Identification of studies

To identify relevant studies, we conducted a search of the Cochrane Central Register of Controlled Trials, PubMed and PsycINFO up to October 2008. We used the text keywords depressive disorder, depression, major depressive disorder or depressive sympto* combined with the keywords psychotherapy, cognitive therapy, behavi* therapy or CBT. We did not include any terms related to underlying somatic diseases to prevent the exclusion of any possible relevant study. This search strategy was repeated using the MESH-terms depressive disorder (exploded), depression (exploded), depressive disorder, major (exploded) AND psychotherapy (exploded), limited to humans, clinical trial or randomised controlled trial. Articles were retrieved for further assessment if the title or abstract suggested that a CBT (broadly defined) was used in a population with a somatic disease. Subsequently, the search was extended by a manual search of the reference lists of all resulting randomised controlled trials (RCTs).

Inclusion criteria

Manuscripts were included if they met the following criteria:

1. (a) use of a randomised controlled research design;

2. (b) inclusion of participants with an underlying somatic disease;

3. (c) inclusion of a treatment arm with CBT, defined as a protocol-based psychological treatment including at least cognitive restructuring and behavioural activation procedures for depression for a minimum of six sessions. We also included problem-solving therapy and cognitive–behavioural stress management on the prerequisite that these interventions met our definition of CBT;

4. (d) results for depressive symptoms were presented separately for each treatment arm at pre- and post-treatment;

5. (e) use of a valid outcome measure (validated self-report questionnaire or clinical interview).

The first two authors (M.W.B. and R.C.O.V.) independently checked the inclusion and exclusion criteria of the identified studies. Disagreement over inclusion was resolved through discussion. When no consensus could be reached, a third investigator (A.E.S.) decided. Excluded were outcome studies without a randomised controlled design (e.g. open trials, case series and case reports), review papers, studies examining depression associated with unexplained somatic symptoms (e.g. fibromyalgia, chronic fatigue syndrome, irritable bowel syndrome, chronic pain), studies of individuals aged 18 years and under, studies in which the CBT could not be distinguished from other elements of the intervention (e.g. depression care management), studies including people with dementia or severe cognitive impairment, and studies limited to a specific psychiatric disorder with depressive symptoms as a second outcome measure. The language of publication was not an exclusion criteria.

Data extraction, selection procedure and quality assessment

The first two authors subsequently coded the selected studies separately on a coding form consisting of the following items: year of publication, total number of patients included, number of completers and number who dropped out, underlying somatic disease, setting (in- or out-patient clinic), age, gender, type of CBT intervention, delivery of treatment (face-to-face, telephone, internet), individual or group treatment, number of sessions, type of control condition (classified as either treatment as usual, a waiting list or other psychotherapy), outcome measures, whether depressive symptoms were the primary or secondary outcome measure, completer or intent-to-treat analysis, the results on post-treatment measures of interest (see Statistical analysis), and finally the psychiatric diagnostic criteria used at the time of inclusion (depressive disorder or depressive symptoms). Depressive disorder was defined as meeting DSM–IV ¹⁹ criteria or scoring above a predefined cut-off level on a screening questionnaire for depression. Depressive symptoms were used to indicate the dimensional score on a self-report questionnaire for depressive symptoms. Discrepancies in the two coding forms were resolved by discussion between both authors (M.W.B. and R.C.O.V.), and when no consensus could be reached the third author decided (A.E.S.).

When data on means, standard deviations or number of participants in the experimental or control groups at the end of the treatment were missing, we contacted the trial authors. Where standard deviations were not available from trial authors, they were calculated from t-tests, confidence intervals or standard errors, where reported in articles. ^{Reference Deeks20,Reference Deeks21} If these additional data were not available, the study data were not included in the meta-analysis.

