Assessment of a Brief CES-D Measure for Depression in Homebound Medically Ill Older Adults

Abstract

Depressive symptoms are highly prevalent among medically ill homebound elderly and are associated with significant functional decline, lower quality of life, and increased health care utilization. Despite this, depression is generally under-diagnosed and undertreated among medically ill homebound older adults. The objective of this study was to determine the validity of a brief depression measure (CES-D) and examine the nature of depressive symptoms reported by the older sample. Using confirmatory factor and rating scale analysis, the factor structure of responses in a cross-sectional home care sample (n = 618) was examined with a further analysis of item responses from identified urban and rural subsamples. Radloff’s (1977) four factor depression model fit the data well. Some symptom items were expressed differently and this offers an opportunity to understand the unique clinical aspects of depression in homebound older adults.

Recent comprehensive reviews have emphasized that late life depression is one of the most common mental disorders to present in older adults though frequently under-diagnosed and rarely adequately treated (Frederick, Steinman, Prohaska et al., 2007; Gellis, 2009). Prevalence estimates of depressive disorders in community-dwelling older adults range from 5 – 10% in primary care (Tai- Seale, Bramson, Drukker, et al.,2005), 14% in home health care (Bruce et al, 2002) and 24% in long term care settings (Hyber, Carpenter, Bishmann, & Wu, 2005).

Elevated symptoms of depression have been correlated with negative outcomes in older persons including increased mortality related to suicide (Cowell & Thompson, 2008; Raue, Meyers, Rowe, Hao, & Bruce, 2007), medical illness (Gellis & Bruce, 2010; Lyness, 2008), and disability due to related cognitive disorders (Gellis, McClive, & Brown, 2009; Li & Conwell, 2009). Left untreated, depression leads to exacerbation of both physical and psychiatric symptoms, impairment in functioning, and increased health costs (Crystal, Sambamoorthi, Walkup, & Akincigil, 2003; Unutzer, Katon, Callahan, et al., 2003).

The measurement of depressive symptoms is a major concern to social work practitioners seeking brief and valid indicators of depression status in medically ill elderly. The Center for Epidemiologic Studies Depression Scale (CES-D: Radloff, 1977) 20-item index is one of the most common self-report depression instruments designed for use in large scale surveys and has become one of several standard measures of depressive symptomatology in older persons. Despite the importance of providing geriatric mental health care, little is known about depressive disorders in homebound elderly receiving home healthcare services, and is compounded by a lack of attention to standardized screening in this community service sector (Choi & McDougall, 2007; Gellis, 2010). The under-detection of depression in older adults creates many challenges for the social work practitioner and is exacerbated by several sets of factors. First, at the client level, older home healthcare recipients often have intricate medical needs that may obscure depression detection. Also, personal factors such as stigma, amotivation and pessimism, distrust, avolition, financial worries, isolation, and fear of losing independence may reduce an individual’s willingness to seek mental health care when depression is detected. Lack of financial resources and transportation may also impede access to mental health care. Service provider factors include time constraints and knowledge and skill deficits in mental health screening that may discourage providers from assessing depression (Brown, Kaiser, & Gellis, 2007). Finally, service providers may simply be uncomfortable in dealing with depressed patients, sharing similar preconceptions and biases about mental illness as is seen in society.

Likewise, home healthcare agencies located in smaller cities (less than 100,000 persons) provide services to large geographic areas that may include rural communities with limited, insufficient, or inaccessible mental health services in comparison to urban counties. Examination of urban and rural differences in the mental health status of older persons is important since mental health care resources are inadequate and limited published reports exist on the topic (Probst, Laditka, & Moore, 2008; Wang, 2004). Our investigation adds to the knowledge base by examining depressive symptoms in an older homebound sample from New York State who reside in rural and urban areas.

Mental health services researchers noted that a majority of homecare providers do not detect depression in their older clients (Bruce et al., 2002; Brown, Kaiser, & Gellis, 2007). Other researchers noted that home healthcare staff had difficulty in reliably identifying older adults with depression and assessing the severity of their symptoms, and generally did not use standardized depression screening instruments in their practice (Brown, McAvay, Raue, Moses, & Bruce 2003). Another study reported that home healthcare providers missed 43% of their depressed clients identified by the Geriatric Depression Scale (Flaherty, et al, 1998).

