Somatoform disorders are characterised by persistent physical symptoms that suggest the presence of a medical condition, but are not explained fully by that condition or by the direct effects of substance misuse or mental disorder (DSM-IV).1 The prevalence of somatoform disorders is estimated at 6% in the general population.Reference Wittchen, Jacobi, Rehm, Gustavsson, Svensson and Jonsson2 Patients with such disorders usually have high functional impairment,Reference De Waal, Arnold, Eekhof and Van3,Reference Schneider and Rief4 are difficult to treat,Reference Hahn5,Reference Sattel, Lahmann, Gündel, Guthrie, Kruse and Noll-Hussong6 and show high utilisation of medical care.Reference Tschuschke, Weber, Horn, Kiencke and Tress7 Moreover, it typically takes years before they are referred to mental healthcare.Reference Sattel, Lahmann, Gündel, Guthrie, Kruse and Noll-Hussong6,Reference Allen, Woolfolk, Escobar, Gara and Hamer8,Reference Nanke and Rief9 A strictly somatic approach and unnecessary diagnostic examinations may increase somatising behaviours,Reference Allen, Woolfolk, Lehrer, Gara and Escobar10 and lead to chronic symptoms and high medical costs:Reference Tschuschke, Weber, Horn, Kiencke and Tress7,Reference Barsky, Orav and Bates11,Reference Hiller, Fichter and Rief12 these findings emphasise the need for early intervention.Reference Bouton, Roche, Roquelaure, Legrand, Penneau-Fontbonne and Dubus13 Psychotherapy may be a viable treatment option given the role of behavioural, cognitive and emotional processes in these disorders and their high degree of comorbidity with mental disorders.14-Reference Sar, Akyuz, Kundakci, Kiziltan and Dogan16 Some reviews and meta-analyses suggest that psychotherapy may be effective in patients with somatoform disorder.Reference Abbass, Kisely and Kroenke17-Reference Kroenke19 However, these reviews were restricted to psychodynamic psychotherapy only,Reference Abbass, Kisely and Kroenke17 or predominantly involved groups with less severe disorder, with functional neurological or conversion disorder generally being excluded.Reference Kleinstauber, Witthoft and Hiller18 Hypochondriasis and body dysmorphic disorder were typically included in these reviews,Reference Kroenke19 although it is still a matter of debate whether these conditions should be classified as somatoform disorder.14,Reference Mayou, Kirmayer, Simon, Kroenke and Sharpe20 The results of previous reviews cannot always be generalised to patients with strictly defined somatoform disorder in secondary and tertiary care, as these patients are generally more impaired than those seen in primary care.Reference Van der Feltz-Cornelis, Van Oppen, Ader and Van Dyck21 Finally, previous meta-analyses typically included only randomised trials, often excluding effectiveness studies,Reference Morrison, Bradley and Westen22-Reference Westen, Novotny and Thompson-Brenner24 whereas the inclusion of both randomised and non-randomised studies allows the meta-analytic comparison of effect sizes between these designs.
The aim of our meta-analysis therefore was to examine the effectiveness of psychotherapy for patients with strictly defined, severe somatoform disorder treated in secondary and tertiary care. To that aim, we compared effect sizes from pre- to post-treatment and from post-treatment to follow-up of psychotherapy and treatment as usual, excluding waiting-list control groups. This study focused on pre- to post-treatment contrasts, and not on between-group contrasts, given the limited number of controlled treatment studies in this context. Given the small number of studies included, moderators of treatment effect were examined only exploratively. We examined methodological quality of the studies,Reference Thoma, McKay, Gerber, Milrod, Edwards and Kocsis25 intervention characteristics (type, modality, frequency and length),Reference Trijsburg, Frederiks, Gorlee, Klouwer, den Hollander and Duivenvoorden26 and whether treatment was offered in tertiary (multimodal and integrative) or secondary care settings,Reference Kleinstauber, Witthoft and Hiller18 as potential factors influencing treatment effectiveness.
