Regular Articles - CiteSeerX

Regular

and

Articles

Patient Predictors of Response to Psychotherapy Pharmacotherapy: Findings in the NIMH Treatment of Depression Collaborative Research Program

Stuart

M. Sotsky, M.D., M.P.H., David R. Glass, Ph.D., M. Tracie Shea, Ph.D., Paul A. Pilkonis, Ph.D., Joseph F. Collins, Sc.D., Irene Elkin, Ph.D., John T. Watkins, Ph.D., Stanley D. Imber, Ph.D., William R. Leber, Ph.D., Janet Moyer, M.S., and Mary Ellen Oliveri, Ph.D.

Objective: The authors investigated patient characteristics predictive of treatment response in the National Institute of Mental Health (NIMH) Treatment of Depression Collaborative Research Program. Method: Two hundred thirty-nine outpatients with major depressive disorder according to the Research Diagnostic Criteria entered a 1 6-week multicenter clinical trial and were randomly assigned to interpersonal psychotherapy, cognitivebehavior therapy, imipramine with clinical management, or placebo with clinical management. Pretreatment sociodemographic features, diagnosis, course of illness, function, personality, and symptoms were studied to identify patient predictors of depression severity (measured with the Hamilton Rating Scale for Depression) and complete response (measured with the Hamilton scale and the Beck Depression Inventory). Results: One hundred sixtytwo patients completed the entire 1 6-week trial. Six patient characteristics, in addition to depression severity previously reported, predicted outcome across all treatments: social dysf unction, cognitive dysfunction, expectation of improvement, endogenous depression, double depression, and duration of current episode. Significant patient predictors of differential treatment outcome were identified. 1) Low social dysfunction predicted superior response to interpersonal psychotherapy. 2) Low cognitive dysfunction predicted superior response to cognitive-behavior therapy and to irniprarnine. 3) High work dysfunction predicted superior response to irnipramine. 4) High depression severity and impairment of function predicted superior response to imipramine and to interpersonal psychotherapy. Conclusions: The resuits demonstrate the relevance of patient characteristics, including social, cognitive, and work function, for prediction of the outcome of major depressive disorder. They provide indirect evidence of treatment specificity by identifying characteristics responsive to different modalities, which may be of value in the selection of patients for alternative treatments. (Am J Psychiatry 1991; 148:997-1008)

Presented in part at the 139th annual meeting of the American Psychiatric Association, Washington, D.C., May 10-16, 1986. Received Aug. 13, 1990; revision received Feb. 20, 1991; accepted March 20, 1991. From the Department of Psychiatry and Behavioral Sciences, George Washington University Medical Center; the Mood, Anxiety and Personality Disorders Research Branch, NIMH, Rockville, Md.; Western Psychiatric Institute and Clinics, Pittsburgh, Pa.; the VA Medical Center, Perry Point, Md.; the University of Chicago School of Social Service Administration; and the Department of Psy-

Am

J Psychiatry

1 48:8,

August

1991

chiatny and Behavioral Sciences, University of Oklahoma Health Sciences Center, Oklahoma City. Address reprint requests to Dr. Sotsky, Department of Psychiatry and Behavioral Sciences, George Washington University Medical Center, 2150 Pennsylvania Ave., N.W., Washington, DC 20037. The authors thank David Reiss, M.D., for scientific consultation and manuscript review, Samuel Simmens, Ph.D., for statistical consultation, Victoria L. Herzberg for administrative assistance, and the clinical evaluators and therapists at the three research sites.