The methodological quality of the included studies was assessed using the Amsterdam–Maastricht consensus list ranging from 0 (poor quality) to 19 (excellent quality). The Amsterdam–Maastricht consensus list is accepted by the Cochrane Review Group ^{Reference van Tulder, Assendelft, Koes and Bouter22} and has been used previously in systematic reviews of RCTs in psychiatric disorders. ^{Reference van Boeijen, van Balkom, van Oppen, Blankenstein, Cherpanath and van Dyck23,Reference Oude Voshaar, Couvée, Van Balkom, Mulder and Zitman24} It covers the Chalmers criteria usually applied in the assessment of study quality. ^{Reference van Tulder, Assendelft, Koes and Bouter22,Reference Chalmers, Smith, Blackburn, Silverman, Schroeder and Reitman25}

Statistical analysis

For the main analyses, studies were subdivided in studies restricted to participants with depressive disorder according to predefined criteria and studies of participants with depressive symptoms. The primary outcome measure for both categories was the severity of depressive symptoms. We a priori decided to use the Beck Depression Inventory (BDI) ^{Reference Beck, Ward, Mendelson, Mock and Erbaugh26} or Beck Depression Inventory–II (BDI–II) ^{Reference Beck, Steer and Brown27} as the primary outcome measure and, if not available (in order of preference), the Hamilton Depression Rating Scale (HDRS) ^{Reference Hamilton28} or the Center for Epidemiological Studies – Depression Scale (CES–D). ^{Reference Radloff29} If a separate measure of depressive symptoms was lacking, a subscale of a validated generic psychiatric instrument was used (in order of preference): the depression subscale of the Hospital Anxiety and Depression Scale (HADS), ^{Reference Zigmond and Snaith30} the Profile of Mood Symptoms (POMS), ^{Reference McNair, Lorr and Drappelman31} the Symptom Checklist 90-item version (SCL–90–R) ^{Reference Derogatis32} or the Impact of Rheumatic Diseases on Health and Lifestyle (IRGL). ^{Reference Huiskes, Kraaimaat and Bijlsma33}

Data were analysed with Review Manager (RevMan) version 5.0.17 for Windows. At the moment, Cochrane methodology is widely advised and accepted as the preferred method in meta-analytic studies. We used the standardised mean difference (SMD) as the summary statistic in our meta-analysis, which expresses the size of the treatment effect in each trial relative to the variability observed in that trial, enabling us to pool different scales assessing one outcome measure. The SMD thus reflects the difference in the mean outcome between groups divided by the standard deviation of outcomes among participants. We applied the chi-squared test for heterogeneity. Since the chi-squared test has low power to assess heterogeneity where a small number of participants or trials are included, the P-value was conservatively set at 0.10. ^{Reference Higgins and Green34} We used a fixed-effects model if homogeneity was found and a random-effects model if not (although the latter model slightly compromised the statistical power of our analysis). Summary statistics were based on intention-to-treat data and when missing, on available case analyses. It is generally assumed to be more realistic to include papers based on their completer data than excluding papers by not providing intention-to-treat data, on the prerequisite that potential difference between studies will be discussed and taken into account as a source of heterogeneity. ^{Reference Higgins and Green34}

The following subgroup analyses were performed to examine whether specific characteristics of the studies were related to the effect sizes:

1. (a) type of CBT condition, classified as classic CBT, problem-solving therapy and cognitive–behavioural stress management;

2. (b) control condition, classified as treatment as usual, waiting-list control condition and other psychotherapy;

3. (c) treatment delivery in group versus individual format.

Sensitivity analyses were undertaken to assess whether the quality of the studies was related to the magnitude of the observed effect sizes. Studies were limited to those of higher quality as determined by risk of: drop-out rate lower than 20%, studies providing intention-to-treat data, outcome assessment by validated questionnaire (excluding subscales) or whether depressive symptoms was the primary or secondary outcome measure. Publication bias was explored by preparing funnel plots for all outcome measures.

