Combined Pharmacotherapy and Psychological Treatment for Depression: Title and subTitle BreakA Systematic Review

Adherence to treatment with antidepressant drugs is an issue of clinicalrelevance. A review of controlled therapeutic studies suggests a dropout rateof up to 33% irrespective of drug class.¹ Inclinical practice, higher rates are observed.² Themain factors affecting adherence to treatment pertain to the length of psychiatrichistory, latency and safety profile of the drug, and quality of the patient-physicianrelationship.³ However, this knowledge hasnot been integrated into pharmacological research: 92% of antidepressant trialsconsider assessment of depression as the only outcome measure, and 87% ofthem present results without application of the intention-to-treat approach.⁴ Inadequate attention to adherence is evident in thechoice of a solely therapeutic intervention, in the lack of additional outcomemeasures to complement efficacy such as patient satisfaction, and in the removalof dropouts from the analysis. A review of the few controlled studies in whichtreatment adherence was the main end point does not provide clear evidenceon how to affect adherence, as they often adopt composite interventions and,in many cases, cannot relate efficacy to adherence because they do not systematicallymeasure clinical outcome.²

Psychological treatment is the most common nonpharmacological interventionadded to pharmacotherapy, with the objective to enhance efficacy, improvepatients' social functioning and quality of life, and prevent or delay recurrence.Some authors have suggested that psychological treatment may also affect adherenceto pharmacotherapy.^{5 - 6} So far,systematic reviews of combined treatment vs pharmacotherapy alone have onlyfocused on efficacy data.^{7 - 9} However,evidence on adherence-enhancing effects of psychological interventions couldbe obtained as a by-product of studies in which pharmacotherapy is commonto all arms and at least 1 of the arms being compared is pharmacotherapy alone.Although treatment adherence includes several behaviors, we focused in thisstudy on its ultimate expression, the premature interruption of treatment(ie, dropout). To investigate the relationship between completion of treatmentand efficacy, we reviewed all randomized clinical trials (RCTs) contrastingantidepressant pharmacotherapy vs combined psychological treatment, to determinewhether combined treatment is more efficacious than pharmacotherapy aloneand whether combined intervention affects nonresponse to and dropping outof drug treatment.

To identify RCTs contrasting antidepressant pharmacotherapy with combinedtreatment (pharmacotherapy and psychological treatment), we searched MEDLINE,Current Contents, and PsychInfo, from January 1980 to November 2002. The searchstrategy included all key words ordered by MEDLINE under the following medicalsubject headings: patient compliance, depressive disorder, antidepressiveagents, psychotherapy, mental health services, and RCTs. The Cochrane Librarywas similarly searched. Numerous review articles, textbooks of treatment ofdepression, and treatment guidelines for depression were also screened toidentify potential studies. Two of us (B.K. and P.B.) read each abstract andretained RCTs that contrasted pharmacotherapy alone vs pharmacotherapy combinedwith psychological treatment. To classify the latter, we followed the systemof Rush and Thase,⁸ who consider problem-solving,interpersonal, cognitive, behavioral, marital, and psychodynamic therapies.Although these therapies have different conceptual backgrounds, the rationaleof considering them together is that they aim at the reduction in symptomsof depression and prevention of recurrence.

Copies of articles regarding RCTs were obtained, and their referencelists were screened to identify additional studies of interest. We excludedarticles that had additional nonpharmacological intervention besides psychotherapyin the combined treatment arm. We also excluded articles that were a subsequentpublication regarding an original trial, that concerned a subset of patientsof the original trial, or that presented long-term follow-up results of theoriginal trial. When data could not be extracted from the article, we contactedthe authors by e-mail to clarify aspects of the study design or to obtainoutcome data in a form suitable for the meta-analysis.