Investigating the usefulness of depression screening measures and how depressive symptoms affect medically ill homebound older adults represents a significant step toward improving evidence-based clinical care and outcomes in this population (Gellis & Reid, 2004). Therefore, it is important to consider brief depression screening tools that are not burdensome for older persons in order to improve the ease of administration. One instrument, the Center for Epidemiological Studies-Depression Scale (CES-D: Radloff, 1977) has been used in epidemiological studies such as the Established Populations for the Epidemiological Studies of the Elderly (EPESE; Cornoni-Huntley et al., 1986) and the Health and Retirement Study (Yang & Jones, 2008). Of great value is whether brief instruments such as the CES-D that have been shown to identify depressive symptoms in the general adult population, are measuring the same constructs when used with older persons in home care settings. Moreover, researchers have called for closer examination of items in brief versions of the CES-D (Grayson, Mackinnon, Jorm, Creasey, & Broe, 2000).

This study is the first, to our knowledge, to psychometrically examine the brief 11-item version CES-D in medically ill homebound older adults. The first goal of this research was to test the four factor model for the 11-item version CES-D in a cross-sectional sample. The second goal was to determine the adequacy of a single second-order Depression factor in fitting correlations among the four first-order item factors. We also compared urban and rural subsamples of medically ill homebound older adults to examine differences in item responses on the depression screening instrument.

Method

Subjects

Table 1 describes the sample demographics. The sample (n = 618) of older adults came from a range of urban and rural areas and were treated for various medical conditions by a large urban-based certified home healthcare agency affiliated with a local university. This study was approved by agency and university institutional review boards. Mean age for this older sample was 76.1 years (range 65–101 years, SD = 6.9); 89% of the patients were women; most were widowed (64%) and Caucasian (87%). It was generally an educated group with a mean of 10.9 years (range 9 – 20 years, SD =2.8). Two-thirds of the participants were living in an urban setting (defined as a Metropolitan Statistical Area of 100,000 or more persons). One-third of the sample received disability benefits and a majority lived alone in their own residence. More than half had reported poor health status.

Table 1.

Demographic and Clinical Characteristics of the Home Care sample (n = 618)

Characteristics	n	%
Age
65–84	517	83.66
85–101	101	16.34
Gender
Male	67	10.9
Female	551	89.1
Race
White	536	86.7
Non-White	82	13.3
Marital Status
Married	126	20.4
Widowed	394	63.8
Divorced or separated	98	15.8
Living Arrangement
Living alone	514	83.2
Living with someone	104	16.8
Setting
Urban	411	66.51
Rural	207	33.49
Health Status
Excellent	10	.02
Good	91	14.74
Fair	204	33.01
Poor	313	52.23
IADL Limitations
0–1	316	51.07
≥ 2	302	48.93
Health Condition*
Heart Disease	267	43.27
Stroke	203	32.91
Chronic Lung Disease	94	15.24
Osteoporosis	171	27.68
High Blood Pressure	286	46.30
Diabetes	173	28.05
Arthritis	316	51.23

Note:

^*Patients reported more than one medical condition. IADL = Instrumental Activities of Daily Living

Procedures and Measure

The short-form 11-item version CES-D was administered as part of the home care admission phase by the home care intake coordinator in the older patient’s home. The depression assessment was a component integrated into the ‘Outcome and Assessment Information Set’ (OASIS) home healthcare form mandated by the Centers for Medicare and Medicaid Services for all home health care agencies across the U.S.. This procedure was part of a larger quality improvement initiative for mental health care introduced into an urban-based home healthcare agency that served a large geographic region comprised of urban and rural counties in New York.

The brief revised version of the CES-D used in the Established Populations for Epidemiologic Studies for the Elderly (EPESE) study is preferable to the original 20-item version because pretests with a geriatric sample revealed that the full version was taxing for older people (Kohout, 1992) and the modified CES-D scale developers recommended a shorter but reliable and valid screen for older adults thus reducing the likelihood of false positives. In addition, data from the EPESE study demonstrated that the 11-item scale correlated highly (Pearson r = .95; n = 2339) with the full CES-D (Kohout et al., 1993). The revised 11-item CES-D contained a three response format (0 = hardly ever or never; 1 = some of the time; 2 = much or most of the time) for each item. Each subject endorsed the item if he or she experienced the symptom “much of the time during the past week.” The established cut-point of 16 or more on the full scale can correspond to anywhere from six symptoms scored at a 2 or 3 level to eight symptoms scored at a 2 level using the four-level response choice of the original scale. Therefore, this study used the six-symptom cut-point. Eleven items are included in the four subscales (Depressed Affect-3 items, Positive Affect-2 items, Somatic-4 items, Interpersonal-2 items). The sample completed the brief version CES-D as part of a group of self-report instruments in a study to identify and treat depressed medically ill older adults. Item response data from those 11 CES-D items were used in this study.

