Social support and the outcome of major depression. L K George, D G Blazer, D C Hughes and N Fowler BJP 1989, 154:478-485. Access the most recent version at DOI: 10.1192/bjp.154.4.478

References This article cites 0 articles, 0 of which you can access for free at: http://bjp.rcpsych.org/content/154/4/478#BIBL Reprints/ To obtain reprints or permission to reproduce material from this paper, please write permissions to [email protected] You can respond http://bjp.rcpsych.org/cgi/eletter-submit/154/4/478 to this article at Downloaded http://bjp.rcpsych.org/ on December 4, 2013 from Published by The Royal College of Psychiatrists

To subscribe to The British Journal of Psychiatry go to: http://bjp.rcpsych.org/site/subscriptions/ British Journal of Psychiatry (1989), 154, 478—485

Social Support and the Outcome of Major Depression

LINDA K. GEORGE, DAN G. BLAZER, DANA C. HUGHES and NANCY FOWLER

One hundred and fifty middle-aged and elderly adults with a diagnosis of major depression wereassessedinitiallyasin-patients,and were reinterviewed6-32 monthslater.Bothsize of socialnetworkand subjectivesocialsupportwere significantpredictorsofdepressive symptomsatfollow-up,withbaselinedepressionscoresandotherpredictorsofoutcomestatus statisticallycontrolled.Subjectivesocialsupportwas most stronglyassociatedwithmajor depression; this effect was significantly stronger for middle-aged than older adults, and for men than women. Differences in the effects of marital status, size of social network, and subjective social support also suggest the importance of distinguishing between involvement in and quality of interpersonal relationships.

A review of the literature reveals an increasing Few investigators have studied the predictive interest in the outcome of depressive illness. power of patients' personal characteristics or the Although evidence is limited, previous research experience of life events for recovery status, although consistently indicates that the outcome of depressive the limited research available suggests their potential illness is variable, and that sustained recovery relevance. Baldwin & Jolley (1986), in a study of 100 is less common than had been expected (Keller severely depressed elderly in-patients, reported that & Shapiro, 1981; Keller et a!, 1982; Murphy, only two factors were associated with recovery status: 1983; Baldwin & Jolley, 1986; Carney et a!, males and persons in poor physical health exhibited 1987). Keller et a! (1982), for example, found significantly lower rates of recovery. In another study that only 50°loof 101 in-patients had recovered of elderly depressed in-patients, Murphy (1983) from their index episodes of depression at one reported that the experience of stressful life events year follow-up. Similarly, Murphy (1983) reported between the baseline and follow-up assessments was that only a third of a sample of 124 elderly related to poorer outcomes. depressed patients had recovered one year after Social support has been perhaps the most fre their index episodes. Given this variability in quently examined factor in relation to differential outcome, and the poor prognosis for many depressed outcome of psychiatric disorder, although the results patients, it is important to identify factors associated of previous studies are inconsistent. In a longitudinal, with different outcomes. community-based study, Holahan & Moos (1@l) Only limited information is available concerning reported that decreases in social support over a the risk factors for recovery from depressive illness. one-year interval were associated with increased The types of risk factors examined in one or more psychiatric symptoms. Pattison et a! (1979) reported previous studies include clinical features of the index that social support can help to alleviate symptoms episode, patients' personal characteristics, life events of anxiety, but added that the salubrious effects of during the interval between baseline and follow-up supportive relationships are contingent upon the assessments, and the availability of supportive support network being “¿normal―rather than interpersonal relationships. At this point, no con “¿pathologic―.In a longitudinal, community-based sistent markers of recovery have been identified. study, Henderson examined the effects of both Carney et a! (1987) reported that the presence or objective (i.e. network size) and subjective (i.e. absence of melancholia failed to distinguish between perceptions of adequacy) social support on neurotic poor and good outcome. In a six-month prospective symptoms (Henderson & Moran, 1983; Henderson, study, Zimmerman et a! (1987) found that the 1984). Henderson & Moran (1983) reported that dexamethasone suppression test (DST), which was objective social support is unrelated to changes in administered during the first week after admission symptoms, but perceptions of inadequate support are to primary unipolar depressed patients, was not associated with both the onset and continuation associated with any of the outcome variables studied of symptoms. Henderson (1984) suggested, how (including the Hamilton Rating Scale, the Beck ever, that conclusions about the impact of social Depression Inventory, and the Global Assessment support be restricted to evidence based on objective Scale). dimensions, because subjective measures may be