Method
A multiple-phase search was conducted in March 2010 to retrieve as many studies as possible focusing on the effectiveness of psychological treatments of severe somatoform disorder. ‘Severe’ disorder was defined as a diagnosis of somatoform disorder according to established criteria and treatment offered in secondary or tertiary care settings. First, studies were retrieved from the Scopus and Web of Science databases with the following search terms: (somatoform* OR somatisation OR (conversion AND (disorder OR symptoms)) OR (somatoform* AND pain disorder) in the title AND (treatment OR therapy OR intervention OR outcome OR effect* OR efficacy OR evaluation) in the title, abstract or keywords. Second, the reference lists of previous reviews and meta-analyses were hand-searched for additional studies. Third, several experts in the field of somatoform disorders were contacted for other studies. Inclusion criteria for the studies were as follows:
(a) a somatoform disorder was present according to DSM-III-R, DSM-IV-TR or ICD-10 diagnostic criteria, with exclusion of hypochondriasis and body dysmorphic disorder,1,27 or according to the Somatic Symptom Index, with a lifetime history of eight or more somatic symptoms criterion;Reference Hiller, Rief and Fichter28
(b) patients received psychotherapy in secondary or tertiary care;
(c) the study reported outcome statistics of validated assessment instruments to allow for effect size calculation;
(d) a prospective design was used;
(e) the publication was in English, Dutch or German;
(f) at least ten patients were included.
Two independent judges applied these selection criteria to the full-text papers that were retrieved, discussed disagreements and decided about definite inclusion. The study selection process is shown in online Fig. DS1. Both randomised and non-randomised (controlled) trials were included.
Data extraction and assessment of study quality
For each study included, basic within-group (pre- to post-treatment and post-treatment to follow-up) data necessary for effect size calculation were extracted. If necessary, we contacted the authors to obtain additional information. No study had to be excluded because of unavailable data. Effect sizes were calculated for three outcome domains separately. The primary outcome was ‘physical symptoms’; secondary outcomes were ‘psychological symptoms’ (depression, anxiety, anger and general symptoms) and ‘functional impairment’ (health, life satisfaction, interpersonal problems and maladaptive cognitions and behaviour), as these outcome domains have been commonly used in evaluations of treatments in somatoform disorder.Reference Kroenke19 When a study included multiple variables for the same domain, the means of z-transformed variables were used to calculate an average effect size. Self-report measures were included as most studies used self-report instruments only.
Participant information and formal characteristics of the treatments and therapeutic interventions were extracted by a standardised coding scheme. Formal characteristics included setting, duration, frequency, treatment modality, adherence and structure.Reference De Groot, Verheul and Trijsburg29 To classify therapeutic interventions, categorisation was based on that of De Groot et al,Reference De Groot, Verheul and Trijsburg29 and the Comprehensive Psychotherapeutic Interventions Rating Scale (CPIRS) which comprised interventions representative of the following five psychotherapeutic orientations: experiential, psychodynamic, directive-behavioural, cognitive and systemic.Reference Trijsburg, Frederiks, Gorlee, Klouwer, den Hollander and Duivenvoorden26 We adapted this scale for the treatment of somatoform disorder by adding ‘body-directed interventions’. The intervention was rated as ‘important’ (explicitly mentioned as core component), ‘possibly important’ (implicit as core component), ‘possibly not important’ (uncertain whether included or not) and ‘not important’ (not mentioned and definitely not a core component).
Studies included were rated by two independent judges on methodological quality with the revised Psychotherapy Quality Rating Scale (PQRS).Reference Kocsis, Gerber, Milrod, Roose, Barber and Thase30 This scale comprised 25 items covering six domains: description of patients (four items); definition and delivery of treatment (five items); outcome measures (five items); data analysis (five items); treatment assignment (three items); and overall quality (three items). The last item was scored on a seven-point scale and the other items on a three-point scale. Some minor adaptations to the scale were made based on the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement recommendations for cohort studies to render it suitable for quasi-experimental and uncontrolled studies.Reference Von Elm, Altman, Egger, Pocock, Gotzsche and Vandenbroucke31 Items were added to yield information about repeated measurements, to evaluate sources of bias in uncontrolled trials as a consequence of recruitment procedures and non-random allocation to treatment groups and their impact on study generalisability, and concerning the matching of groups on clinical and demographic variables.
Two independent judges coded all studies to establish inter-rater reliability (across 14 studies). The one-way random intraclass correlation (ICC) single measures was 0.91 and the ICC average measures was 0.95, which is excellent.Reference Shrout and Fleiss32
Statistical analysis
Effect sizes
Standardised mean pre- to post-psychotherapy changes were evaluated. Effect sizes (Cohen’s d) were calculated by subtracting post-treatment scores from pre-treatment scores and dividing these by the pooled standard deviations. To calculate the pooled s.d. the test-retest reliability for the most common instrument in that domain was taken into account. Follow-up effect sizes were calculated by subtracting follow-up scores from post-treatment scores and dividing by the pooled s.d. to determine the sustainability of effects. We selected the longest time frame between treatment termination and follow-up. Effect sizes of 0.20, 0.50 and 0.80 are considered small, medium and large respectively; effect sizes below 0.20 are considered very small.Reference Cohen33 Effect sizes were weighted by their inverse variance.Reference Lipsey and Wilson34 This procedure gives more weight to larger studies (with smaller sampling error), thus increasing the reliability of the effect estimates.