Analytic Procedures

Descriptive and Reliability Analyses

Several psychometric properties of the 11 CES-D scale items were examined. Cronbach’s alpha coefficients were calculated for each subscale. Intercorrelations between subscales were also obtained. For rating scale analysis we compared the urban and rural sub-samples of home care older adults. In this study, rating scale analysis using Rasch modeling is a practical method of understanding data from a set of individuals (older adults) responding to an assessment of items, and how well the responses fit a model, in this case depression items on a scale.

Confirmatory Factor Analysis Strategy

A Confirmatory Factor Analysis (CFA) model was estimated using AMOS 5.0 structural equation modeling software (Arbuckle, 1999) using maximum likelihood parameter estimates based on the sample covariance matrix to determine the extent to which the original Radloff (1977) four-factor model of depression (CES-D scale) fit the sample of home care older adults. The use of CFA permitted testing whether the depression items tapped the latent dimensions thought to underlie self-reported depression as measured by the CES-D short form. The analysis was based on a prior specification that the four depression factors characterized the data, and the model investigated corresponded to the expected structure. This allowed for empirically validating the factor structure of the 11 CES-D items when applied to the older adult home care sample.

A second-order factor analysis allows one to investigate the relationship between hierarchically nested factors (Rindskopf & Rose, 1988). The second-order models evaluated the ability of a single second-order Depression factor to account for the covariances among the four first-order factors. The fit of a second-order model is relative to the fit of the first order model on which it is based (Hertzog, 1989).

Goodness of fit indices were assessed using the Comparative Fit Index (CFI; Bentler, 1990); the Goodness of Fit index (Arbuckle, 2003), and the root mean square error of approximation (RMSEA; Brown & Cudeck, 1992). CFI and GFI values in excess of 0.90 are indicative of well-fitting models (Marsh, Balla, & McDonald, 1988), whereas RMSEA values less than 0.08 indicate a reasonable fit, with those below 0.04 indicating close fit of a model (Brown & Cudeck, 1992).

Rating Scale Analysis Strategy

The rating scale model was used to evaluate the construct validity of the subscales in greater detail and to evaluate specifically the extent and nature of differential item functioning across samples. Rating scale model is an extension of Rasch measurement models which are mathematical probability models that transform ordinal observations into measurements (Rasch, 1980). Items and persons are measured on a common interval scale. The rating scale specifies that the log odds of scoring in two adjacent categories is a function of two facets (person measure and item difficulty) and a response structure (step difficulty). The log odds is given by:

$$\text{Log}[P_{\mathit{nij}}/P_{ni(j-1)}]=B_n-D_i-F_j$$

where P_nij is the probability of person n responding in category j of item i, P_ni(j-1) is the probability of person n responding in category j-1 of item i, B_n is the latent trait measure (depression) of person n, D_i is the difficulty of the item i, and F_j is the step difficulty of the threshold between categories j-1 and j. In the present study, F₁ is the transition from category 1 (rare or never) to category 2 (some of the time) and F₂ is the transition from category 2 (some of the time) and category 3 (almost always). This linear relationship allows the establishment of an underlying dimension on a continuum, along which home care patients and items can be jointly placed, ordered, and compared.

Rating scale analysis was used to assess the unidimensionality of CES-D subscales and to establish whether items on the subscales measured the same underlying construct for the original Radloff (1977) sample, and to help define the CES-D subscales operationally. A mean-square outfit statistic with expectation 1, was used to identify misfitting items. Values substantially less than 1 indicated dependency in the data. Values greater than 1 indicated the presence of unexpected outliers (Wright & Linacre, 1995). The standardized differential item functioning (DIF) index or z statistic (Wright & Masters, 1982) was used to determine if item difficulty hierarchies were statistically different for the urban and rural home care sample groups. The WINSTEPS computer program (Wright & Linacre, 2001) was used for rating scale analyses. Data from the urban and rural home care samples were analyzed jointly and then separately. Separate item calibrations anchored on step measures obtained from joint calibrations were carried out for Depressed Affect, Positive Affect, Somatic, and Interpersonal subscales.

