Tang et. al.

< - - Back to Measurement matrix         MEGs Search

Quick Links:   Sample   Response Rate   Internal Consistency    DIF Method    DIF Evidence    DIF Impact    Expert Opinion

Tang et. al.   ( see abstract )

Geriatric Depression Scale (GDS)

Name/ Reference

Tang WK, Wong E, Chiu HFK, Lum CM, Ungvari, GS. The Geriatric Depression Scale should be shortened: results of Rasch analysis. International Journal of Geriatric Psychiatry. (2005) 20:783–789.

Source contact info

Correspondence to: W. K. Tang, Department of Psychiatry, Shatin Hospital, Shatin, NT, Hong Kong SAR, China. Tel: þ852-2636-7760. Fax: þ852-2648-3394.
E-mail: tangwk@cuhk.edu.hk

Availability (private or public)

public

Conceptual framework

Depression symptoms are causally emergent from a posited underlying trait or state of depression. The Rasch model is based on the premise that item response is a function of a person disability and item characteristics. The aim of this study was to examine the unidimensionality, item fit, and DIF of the GDS with respect to age, education and Cognitive Impairment (CI) in a community sample of Hong Kong Chinese patients with pneumoconiosis.

Purpose of measure & application (clinical, research, survey, screening)

Clinical: GDS has been used to identify depressive symptoms and depression in different types of populations.

Sample characteristics

A community sample of 300 Hong Kong Chinese patients with pneumoconiosis.

Age ranged from 42-94 years; <60 (33%), 60-74 (45.3%), >75(20.7).

DSM-IV diagnosis of depressive disorders: 37 (12.3%) pot pf 300 participants; minor dep (6%), major dep (3.67%), dysthymia (2.67%).

Educ level: mean (sd) 3.1(2.9); no (27.9%), primary (62.3%), sec (8.8%);

The MMSE score ranged from 8-30, average=25.6+4.5. Thirty-six (12.0%) patients had CI; the mean MMSE of those with CI= 16.6+3.6. Fourteen patients had a MMSE score less than 15.

Recruitment methods

421 participants were randomly selected by computer generated random numbers from 1442 patients representing 75% of all cases registered in the Pneumoconiosis Compensation Fund Board of Hong Kong.

Recruitment of subjects took place from September to November 2003.

Data collection method

A trained research assistant administered the GDS to all participants. A psychiatrist, who was blind to the GDS scores conducted a structured clinical interview to diagnose depressive disorders according to the locally validated Chinese version of the Structured Clinical Interview for Diagnostic and Statistical Manual for Mental Disorders, Version IV (DSM-IV) criteria.

Response rate

Of the 421 patients selected, 300 (71.3%) participated and 121 (28.7%) patients were excluded from the study.

No significant difference found between the participants & non participants in terms of male sex (99.0% vs 98.3%, p=0.628) and age (65.1+10.7 vs 64.1+11.1, t=0.896, p=0.371) distribution.

Exclusion criteria: absence from the annual interview (n=81, 66.9%), refusal to participate in the study (n=36, 29.8%) and deafness (n=4, 3.3%).

Format & design (readability, # of items, time to complete, response categories)

The GDS consists of 15 items measuring depression symptoms. Questions are answered ‘yes’ or ‘no’ which are selected to measure depressive symptoms. Based on their GDS scores, patients are classified as depressed or non-depressed, using a cut score of 8 or more.

Type of measurement (nominal, ordinal, interval, ratio)

ordinal

Scoring (range, direction, rules, missing data)

The theoretical range is from 0-15. Missing data treatment was not discussed.

Availability of translations & source

Not provided

Psychometric Properties:

Scale construction

Not provided

Basic summary statistics

GDS for sample mean (sd)= 7.27(3.81)

Variability

Not provided

Test-retest reliability

See ref #4 below

Interrater reliability

Not provided

Internal consistency

See ref #1 (below)

Content validity

Not provided

Construct validity

See ref #1, 2, & 9 (below)

Concurrent validity

The locally validated Chinese version of the Structured Clinical Interview for DSM-IV served as a benchmark to judge the performance of the GDS as a screening instrument.