Search strategy

The study extraction and selection process as recommended by the Quality of Reporting of Meta-analyses (QUOROM) statement ^{Reference Moher, Cook, Eastwood, Olkin, Rennie and Stroup35} is shown in Fig. 1. The initial search yielded 1361 reference titles in PubMed, 2677 in PsycINFO and 604 in the Cochrane Library of which 305 titles were identified as having a possible relevance to the treatment of depression in people with an underlying somatic disease by M.W.B. or R.C.O.V. After screening of the abstracts and the full text, we excluded 266 papers. Of the 39 articles of interest, 13 papers did not provide sufficient information for the meta-analyses and we contacted the corresponding author for this. Of the 13 authors contacted, ^{Reference Lustman, Griffith, Freedland, Kissel and Clouse2,Reference Tan and Bruni36–Reference Doering, Cross, Vredevoe, Martínez-Maza and Cowan47} 3 authors responded to our request and were included. ^{Reference Edelman, Bell and Kidman42,Reference Lincoln and Flannaghan44,Reference Given, Given, Rahbar, Jeon, McCorkle and Cimprich45} Two authors claimed that the relevant study data were not available anymore ^{Reference Tan and Bruni36,Reference Mulder, Emmelkamp, Antoni, Mulder, Sandfort and de Vries38} and 8 authors did not respond to our request, leaving 29 studies for the meta-analysis. ^{Reference Edelman, Bell and Kidman42,Reference Lincoln and Flannaghan44,Reference Given, Given, Rahbar, Jeon, McCorkle and Cimprich45,Reference Larcombe and Wilson48–Reference Safren, O'Cleirig, Tan, Raminani, Reilly and Otto73}

Study quality

Online Tables DS1 and DS2 list the scores for methodological quality of the studies as assessed with the Amsterdam–Maastricht consensus list. The sum score (range 0–19) is taken to reflect study quality but as neither participants nor therapists can be masked to a psychological treatment condition, the maximum score in psychotherapy studies is 17. Interrater discrepancies were limited to a one-point difference in eight studies and a two-point difference in two studies. The broad range from 8 to 16 indicates a high variation of study quality, which emphasises the need for the (planned) sensitivity analyses to evaluate the effect of the most important issues of study quality.

Fig. 1 Flow diagram of randomised controlled trials (RCTs) included and excluded in meta-analysis. CBT, cognitive–behavioural therapy.

Study characteristics

The study characteristics are summarised in Table 1 for studies restricted to individuals with depressive disorder and Table 2 for studies of individuals with depressive symptoms. The 13 trials that included participants with depressive disorder ^{Reference Lincoln and Flannaghan44,Reference Larcombe and Wilson48,Reference Kelly, Murphy, Bahr, Kalichman, Morgan and Stevenson51,Reference Evans and Connis52,Reference Markowitz, Kocsis, Fishman, Spielman, Jacobsberg and Frances55,Reference Mohr, Likosky, Bertagnolli, Goodkin, van der Wende and Dwyer57,Reference Mohr, Boudewyn, Goodkin, Bostrom and Epstein58,Reference Nezu, Nezu, Felgoise, McClure and Houts63,Reference Mohr, Hart, Julian, Catledge, Honos-Webb and Vella66,Reference Savard, Simard, Giguère, Ivers, Morin and Maunsell68,Reference Gellis, McGinty, Horowitz, Bruce and Misener69,Reference Kunik, Veazy, Cully, Souchek, Graham and Hopko72,Reference Safren, O'Cleirig, Tan, Raminani, Reilly and Otto73} comprised a total of 1139 people with a mean age of 51 years. These studies all had treatment of depression as their primary goal. In total, 252 participants (22%) dropped out prematurely; 112 (21%) in the CBT condition and 140 (23%) in the control condition (χ² = 0.62, d.f. = 1, P = 0.43).

