| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||
Correspondence: Address correspondence to Kathryn Betts Adams, Mandel School of Applied Social Sciences, Case Western Reserve University, 10900 Euclid Avenue, Cleveland, OH 44106. E-mail: kathryn.adams{at}case.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Key Words: Assessment tools • Psychometrics • Mental health
The GDS has been used to screen for depressive symptoms in a wide range of research studies in a number of health and service settings (Peach, Koob, & Kraus, 2001). It is an indictor of its breadth of application that the public access Web site maintained by the scale's authors currently offers links to translations of the scale in 24 languages (Yesavage, 2003), and a recent review of the reliability of the GDS cites 338 published articles using the scale (Keiffer & Reese, 2002).
A number of studies have examined the properties and demonstrated the utility of the GDS with older adults of diverse cultures or nationalities, such as Japanese American, Chinese, and Turkish (e.g., Boey, 2000; Ertan, & Eker, 2000; Iwamasa, Hilliard, & Kost, 1998), although recent review has cautioned that cultural factors may affect the reliability or validity of the GDS (Mui, Burnette, & Chen, 2001; Mui, Kang, Chen, & Domanski, 2003). Some investigators have found that the GDS does not adequately assess depression in persons who have mild to moderate dementia (Montorio & Izal, 1996). Despite some of these concerns, the GDS has been recommended for standard use in research studies with older adults to facilitate cross-comparisons and meta-analyses (Koder, Brodaty, & Anstey, 1996), for standard screening by geriatric social workers (Peach et al., 2001) and for screening in geriatric settings by the Geriatric Review Syllabus (Beck, 1991), and by the British Geriatrics Society and the Royal College of General Practitioners (Anderson, 2001).
One of the major strengths of the GDS is its simplicity of administration and scoring. It uses a simple "yes–no" response format; therefore, each question counts as 1 point when scored in the depressed direction, for a possible total score of 0 to 30 points. The GDS has well-established internal consistency, with high Cronbach's alpha reliabilities reported, such as.94 (Yesavage et al., 1983),.91 (Parmelee, Lawton, & Katz, 1989), and.87 (Adams, 2001). The scale was designed to yield a single score without subscales. Scoring cutoff points have been suggested in the literature, beginning with the scale's authors, who determined that a score of 11 or above for designating possible depression obtained reasonable specificity and sensitivity with a relatively small sample of older adults (Brink et al., 1982; Yesavage et al., 1983). Subsequent studies have suggested other cutoff points; one study found that 14 points was the optimal cutoff when results on the GDS are compared with those of a clinical psychiatric evaluation (Burke, Nitchter, Roccaforte, & Wengel, 1992). A shorter form of the GDS, the 15-item GDS-S, (Sheikh & Yesavage, 1986) is also widely used by clinicians and researchers and has been found to offer an acceptable substitute for the long form (Lesher & Berryhill, 1994).
Despite its prominence and widespread use, few principal components analyses (PCAs) and, to our knowledge, no confirmatory factor analysis (CFA) of the GDS are published in the literature. Four PCAs of the English-language GDS and one of the GDS-S are found in research journals. A summary of the assignment of the scale's items to factors in these studies is shown in Table 1. In the earliest of these published studies, Parmalee and colleagues (1989), using data from a group of 417 institutionalized elders, proposed a factor structure including six factors. They identified the first 14-item factor as Dysphoria, and the subsequent factors as Worry (4 items), Withdrawal–Apathy (4 items), Vigor (3 items), Decreased Concentration (2 items), and Anxiety (3 items). Sheikh, Yesavage, Brooks, Friedman, Gratzinger, Hill, Sadeik and Crook (1991) subsequently published a proposed factor structure based on a sample of 326 community-dwelling adults over the age of 65. These authors obtained five reliable factors, from which four of the scale's items were omitted. Although they did not label their factors, their discussion describes the first 9-item factor as "sad mood and pessimistic outlook," comparable with the Dysphoria dimension of Parmalee and colleagues (1989); the second as characterized by a "lack of mental and physical energy," comparable with the Vigor factor, with some overlap with a cognitive impairment–reduced concentration concept (6 items); the third as relating to "positive mood and optimism" (6 items); the fourth as "agitation or restlessness" (3 items); and finally, a "social withdrawal" factor of 2 items (Sheikh et al., 1991, p. 27).
|
A factor analysis of the 15-item GDS-S (Mitchell, Mathews, & Yesavage, 1993) obtained three dimensions: General Depressive Affect, again resembling Dysphoria (7 items); Life Satisfaction, consisting of the positively worded questions (4 items); and Withdrawal (3 items). Subjects were 868 noninstitutionalized older persons in North Carolina. That study's authors noted that the three factors had low-moderate to moderate intercorrelations with one another, and that the effects of other demographic and social variables on the three were quite different, leading them to conclude that depression in older adults is multidimensional.