478 SOCIAL SUPPORT AND DEPRESSION 479 contaminated by depressed affect. Finally, Murphy a check of patients' charts for psychiatric diagnoses, and (1983) reported that the presence of a confidant administration of the Center for Epidemiologic Studies neither protected the elderly depressed patients in her Depression Scale (CES-D) (Radloff, 1977). Those patients sample against relapse, nor promoted recovery. with a diagnosis of affective disorder and/or a CES-D score Overall, then, previous findings fail to identify of 16or greater were enrolled into the research programme. A score of 16 or more on the CES-D has been shown to factors consistently related to recovery from be indicative of clinically significant depressive symptoms depressive illness. The methodological limitations of (Radloff, 1977). these previous studies also must be recognised, After a patient ‘¿passed'one or both of the screening however. First, a number of them, especially those criteria, a clinical psychologist administered the Duke examining the effects of social support, were based Depression Evaluation Schedule for the Elderly (DDESE) on community samples, in which the prevalence of (a composite interview schedule). Diagnostic information significant depression is low. Second, several studies was obtained using several sections of the Diagnostic reviewed (again, primarily those examining the Interview Schedule (DIS) (Robins et a!, 1981), including effects of social support) focused on measures of the sections on depression, mania, alcohol abuse/ dependence, generalised anxiety disorder, and the Mini neurotic symptoms, general psychopathology, and mental State Exam (Folstein et a!, 1975). We enriched anxiety. Consequently, additional effort is needed the DIS depression section by adding items on sleep to examine the factors associated with recovery from problems, and items to distinguish both major depression depressive illness. Third, in all previous studies, with and without melancholia, and depression with and examination has been restricted to one or two without psychotic features. DDESE and screening data were potential correlates or predictors of outcome —¿no reviewed by a diagnostic team, and standardised DSM effort has been made to examine a broader range of III (American Psychiatric Association, 1980) diagnoses the potential factors. A corollary of this pattern is assigned to all participants. These standardised diagnoses failure to examine the potential predictors of have been compared with diagnoses derived from interviews usingthe Schedulefor Affective Disorders and Schizophrenia outcome in a multivariate framework. (SADS) (Endicott & Spitzer, 1978), in which the clinical The present study was designed to overcome some interviewers were blind to the DDESE and screening data, of these limitations. Six categories of factors as wellas to the standardised diagnoses based on those data. potentially related to the outcome of depressive Comparisons of the standardised and SADS diagnoses illness were examined: patients' demographic indicate high concordance for the diagnosis of major characteristics, patients' psychiatric histories, clinical depression (x >0.80), and acceptable concordance for the features of the index episode of depression, distinction between major depression with and without psychiatric comorbidity, life events, and social melancholia (x >0.60). support (as measured in multiple dimensions). Follow-up interviews were conducted by telephone after 6—32months. The interview schedule included the CES Subjects consisted of 150 middle-aged and elderly D, other questions about depressive symptoms and depressed in-patients who were administered baseline symptom episodes, questions about medication and use of interviews during their index hospital stay, and were mental health services, and measures of social functioning. reinterviewed 6—32months later. Although a wide Of the 200 patients contacted, 177participated in the follow range of potential predictors of recovery status was up interview. Four subjects had died, five were not examined, our primary interest was in the effects of competent to report reliable data, and 14 refused. One social support. Consequently, our major hypotheses hundred and fifty of the 177 participating subjects had were that impaired social support at baseline would received a standardised diagnosis of major depressive be associated with poorer outcome, and that episode, with (n = 56) or without (n = 94) melancholia at baseline. These 150 subjects comprise the sample used in subjective social support would be more strongly thi@spaper. Compatible with the research design, these 150 associated with recovery status than more objective subjects are in two age groups: 35—50years (n = 77) and social support dimensions. 60+ (n=73).