Within the same study, effect sizes were calculated separately for one or more treatments and for the treatment as usual (TAU) control condition. To be regarded as TAU, treatments had to be specified by the researchers as such or as (standard or enhanced) medical care, psychiatric advice to the general practitioner or TAU with a small component added (e.g. relaxation or gymnastics). Waiting-list control groups were not considered to be TAU. When the alleged comparison condition clearly included psychotherapy as a main component, the condition was categorised as psychotherapy.
Pooled effect sizes across studies
Pooled effect sizes were calculated with the Comprehensive Meta-analysis (CMA) software version 2.0 on Windows (Biostat, Englewood, New Jersey, USA). A random effects model was used to compute weighted mean effect sizes, because we expected true population effect sizes to vary across studies owing to differences in sample, methodology and treatment. The random effects model results in more conservative results and broader 95% confidence intervals than the fixed effects model.Reference Borenstein, Hedges, Higgins and Rothstein35 Heterogeneity of effect sizes within and between subsamples was calculated using the Q statistic.Reference Borenstein, Hedges, Higgins and Rothstein35 Pooled effect sizes were also recalculated with outliers removed when the individual effect sizes did not overlap with the confidence interval.
Moderator analysis
To examine potential moderators of treatment outcome, associations between effect sizes and the following dichotomous (yes/no) variables were calculated with CMA: group psychotherapy (with or without additional individual psychotherapy); treatment setting (out-patient v. in-patient); high or low treatment frequency (more than two sessions per week); treatment manual available; adherence check; supervision of therapists; additional pharmacotherapy allowed; randomised or non-randomised trial; intention-to-treat analysis; multicomponent (treatment components besides psychotherapy); cognitive interventions; behavioural interventions; psychodynamic interventions; affective interventions; interpersonal interventions; experiential interventions; and body-directed interventions. For the purpose of moderator analyses, studies were divided into subgroups. For each subgroup the pooled mean effect size was calculated, and differences in effect sizes between the subgroups (with a minimum of three studies) were examined for statistical significance using the Q statistic. Meta-regression analyses were conducted to explore whether gender, age, prevalence of diagnosis of somatisation disorder, methodological quality of the study (according to the Kocsis et al checklist) and length of treatment predicted effect sizes.Reference Kocsis, Gerber, Milrod, Roose, Barber and Thase30 Moderator analyses were performed only for pre-post contrasts and not for follow-up, because the samples were too small to yield meaningful effects.
Publication bias
Meta-analyses typically contain an overrepresentation of published studies that are biased towards statistically significant findings.Reference Higgins, Thompson, Deeks and Altman36 We tested potential publication bias by means of the iterative non-parametric trim and fill procedure as implemented in CMA.Reference Duval and Tweedie37 This procedure controls for the association between individual effect sizes and their sample sizes (sampling error) by inspecting funnel plots: publication bias is present when the effect sizes of small studies with larger sampling variation than large studies are represented asymmetrically within and around the funnel. The Duval & Tweedie procedure provides a correction of the effect size after publication bias has been taken into account by trimming away studies suggesting asymmetry.Reference Duval and Tweedie37 We used the random effects model. Publication bias was also evaluated with Orwin’s fail-safe N statistic, which expresses how many studies would bring the combined effect to a specified level other than zero.Reference Borenstein, Rothstein, Sutton and Borenstein38 The criterion was set to d = 0.10.