Results

All intercorrelations were significant at the p < 0.001 level. Alpha reliabilities for the four subscales were 0.86 for the Depressed Affect, Positive Affect (0.81), Somatic (0.79), and Interpersonal (0.81) respectively. Cronbach’s alpha for the 11 items in the home care elderly sample was 0.83.

Confirmatory Factor Analysis

A common factor model was fitted to the 11 CES-D items. The maximum likelihood method of factor estimation in AMOS 5.0 (Arbuckle, 1999) was used to test whether the items of the CES-D-11 reflected the latent constructs of Depressed Affect, Positive Affect, Somatization, and Interpersonal Relations as hypothesized. Prior to analysis, the data were evaluated for multivariate outliers by examining leverage indices for each individual and defining an outlier as a leverage score 5 times greater than the mean leverage. No outliers were detected. Multivariate normality was evaluated using Mardia’s test for multivariate kurtosis and the test yielded a statistically non-significant result. Univariate indices of skewness and kurtosis all suggested reasonably behaved data in terms of normality. This model provided an adequate fit to the original structure posited by Radloff (1977). The chi-square was not statistically significant for the four factor model χ²(545) = 301.62. Multiple fit statistics were used to examine the fit of the data to the CFA model. The Comparative Fit index was 0.97, the GFI was 0.95, and the RMSEA was 0.046. The indices uniformly point towards good model fit. Inspection of the residuals and modification indices revealed no theoretically meaningful and significant points of ill-fit in the model.

Table 2 presents the standardized factor loadings and the factor correlations for the elderly home care sample. All 11 items loaded significantly (p < 0.01) only on those factors that corresponded to their a priori subscales. The fit indexes and the fact that all hypothesized factor loadings were significant suggest that the model provides an adequate approximation to the data. All three Depressed Affect items loaded high on the Depressed Affect subscale. The item with the highest loading on the Positive Affect subscale was ‘happy’ while the item ‘enjoyed’ had the lowest factor loading. Three somatic items had slightly lower factor loadings (0.61–0.67) on the Somatic subscale. Items on the Interpersonal subscale were relatively high. The tabled results indicated that the items generally behaved as expected.

Table 2.

Factor Loadings of the Short Version CES-D 11-items for the Homecare Sample

Subscale and items	Factor
	I	II	III	IV
(I) Depressed Affect
Depressed	0.86	-	-	-
Lonely	0.84	-	-	-
Sad	0.81	-	-	-
(II) Positive Affect
Happy	-	0.80	-	-
Enjoyed	-	0.78	-	-
(III) Somatic
Appetite	-	-	0.78	-
Effort	-	-	0.67	-
Sleep	-	-	0.61	-
Get going	-	-	0.63	-
(IV) Interpersonal
Unfriendly	-	-	-	0.83
Disliked	-	-	-	0.80

Note: All 0 loadings and standardized factor variances were fixed by hypothesis. All nonzero parameter estimates were significantly different from 0 beyond the .001 level of confidence.

A second-order confirmatory analysis, as depicted in Figure 1, was performed to determine whether or not the four first-order dimensions could be modeled using a single-second order depression factor. The fit of the second-order model on the entire home care sample was very good, χ²(545) = 339.71, the CFI was 0.92, and the traditional GFI was 0.91. The loss of fit from the first-order model with unconstrained factor correlations was not significant at the 1% level of confidence. Figure 1 presents the standardized parameter estimates for the measurement model. The residuals for each of the observed items were generally low, suggesting that the measure represents reasonable indicators of the depression construct.

Figure 1.

Confirmatory factor analysis of four-factor model of depression.