Method Measures

Sample Outcome of depression. The dependent variable is outcome of depression as measured by CES-D score at follow-up. Our sample was taken from in-patients participating in the For some analyses, the CES-D is used in continuous form Duke UniversityCenter for the Study of Depression in Later (potential range= 0-60). For other analyses, the CES-D was Life, a multidisciplinary research programme, designed to split into two categories: ‘¿recovered'(CES-D@15)and ‘¿non contrast the phenomenology of depression in middle (age recovered' (CES-D 16). As noted previously, this cut-off 35—50years) and later (older than 60) life. All patients were point is compatible with that used in previous studies screened for significant depressive symptoms: this involved (Radloff, 1977). 480 GEORGE ET AL

Six categories of independent variables were included in tapping tangible services that the respondent receives from the analyses. network members. Examples of the tangible services included are help when sick; financial assistance; help with (a) Demographic characteristics. Three were examined: sex, housework, home repairs, and meals; transportation; gifts; marital status, and age (35—50,and 60 + years). and advice about problems. (iv) Subjective social support is a nine-item scale, tapping the respondent's perceptions (b) Psychiatric history. Three measureswere examined: of: his/her inclusion as a valued and useful member of a presence v. absence of (i) family history of suicide, (ii) social network; whether the network would provide help family history of mental disorder, and (iii) a previous and support if needed; and general satisfaction with the episode of major depression. All three measures were based quantity and quality of social support available. Each on patients' self-reports in the DDESE. dimension was categorised as impaired (scores in bottom fifth of scale) v. unimpaired support. Note that the first (c) Clinical features of the index episodes. Three features three measures tap relatively objective facets of social were examined as potential predictors of recovery status: support, whereas the fourth dimension is a purely subjective baseline CES-D score, duration of the depressive episode evaluation. before the baseline interview, and presence v. absence of It is important to note that all of the independent major depression with melancholia. Baseline CES-D scores variables were measured at baseline, 6—32months before were included as indicators of disease severity; in addition, assessment of recovery status. Because of the wide variation in the multivariateanalyses,wecontrolledon baselineCES in the intervals between the two interviews, time to follow D scores to highlight the effects of other predictors on up was included as a control variable in the multivariate changes in levels of depressive symptoms. Duration of the analyses, and was examined as a potential correlate of depressive episode was viewed as an indicator of chronicity recovery status. and was assessed via patients' self-reports at the baseline interviews. Methods of analysis (d) Psychiatric comorbidity. To determine whether First, contingency table analysis was used to examine the coexisting psychiatric conditions decreased the likelihood bivariate relationships between the independent variables of recovery, we examined presence v. absence of three other and outcome (recovered v. non-recovered). Subsequently, DSM—IlIdiagnoses: dysthymia, alcohol abuse and/or multiple regression was used (a) to examine the effects of dependence, and generalised anxiety disorder. Diagnoses were based on the DIS sections included in the DDESE. the independent variables on outcome levels of depressive symptoms, with the effects of other predictors statistically Presence v. absence of mild cognitive impairment was controlled, and (b) to determine the efficacy of the determined on the basis of responses to the Mini-mental independent variables, as a set, in predicting depressive State Exam. symptoms at follow-up. In the regression analyses, follow (e)Life events. The DDESEincludeda 19-itemlife-events up CES-D score (in continuous form) was the dependent variable. Hierarchical regression was used to enter the checklist. This list, previouslyusedin theDukeEpidemiologic Catchment Area surveys (Blazer et a!, 1987; Hughes et a!, baselineCES-Dscore(alsoin continuousform)as the first predictor examined; the other independent variables were 1988a,b), asks respondents about the occurrence of life entered in a subsequent step. This procedure, known as events during the past year. The events listed include health residualisedchange analysis, estimates the effects of the events, marriage and family events, work-related events, independent variables on changes in depressive symptoms financial problems, residential relocation, interpersonal between test dates. Finally, potential interactions among disruptions, and legal problems. For each event reported, the independent variables were examined in the regression respondents also were asked to rate the event as negative models. or positive, as important or unimportant, and as expected or unexpected. For the purposes of this paper, two measures of life events were examined: number of positive events Results (coded as none or one v. two or more) and number of unexpected, very important, negative events (coded as none Table I presents the bivariate relationships between the v, one or more). These measures permiued us to determine independent variables and recovery status at follow-up. whether positive experiences promote recovery and whether Only a few of the independent variables were significantly stressful events predict poorer prognosis. related to outcome of depressive illness. Among the demographic characteristics, only marital status was (1) Social support. Finally, the Duke Social Support Index significantly related to recovery status. Unexpectedly, the (Landerman et a!, 1989) was used to generate information married were less likely to have recovered. None of the about four dimensions of social support. (i) Size of measures of psychiatric history were significantly related social network is measured by four items, including an to recovery. Of the three clinical features of the index indication of complexity. (ii) Amount of social interaction episode, as might be expected, patients with baseline CES-D is a four-item index, measuring the frequency of the scores of 16 or more, and patients diagnosed as having respondent's interaction with members of the support major depression with melancholia were less likely to have network. (iii) Instrumental support is a 13-item index, recovered than their respective counterparts. Two of the SOCIAL SUPPORT AND DEPRESSION 481