Table 1 Study characteristics of participants with depressive disorder

Study	Included, n	Dropped out, n (%)	Completers, n	Age, years: mean	Male, %	Somatic disease	Sessions, n	Group therapy	Control condition(s)	Setting
Larcombe (1984) Reference Larcombe and Wilson48	21	2 (10)	19	43	32	Multiple sclerosis	6	Yes	WLC	Out-patients
Kelly (1993) Reference Kelly, Murphy, Bahr, Kalichman, Morgan and Stevenson51	115	47 (41)	68	34	100	HIV	8	Yes	TAU and OP	Out-patients
Evans (1995) Reference Evans and Connis52	78	6 (8)	72	54	65	Cancer	8	Yes	TAU and OP	Out-patients
Markowitz (1998) Reference Markowitz, Kocsis, Fishman, Spielman, Jacobsberg and Frances55	75	23 (31)	52	37	83	HIV	16	No	OP	Out-patients
Mohr (2000) Reference Mohr, Likosky, Bertagnolli, Goodkin, van der Wende and Dwyer57	32	9 (28)	23	42	28	Multiple sclerosis	8	No	TAU	Out-patients
Mohr (2001) Reference Mohr, Boudewyn, Goodkin, Bostrom and Epstein58	42	3 (7)	39	44	27	Multiple sclerosis	16	No	OP	Out-patients
Lincoln (2003) Reference Lincoln and Flannaghan44	123	5 (4)	118	66	51	Heart disease	10	No	TAU and OP	In-patients
Nezu (2003) Reference Nezu, Nezu, Felgoise, McClure and Houts63	150	18 (12)	132	47	33	Cancer	10	No	WLC	In-patients
Mohr (2005) Reference Mohr, Hart, Julian, Catledge, Honos-Webb and Vella66	127	5 (4)	122	48	23	Multiple sclerosis	16	No	OP	Out-patients
Savard (2006) Reference Savard, Simard, Giguère, Ivers, Morin and Maunsell68	45	8 (18)	37	52	0	Cancer	8	No	WLC	Out-patients
Gellis (2007) Reference Gellis, McGinty, Horowitz, Bruce and Misener69	48	8 (17)	40	80	15	Medically Ill	6	No	TAU	Out-patients
Kunik (2008) Reference Kunik, Veazy, Cully, Souchek, Graham and Hopko72	238	115 (48)	123	66	96	COPD	8	Yes	OP	Out-patients
Safren (2008) Reference Safren, O'Cleirig, Tan, Raminani, Reilly and Otto73	45	3 (7)	42	n/a	n/a	HIV	12	No	TAU	Out-patients
Summary statistics	1139	252 (22)	887	51

