37, 38]. Adjusted for a dropout of 30 percent, this resulted in a total sample size of 161. The total sample size of 162 in the randomized controlled trial was thus sufficient for the hierarchical multiple regression analyses of the present study.
Analyses
The statistical analyses concern drop-out and prediction. Drop-out analyses explored whether the patients who dropped out differed at baseline from study completers (patients who could be followed over a year). This was analyzed using two-tailed t-tests for independent samples for the continuous variables, two-tailed Mann–Whitney U-tests for the ordinal variables, and chi-square tests for the categorical variables.
The prediction analyses included a preliminary exploration of the relationships between the individual predictors and CBT outcomes, and a full exploration of the predictive power of the predictor set while controlling for pre-treatment score of the outcome measure and socio-demographic variables. For the preliminary exploration, a correlation matrix was composed. For the full exploration, hierarchical multiple regression analyses were used. In the first step of these regression analyses, pretreatment score on the outcome measure and socio-demographic variables were simultaneously entered as a block to statistically control for their impact on outcome. In the second step of these regression analyses, the predictors were simultaneously entered as a block to evaluate their impact as a set and as individual predictors on outcome. Since predictors have clinical relevance only if they are stable over time, these analyses were conducted for immediate, short-term and long-term CBT outcomes.
Five checks were used to verify whether the assumptions of hierarchical multiple regression analysis had been met and how accurate the resulting model was [33, 39]. The first check used was Cook’s distance to explore whether the model was highly influenced by a small number of cases. The second check used was tolerance to confirm non-multicollinearity. The third check used was the Durbin-Watson statistics to confirm the independency of errors. The fourth check used was residual plots to explore for linearity and homoscedasticity. The fifth check used was the Shapiro-Wilk test to confirm the normality of standardized residuals.