Note. e = error term

Rating Scale Analysis

Based on rating scale analytic methods, the item difficulty in each scale operationalizes the construct definition and can be examined to confirm the stability of the construct definitions across different populations. Summaries of item difficulties with standard errors and outfit mean square by subscale (Depressed Affect, Positive Affect, Somatic, and Interpersonal) are presented in Table 3. The item difficulties within each subscale identify which items were harder or easier for participants to endorse. For example, the ‘lonely’ item on the Depressed Affect subscale was the least likely to be endorsed item, that is, the most difficult (item difficulty = −0.36 logits), while sad was the most likely to be endorsed item (least difficult) for the urban home care patient sample (item difficulty = −0.61 logits). On the Positive Affect subscale, urban home care patients were most likely to endorse ‘happy’ and less likely to endorse ‘enjoyed’ in comparison to the rural home care patient group. On the Somatic subscale, ‘getgoing’ was the least likely to be endorsed and ‘sleep’ was the most likely to be endorsed item for the urban home care patients.

Table 3.

Item difficulties (D), SE, and MNSQ for the 11-item Version CES-D subscales by group (Urban and Rural Homecare Samples)

Subscale/item	Urban (n=411)				Rural (n=207)		ZR
	D	SE	MNSQ	D	SE	MNSQ
Depressed Affect
Depressed	−0.59	0.08	0.93	−0.57	0.09	0.82	−1.47
Lonely	−0.36	0.09	1.10	−0.10	0.10	1.04	2.06a
Sad	−0.61	0.09	1.18	−0.98	0.10	0.93	2.39a
Positive Affect
Happy	0.57	0.08	1.02	0.53	0.10	0.95	0.98
Enjoyed	0.47	0.07	0.87	0.41	0.11	0.89	0.86
Somatic
Appetite	−0.73	0.07	0.73	−0.68	0.06	0.84	−1.45
Effort	−0.51	0.07	0.69	−0.92	0.06	0.69	−3.96a
Sleep	−0.86	0.07	1.48	−0.74	0.07	0.91	−0.84
Get going	−0.42	0.07	0.64	0.18	0.06	0.55	−2.78a
Interpersonal
Unfriendly	−0.77	0.08	1.05	−0.66	0.10	0.66	−1.02
Disliked	−0.35	0.07	1.12	0.12	0.09	1.14	3.04a

^aP < 0.05.

Note: Low item difficulty measures within each subscale indicate the items were more likely to be endorsed.

Potential misfitting items were examined by subscale according to the rule that items with an unweighted item fit mean square (MNSQ) value higher than 1.3 or lower than 0.7 were misfitting items (Wright & Linacre, 1995). None of the items in the Depressed Affect, Positive Affect, or the Interpersonal subscales were identified as misfitting among the home care patients. There are two clusters in the Somatic subscale in our sample; items have MNSQ higher than 1.3 and lower than 0.7, and in between. One item in the Interpersonal subscale had MNSQ values less than 0.7.

Since items within each scale were modeled to have the same relative difficulty (invariant item difficulty parameter) for both the urban and rural home care subsamples, items that exhibited differential item functioning (DIF) could be examined further. As can be seen in Table 3, the hierarchies of item difficulties were similar for the two homecare groups (urban and rural) on the Depressed Affect, Interpersonal, and Somatic subscales with some exceptions. Two items (“sad” and “lonely”) in the Depressed Affect subscale exhibited DIF, indicating that their items operate significantly differently between groups. Urban elderly patients endorsed “sad” more likely and “lonely” less likely when compared with members of the rural elderly home care group. One DIF item was identified in the Interpersonal subscale. The urban sample group endorsed “dislike” relatively fewer times than did the rural group. On the Somatic subscale, “unable to get going” was less likely to be endorsed in the urban home care group, while “effort” was easier for the urban group to endorse than the rural group of homebound older adults.

Discussion

Our aim in this study was to examine the short version CES-D scale in a home healthcare older population. Screening for depression is an important first step in overall depression management. Screening for the detection of depressive disorders involves the use of easily administered inexpensive measurement procedures to identify older adults who may be experiencing psychological distress.