TABLE I Correlates of recovery status Independent variable Recovered Not (n = 73) (n = 77) (n = 150) n nrecovered %Total n Sex Male 34 46 26 33 60 40 Female 39 54 51 67 90 60 Marital status Married 51 69 62 80 113 75 Widowed 6 8 8 10 14 9 Divorced/separated 10 13 7 10 17 11 Never married 6 10 0 0 6 5 Age: years 35—50 39 54 38 49 77 51 60 and older 34 46 39 51 73 49 Family history of suicide No 38 52 50 64 88 58 Yes 35 48 27 36 62 42 Family history of mental disorder' No 41 56 36 46 77 51 Yes 31 44 38 54 69 49 Baseline CES-D score <16 9 13 1 2 10 7 16 64 87 76 98 140 93 Duration of depressive episode before admission' <13 months 57 78 52 67 109 73 @ 13 months 16 22 24 33 40 27 Previous episode of depression No 10 13 8 11 18 12 Yes 63 87 69 89 132 88 Major depression with melancholia** No 53 73 41 53 94 62 Yes 20 27 36 47 56 38 Dysthymia** No 39 54 22 29 61 40 Yes 34 46 55 71 89 60 Mild cognitive impairment No 69 94 71 92 140 93 Yes 4 6 6 8 10 7 Alcohol abuse/dependence No 60 82 64 83 124 82 Yes 13 18 13 17 36 18 Generalised anxiety disorder No 24 32 11 15 35 24 Yes 49 68 66 85 115 76 Unexpected, very important negative life events None 53 68 57 74 110 73 One or more 20 32 20 26 40 27 Positive life events None or one 8 11 7 10 15 10 Two or more 65 89 70 90 135 90 Social network' Impaired 42 57 32 43 74 51 Not impaired 31 43 41 57 72 49 Social interaction' Impaired 11 16 21 29 32 23 Not impaired 61 84 52 71 113 77 Instrumental support' Impaired 8 11 9 13 17 12 Not impaired 65 89 64 87 129 88 Subjective social support***@ Impaired 4 6 19 27 23 16 Not impaired 69 94 53 73 122 84 Length of time to follow-up 12—17months 53 68 54 70 107 71 @ 18 months 20 32 23 30 43 29 1. Total does not equal 150 because of missing data. *[email protected], @[email protected], [email protected] for differences between recovered and not recovered. 482 GEORGE ET AL

TABLE II Residualisedchangeanalysisof recoveryfrom depression

Independent2bBbBBaselinevariableModel 1Model score0.320.26**0.320.25**SexCES-D (female=l)2.440.075.400.16*Age =1)0.980.032.500.08Widowed'—8.43—7.29—0.12Divorced/separated'—7.23—8.70Never(60 or older

married'—10.66—10.47Dysthymia5.980.18*5.650.17*Unexpected,

0.04-0.000.650.02Social very important negative life events-

0.20**Subjectivenetwork (high score = impaired)—7.86— 0.24**—6.46— score=impaired)8.880.20*21.940.49***Agesocial support (high