997

PATIENT

PREDICTORS

OF RESPONSE

I

n the effort to understand the variability of response to treatments for depression, the relevance of patient characteristics has received much attention. Nevertheless, relatively little systematic research has exammed the predictive value of these characteristics in studies comparing different forms of psychotherapy and pharmacotherapy. Certain characteristics of the patient and the nature of the depression may be general indicators of prognosis irrespective of treatment, while others may be indicators of response to individual treatments alone or of differential treatment outcome, that is, preferential response to one or more treatments compared to others. The elucidation of predictors of response addresses an important aspect of treatment specificity by characterizing the type of patients for whom a treatment is most or least effective; it could also have direct clinical applicability in the selection of patients for the most appropriate treatment modality. In addition, some predictors of response to treatment may provide indirect evidence about the putative specific mechanism of a particular modality

by indicating

patient

characteristics

that

are

especially responsive to it, such as functional and relational capacities or type of depression. This may be especially important because only limited evidence about differential treatment effects on measures of outcome hypothesized to be specific to each modality has emerged from this study (1, 2). Although we chose to consider patient characteristics initially, since there is some indication from the literature (3, 4) that they have greater influence on outcome, we plan to address therapist, relationship, and process variables in future analyses. Comprehensive reviews of general predictors of response to psychotherapy (5, 6) and to tricyclic antidepressants (7-9) reveal that the literature is characterized by lack of consensus about or replication of many findings. The general predictors of response to psychotherapy that have emerged have been derived from many studies based on various psychotherapies, often in small and heterogeneous samples of patients with diverse disorders. Other major reasons for the inconsistent findings among studies include lack of specification of the treatment, variability of inclusion and diagnostic criteria, lack of standardization of outcome measures, variability of criteria for improvement, and differences in selection of patient variables and measures investigated. The purpose of this article is to report the patient characteristics that predicted treatment response in the National Institute of Mental Health (NIMH) Treatment of Depression Collaborative Research Program (10), which was the first multicenter, comparative dinical treatment trial in the field of psychotherapy research initiated by NIMH. This study compared the efficacy of interpersonal psychotherapy, cognitive-behavior therapy, imipramine with clinical management (as a standard reference condition), and placebo with clinical management (as a control condition) for out-

998

patients with nonbipolar, nonpsychotic major depressive disorder. Several features of the Collaborative Research Program clinical trial design were considered advantageous for the study of the relation of patient predictors

to treatment

outcome.

First,

the

multisite,

common

protocol design permitted study of a larger sample than do most single-site studies, but with uniformity of diagnosis and severity of depression across sites. Second, the extraordinary standardization of the thera-

pies-with

treatment

manuals,

therapist

training,

and

monitoring of the quality of performance-would enhance the prediction of response to specific treatments by reducing the variability of the treatments themselves. Third, the use of a control condition, placebo with clinical management, would help differentiate predictors of response to specific treatments from predictors of general response or response to nonspecific treatment. Fourth, the use of standard outcome measures and response criteria would permit better comparison with the results of other studies. We examined predictors in two groups of patients: 1) those who completed the course of treatment, in order to search

for predictors

of outcome

sure to the all of those

specific ingredients who entered the

those

who

search

for

withdrew

of the

was

full expo-

therapies, trial,

completion, full

range

and 2) including

in order

to

of outcomes.

single-site comparative treatment studies psychotherapy and cognitive-behavior

therapy whose results would those of the NIMH Treatment

orative

there

of the treatment

before

predictors

In previous of interpersonal

when

Research

Program,

be

most comparable to of Depression Collab-

some

patient

predictors

of

response to psychotherapy and tricyclic antidepressants were reported. Among the six studies that reported patient predictors of response to cognitive-behavior therapy, one (1 1) found that pretreatment symptom severity was significantly associated with negative outcome at termination, while another (12) reported an association with positive outcome. Inconsistent findings were reported for endogenous depression (13, 14). Learned resourcefulness, assessed by the

Self-Control itive

ured poor

Schedule, ,

by the response

havior

Dysfunctional at termination

therapy

follow-up,

(16)

though

and

not

cognitive-behavior

Among

was

found

to be related to posdysfunction, as measAttitude Scale, predicted for group cognitive-be-

( 1 1 1 5). Cognitive

outcome

negative

therapy

studies

of patient

interpersonal

psychotherapy,

on the Social

Adjustment

outcome

at termination,

Am

1-year

individual

(17).

predictors initial

Scale

of outcome social

and

interpersonal (19).