An ad hoc form was designed for data extraction, including diagnosis,sex (percentage female), and mean age of the study sample; antidepressantdrug administered; type of combined intervention; and duration of the study.For each treatment arm, we extracted the number of patients randomized andthe clinical outcome according to 3 categories: (1) the number with full response(according to the most stringent criteria if more than 1 categorization wasused), (2) the number with partial or no response, and (3) the number whodropped out of the study treatment for any reason. The mean daily doses ofantidepressants were converted to milligram equivalents of imipramine hydrochloride,according to conversion factors available in the literature.^{10 - 11} Weassessed the methodological quality of the studies in terms of treatment allocationconcealment (adequate vs inadequate) and independence of evaluators from thoseproviding the interventions (yes vs no).¹² Datawere extracted by 3 of us (P.B., G.T., and S.P.) and controversies resolvedin a meeting.

In this meta-analysis, we estimated the effect of psychotherapy as anaddition to pharmacotherapy of depression to increase the response rate, andwe investigated how it affected dropout and nonresponse rates. For descriptivepurposes, and to guide the interpretation of other results, a random-effectsestimate¹³ (and corresponding 95% confidenceinterval [CI]) of the rate difference (combined treatment minus pharmacotherapyalone) of each of the 3 categories listed in the "Data Extraction" subsectionwas computed. (The estimated pooled rate differences of responders, dropouts,and nonresponders would sum to zero only if the same weights were used topool the estimates from each of 16 trials. Herein, each pooled estimate isgenerated by a separate weighted analysis, using a different set of weights.Therefore, the estimated rate differences of the categories do not sum tozero.) However, these quantities being interdependent and sensitive to theabsolute rate observed in the drug-alone arm, they are not directly amenableto an unbiased analysis. We therefore followed the perspective used for partitioningthe χ² test.¹⁴ (For a 2-row× J-column table, the χ² statistic can be divided intoJ − 1 independent components: in the present context, J = 3 identifiesthe 3 possible outcomes [responder, dropout, and nonresponder] of each ofthe 2 treatments being compared. We first considered the series of 2 ×2 tables [1 for each of the 16 studies], in which the first column correspondedto responders and the second column combined dropouts and nonresponders. Thedistribution of dropouts and nonresponders, conditional on response, was thenstudied, restricting the analysis to the series of 2 × 2 tables in whichthe first column corresponded to dropouts and the second to nonresponders.)

Considering all patients in each study, we first analyzed the odds ratio(OR) of response in the combined arm relative to the pharmacotherapy-onlyarm. This comparison was intended to investigate whether the combined treatmentinduced an increase in responders. We then restricted our analysis to patientswho dropped out or failed to respond and analyzed the OR of dropping out inthe combined arm relative to the pharmacotherapy-only arm. This comparisonwas intended to investigate whether, irrespective of the level of responseachieved in each arm, the relative distribution of dropouts and nonresponderswas the same in the 2 arms. A test for publication bias¹⁵ wasperformed for the analysis of the OR of response. The χ² testfor heterogeneity of ORs (on the natural logarithm scale) was applied, usingas weights the inverse of the variance from the fixed-effects model.¹⁶ Random-effects¹³ pooledestimates of ORs (on the natural logarithm scale) were computed, with their95% CIs (back-transformed using the inverse logarithm transformation for presentationpurposes). We conducted a sensitivity analysis of overall study quality, separatingtrials with appropriate allocation concealment and independent evaluatorsfrom the other trials. The effect of each study on the overall estimates wasalso assessed, computing the pooled estimates after removing each study inturn. Meta-regression analysis on the random-effects estimates was performedusing overall study quality and study duration (≤12 weeks vs >12 weeks)as covariates. All statistical analyses were performed with the computer packageStata.¹⁷

The abstracts of 1123 articles retrieved by the search were screened,and the full publications of 94 articles were examined for pertinence to ourmeta-analysis. Of these, we retained 29 trials of antidepressant use combinedwith psychological treatment vs antidepressant use alone. We excluded 13 articlesfrom the final analysis for the following reasons: treatment was not assignedrandomly,^{18 - 19} an additional nonpharmacologicalintervention or medication affecting outcome and adherence was administeredin one of the study arms,^{20 - 22} resultswere not reported by treatment arm,²³ outcomedata could not be extracted,^{24 - 28} anddrug treatment was not prescribed to all patients.^{29 - 30} Sixteenoriginal studies (reported in 18 articles) were therefore available for analysis.^{31 - 48} Themain characteristics of the studies are listed in Table 1.