14.04Sex by subjective social support———

15.90Intercept6.123.09R20.3l***0.37***by subjective social support———

1. Marital status was enteredas a seriesof ‘¿dummy'variables;the omitted, referencecategoryis the married. •¿Ps0.05,**[email protected], ***p0.OrJl3l. four measures of psychiatric comorbidity were significantly Unexpectedly, however, impaired social network at baseline related to outcome status: persons with a diagnosis of was associated with lower numbers of depressive symptoms dysthymia, as well as those with a diagnosis of generalised at follow-up. Overall, the main effects model explains 31% anxiety disorder, were less likely to have recovered. Neither of the variance in follow-up CES-D scores. This explanatory measure of life events was significantly related to outcome. power is highly significant. Two of the four dimensions of social support, however, Model 2 is nearly identical to model 1 in terms of the were significantly associated with recovery status: as main effects of the independent variables. Note, however, expected, patients who were impaired in social interaction that sex does not emerge as a significant predictor of and subjective social support at baseline were less likely to depressive symptoms at follow-up until the significant have recovered. Length of time to follow-up was unrelated interaction terms are added to the model. Inclusion of the to recovery. interaction terms also increases the explained variance of Table II presents the multivariate results from the the model to 37%. Before interpreting the significant residualised change analysis. Before describing the findings, interaction terms, it should be noted that the stress-buffering however, the contents of Table II merit brief description. hypothesis (Cohen & Wills, 1985; Alloway & Bebbington, First, two regression models are presented. Model 1presents 1987)was also tested. This hypothesis posits that life events the main effects model; model 2 presents the results for have more negative effects under conditions of inadequate the regression equation including significant interaction social support than under conditions of adequate terms. Second, the configuration of independent variables support. We tested the stress-buffering hypothesis for in Table II merits comment. Regression models including the four dimensions of social support included in the all of the independent variables in Table I were estimated database. In no case was the interaction term statistically initially. Subsequently, most of the independent variables significant. that did not significantlypredictoutcomeCES-Dscorewere As model 2 indicates, two significant interactions were deleted from the models. Both age and sex were retained identified; the effects of subjective social support were in the models, however, because the main effects for any conditional on both sex and age. Interpretation of the variables in the interaction terms must be estimated. interaction terms is facilitated by examination of Table III, Theresultsfor modd 1indicatethat higherbaselineCES which provides mean CES-D follow-up scores for study D scores, being married, a baseline diagnosis of dysthymia, participants cross-classified by age and subjective support, a small social network, and an impaired subjective social and by sex and subjective support. The mean CES-D support are significant predictors of higher numbers of outcome score for middle-aged patients with impaired CES-D symptoms at follow-up, as expected on the basis subjective social support at baseline is twice as high as the of the bivariate results. Several predictors that were means for the middle-aged patients with unimpaired significantlyrelatedto outcomeCES-Dscoreinthebivariate subjective social support, and for older patients regardless analyses (see Table I) were reduced to non-significance in of baseline levels of subjective social support. Subjective the regression analysis (e.g. major depression with social support is more strongly related to CES-D symptom melancholia, generalised anxiety disorder). In addition, one levelsat follow-upfor menthan for women.Thus,although predictor was significant in the regression analysis that was perceptions of inadequate social support generally predict not significant in the bivariate analyses: social network. higher levels of depression at follow-up, they are especially SOCIAL SUPPORT AND DEPRESSION 483