J Psychiatry

of emo-

predicted not depreswas associpsychotreatment with favor-

psychotherapy

148:8,

in

adjustment

an index

tional freedom, education, and occupation positive outcome in social adjustment but sion severity (4). Endogenous depression ated with poorer outcome for interpersonal therapy than for tricyclic or combination (18). Situational depression was associated

able response to either a tricyclic antidepressant

at

for

August

or

1991

SOTSKY,

Among these studies, there were several patient characteristics that predicted response to tricyclic antidepressants. Endogenous depression was associated with favorable outcome with an antidepressant alone or in combination with interpersonal psychotherapy (18), as well as good outcome for patients who completed treatment, but poor outcome for those who dropped out (14). Depression severity was related to negative outcome ( 1 1 ), and learned resourcefulness was related to poor response (15).

METHOD

A common

protocol

was

conducted

at three

clinical

research sites with a prospective, random-assignment, placebo-controlled design, double-blind for pharmacotherapy and with independent, blind clinical evaluation. A detailed description of the background and design of the main treatment study and the pilot training study has been previously published (10). The subjects were male and female outpatients between the ages of 21 and 60 years who met the Research Diagnostic Criteria (RDC) for a current, definite episode of major depressive disorder on the basis of the Schedule for Affective Disorders and Schizophrenia (SADS) (20) structured interview and who had

a score

of 14 or more

on the 17-item

modified

version

of the Hamilton Rating Scale for Depression (2 1) for at least 2 weeks before initial screening and again at rescreening after a 1- to 2-week wait or drug washout period. Exclusion criteria consisted of other specific psychiatric disorders, medical contraindications for the use of imipramine, concurrent psychiatric treatment, current active suicide potential, and need for immediate treatment. Patients were clinically referred, voluntary research subjects who gave written informed consent to participate.

All 28 therapists

were

experienced

psychiatrists

and

psychologists who had received further clinical training by independent expert trainers and had met competence criteria in order to participate. Their performance was monitored throughout the study (10, 2224). Treatments were conducted in accord with detailed manuals that specified the theoretical rationale, strategies, techniques, and boundaries of each modality (2527). All treatments were planned to be 16 weeks in duration and to consist of 16-20 sessions. The pharmacotherapy dosage schedule was flexible, with a goal

of 200 mg management

and

a maximum component

for

of 300 mg. The clinical the pharmacotherapy

conditions provided not only guidelines for medication monitoring and management but also support, encouragement, and advice as necessary, although specific psychotherapy interventions were proscribed.

The total who actually

sample entered

(N=239) treatment

comprised all patients from among the 250

patients assigned.

who met inclusion The completer

criteria sample

and were comprised

patients

who

at least

12 sessions

Am

J Psychiatry

completed

148:8,

August

1991

randomly only the

and

15

GLASS,

SHEA,

ET AL.

weeks of treatment and had clinical evaluations at termination. Clinical evaluations were available for 155 patients on the Hamilton depression scale for analyses of depression severity and for 156 patients for analyses of complete response (because of the inclusion of an additional patient for whom the Beck inventory score only was available). There were no statistically significant differences between treatments in overall attrition, treatment-related attrition, or symptomatic failure. Seven patients did not have termination evaluations (1). The findings on comparative treatment efficacy with regard to depressive symptoms and general functioning, modality-specific measures of change, and the temporal course of treatment effect have been reported elsewhere (1, 2, and manuscript by J.T. Watkins et al., submitted for publication).