Results
Sixteen studies were included in the meta-analysis (online Fig. DS1).Reference Schneider and Rief4,Reference Sattel, Lahmann, Gündel, Guthrie, Kruse and Noll-Hussong6-Reference Allen, Woolfolk, Lehrer, Gara and Escobar10,Reference Hiller, Fichter and Rief12,Reference Leichsenring, Kreische, Biskup, Staats, Rudolf and Jakobsen39-Reference Zaby, Heider and Schroder47 The main characteristics are summarised in online Table DS1. Fourteen studies focused exclusively on somatoform disorder; two studies included patients with a range of mental disorders but were eligible because statistics for the somatoform disorder subgroup were reported separately.Reference Leichsenring, Kreische, Biskup, Staats, Rudolf and Jakobsen39,Reference Junkert-Tress, Schnierda, Hartkamp, Schmitz and Tress40 All studies but one were published after 2000.Reference Dongfen, Shizong, Ono, Janca, Asai and Sartorius41 Ten studies were conducted in Germany,Reference Schneider and Rief4,Reference Sattel, Lahmann, Gündel, Guthrie, Kruse and Noll-Hussong6,Reference Tschuschke, Weber, Horn, Kiencke and Tress7,Reference Nanke and Rief9,Reference Hiller, Fichter and Rief12,Reference Leichsenring, Kreische, Biskup, Staats, Rudolf and Jakobsen39,Reference Junkert-Tress, Schnierda, Hartkamp, Schmitz and Tress40,Reference Bleichhardt, Timmer and Rief45-Reference Zaby, Heider and Schroder47 two in the USA,Reference Allen, Woolfolk, Escobar, Gara and Hamer8,Reference Allen, Woolfolk, Lehrer, Gara and Escobar10 two in The Netherlands,Reference Moene, Spinhoven, Hoogduin and Van Dyck43,Reference Moene, Spinhoven, Hoogduin and Van Dyck44 one in Turkey,Reference Ataoglu, Ozcetin, Icmeli and Ozbulut42 and one in China.Reference Dongfen, Shizong, Ono, Janca, Asai and Sartorius41
Characteristics of treatment
In the included studies the intention-to-treat sample comprised 936 participants in the psychotherapy conditions of whom 854 (91%) completed treatment and 890 (95%) were eventually included in analyses. In the TAU conditions the intention-to-treat sample comprised 568 patients of whom 519 (91%) patients actually completed their treatment and 548 (96%) were included in analyses. The median duration of psychotherapy was 9.2 weeks (range 2-25) for all but one study which provided more than 250 sessions over on average 178 weeks.Reference Nickel, Cangoez, Bachler, Muehlbacher, Lojewski and Mueller-Rabe46 In the TAU groups, patients received treatment for a median of 9 weeks (range 5.5-12). Other relevant information regarding the study characteristics and therapeutic interventions is reported in online Tables DS1 and DS2.
Characteristics of participants
The mean age of the treated patients was 39.7 years (s.d. = 8.1). All studies included adults except for one study in which the age range was 16-30 years.Reference Ataoglu, Ozcetin, Icmeli and Ozbulut42 The percentage of women across the studies was 67% (n = 950). The mean duration of physical symptoms (reported in ten studies) was 3135 days (s.d. = 2564), i.e. about 8.6 years. Eleven studies provided information about comorbid diagnoses: the most common were mood and anxiety disorders. Two studies excluded one patient with depression.Reference Moene, Spinhoven, Hoogduin and Van Dyck43,Reference Moene, Spinhoven, Hoogduin and Van Dyck44
Methodological characteristics
The mean methodological quality across 16 studies according to the revised PQRS was 21.0 (range 10-36), indicating on average a moderately poor study quality (ranging from very poor to very good). The mean methodological quality of the ten randomised controlled trials (RCTs) was rated 25.1 (range 17-36) v. 15.0 (range 12-21) for the six non-RCTs. Twelve studies used manualised treatments or provided sufficient information about the treatment to allow replication. Four of these studies conducted a treatment integrity check.Reference Sattel, Lahmann, Gündel, Guthrie, Kruse and Noll-Hussong6,Reference Allen, Woolfolk, Escobar, Gara and Hamer8,Reference Moene, Spinhoven, Hoogduin and Van Dyck43,Reference Moene, Spinhoven, Hoogduin and Van Dyck44 Six studies with manualised treatments provided regular supervision of the therapists.Reference Sattel, Lahmann, Gündel, Guthrie, Kruse and Noll-Hussong6,Reference Allen, Woolfolk, Escobar, Gara and Hamer8,Reference Junkert-Tress, Schnierda, Hartkamp, Schmitz and Tress40,Reference Moene, Spinhoven, Hoogduin and Van Dyck43,Reference Moene, Spinhoven, Hoogduin and Van Dyck44,Reference Zaby, Heider and Schroder47 Other treatments such as pharmacotherapy were stopped during the treatment period in only two studies.Reference Dongfen, Shizong, Ono, Janca, Asai and Sartorius41,Reference Ataoglu, Ozcetin, Icmeli and Ozbulut42 One study provided paradoxical therapy, which was compared with diazepam.Reference Ataoglu, Ozcetin, Icmeli and Ozbulut42 One study examined the effect of acupuncture in addition to cognitive-behavioural therapy (CBT).Reference Dongfen, Shizong, Ono, Janca, Asai and Sartorius41 In six studies no explicit mention was made of pharmacotherapy. In one study, outcome assessments were made by a psychiatrist who was masked to the treatment conditions of the patients throughout the study.Reference Ataoglu, Ozcetin, Icmeli and Ozbulut42 Three studies used partial masking (only on some outcome instruments),Reference Tschuschke, Weber, Horn, Kiencke and Tress7,Reference Moene, Spinhoven, Hoogduin and Van Dyck43,Reference Moene, Spinhoven, Hoogduin and Van Dyck44 and one study contained no information as to whether assessments were conducted masked to treatment condition.Reference Dongfen, Shizong, Ono, Janca, Asai and Sartorius41 Most studies conducted intention-to-treat analyses including patients who withdrew from treatment in the analyses. Ten studies carried out follow-up assessments; the mean follow-up period across these studies was 10.7 months (range 5-22) after treatment termination.