TABLE III substantial time interval (6—32months) between Comparisons of mean CES-D scoresfor age and sex groups measurements argues against a simple confounding by subjective social support due to affective or mood states at the time of the interview. Second, subjective social support at baseline SubgroupMean CES-D follow-upMiddle-aged, scoreat is a strong predictor of follow-up CES-D scores, controllingfor baselineCES-Dscores.Consequently, subjective36.11socialimpaired any long-term effects of affective states at the time supportMiddle-aged, of the baseline measurements should be eliminated subjective16.69socialunimpaired or substantially reduced. Finally, the presence supportOld, of statistically significant and substantively large social16.40supportOld,impaired subjective interactions is hard to reconcile with contamination. That is, if confounding is taking place, it would be social17.03supportMale,unimpaired subjective necessary to determine why that confounding is especially likely among middle-aged rather than older social39.30supportMale,impaired subjective adults, and among men rather than women. Although the measure of subjective social support social13.02supportFemale,unimpaired subjective operated in line with theoretical expectations, results social26.08supportFemale,impaired subjective based on other dimensions of social support did not. In both the bivariate and multivariate analyses, a social19.54supportunimpaired subjective smaller social network at baseline was related to fewer depressive symptoms at follow-up, and being married was associated with higher levels of depressive symptoms at follow-up. There is no potent predictors of outcome for middle-aged patients and obvious explanation for these findings. However, for men. they do support the warnings of some investigators (e.g. Pattison et a!, 1979; Cohen & Wills, 1985; Alloway & Bebbington, 1987) that not all social Discussion relationships are supportive and beneficial. Conse The results of this study support the hypothesis that quently, the quality of interpersonal relationships social support affects the outcome of depressive is an important component of social support illness. The subjective social support measure assessment, and clearly, in terms of both the exhibited the strongest and most complex relation statistical significance and theoretical meaning of the ships with CES-D outcome scores. In the main findings, subjective support is the most potent effects model, subjective social support was the dimension. strongest predictor of CES-D outcome scores In an effort to understand why subjective support with the exception of baseline CES-D score. The plays this prominent role, additional analyses were interactive model suggested that the effects of performed, to see which of the nine items in the scale subjective social support on CES-D follow-up scores of subjective social support were significantly related were even more complex. to recovery status (P

Our results also suggest that factors other than on a single follow-up interview that varied in social support may affect the course and outcome time since baseline assessment, although our of depressive illness. In the bivariate analyses, results suggest that the length of this interval the presence of both melancholia and other was not related to follow-up CES-D scores. None psychiatric disorders at baseline were associated with theless, use of a single follow-up, as opposed to recovery status. Dysthymia was the only of these follow-ups obtained at standardised intervals, characteristics to remain statistically significant in precludes a dynamic view of illness course and the multivariate analyses. It is possible, however, that outcome. We now have instituted follow-up clinical characteristics contribute to the severity of examinations at defined intervals, and hope to the major depressive episode, and thus indirectly examine illness course in more complex ways in affect its course and outcome. This pattern of effects future analyses. might be missed in multivariate analyses (such as From both research and clinical perspectives, those presented in this paper) in which baseline levels better understanding of the course and outcome of of depressive symptoms are statistically controlled. depressive illness (and other psychiatric disorders) Life events have been implicated in the onset of is needed. Additional effort also is needed to identify depressive symptoms in numerous studies. In this the personal, social, and clinical factors that affect study, however, neither positive nor negative events the course, duration, and outcome of illness. These were significant predictors of outcome. It should be results suggest that social support may be one such noted, however, that the measures of life events used factor and that it is crucial to distinguish between may not have been the best ones possible for testing patients' involvements in interpersonal relationships the relationship between life events and recovery and the quality of the social support provided by from depression. Our measures of life events were those relationships. based on baseline data and inquired about events experienced during the year before admission. In Acknowledgements studies suggesting that life events are important for outcome (e.g. Murphy, 1983), measures of life events This work was supported by center grant P50 MH40159 from the were based on the interval between baseline and National Institute of Mental Health, Center for Studies of Mental follow-up assessments. Consequently, we are Health and Aging. unwilling to conclude that life events are irrelevant to the course and outcome of major depression. References Relative to previous studies, this research had several advantages. First, these results are based on ALLOwAY, R. & BEBBINGTON, P. (1987) The buffer theory of social support—¿a reviewof the literature.PsychologicalMedicine,17, a relatively large clinical sample, in which the 91—108. diagnosis of major depression at baseline was AMERICAN PSYCHiATRIC ASSOCIATION (1980) Diagnostic and carefully performed. Second, this was a prospective Statistical Manual of Mental Disorders (3rd edn) (DSM—III). study, in which the predictors of outcome were Washington, DC: APA. ascertained before and independent of follow-up BALDWIN, R. C. & Joiiay, D. 3. (1986) The prognosis of depression in old age. British Journal of Psychiatry, 149, status (a procedure that was advantageous for all of 574—583. the variables except life events, as just noted). BLAZER, D. 0., HUGHES, D. & GEORGE, L. K. (1987) Stressful Finally, the database permitted simultaneous life events and the onset of a generalized anxiety syndrome. American Journal of Psychiatry, 144, 1178- examination of six categories of factors potentially 1183. affecting course and outcome of illness. CARNEY,M. W. P., REYNOLDS,E. H., & SHEFFIELD,B. F. (1987) Although this study has a number of strengths, Prediction of outcomein depressiveillnessby the Newcastle it also has weaknesses. First, all patients enrolled in Diagnosis Scale: its relationship with the unipolar/bipolar and DSM—IIIsystems. British Journal of Psychiatry, 150, the study were in-patients at a single treatment 43—48. facility at baseline. In addition, rather narrow COHEN, S. & WILLS, T. A. (1985) Stress, social support, and inclusion criteria were used in defining the sample the buffering hypothesis. PsychologicalBulletin, 98, 310— and, consequently, generalisation of our fmdings is 357. F0LsTRIN,M. F., FOLSTEIN,S. E., & MCHUGH,P. R. (1975) “¿Mini necessarily restricted. Second, although careful MentalState―:a practicalmethodforgradingthecognitivestate diagnostic procedures were implemented during of patients for the clinician. Journal of Psychiatry Research, 12, baseline assessments, follow-up status was based 189—198. solely on a 20-item screening scale. Although the HENDERSON, A. S. (1984) Interpreting the evidence on social CES-D is a widely used instrument, with documented support. Social Psychiatry, 19, 49-52. —¿ & MORAN, P. A. P. (1983) Social relationships during the psychometric properties, it is clearly not comparable onset and remission of neurotic symptoms: a prospective with a diagnostic evaluation. Third, results are based community study. British Journal of Psychiatry, 143, 467-472. SOCIAL SUPPORT AND DEPRESSION 485