Most recently, Adams (2001), working with data from a sample of 327 community-dwelling adults over the age of 65, obtained a six-factor solution of the GDS. This solution most resembled the results of Parmalee and colleagues (1989) already discussed, albeit with a split in the Dysphoria items into two separate factors, one that included those items that involve negative affect and the lack of positive affect (9 items) and the other that included items about hopelessness, worthlessness, and helplessness (4 items). This exploratory factor analysis also obtained a combined Withdrawal–Apathy and (Lack of) Vigor factor (6 items), an Anxiety factor (4 items), a Mental Impairment factor (4 items), and an Agitation factor (3 items). In accord with the study by Mitchell and associates (1993), Adams also found differences in endorsement rates and in the zero-order correlations of the factors with demographic and health variables that suggest the GDS could be tapping several distinct clusters of symptoms. As a case in point, Adams has provisionally proposed that the withdrawal–apathy and lack of vigor dimension may be indicative of a depletion syndrome related to advanced age and functional impairment (2001).
Both commonalities and differences are apparent among the five extant factor analyses of the GDS or the GDS-S. Each of these solutions includes at least one dimension that represents depressive affect or dysphoria, and one that represents social withdrawal. In addition, several of the analyses obtained a dimension involving cognitive impairment or lack of concentration, and an anxiety, worry, or agitation dimension. The analysis by Parmalee and colleagues (1989) grouped positively stated items into a Vigor factor, and two other studies reported that positive items clustered into a Life Satisfaction and a Positive Mood–Optimism factor (Mitchell et al., 1993; Sheikh et al., 1991). The two remaining studies obtained results with mixed positive and negative items within factors. Regardless of the obvious similarities in the results from these studies, however, without the next step of a CFA on the GDS to provide evidence that the factor structure remains tenable in a new independent sample, investigators are left without a way of identifying reliable subscales that may represent distinct dimensions or subtypes of depression in older adults.
The current study describes a CFA of the GDS. The authors compared the PCA structure of the GDS obtained from a community-dwelling sample in an earlier study (Adams, 2001) with GDS data from a second sample of older adults in independent-living communities (Adams, Sanders, & Auth, in press) to determine how well the original factor structure fits the new data. We chose to use the first set of data as the criterion data set for several reasons. First, we have complete access to the data and the earlier analyses that were done; second, a sample of essentially "well" noninstitutionalized elders was appropriate for examining the measurement properties of the scale with a minimum of confounding variables; and lastly, as noted herein, the first sample's PCA obtained was quite similar to that published earlier by Parmalee and associates (1989), thus lending credence to our results. We discuss potential implications of the resulting CFA for use and interpretation of the GDS.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
The comparison sample consisted of residents of two Methodist-affiliated retirement facilities in Pennsylvania. The facilities were age-segregated apartment communities with optional activities, access to health care, and an emergency response system provided for residents. We gathered these data by use of a one-time questionnaire placed into the mailboxes of residents. Our goals in this second study were to examine the prevalence of loneliness and depressive symptoms and relationships among these two constructs, social network, and participation in social and leisure activities for residents in independent living (Adams et al., in press). For the current study, 60 additional cases from assisted living were also used. The combined N = 294. All of these respondents were over the age of 60, about two thirds were female, and nearly all were Caucasian and Protestant. Mean GDS score for this comparison sample was 7.47, with 26.2% of the sample scoring at or above the depression cutoff.