Table 2 Study characteristics of participants with depressive symptoms

Study	Included, n	Dropped out, n (%)	Completers, n	Age, years: mean	Male, %	Somatic disease	Sessions, n	Group therapy	Control condition(s)	Setting
Foley (1987) Reference Foley, Bedell, LaRocca, Scheinberg and Reznikoff49	41	5 (12)	36	39	15	Multiple sclerosis	6	Yes	OP	Out-patients
Greer (1992) Reference Greer, Moorey, Baruch, Watson, Robertson and Mason50	174	23 (13)	151	52	21	Cancer	6	No	TAU	Out-patients
Kraaimaat (1995) Reference Kraaimaat, Brons, Geenen and Bijlsma53	77	6 (8)	71	57	32	Rheumatoid arthritis	10	Yes	WLC and OP	Out-patients
Henry (1997) Reference Henry, Wilson, Bruce, Chisholm and Rawling54	21	2 (10)	19	60	47	Diabetes mellitus	6	Yes	WLC	Out-patients
Edelman (1999a) Reference Edelman, Bell and Kidman42	119	27 (23)	92	50	0	Cancer	11	Yes	TAU	Out-patients
Edelman (1999b) Reference Edelman, Bell and Kidman56	60	13 (22)	47	48	0	Cancer	12	Yes	OP	Out-patients
Sharpe (2001) Reference Sharpe, Sensky, Timberlake, Ryan, Brewin and Allard59	53	8 (15)	45	55	29	Rheumatoid arthritis	8	No	TAU	Out-patients
Cruess (2002) Reference Cruess, Antoni, Hayes, Penedo, Ironson and Fletcher60	125	25 (20)	100	36	100	HIV	10	Yes	WLC	Out-patients
Evers (2002) Reference Evers, Kraaimaat, van Riel and de Jong61	64	5 (8)	59	54	29	Rheumatoid arthritis	20	No	TAU	Out-patients
Kissane (2003) Reference Kissane, Bloch, Smith, Miach, Clarke and Ikin62	303	24 (8)	279	46	0	Cancer	20	Yes	OP	Out-patients
Given (2004) Reference Given, Given, Rahbar, Jeon, McCorkle and Cimprich45	237	72 (30)	165	n/a	20	Cancer	10	No	TAU	Out-patients
Blumenthal (2005) Reference Blumenthal, Sherwood, Babyak, Watkins, Waugh and Georgiades64	134	10 (7)	124	63	69	Heart disease	16	Yes	TAU and OP	Out-patients
Chan (2005) Reference Chan, Kong, Leung, Au, Li and Chung65	16	3 (19)	13	38	100	HIV	7	Yes	WLC	Out-patients
Antoni (2006) Reference Antoni, Carrico, Duran, Spitzer, Penedo and Ironson67	130	52 (40)	78	50	100	HIV	10	Yes	OP	Out-patients
Lii (2007) Reference Lii, Tsay and Wang70	60	12 (20)	48	n/a	48	Renal failure	8	Yes	TAU	Out-patients
Koertge (2008) Reference Koertge, Janszky, Sundin, Blom, Georgiades and László71	247	83 (34)	164	62	0	Heart disease	20	Yes	TAU	In-patients
Summary statistics	1861	370 (20)	1491	51

The 16 trials including participants with depressive symptoms ^{Reference Edelman, Bell and Kidman42,Reference Given, Given, Rahbar, Jeon, McCorkle and Cimprich45,Reference Foley, Bedell, LaRocca, Scheinberg and Reznikoff49,Reference Greer, Moorey, Baruch, Watson, Robertson and Mason50,Reference Kraaimaat, Brons, Geenen and Bijlsma53,Reference Henry, Wilson, Bruce, Chisholm and Rawling54,Reference Edelman, Bell and Kidman56,Reference Sharpe, Sensky, Timberlake, Ryan, Brewin and Allard59–Reference Kissane, Bloch, Smith, Miach, Clarke and Ikin62,Reference Blumenthal, Sherwood, Babyak, Watkins, Waugh and Georgiades64,Reference Chan, Kong, Leung, Au, Li and Chung65,Reference Antoni, Carrico, Duran, Spitzer, Penedo and Ironson67,Reference Lii, Tsay and Wang70,Reference Koertge, Janszky, Sundin, Blom, Georgiades and László71} included 1861 people with a mean age of 51 years. A total of 370 participants (20%) dropped out before the outcome assessment; 192 (21%) in the CBT condition and 178 (19%) in the control condition (χ² = 0.93, d.f. = 1, P = 0.34). In six (38%) of these studies the treatment of depressive symptoms was the primary goal ^{Reference Edelman, Bell and Kidman42,Reference Given, Given, Rahbar, Jeon, McCorkle and Cimprich45,Reference Edelman, Bell and Kidman56,Reference Cruess, Antoni, Hayes, Penedo, Ironson and Fletcher60,Reference Lii, Tsay and Wang70,Reference Koertge, Janszky, Sundin, Blom, Georgiades and László71} and in 10 (62%) a secondary aim. ^{Reference Foley, Bedell, LaRocca, Scheinberg and Reznikoff49,Reference Greer, Moorey, Baruch, Watson, Robertson and Mason50,Reference Kraaimaat, Brons, Geenen and Bijlsma53,Reference Henry, Wilson, Bruce, Chisholm and Rawling54,Reference Sharpe, Sensky, Timberlake, Ryan, Brewin and Allard59,Reference Evers, Kraaimaat, van Riel and de Jong61,Reference Kissane, Bloch, Smith, Miach, Clarke and Ikin62,Reference Blumenthal, Sherwood, Babyak, Watkins, Waugh and Georgiades64,Reference Chan, Kong, Leung, Au, Li and Chung65,Reference Antoni, Carrico, Duran, Spitzer, Penedo and Ironson67}