A comprehensive

review

of the literature

was

under-

taken to identify patient predictors of response to cognitive-behavior therapy, interpersonal psychotherapy, and psychotherapy in general, as well as to tricyclic antidepressants and imipramine in particular. Among the predictor variables for which there was evidence of stability and replication, a limited set of independent variables for the multivariate analyses was selected on the basis of the distribution, intercorrelation, reliability, and face validity of the relevant measures. Personality disorders were added on the basis of experience in the pilot study. Because of the potential importance of a cognitive function predictor for cognitive-behavior therapy, the Dysfunctional Attitude Scale total score was selected on an exploratory basis. Two subsequent reports suggesting its relationship to outcome (see the beginning of this article) confirmed that decision. The 26 independent variables were grouped within three domains: 1) sociodemographic variables-age,

sex,

marital

status,

social

class;

2)

diagnostic

and

course variables-endogenous, recurrent, primary, situational, and double depression, melancholia, family history of affective disorder, age at onset of first episode, duration of current episode, acuteness of onset of current episode, and number of previous episodes; and 3) function, personality, and symptom variables-social dysfunction, work dysfunction, cognitive dysfunction, social satisfaction, expectation of improvement, number of personality disorders, dramatic personality disorder, odd personality disorder, anxiety, somatization/hypochondriasis, and interpersonal sensitivity. The dependent measures of depression outcome at termination were 1) complete response, a stringent categorical measure based on the combination of a 17item Hamilton depression scale score less than or equal to 6 and a Beck Depression Inventory (28) score less than or equal to 9, and 2) depression severity, a continuous measure based on the 23-item Hamilton depression scale score, which included cognitive and atypical vegetative symptom items. Two dependent variables were selected to examine predictors of complete response or remission and change in the severity of depression, which would reflect partial response as well. One measure was based solely on clinical evalu-

999

OF RESPONSE

PREDICTORS

PATIENT

ator ratings, while the other also incorporated patient self-reports. Scores were based on ratings at termination of treatment for completers and on the last rating obtained for patients who were withdrawn or dropped out (generally, at either interim or early termination evaluation, but for 20 early dropouts, at rescreening evaluation). The interrater reliability, as assessed by intraclass correlation coefficients, for the Hamilton 17item scale score ranged from 0.92 to 0.96 across sites and from 0.90 to 0.98 within sites; for the Hamilton 23-item scale score it ranged from 0.93 to 0.96 across sites and from 0.89 to 0.98 within sites. The Beck inventory total score showed good internal consistency, with an alpha coefficient equal to 0.78. The percentages of patients achieving complete response were 43.6% (N=68) and 31.4% (N=7S) in the completer and total samples, respectively. The mean±SD depression severity scores at termination were 9.43 ± 7.33 and 13.92±10.19 in the completer and total samples, respectively.

Data

Analysis

To reduce

further

the number

of independent

pa-

tient variables from the initial set of 26 for the main predictor analyses across treatments, preliminary multiple regression analyses were conducted within each individual treatment condition, and only the most consistent predictors of outcome within individual treatments were selected for the main analyses. In both the total sample and the completer sample, initial analyses

were conducted the three

single

of predictor

for independent domains,

variables

and

within

variables then

in each of

the “best”

each domain

subsets

were corn-

bined, with the pretreatment Hamilton depression score included, in final analyses conducted for each outcome measure. For these preliminary analyses we used the method of all possible subsets regression (29), which examines all possible models containing the independent variables and selects the “best” subsets on the basis of an algorithm that uses the criterion of minimal Mallow’s C (30) to select the regression

model

which

minimizes

the total mean squared

the relative

however, importance

that the results do not identify of any single

variable,

since

the

effect of each variable on outcome variance takes into account the partial effects of all of the other variables in the

model.

We

recognize

that,

given

the

number

of

initial patient variables and the size of the groups in the individual treatment conditions, there is an increased risk of chance findings because of the inherent limita-

1000

of the regression

analytic

approach.

We therefore

consider these analyses exploratory in nature. While certain predictor relationships replicated earlier findings, other predictors will require further attempts to replicate.

The main data analyses examined the evidence for possible predictors of outcome across all treatments and of differential outcome among treatments. Only 13 variables were selected for these analyses, which were the most consistent predictors of outcome in the final

regression

having

been

analyses

within

predictors

individual

in at least

treatments,

two

of the four analyses and at a level of significance of p