Pre-post pooled effect sizes
Figures 1, 2, 3 summarise the pooled weighted pre-post effect sizes, test statistics and heterogeneity indices for psychotherapy and TAU for three outcome domains separately.
Physical symptoms
The pooled effect size of physical symptoms for psychotherapy across nine studies was large (d = 0.80). Treatment as usual (4 studies) yielded a small pooled effect size (d = 0.31). Heterogeneity based on the Q statistic was high and significant in both conditions. Analyses with outliers removed yielded similar results. Comparison of pooled effect sizes yielded a superior outcome for patients receiving psychotherapy (9 studies) compared with TAU (4 studies): Q = 5.4, P<0.05. Moderator analyses for this domain showed lower effect sizes for studies with supervision of therapists (3 studies) v. studies without (6 studies) (online Fig. DS2). Meta-regression showed a better outcome for younger patients (slope –0.036, P<0.01), women (slope 0.007, P<0.03), patients with somatisation disorder (slope 0.004, P<0.01) and studies with lower methodological quality (slope –0.016, P<0.01). Treatment length was not a moderator (slope 0.001, not significant).
Psychological symptoms
The pooled effect size of psychological symptoms for psychotherapy across 14 studies was moderate to large (d = 0.75). Treatment as usual (5 studies) yielded a moderate pooled effect (d = 0.51). Heterogeneity was high and significant in both conditions. Analyses with outliers removed yielded similar results. Comparison of pooled effect sizes suggested a similar effect for psychotherapy (14 studies) and TAU (5 studies): Q = 1.6, P = 0.21. Moderator analyses for the domain of psychological symptoms yielded significant results for five methodological features, and interpersonal techniques were associated with lower effect sizes (online Fig. DS2). Meta-regression showed better results for younger patients (slope –0.009, P<0.03), men (slope –0.007, P<0.01), patients with somatisation disorder (slope 0.023, P<0.01) and studies with lower methodological quality (slope –0.021, P<0.01). Treatment length was not a moderator (slope 0.002, not significant).
Physical symptoms: pre- to post-treatment effects.
a. The study by Bleichhardt et al is listed twice because both conditions were analysed as active psychotherapeutic treatments. *P<0.05, **P<0.01.
Psychological symptoms: pre- to post-treatment effects.
a. The studies by Bleichhardt et al and Dongfen & Shizong are listed twice because both conditions were analysed as active psychotherapeutic treatments. *P<0.05, **P<0.01.
Functional impairment
The pooled effect size of functional impairment for psychotherapy across 12 studies was moderate (d = 0.45). Treatment as usual (5 studies) showed a very small pooled effect size (d = 0.15). Heterogeneity was high and significant in both conditions. Analyses with outliers removed yielded similar results. Comparison of the pooled effect sizes indicated a better outcome for patients receiving psychotherapy (12 studies) than those receiving TAU (5 studies): Q = 6.8, P<0.01. Moderator analyses in this outcome domain showed a better outcome for studies without cognitive and behavioural interventions, studies without body-directed therapeutic interventions and non-RCTs (online Fig. DS2). When compared head to head, psychodynamic interventions were more effective than cognitive interventions: d = 0.79 (4 studies) v. d = 0.27 (7 studies), P<0.05. Meta-regression yielded better results for younger patients (slope –0.027, P<0.01), men (slope –0.012, P<0.01), patients with somatisation disorder (slope –0.004, P<0.01), studies with lower methodological quality (slope –0.016, P<0.01) and longer treatments (slope 0.002, P<0.05).