Underlying somatic diseases in the different studies were: cancer (n = 8), HIV infection (n = 6), multiple sclerosis (n = 5), rheumatoid arthritis (n = 3), vascular disease (n = 3), diabetes mellitus (n = 1), chronic obstructive pulmonary disease (n = 1), chronic renal failure (n = 1) and various somatic diseases (n = 1).

Six studies ^{Reference Lincoln and Flannaghan44,Reference Kelly, Murphy, Bahr, Kalichman, Morgan and Stevenson51–Reference Kraaimaat, Brons, Geenen and Bijlsma53,Reference Markowitz, Kocsis, Fishman, Spielman, Jacobsberg and Frances55,Reference Blumenthal, Sherwood, Babyak, Watkins, Waugh and Georgiades64} compared an active treatment with two control conditions (e.g. an attention control, as well as a standard care group). We combined data from these control conditions and compared this with data for the treatment group. In the subgroup analysis of the control condition we used both control conditions separately. One study ^{Reference Nezu, Nezu, Felgoise, McClure and Houts63} compared two active treatments (problem-solving therapy and problem-solving therapy with a significant other) with a waiting-list control condition. In this case we combined the two treatment conditions and compared this with the control condition.

Most studies evaluated a classic CBT intervention, five studies evaluated cognitive–behavioural stress management, ^{Reference Henry, Wilson, Bruce, Chisholm and Rawling54,Reference Cruess, Antoni, Hayes, Penedo, Ironson and Fletcher60,Reference Blumenthal, Sherwood, Babyak, Watkins, Waugh and Georgiades64,Reference Antoni, Carrico, Duran, Spitzer, Penedo and Ironson67,Reference Koertge, Janszky, Sundin, Blom, Georgiades and László71} and two studies problem-solving therapy ^{Reference Nezu, Nezu, Felgoise, McClure and Houts63,Reference Gellis, McGinty, Horowitz, Bruce and Misener69} (all studies fulfilling our inclusion criteria for CBT). The number of treatment sessions ranged from 6 to 20 weeks, delivered either in a group format (16 studies) or individually (13 studies). In two studies, the intervention was delivered exclusively by telephone. ^{Reference Mohr, Likosky, Bertagnolli, Goodkin, van der Wende and Dwyer57,Reference Mohr, Hart, Julian, Catledge, Honos-Webb and Vella66}

Meta-analyses

The results for the primary outcome variable are presented in Fig. 2 for studies of individuals with depressive disorder and Fig. 3 for studies of individuals with depressive symptoms. Because of significant heterogeneity in the studies restricted to participants with depressive disorder (χ² = 170.12, d.f. = 12, P<0.001, I ² = 93%), we used a random-effects model. We found a significant pooled standardised mean difference of –0.83 (95% CI –1.36 to –0.31) in favour of the CBT condition (Z = 3.13, P = 0.002). Inspection of the data showed that the heterogeneity of results could not be explained by the underlying somatic disease, but was caused by three positive outliers. ^{Reference Larcombe and Wilson48,Reference Nezu, Nezu, Felgoise, McClure and Houts63,Reference Gellis, McGinty, Horowitz, Bruce and Misener69} Removal of these studies led to more homogeneous results, but reduced the effect size by –0.64 yielding a still significant pooled SMD of –0.19 (95% CI –0.33 to –0.05, Z = 2.69, P = 0.007).

The studies of individuals with depressive symptoms yielded a smaller but still significant pooled SMD of –0.16 (95% CI –0.27 to –0.06) in favour of the CBT condition (Z = 3.19, P = 0.001).