Data Analysis
The PCA for the original sample's data explained 50.4% of the variance in the GDS items. Although nine eigenvalues over 1 were initially obtained, Cattell's scree plot suggested a final solution of six components. Varimax rotation converged in nine iterations and the final eigenvalues were as follows: 3.524, 2.818, 2.387, 2.344, 2.312, and 1.737 (Adams, 2001). We estimated the proposed six-factor measurement model with data from the second sample by using a CFA method with the LISREL 8 statistical program. Following the suggestion of Kline (1998), we performed a tetrachoric transformation of the covariance matrix prior to estimation of the measurement model to provide an asymptotic covariance matrix to account for the nonnormal data distribution that was due to the dichotomous nature of the item response set. It is customary practice to present the chi-square significance test along with other measures of overall model fit that represent total variance accounted for in the sample matrix by the estimated model, the proportion of variance accounted for after adjusting for degrees of freedom, and the extent of error represented between the sample and estimated matrices (Bollen, 1989; Kline, 1998). Therefore, we selected the chi-square test, goodness-of-fit index (GFI), adjusted GFI (AGFI), and standardized root mean square residual as fit indices. Our process of analysis started with the full six factors and all of the items, and then we used a nested models approach to test alternative nested structures with reduced parameters to test fit improvement. We used the chi-square difference statistic to test the significance of the change in the chi-square test for each alternative model over the full model.
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
=.64), Item 3, "feeling life is empty" ( =.61), and Item 9, "happy most of the time" ( =.54), showed the highest validity coefficients for this factor. The standardized lambda coefficients for the Withdrawal–Apathy–Lack of Vigor factor (six items), labeled WAV, ranged from.45 to.55; Item 21, "full of energy" ( =.55), Item 28, "avoid social gatherings" ( =.51), and Item 20, "hard to start new projects" ( =.51) showed the highest validity coefficients for this factor. The ANX or Worry factor, composed of four items, showed lambda estimates that ranged from.48 to.54; Item 13, "worry about the future" ( =.54), and Item 6, "bothered by thoughts" ( =.52), were the best-fitting items on this construct. The MENT or Cognitive Impairment factor (four items) showed lambda estimates that ranged from.30 to.55. Item 26, "trouble concentrating" ( =.55), and Item 30, "mind as clear" ( =.48), were the strongest items, and Item 14, "problems with memory" ( =.30), showed poor construct fit and was the lowest validity coefficient for all GDS items. DYS 2 or Hopelessness (four items) showed lambda estimates that ranged from.46 to.60; Item 22, "feel situation is hopeless" ( =.60), and Item 10, "often feel helpless" ( =.57), showed the best construct fit. The sixth factor that we identified in the PCA, AGIT (Agitation; three items) showed consistently poor-fitting items when we analyzed it with confirmatory methods. The lambda coefficients for all three items were below.40. These items, "restless and fidgety," "frequently get upset," and "enjoy getting up in the morning" do not appear to be empirically related to this Agitation construct.
|
|
2 = 462.82, df = 399 (p =.015), standardized root mean square residual =.056, GFI =.99, and AGFI =.86. This did not appear to be too different from the first alternative. We tested a third alternative that was identical to this but with AGIT removed. This resulted in values of 2 = 369.72, df = 321 (p =.031), standardized root mean square =.055, GFI =.89, and AGFI =.87. Neither of the chi-square differences for these two additional alternatives showed a statistically significant improvement over the first alternative, and, conceptually, some meaning was lost by combining the mood items; thus we selected the first alternative model for use.
Interitem Reliabilities and Zero-Order Correlations of the Five Factors
Cronbach's alpha reliability coefficients for the five remaining factors in the alternate model are shown in Table 4. The Dysphoric Mood, Withdrawal–Apathy–Vigor, and Hopelessness factors each have alpha reliabilities that suggest they could be considered as standalone variables. Worry and Cognitive factors have lower alpha coefficients, in part due to lower item counts. The zero-order correlation matrix for the five factors is shown in Table 5. Each of the correlations is significant at the.01 level. As may be expected, the factors labeled Dysphoric Mood and Hopelessness have the highest correlation at r =.722. The Cognitive factor has the lowest absolute correlations except to Withdrawal–Apathy–Vigor, with which it is moderately correlated. Worry is quite highly correlated with both of the Dysphoric Mood factors, but less with Cognitive or Withdrawal–Apathy–Vigor.
|
|
|
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
The DYS 1 or Dysphoric Mood factor consists of 9 items that describe depressed mood, sadness, or emptiness as well as a lack of satisfaction with life and a lack of happiness. There is a mix of positively and negatively stated items. Each of the other proposed factor structures of the GDS in the literature have an initial factor that contains many of these same items and appears to represent the core of depressed mood within the scale. Comparing these results with the 15 items that comprise the GDS–Short Form (Sheikh & Yesavage, 1986), we note that 7 of the items that were selected for the GDS-S fall into this Dysphoric Mood factor.