Predictive validity

For Specificity see ref #12 below.

The optimal cutoff, sensitivity, specificity, and positive and negative predictive values, and the area under the receiver operating characteristic (ROC) curve of the original and revised versions of the GDS, were measured against the clinical diagnoses.

Sensitivity to change

Not provided

Differential Item Functioning (DIF):

Variable studied (e.g., groups)

Age, education and cognitive impairment were studied.

Sample size

Age: <60 n=99 (33%), 60-74 n=136 (45.3%), >75 n=62 (20.7%).

Educ. level: no n=84 (27.9%); primary n=187 (62.3%), sec n=26 (8.8%)

With CI n=36 (12.0%), no CI n=264.

MMSE: Cut scores of < 18 for illiterate subjects; < 20 for those with 1-2 yrs; < 22 for those with > 2 yrs of educ.

DIF method used

(e.g., MH, IRT, Logistic regression, MIMIC, other factor analysis)

“The GDS was assessed for its unidimensionality, item fit, redundancy and differential item functioning using the Rasch models in the Winsteps software package, Version 3.04 which implements an unconditional maximum likelihood procedure.” As a control for the performance of multiple tests the level of significance of 0.05 was adjusted by Bonferroni correction to 0.0033 (0.05/15).

For each item, standardized residuals of the observed from model predicted scores are calculated. “The test of DIF is an ANOVA of the person –item deviation residuals with person factors (e.g., age & education) and class intervals (e.g., group along the trait) as factors.

“The adequacy of the fit of each item to the Rasch model is assessed by the information weighted mean-square residual goodness of fit statistics [information weighted fit (INFIT) and outlier-sensitive fit (OUTFIT)]” (see ref #19 below), which measures the information about the responses given to items around the same difficulty endorsement level as the person’s ability.”

Test of model assumptions

“Unidimensonality refers to the single underlying construct measured by items that form a scale; for the GDS it is the depressive symptoms that result from the depressive illness.

“The mean logits (M) of the sample (-0.05+1.35) and the items (0.00+1.35) were adjacent to each other which indicated that the difficulties of the 15 items and the sample’s severity of depression are quite well matched. Eleven out of 15 items (73.3%) had INFIT/OUTFIT statistics between 0.7–1.3, reflecting that GDS was unidimensional, that is, these items contributed to a single underlying construct. The extreme INFIT/OUTFIT values of items #3, #4, #9, and #10 indicated that these items did not fit the model well and were not closely related to the overall construct.”

“The mean logits (M) of the sample (-0.05+1.35) and the items (0.00+1.35) were adjacent to each other which indicated that the difficulties of the 15 items and the sample’s severity of depression are quite well matched. Eleven out of 15 items (73.3%) had INFIT/OUTFIT statistics between 0.7–1.3, reflecting that GDS was unidimensional, that is, these items contributed to a single underlying construct. The extreme INFIT/OUTFIT values of items #3, #4, #9, and #10 indicated that these items did not fit the model well and were not closely related to the overall construct.

Four items did not fit the model well (Had INFIT/OUTFIT values outside of the acceptable range of .7 to 1.3: “ Do you feel that your life is empty”, “Do you often get bored”, “ Do you frequently worry about your future”, “ Do you feel you have more problems with memory than most”.

Purification

Not performed

Evidence of uniform DIF

Uniform DIF was defined by the authors as a “constant difference between groups in the probability of affirming an item (or category) across the trait (ANOVA main effect).”

Person and item separation index were 1.80 and 4.54, respectively. No items had a significant DIF for age, education and CI.”

Evidence of non-uniform DIF

The authors defined non-uniform DIF as the difference between the groups in the probability of a positive item response that varies across the trait (ANOVA interaction).

No evidence of significant interaction in the ANOVA of residuals; the authors conclude that non-uniform DIF is not detected for any comparisons.

Magnitude of DIF

Not provided

Impact of DIF

The four items exhibiting DIF were deleted in order to create a new version of the scale. There was no significant difference between the area under the ROC curves (AUC) of the original and revised versions of the GDS.