Functional impairment: pre- to post-treatment effects.
a. The study by Nanke & Rief is listed twice because both conditions were analysed as active psychotherapeutic treatments. *P<0.05, **P<0.01.
Publication bias
Some evidence of publication bias was found for physical symptoms and functional impairment, but not for psychological symptoms. The pooled pre-post treatment effect size for physical symptoms remained large after adjustment for publication bias (d = 1.03, 95% CI 0.65-1.41); three studies were trimmed. Orwin’s fail-safe N showed that 44 missing studies with an average effect size of zero would need to be published to render the significant results clinically irrelevant (mean effect size d = 0.10). The pre-post treatment effect size for functional impairment remained moderate after correction for publication bias (d = 0.57, 95% CI 0.39-0.76); three studies were trimmed. Orwin’s fail-safe N indicated that 40 missing studies with a zero effect size would need to be published to bring the significant results to the specified irrelevant level of d = 0.10. In sum, although there was some evidence of publication bias, the results essentially remained similar when adjusted for this bias.
Post-treatment to follow-up pooled effect sizes
Table 1 summarises the effect sizes and test statistics from post-treatment to follow-up for psychotherapy and TAU. For physical and psychological symptoms there was no significant change between treatment termination and follow-up. On measures of functional impairment patients who had received psychotherapy rather than TAU continued to improve. There was no post-therapy to follow-up difference between psychotherapy and TAU on any outcome domain (physical symptoms Q = 0.0, P = 0.93; psychological symptoms Q = 0.0, P = 0.83; functional impairment Q = 0.7, P = 0.39). These findings indicate that after treatment the psychotherapy and TAU patient groups changed in a similar fashion. Our findings further imply that all gains immediately after treatment were sustained at follow-up, for both psychotherapy and TAU, with continuing improvement in the domains of health and interpersonal functioning for psychotherapy.
Discussion
This meta-analysis examined the effectiveness of psychotherapy for severe somatoform disorder in secondary and tertiary care. Within specific limitations, such as heterogeneity of studies and limited methodological quality, overall findings suggest that psychotherapy is effective in severe somatoform disorder. Pre- to post-therapy improvements were large for physical symptoms, medium to large for psychological symptoms and small to medium for functional impairment. For TAU, effect sizes were small for physical symptoms, medium for psychological symptoms and very small for functional impairment. The differences in effect size between psychotherapy and TAU were significant for physical symptoms and functional impairment. These findings suggest that psychotherapy is more effective than TAU in severe somatoform disorder, particularly in terms of reduction of physical symptoms and functional impairment. This meta-analysis extends previous research in at least three ways. It is the first to indicate that psychotherapy is more effective than TAU for strictly defined, severe and chronic somatoform disorder, which is considered a debilitating and difficult-to-treat disorder.Reference Sattel, Lahmann, Gündel, Guthrie, Kruse and Noll-Hussong6,Reference Hiller, Fichter and Rief12 Second, since the previous meta-analysis new empirical data have been published that are included in our meta-analysis. Finally, the inclusion criteria were different from those used in previous reviews and meta-analyses - for example, functional neurological or conversion disorder was not included in earlier meta-analyses.
Post-treatment to follow-up: weighted effect sizes and relevant test statistics
| Psychotherapy | TAU | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| k Footnote a | d Footnote b | 95% CI | Z Footnote c | Q Footnote d | k Footnote a | d Footnote b | 95% CI | Z Footnote c | Q Footnote d | |
| All studies | ||||||||||
| Physical symptoms | 7 | 0.04 | −0.14 to 0.23 | NS | 110.2Footnote ** | 2 | 0.05 | −0.02 to 0.13 | NS | NS |
| Psychological symptoms | 9 | 0.01 | −0.17 to 0.18 | NS | 41.7Footnote ** | 2 | −0.02 | −0.16 to 0.12 | NS | NS |
| Functional impairment | 9 | 0.20 | 0.04 to 0.35 | 2.5Footnote * | 56.0Footnote ** | 3 | 0.09 | −0.12 to 0.29 | NS | 9.5Footnote * |
| Outliers excludedFootnote e | ||||||||||
| Physical symptomsReference Hiller, Fichter and Rief12, Reference Leichsenring, Kreische, Biskup, Staats, Rudolf and Jakobsen39 | 5 | 0.01 | −0.14 to 0.16 | NS | 31.2Footnote ** | 2 | 0.05 | −0.02 to 0.13 | NS | NS |
| Psychological symptomsReference Hiller, Fichter and Rief12, Reference Leichsenring, Kreische, Biskup, Staats, Rudolf and Jakobsen39 | 7 | 0.00 | −0.13 to 0.13 | NS | 12.8Footnote ** | 2 | −0.02 | −0.16 to 0.12 | NS | NS |
| Functional impairmentReference Hiller, Fichter and Rief12 | 8 | 0.25 | 0.17 to 0.34 | 5.7Footnote ** | NS | 3 | 0.09 | −0.12 to 0.29 | NS | 9.5Footnote * |
NS, not significant; TAU, treatment as usual.