HOLAHAN, C. J. & Moos, R. H. (1981) Social support and stress buffering hypothesis. American Journal of Community psychological distress: a longitudinal analysis. Journal of Psychology(in press). Abnormal Psychology, 90, 365-370. MURPHY,E. (1983) The prognosis of depression in old age. British HUGHES, D. C., BLAZER, D. 0., & GEORGE, L. K. (l988a) Age Journal of Psychiatry, 142, 111-119. differences in life events: a multivariate controlled analysis. PATTISON, E. M., LLEAMAS, R., & HURD, 0. (1979) Social International Journal of Aging and Human Development, 27, network mediation of anxiety. Psychiatric Annals, 9, 56- 207—220. 67. —¿, GEORGE, L. K., & BLAZER, D. 0. (1988b) Age differences RADLOFF,L. S. (1977) The CES-D scale: a self-report depression in life event qualities: multivariate controlled analyses. Journal scale for research in the general population. Applied of Community Psychology, 16, 161-174. PsychologicalMeasurement,1, 385—401. KELLER, M. B. & SHAPIRO, R. W. (1981) Major depressive disorder: ROBINS,L. N., HItLZER,3. E., CROUGHAN,J., ci al(l98l) National initialresultsfroma one-yearprospectivenaturalisticfollow-up Institute of Mental Health Diagnostic Interview Schedule: its study. Journal of Nervous and Mental Disease, 169, 761-767. history, characteristics, and validity. Archives of General —¿, SHAPIRO, R. W., LAY0RI, P. W., et al (1982) Recovery in Psychiatry, 38, 381—389. major depressive disorder: analysis with the life table and ZIMMERMAN, M., CORYELL, W., & PFOHL, B. (1987) Prognostic regression models. Archives of General Psychiatry, 39,905-910. validity of the dexamethasone suppression test: results of a six LANDERMAN,R., GEORGE,L. K., CAMPBELL,R. T., ci al (1989) month prospective follow-up. American Journal of Psychiatry, Social support, stress, and depression: alternative models of the 144, 212—214.

*Linda K. George, PhD,Professor of Medical Sociology, Associate Director, Center for the Study of Aging and Human Development, Duke University Medical Center, Box 3003, Durham, North Carolina 27710, USA; Dan G. Blazer, MD, PhD, Professor of Psychiatry, Director, Affective Disorders Program, Duke University Medical Center; Dana C. Hughes, RN, PhD, ResearchAssociate, Division of Community and Social Psychiatry, Duke University Medical Center; Nancy Fowler, MEd,Associate in Research, Division of Community and Social Psychiatry, Duke University Medical Center

*Correspondence