The funnel plots of both categories of papers were not suggestive of publication bias. The effect size was neither correlated with publication year nor with the quality score (all P>0.22). A strong correlation was found between publication year and quality score among studies restricted to participants with depressive disorder, showing an increase of study quality over the years (Spearman's rho = 0.67, P = 0.013). Based on Cochrane methodology, differences in quality score are addressed in sensitivity analyses as planned (see below) instead of excluding low-quality studies. We did not expect to find studies delivering CBT by telephone. This finding urged us to perform a post hoc subgroup analyses excluding these two studies, which yielded a higher effect size of –0.92 (95% CI –1.55 to –0.29) in favour of the CBT condition (Z = 2.86, P = 0.004) using a random-effects model.

Furthermore, the effect size was significantly correlated with the number of CBT sessions provided in the studies of participants with depressive symptoms (Spearman's rho = 0.53, P = 0.033), but not in the studies of participants with depressive disorder. In the studies of people with depressive symptoms, a post hoc subgroup analysis showed that the effect size in studies providing less than ten CBT sessions was –0.39 (95% CI –0.61 to –0.16) in favour of the CBT condition (Z = 3.34, P = 0.00008), compared with –0.11 (95% CI –0.22 to 0.00, Z = 1.89, P = 0.06) in studies providing ten or more sessions.

Fig. 2 Forrest plot of standard mean difference (with 95% CI) of effect of cognitive–behavioural therapy on depressive symptoms in studies including participants with depressive disorder, with overall effect (black diamond), based on meta-analysis. Std, standard.

Fig. 3 Forrest plot of standard mean difference (with 95% CI) of effect of cognitive–behavioural therapy on depressive symptoms in studies of participants with depressive symptoms, with overall effect (black diamond), based on meta-analysis. Std, standard.

Planned subgroup analyses

Sensitivity analyses

Methodological considerations

The study has some limitations. First, the methodological quality of the included studies was highly variable. The sensitivity analyses showed that the results were partly biased by methodological shortcomings, but the overall conclusions seem valid. For example, exclusion of studies with a drop-out rate of 20% or more led to an increased effect size in studies restricted to individuals with depressive disorder (from –0.83 to –1.18), whereas in studies of participants with depressive symptoms the effect size didn't change very much (–0.17 v. –0.13). Second, analysing only studies providing intent-to-treat data reduced the effect size in the studies of people with depressive disorder from –0.83 to a non-significant –0.11. This was also the case in studies of people with depressive symptoms. Third, neither exclusion of studies that did not use a specific instrument for assessing depression, nor exclusion of studies that used depressive symptoms as a secondary outcome measure influenced the effect size. In addition to the high variability of the quality, we did find heterogeneity among studies restricted to participants with depressive disorder, caused by three studies with extreme positive results. After removal of these outliers the results remained significant. As a last limitation, we should mention that we could not control for the stage of the underlying disease.

Implications

Cognitive–behavioural therapy significantly reduces depressive symptoms in people with an underlying somatic disease. The effect size, however, was clearly dependent on the severity of depressive symptoms at the time of inclusion in the study, with less of an effect in individuals with a lower level of depressive symptoms. A significant gap in knowledge, that might contribute to this difference, is the lack of studies that address which components of CBT might be most helpful in this population and by which personality characteristics or coping styles these effects are mediated. Currently, we cannot conclude that the positive effects are specific for CBT-oriented interventions. However, the findings of our meta-analysis support the effectiveness of CBT and therefore as the treatment of choice in people with underlying somatic diseases. The results also suggest that individual treatment might be more effective than group therapy in somatically ill people with depressive disorder. This hypothesis, however, should be further examined in future research. In addition to the need to focus more specifically on which CBT-components are most efficacious, further research should also include long-term follow-up data with respect to both depressive symptoms and the prognosis of the underlying somatic disease.