Strengths:

1. Use of ANOVA with residuals provides additional information about DIF, beyond that provided by fit statistics and t-test of the difficulty parameters.

Limitations according to authors:

1. The authors recognized the limited generalizability of the results due to the study sample (“findings may not be applicable to non-Chinese populations or patients with other forms of medical disease, such as stroke”) (pg 787). Additionally the DIF analyses included only small numbers of individuals with low MMSE’s (below 15) & should be replicated.

2. The authors noted that study results should be regarded as preliminary until a confirmation study supports the superiority of the revised version of GDS in the general population or other disease groups.

Additional limitations noted by expert reviewers:

Possible limitations of this study include the fact that:

1. The sample sizes of all of the group variables (age, education, CI) were small.

2. Purification was not performed.

3. Model fit was not compared to a 2-parameter IRT model; the Rasch model may not have been the best choice if parameters differ across groups. Thus, the test of non-uniform DIF may not be adequate.

Key references:

1. Lim PP, Ng LL, Chiam PC, Ong PS, Ngui FT, Sahadevan S. 2000. Validation and comparison of three brief depression scales in an elderly Chinese population. Int J Geriatr Psychiatry 15: 824–830.

2. Lai WKD. 2000. Measuring depression in Canada’s elderly Chinese population: use of a community screening instrument. Can J Psychiatry 49: 279–284.

3. Bedard M, Molloy DW, Squire L, et al. 2003. Validity of self reports in dementia research: The Geriatric Depression Scale. Clin Gerontol 26: 155–163.

4. Cannon BJ, Thaler T, Roos S. 2002. Oral versus written administration of the Geriatric Depression Scale. Aging Ment Health 6: 418–422.

5. Cialdella P, Guillaud-Bataille JM, Gausset MF, et al. 1992. A study about the unidimensionality of the Geriatric Depression Scale (Yesavage and Brink): comparison between classical methods and the Rasch model. Encephale 18: 537–544.

6. Cwikel J, Ritchie K. 1989. Screening for depression among the elderly in Israel : an assessment of the Short Geriatric Depression Scale (S-GDS). Isr J Med Sci 25: 131–137.

7. Ferraro FR, Chelminski I. 1996. Preliminary normative data on the Geriatric Depression Scale-Short From (GDS-SF) in a young adult sample. J Clin Psychol 52: 443–447.

8. Galaria II, Casten RJ, Rovner BW. 2000. Development of a shorter version of the geriatric depression scale for visually impaired older patients. Int Psychogeriatr 12: 435–443.

9. Gilley DW, Wilson RS. 1997. Criterion-related validity of the Geriatric Depression Scale in Alzheimer’s disease. J Clin Exp Neuropsychol 19: 489–499.

10. Jackson R, Baldwin B. 1993. Detecting depression in elderly medically ill patients: the use of the Geriatric Depression Scale compared with medical and nursing observations. Age Ageing 22: 349–353.

11. Kim JM, Prince MJ, Shin IS, Yoon JS. 2001. Validity of Korean form of Geriatric Depression Scale (KGDS) among cognitively impaired Korean elderly and development of a 15-item short version (KGDS-15). Int J Methods Psychiatr Res 10: 204–210.

12. Lam CK, Lim PP, Low BL, Ng LL, Chiam PC, Sahadevan S. 2004. Depression in dementia: a comparative and validation study of four brief scales in the elderly Chinese. Int J Geriatr Psychiatr 19: 422–428.

13. Lyness JM, Noel TK, Cox C, King DA, Conwell Y, Caine ED. 1997. Screening for depression in elderly primary care patients. A comparison of the Center of Epidemiologic Studies Depression Scale and the Geriatric Depression Scale. Arch Intern Med 157:449–454.

14. McGivney SA, Mulvihill M, Taylor B. 1994. Validating the GDS depression screen in the nursing home. J Am Geriatr Soc 42: 490–492.

15. Montorio I, Izal M. 1996. The geriatric depression scale: a review of its development and utility. Int Psychogeriatr 8: 103–112.

16. Rinaldi P, Mecocci P, Benedetti C, et al. 2003. Validation of the five-item Geriatric Depression Scale in elderly subjects in three different settings. J Am Geriatr Soc 51: 694–698.