In home healthcare, the challenge of screening is to understand the experience of depression separate from the medical condition in older adults. While it is important not to miss detecting depression, it is likewise critical to not overdiagnose depression. Indeed, the results of the present study suggest that the brief 11-item version CES-D possesses reliable and valid psychometric properties when used with elderly home care adults. The internal consistency for the 11-item version CES-D was acceptable and consistent with previous findings (Kohout, 1993). The findings replicate the original four factor solution for the CES-D 11-items on the basis of confirmatory factor analyses. The results also demonstrate that the four item factors are highly intercorrelated. Furthermore, a model with a single second-order Depression factor fit the first-order item factor covariances relatively well in this sample. Items appear to fit scale assignments as expected corroborating prior findings (Radloff, 1977). Results using rating scale analysis indicate acceptable fit to the model. Evaluation of the structure underlying response to Depressed Affect, Positive Affect, Somatic, and Interpersonal subscales demonstrated reasonable construct validity for each of the subscales, suggesting that the 11-item version CES-D is a useful and conceptually valid instrument for depressive symptoms in elderly home care adults.

The results of this study suggest that DIF differences are potentially important clinical markers to understand the differing constructs underlying items when scales are used in urban and rural home care populations. For example, the DIF for the ‘effort’ and ‘get going’ items suggest that urban and rural older adults are responding to the items differently in light of their health rather their psychological status. The Depressed Affect items ‘sad’ was also reported relatively more frequently in the urban sample, yet ‘lonely’ less reported in the urban group in comparison to the rural group. This may be an indicator of the availability of resources in an urban setting, and perhaps, a potential psychological response to functional decline. Examination of the urban/rural sample differences in item response suggests that the urban group compresses the scale with little differences at the individual level while the rural group expanded the scope of the items.

The results of the two types of analyses done were generally congruent, extending the knowledge base on the theoretical features of the depression construct and offering support to the consistency of measurement in a home healthcare elderly sample. It also supports the value of using complimentary statistical approaches of structural equation modeling and rating scale analysis in the same confirmatory study. CFA analysis suggests that the underlying factorial structure of homecare older adult’s responses is similar to those of the general adult population. The rating scale analysis suggests that although most items have similar difficulty relative to the underlying construct being measured, some function uniquely in home care older adults, particularly in the Depressed Affect and Somatic subscales. The rating scale analysis data helps to improve our understanding of the nature of depression in home healthcare settings and the extent to which it differs from clinical reports of older hospitalized and primary care patients.

The study supports the utility of the 11-item version CES-D for a home care elderly sample. Use of the CES-D in the context of routine mental health screening has the potential to positively identify those older persons experiencing depressive symptoms and may contribute to immediate individualized treatment planning. The 3-item response format and the shortened length of the CES-D do not appear to compromise the psychometric properties of the instrument and patients reported ease of use. The measure shows adequate internal consistency and a factor structure similar to that reported in earlier studies of the original CES-D scale. Despite the reduced set of items, the Cronbach alpha of 0.83 is comparable to internal consistency estimates for the original scale (Radloff, 1977) which ranged from 0.84 – 0.85 for a general adult population sample.

Several methodological issues have to be considered. The revised 11-item CES-D is based on a measure with a strong track record, but it has not been validated to a great extent in home care elderly populations. In addition, this study did not use a DSM-based measure of depression due to the nature of the home care environment and time constraints on the home care provider. Our next phase of research is to continue screening for depressive symptoms in older home care patients and to develop a protocol for using a relatively easy to administer DSM-based measure such as the patient health questionnaire.

In conclusion, the goal of screening for late life depression is early identification and management through early intervention. This is critical since depression is a treatable mental health disorder with a variety of effective pharmacological and psychosocial treatments available to the gero-social work clinician.

We recommend a set of criteria for gero-social work managers and clinicians to consider and justify implementation of evidence-based depression screening, and they are the following:

1. Is the incidence of depression high enough to justify the cost of screening in an agency?

2. Does the problem have a significant effect on the quality of life of the older adult?

3. Are depression screening instruments available that are valid and cost-effective?

4. Are effective treatments available for late life depression when individuals are identified?

5. Is depression screening acceptable to social workers and to older adult clients?

This is the first attempt at validating the short form CES-D depression measure for a medically ill frail homebound elderly population finding some differences in item responses between urban and rural older adults. The psychometric results for the 11-item version CES-D look promising, and are comparable to the original scale. Further validation is needed to offer a briefer version of a long measure that has proven to be taxing and difficult for older persons. Such investigation would assist home care clinicians in timely evaluation of their elderly client’s mental health status. Our future studies will consider differences in age, gender, and medical conditions.