a. Number of studies per outcome domain.
b. Cohen’s effect size.
c. Combined weighted z-score; the direction of effect is indicated by the sign of the z-score.
d. Cochran’s index of heterogeneity, calculated as the weighted sum of squared differences between individual study effects and the pooled effect across studies.
e. Superscript numbers denote excluded studies (outliers).
* P<0.05.
** P<0.01.
The effects found for psychotherapy were maintained nearly a year after treatment. For functional impairment the effect during treatment was smaller than for the symptom domains, yet results suggested a small but significant continuing decrease in functional impairment during follow-up. This finding is congruent with studies suggesting that improvements in patients’ general functioning may take considerable time given the chronic nature of this disorder and the severity of functional impairment.Reference Luyten, Van Houdenhove, Lemma, Target and Fonagy48,Reference Luyten and Van Houdenhove49 Dose-effect studies indicate that significant improvements in characterological, interpersonal and social functioning generally require at least 6 months of psychotherapy.Reference Hilsenroth, Ackerman and Blagys50 Our findings do not confirm this suggestion for functional impairment, but the continued improvement at the long-term follow-up in this domain may reflect that improvement in symptoms often precedes improvement in general functioning.Reference Hilsenroth, Ackerman and Blagys50,Reference Perry and Bond51 Given the chronic nature of somatoform disorders and the high levels of associated functional impairment, it is remarkable that extant research has mainly focused on studies of the effectiveness of brief psychotherapies. It is relevant here that within these brief treatments, longer treatment duration was associated with larger effects in terms of general functioning.
Effect sizes for psychotherapy for somatoform disorder were generally lower than those typically found for mental disorders. For instance, one meta-analysis reported higher pre- to post-treatment effect sizes of short-term psychodynamic therapies for depression (d = 1.30-1.87),Reference Driessen, Cuijpers, Maat, Abbass, De Jonghe and Dekker52 compared with the effect sizes for somatoform disorder in this meta-analysis (d = 0.45-0.80). There are several possible reasons for this. First, treatments for depression have been more systematically developed and examined compared with those for somatoform disorder. Second, clinically relevant changes may be easier to achieve in a syndrome with an episodic nature such as depression than in a more stable syndrome such as somatoform disorder. Third, patients with somatoform disorder, who often are subjected to redundant medical examinations over many years, may show much longer delay in seeking psychotherapy than patients with depression.Reference Bouton, Roche, Roquelaure, Legrand, Penneau-Fontbonne and Dubus13
Moderators
A secondary aim of our meta-analysis was to examine moderators of treatment effects. However, moderator analyses reported here should be interpreted with considerable caution. Because they are intercorrelated owing to the relatively small number of studies, the moderators in this analysis may be statistically associated with outcomes, but the underlying mechanism may be very different and not (fully) captured by the moderator. Within these limitations, findings indicated that interpersonal interventions were less effective in reducing psychological symptoms. Furthermore, cognitive and behavioural therapies and therapies that mainly focus on the body may not be effective in reducing functional impairment. Some studies have emphasised the importance of emotional awareness,Reference Sattel, Lahmann, Gündel, Guthrie, Kruse and Noll-Hussong6,Reference Abbass, Kisely and Kroenke17,Reference Subic-Wrana, Bruder, Thomas, Lane and Kohle53 emotion exposure and expression when dealing with patients with somatoform disorder,Reference Lumley54,Reference Rimes and Chalder55 as well as their interpersonal difficulties;Reference Noyes, Langbehn, Happel, Stout, Muller and Longley56,Reference Waller, Scheidt and Hartmann57 all of these are core features of psychodynamic therapy.Reference Blagys and Hilsenroth58,Reference Diener, Hilsenroth and Weinberger59 Concordant with this notion, post hoc analysis indicated that psychodynamic interventions were more effective in improving functioning than cognitive interventions, although not in improving symptoms. Future research directly comparing these interventions is needed to replicate these findings before any substantial conclusions can be drawn. It is relevant that our meta-analysis indicates that other forms of psychotherapy besides CBT are potentially effective for severe somatoform disorder. This finding suggests that different mechanisms, rather than a single mechanism, may underlie therapeutic change in patients with severe somatoform disorder.Reference Luyten and Van Houdenhove49