Across all study arms considered, 932 patients were randomized to receivepharmacotherapy alone and 910 to combined treatment. The diagnosis was majordepression in 10 studies, dysthymic disorder in 3, and unipolar depressionin 2, while mixed diagnoses were considered in 1. The mean age was 40 yearsor younger in 7 of 13 studies that reported the information and older than40 in 6 studies. The percentage of female patients varied from 50% to 100%(median, 65%). Study length varied from 4 to 24 weeks (median, 12 weeks).The dose equivalent of imipramine varied from 100 to 269 mg daily for the13 studies in which drug data could be extracted. In all but 1 such studies,the dose used in the pharmacotherapy-alone arm was identical to the dose usedin the combined treatment arm. The psychological interventions included thefollowing: cognitive therapy (7 studies), interpersonal psychotherapy (2 studies),psychodynamic therapy (2 studies), and marital therapy, social skills training,problem-solving therapy, a combination of cognitive and behavioral psychotherapy,and a combination of cognitive and interpersonal psychotherapy (1 study each).Most studies described pharmacotherapy (in the pharmacotherapy-only and thecombined treatment arms) as given within the context of the usual care providedby the participating institutions.

In 11 of 16 studies, the randomization procedure was described and appropriate,and in 12 the evaluators were independent from those providing treatment.Nine studies met both quality criteria.

The pooled estimates of the rate differences (combined treatment minuspharmacotherapy alone) of response, dropout, and nonresponse were 12.6% (95%CI, 6.9% to 18.2%), −6.5% (95% CI, −12.1% to −1.0%), and−5.0% (95% CI, −10.8% to 0.8%), respectively. Although this analysiswas biased, it suggests that the increase in response attributable to thecombined treatment is associated with a reduction in dropouts and nonresponderswith respect to pharmacotherapy alone.

Results of the test for publication bias were not significant (P = .32). The borderline significant findings of the testfor heterogeneity (P = .06) led us to choose therandom-effects model, which was applied to all analyses for consistency. Theoverall random-effects estimate of the OR of response was 1.86 (95% CI, 1.38-2.52),indicating a significant advantage of combined treatment over pharmacotherapyalone (Figure 1). No individualstudy showed a particular effect on this result: when pooled estimates wereobtained by removing each of the studies in turn, they were always withinthe CI of the overall estimate (data not shown). A sensitivity analysis wasperformed, stratifying the results according to study quality, which showedno effect (data not shown). Examining the effect of study duration on response,we found no appreciable difference between studies of up to 12 weeks vs thosethat were longer. A formal test of the role of study quality and durationwas performed by means of meta-regression analysis, which confirmed the results,with the P value for the coefficients for study qualityand study duration being P = .79 and P = .58, respectively, when both factors were modeled simultaneously.For further reference, we computed the pooled estimates of response rate differencesfor shorter and longer studies, which were 12.2% (95% CI, 6.0%-18.4%) and15.1% (95% CI, 4.6%-25.6%), respectively.

Having established that the combined treatment is more efficacious thanpharmacotherapy alone, we wanted to assess whether the different responserate corresponded to a different relative frequency of dropout and nonresponsebetween the 2 treatments. We therefore excluded responders and assessed theORs of dropouts, considering only dropouts or nonresponders. Results of thetest for heterogeneity of ORs were not significant (P =.13). The overall random-effects estimate of the risk ratio was 0.86 (95%CI, 0.60-1.24) (Table 2), indicatingno significant difference in distribution of dropouts and nonresponders betweencombined treatment and pharmacotherapy alone (Figure 2). The analyses of effect and sensitivity (data not shown)provided results analogous to those described in the previous subsection.However, longer studies showed a significant reduction in dropouts in thecombined treatment arm (OR, 0.59; 95% CI, 0.39-0.88). A meta-regression analysisconfirmed the lack of effect of study quality (P =.91) and the significant role of study duration (P =.02). As an aid in the interpretation of the results, we also computed thepooled estimates of dropout and nonresponse rate differences for shorter andlonger studies: for studies of up to 12 weeks, these were −1.9% (95%CI, −6.7% to 2.9%) and −8.3% (95% CI, −15.9% to −0.8%),respectively, and for studies longer than 12 weeks, these were −12.9%(95% CI, −3.6% to −2.2%) and 0.4% (95% CI, −6.2% to 7.0%),respectively.