17. Rule BG, Harvey HZ’Anne, Dobbs AR. 1989. Reliability of the Geriatric Depression scale for younger adults. Clin Gerontol 9: 37–43.

18. Sutcliffe C, Cordingley L, Burns A, et al. 2000. A new version of the Geriatric Depression Scale for nursing and residential home populations: the Geriatric Depression Scale (residential) (GDS-12R). Int Psychogeriatr 12: 173–181.

19. Wright BD, Masters GN. 1982. Rating Scale Analysis. Mesa Press: Chicago, IL.

( see abstract )

Back to   TOP    Measurement matrix         MEGs Search

Geriatric Depression Scale (GDS)
Name/ Reference	Tang WK, Wong E, Chiu HFK, Lum CM, Ungvari, GS. The Geriatric Depression Scale should be shortened: results of Rasch analysis. International Journal of Geriatric Psychiatry. (2005) 20:783–789.
Source contact info	Correspondence to: W. K. Tang, Department of Psychiatry, Shatin Hospital, Shatin, NT, Hong Kong SAR, China. Tel: þ852-2636-7760. Fax: þ852-2648-3394. E-mail: tangwk@cuhk.edu.hk
Availability (private or public)	public
Conceptual framework	Depression symptoms are causally emergent from a posited underlying trait or state of depression. The Rasch model is based on the premise that item response is a function of a person disability and item characteristics. The aim of this study was to examine the unidimensionality, item fit, and DIF of the GDS with respect to age, education and Cognitive Impairment (CI) in a community sample of Hong Kong Chinese patients with pneumoconiosis.
Purpose of measure & application (clinical, research, survey, screening)	Clinical: GDS has been used to identify depressive symptoms and depression in different types of populations.
Sample characteristics	A community sample of 300 Hong Kong Chinese patients with pneumoconiosis. Age ranged from 42-94 years; <60 (33%), 60-74 (45.3%), >75(20.7). DSM-IV diagnosis of depressive disorders: 37 (12.3%) pot pf 300 participants; minor dep (6%), major dep (3.67%), dysthymia (2.67%). Educ level: mean (sd) 3.1(2.9); no (27.9%), primary (62.3%), sec (8.8%); The MMSE score ranged from 8-30, average=25.6+4.5. Thirty-six (12.0%) patients had CI; the mean MMSE of those with CI= 16.6+3.6. Fourteen patients had a MMSE score less than 15.
Recruitment methods	421 participants were randomly selected by computer generated random numbers from 1442 patients representing 75% of all cases registered in the Pneumoconiosis Compensation Fund Board of Hong Kong. Recruitment of subjects took place from September to November 2003.
Data collection method	A trained research assistant administered the GDS to all participants. A psychiatrist, who was blind to the GDS scores conducted a structured clinical interview to diagnose depressive disorders according to the locally validated Chinese version of the Structured Clinical Interview for Diagnostic and Statistical Manual for Mental Disorders, Version IV (DSM-IV) criteria.
Response rate	Of the 421 patients selected, 300 (71.3%) participated and 121 (28.7%) patients were excluded from the study. No significant difference found between the participants & non participants in terms of male sex (99.0% vs 98.3%, p=0.628) and age (65.1+10.7 vs 64.1+11.1, t=0.896, p=0.371) distribution. Exclusion criteria: absence from the annual interview (n=81, 66.9%), refusal to participate in the study (n=36, 29.8%) and deafness (n=4, 3.3%).
Format & design (readability, # of items, time to complete, response categories)	The GDS consists of 15 items measuring depression symptoms. Questions are answered ‘yes’ or ‘no’ which are selected to measure depressive symptoms. Based on their GDS scores, patients are classified as depressed or non-depressed, using a cut score of 8 or more.
Type of measurement (nominal, ordinal, interval, ratio)	ordinal
Scoring (range, direction, rules, missing data)	The theoretical range is from 0-15. Missing data treatment was not discussed.
Availability of translations & source	Not provided
Psychometric Properties:
Scale construction	Not provided
Basic summary statistics	GDS for sample mean (sd)= 7.27(3.81)
Variability	Not provided
Test-retest reliability	See ref #4 below
Interrater reliability	Not provided
Internal consistency	See ref #1 (below)
Content validity	Not provided
Construct validity	See ref #1, 2, & 9 (below)