Nearly half of the studies were conducted in clinical settings involving intensive, multimodal treatments; however, these treatments were not superior to out-patient treatment. Hence, high-frequency and integrative multicomponent treatments may not be more effective than less intensive - and probably less expensive - out-patient treatments. Yet, it is also possible that pre-treatment variables associated with differential referral to out-patient and in-patient care are responsible for these findings. Moreover, we found no difference in the effectiveness of group v. individual therapy; this is in contrast to results from a previous meta-analysis that included also patients with less severe somatoform disorder,Reference Kleinstauber, Witthoft and Hiller18 underscoring the need for a separate meta-analysis focusing on patients with severe disorder only. Generally, elderly patients benefited less from psychotherapy, which indicates that some elements of the treatment may need to be adapted to the patient’s age. Women benefited more from treatment than men in terms of physical symptoms, but not psychological symptoms or functional impairment. Because women tend to have more physical symptoms than men, especially in clinical samples,Reference Dekker, Koelen, Peen, Schoevers and Gijsbers-Van Wijk60 there may simply be more room for symptom improvement in female patients. A similar argument might apply to more symptomatic somatoform disorder, as physical and psychological symptoms improved more in studies with a higher percentage of patients diagnosed with somatisation disorder.
Congruent with another meta-analysis,Reference Thoma, McKay, Gerber, Milrod, Edwards and Kocsis25 smaller effect sizes were found in studies of higher methodological quality. In some outcome domains stricter control over the delivery of treatment (e.g. manual or adherence check) or the study design (RCT v. non-RCT) yielded lower effect sizes. Studies of low methodological quality are likely to include fewer experimentally controlled factors that might subsequently result in higher effectiveness.Reference Thoma, McKay, Gerber, Milrod, Edwards and Kocsis25 Second, higher methodological quality might actually lead to therapists relying more rigidly on the treatment manual, which might hamper the flexibility that is needed in applying any treatment protocol.Reference Luyten and Van Houdenhove49 This finding emphasises the importance of including a range of study designs in the examination of the effectiveness of treatments, especially in the early stages of treatment development.Reference Morrison, Bradley and Westen22
Limitations
The results of this meta-analysis should be interpreted within the context of its methodological limitations. First, the small number of studies limited the ability to draw definite conclusions. Second, heterogeneity of studies and outcome measures may have influenced results, although we used conservative confidence intervals for the estimation of effect sizes. Third, publication bias was present, although did not appear to have significantly influenced the results. Fourth, study quality was moderately poor on average. Fifth, most outcome assessments relied only on patients’ self-reports. Sixth, the use of concomitant medications was not reported in each study. Seventh, a meta-analysis of between-group contrasts was not feasible, because three of the retrieved effectiveness studies were non-inferiority trials comparing two forms of CBT and seven studies had no control treatment, thus hampering between-group comparisons. Eighth, as all studies involved patients with severe somatoform disorder, results from the meta-analysis do not generalise to syndromes labelled ‘functional somatic syndrome’, ‘medically unexplained physical symptoms’ or ‘abridged somatisation disorder’, and chronic diseases with multiple physical symptoms. Finally, in light of recent changes in diagnostic classification, it remains uncertain how the findings in this group of somatoform disorders will translate to the revised diagnostic system in DSM-5.14,Reference Creed, Guthrie, Fink, Henningsen, Rief and Sharpe61,Reference Fink and Schroder62
Future research
Future studies should aim to examine cost-effectiveness of various forms of psychotherapy given the excessively high medical costs and loss of productivity associated with somatoform disorder. Given the limited quality of the existing body of studies, future research should include more randomised controlled studies of higher quality. This would allow comparisons between potentially effective psychotherapies for (severe) somatoform disorder or ‘somatic symptom syndromes’,14 and more focused hypothesis testing of specific interventions and potential mechanisms of change.
Acknowledgements
The authors thank F. Stuke for her contributions in rating the methodological quality of the individual studies, E. Driessen for her helpful suggestions concerning the software and the analyses, H. Sattel and F. Leichsenring for providing additional data, and N. Thoma for his advice concerning the Psychotherapy Quality Rating Scale.
eLetters
No eLetters have been published for this article.