We showed that psychotherapy combined with antidepressant use is associatedwith a higher response rate than pharmacotherapy alone. We found borderlinesignificant heterogeneity of the ORs, possibly due to differences in casemix, setting, and duration of treatment among studies. Subgroup analyses accordingto depression severity were not feasible because of unreported data. However,study entry and evaluation criteria were similar across studies. The distributionof dropouts and nonresponders was comparable in the 2 treatment modalities(OR, 0.86; 95% CI, 0.60-1.24). It is difficult to say whether the observedsuperiority of the combined treatment corresponds to an additive benefit ofpsychotherapy or to a nonadditive phenomenon. The results suggest, however,that the addition of psychotherapy reduces nonresponse and helps to keep patientsin treatment.

The subgroup analysis according to study length provided more insightinto this result. Study duration was an obvious effect modifier to be considered,because the number of patients in the dropout group can only increase, ifit changes at all, with increasing duration of treatment. Our results indicatethat in studies of up to 12 weeks the overall results of the meta-analysisare confirmed by the ORs; the rate differences can be considered consistentwith this result based on the width of their corresponding CIs. In studieslonger than 12 weeks, there is a significant reduction in the number of dropoutsin the combined treatment arm, and the rate difference analysis indicatesthat there is no change in nonresponse rate (0.4%) but a strong and significantreduction in dropout rate (−12.9%). For shorter treatment, the combinedmodality may thus be effective in some patients who would not otherwise respondto pharmacotherapy alone, while for longer duration treatments, the additionof psychotherapy induces a stronger compliance with the combined treatment.The design of the trials reviewed did not allow us to determine whether inlonger studies the improved response in the combined treatment arm correspondsto a joint therapeutic effect (additive or otherwise) of pharmacotherapy pluspsychotherapy or to a prolonged exposure to medication in patients who remainon treatment. Nevertheless, one wonders what would be the effect of a compliance-enhancingintervention on the excess dropout rate in longer treatments among patientsreceiving pharmacotherapy only. The possible adherence-enhancing role of psychotherapywas suggested, although not supported by data, as early as 25 years ago byKlerman, as quoted by Paykel.⁵ Other authorshave suggested that patients may be more satisfied with combined therapy thanwith single-modality treatments or that they would discontinue pharmacotherapyif it were not accompanied by psychotherapy.⁴⁹

By showing that dropout rates can be reduced by a combination of pharmacotherapyand psychological treatment, this study calls for the identification of simplebut effective strategies to better respond to the needs of depressed patients,thus keeping them in treatment. Research on adherence-enhancing interventionshas not provided clear evidence in this respect, although a higher intensityof intervention² and multifaceted interventions^{50 - 51} appear promising. If the therapeuticeffect of psychotherapy were only moderate compared with a strong compliance-enhancingeffect, then more feasible and cheaper means of achieving the same goals couldbe explored. Only properly designed studies can help to disentangle the genuinetherapeutic effect of psychotherapy from its compliance-enhancing effect,over and above the effect of drugs. However, psychological treatment may havea stronger effect than pharmacotherapy on patient satisfaction and socialfunctioning or other dimensions of well-being that cannot be measured in termsof improvements in depression scores.⁵² Theresults of this study favor a broader approach to the management of depressedpatients and the adoption of other axes of outcome evaluation besides thecontrol of symptoms. Especially in longer treatment, research should aim atenhancing compliance and response rates, incorporating these principles.

Correspondence: Carmine Munizza, MD, Centro Studi e Ricerche in Psichiatria,Piazza del Donatore di Sangue, 3, 10154 Torino, Italy (cmunizza@tin.it).

Submitted for publication September 25, 2003; final revision receivedJanuary 30, 2004; accepted February 9, 2004.

The study was supported by Centro Studi e Ricerche in Psichiatria; IstitutoSuperiore di Sanità, Rome; and Ravizza Pharmaceuticals, Milan, Italy.

We thank 3 anonymous referees for their helpful comments regarding aprevious version of this article.