The DYS 2 factor consists of four items that deal with hopelessness, helplessness, and worthlessness, and we have labeled it Hopelessness. There is a desperate quality to each of these items that sets them apart. It is noteworthy that the mean endorsement or score for this DYS 2 factor is far lower than that for DYS 1. Studies of suicidal persons have found that hopelessness is predictive of suicidal ideation (Brown, Beck, Steer, & Grisham, 2000; Szanto et al., 2002). Certainly, feelings of worthlessness and helplessness also may be associated with wishes to die, or at least low motivation to continue living. Thus, this Hopelessness factor may be of particular importance in screening older adults for suicidal ideation or intent. Three of the four items related to hopelessness also appear in the GDS-S.
The six items in the WAV factor were originally found, along with one additional item, in two separate factors identified in the PCA by Parmalee and colleagues (1989). Four of the Withdrawal–Apathy–Vigor items also loaded on a withdrawal factor in the PCA by Salamero and Marcos (1992), and three of them are in the withdrawal factor found in the PCA of the GDS-S cited earlier (Mitchell et al., 1996). This WAV factor has the highest mean score for the respondents and the lowest proportion of nonendorsement of any of the factors. Higher endorsement simply means these items are more common among the sample, both those respondents with and without scores in the depressed range for the total scale. A similar result was obtained in the study with the original sample (Adams, 2001), where WAV was also found to have a stronger association than the other GDS factors with age and health problems. Thus, Adams (2001) has suggested that WAV represents behaviors and experiences that are influenced by age and physical frailties and may be akin to a "depletion" syndrome.
The ANX factor, like the Hopelessness (DYS 2) factor, is not highly endorsed within this sample. Its mean score of.7169 is the second lowest of the five factors. The four items each describe some kind of obsessive ruminating or worry, whether about the future or the past, so that Worry appears to be an apt name for the factor. These items concerning worry capture symptoms related to thinking and emotion that are commonly found in depressed persons of all ages. In contrast, the three agitation items, which have not been retained in the alternative model, appear to capture symptoms that are more clearly physical or behavioral. The agitation items may have a number of underlying causes and did not hold together well as a dimension in this sample of older adults.
The Cognitive Impairment factor performed better after we removed one item from it in the alternative model. This item—"Do you feel you have more problems with memory than most?"—did not load highly on the mental impairment construct. Interestingly, the same item did not load on any factor in the PCA proposed by Mitchell and associates (1993) for the GDS-S or the PCA by Sheikh and associates (1991) for the GDS long form. There may be two problems with that item. First, older adults may perceive that they have memory problems for a number of reasons, and thus the item may not be predictive of depression, although this could be said for many of the items on the GDS. The bigger problem may be the ambiguity inherent in deciding whether or not one has more memory problems "than most." Does this mean most others of similar age, most others who live in the same facility, or most other persons of any age? It is likely that interpretation of this item will differ according to the residential setting and the available reference group of the respondent, which would suggest that older persons in an institutional setting who see a number of people suffering with dementia may tend to answer in the negative more often than older persons residing in the community. After removal of that problematic item, the remaining cognitive items address confusion, difficulties with decision making, and reduced concentration. That this Cognitive subscale represents a distinct set of symptoms is illustrated by its stronger association with the Withdrawal–Apathy–Vigor factor than with the Dysphoric Mood, Hopelessness, or Worry factors, all of which represent the core of depressive symptoms. Although these cognitive symptoms may be hallmarks of depression, they may also be caused by other conditions, such as early dementia, strokes, certain medications, or normal changes of later old age. A higher score on the Cognitive subscale could be an indication to pursue more extensive mental status testing.
There are certain limitations inherent in this study. First and foremost is that the CFA presented here was based on two data sets obtained from convenience samples of older adults rather than randomly selected samples of a targeted underlying population. The samples also were not representative of diverse racial or cultural groups. Because these analyses were conducted with homogeneous, nonprobability samples of relatively modest size, and because this is the first CFA of the scale in the literature, we would recommend interpreting these findings with some caution and consider the subscales of the GDS to be preliminary in nature, pending further validation by investigators in other settings. It is possible that these results will not cross-validate in future studies, so we do not recommend their use for clinical decision making as yet. However, the model presented here has obtained adequate GFI indices, appears to have good face validity, lends itself to easy interpretation, and does not stray too far from the previously published exploratory factor structures, as was shown in Table 1. In light of these promising aspects of our results, we see potential utility in beginning to use the three subscale scores that performed best here, that is, Dysphoric Mood, Withdrawal–Apathy–Vigor, and Hopelessness, in research. These may serve as an adjunct to the familiar well-normed total scale score, to provide more detail about symptoms experienced. A second-order CFA, based on representative samples of older adults, will be necessary to establish the appropriateness of subscale scores over the single scale score.
Clinical and Research Implications
This article has described what is, to our knowledge, the first CFA following an exploratory PCA of the Geriatric Depression Scale. The resulting five subscales of the GDS each represents a distinct subdimension of the syndrome we call depression in older adults. As Blazer (2003) has noted in a recent review, there are a number of subtypes of geriatric depression, such as major depressive disorder, minor but clinically significant depression, dysthymia, vascular depression, depression associated with dementia, and melancholia. Each of these presents a different symptom configuration, and each may have a different scoring pattern on depression screening scales. The GDS is already a very popular and well-established screening tool for depression in older adults, but there may be benefits to adopt subscale scoring. This study represents a first step in bringing added precision of measurement to the scale.
More research will be needed to see whether these subscales appear to fit the response patterns of a broad range of clinical and community samples. Investigators using the GDS with groups of respondents may wish to take a stepwise approach to scoring it. As noted earlier, we suggest first obtaining the total score as usual, and then conducting an exploratory factor analysis with the group's GDS data in order to identify areas of similarity and difference from the subscales presented in the measurement model here. If there are few serious differences, then the subscale scores suggested here may be used with greater confidence. It would also be appropriate, as with any measure, to obtain Cronbach's alpha reliability coefficients for the subscale scores. Those scores with alphas that meet the investigator's criteria for reliability may be used as additional variables or assessment tools for the respondents in question.
Although we remain mindful of the preliminary nature of the data, there are some additional clinical or research implications presented by the specific subscales identified here. The Dysphoric Mood and Hopelessness subscales appear to include the most serious and worrisome depressive symptoms in the scale. Clinicians wishing to evaluate the severity of depression in a given individual or among a group of individuals could use the scores of those two subscales as a type of affective severity index. The Withdrawal–Apathy–Vigor subscale has received high endorsement from respondents regardless of their depression level, as indicated by the total GDS score, and thus appears to have less discriminant validity for depression, but this may be an indicator of health- or age-related frailty, depletion, or malaise. All three of these subscales produced Cronbach's alphas in the range of.65–.80 in the study sample, which suggests they may hold up as consistent, reliable variables with future samples. In contrast, the Worry and Cognitive Impairment subscales, each with a small number of items, have obtained lower reliability coefficients here. We have also noted that these two subscales may be tapping symptoms that frequently accompany depression but may also be indicative of a number of other conditions. Depending on their study aims, investigators will now have the opportunity to use only the most reliable subscales or the most clearly depression-related subscales from the total GDS.
The findings of this study also may have implications for use of the GDS–Short Form. The 15 items of that version of the scale include 7 from this study's Dysphoric Mood subscale, 3 from the Hopelessness subscale, 3 from Withdrawal–Apathy–Vigor, 1 from Worry, and 1 from Cognitive Impairment. The distribution of the GDS–Short Form items across these new subscales is quite congruent with the structure we have obtained for the full scale, and the importance we would tend to assign to individual items. One exception is the cognitive item assigned to the GDS–Short Form (Do you have more problems with memory than most?), which unfortunately is one that was omitted from our final model because of very low factor loading. The diagnostic or research utility of the existing GDS–Short Form versus use of two or three of these suggested subscales from the full GDS is a question that future research may examine empirically.
This study used data from two convenience samples of mostly Caucasian older adults to obtain a CFA of the widely used Geriatric Depression Scale. Using data from older adults who live in the community or in independent-living facilities, we have found that three subscales of the GDS representing distinct symptom clusters have performed adequately and have proved consistent with previous factor analytic studies. The two remaining subscales evidenced poor internal consistency for this sample. A second-order CFA and further cross-validation of the subscales presented here with representative samples of older adults is now needed.
|
Footnotes |
|---|
Dr. Matto is now with the School of Social Welfare, University at Albany, State University of New York.
1 Mandel School of Applied Social Sciences and University Memory and Aging Center, Case Western Reserve University, Cleveland, OH.
2 School of Social Work, Virginia Commonwealth University, Alexandria.
3 School of Social Work, University of Iowa, Iowa City.
Decision Editor: Linda S. Noelker, PhD
Received for publication November 20, 2003. Accepted for publication March 18, 2004.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
yesavage/GDS.html.
| ||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