Concurrent validity	The locally validated Chinese version of the Structured Clinical Interview for DSM-IV served as a benchmark to judge the performance of the GDS as a screening instrument.
Predictive validity	For Specificity see ref #12 below. The optimal cutoff, sensitivity, specificity, and positive and negative predictive values, and the area under the receiver operating characteristic (ROC) curve of the original and revised versions of the GDS, were measured against the clinical diagnoses.
Sensitivity to change	Not provided
Differential Item Functioning (DIF):
Variable studied (e.g., groups)	Age, education and cognitive impairment were studied.
Sample size	Age: <60 n=99 (33%), 60-74 n=136 (45.3%), >75 n=62 (20.7%). Educ. level: no n=84 (27.9%); primary n=187 (62.3%), sec n=26 (8.8%) With CI n=36 (12.0%), no CI n=264. MMSE: Cut scores of < 18 for illiterate subjects; < 20 for those with 1-2 yrs; < 22 for those with > 2 yrs of educ.
DIF method used (e.g., MH, IRT, Logistic regression, MIMIC, other factor analysis)	“The GDS was assessed for its unidimensionality, item fit, redundancy and differential item functioning using the Rasch models in the Winsteps software package, Version 3.04 which implements an unconditional maximum likelihood procedure.” As a control for the performance of multiple tests the level of significance of 0.05 was adjusted by Bonferroni correction to 0.0033 (0.05/15). For each item, standardized residuals of the observed from model predicted scores are calculated. “The test of DIF is an ANOVA of the person –item deviation residuals with person factors (e.g., age & education) and class intervals (e.g., group along the trait) as factors. “The adequacy of the fit of each item to the Rasch model is assessed by the information weighted mean-square residual goodness of fit statistics [information weighted fit (INFIT) and outlier-sensitive fit (OUTFIT)]” (see ref #19 below), which measures the information about the responses given to items around the same difficulty endorsement level as the person’s ability.”
Test of model assumptions	“Unidimensonality refers to the single underlying construct measured by items that form a scale; for the GDS it is the depressive symptoms that result from the depressive illness. “The mean logits (M) of the sample (-0.05+1.35) and the items (0.00+1.35) were adjacent to each other which indicated that the difficulties of the 15 items and the sample’s severity of depression are quite well matched. Eleven out of 15 items (73.3%) had INFIT/OUTFIT statistics between 0.7–1.3, reflecting that GDS was unidimensional, that is, these items contributed to a single underlying construct. The extreme INFIT/OUTFIT values of items #3, #4, #9, and #10 indicated that these items did not fit the model well and were not closely related to the overall construct.” “The mean logits (M) of the sample (-0.05+1.35) and the items (0.00+1.35) were adjacent to each other which indicated that the difficulties of the 15 items and the sample’s severity of depression are quite well matched. Eleven out of 15 items (73.3%) had INFIT/OUTFIT statistics between 0.7–1.3, reflecting that GDS was unidimensional, that is, these items contributed to a single underlying construct. The extreme INFIT/OUTFIT values of items #3, #4, #9, and #10 indicated that these items did not fit the model well and were not closely related to the overall construct. Four items did not fit the model well (Had INFIT/OUTFIT values outside of the acceptable range of .7 to 1.3: “ Do you feel that your life is empty”, “Do you often get bored”, “ Do you frequently worry about your future”, “ Do you feel you have more problems with memory than most”.
Purification	Not performed
Evidence of uniform DIF	Uniform DIF was defined by the authors as a “constant difference between groups in the probability of affirming an item (or category) across the trait (ANOVA main effect).” Person and item separation index were 1.80 and 4.54, respectively. No items had a significant DIF for age, education and CI.”
Evidence of non-uniform DIF	The authors defined non-uniform DIF as the difference between the groups in the probability of a positive item response that varies across the trait (ANOVA interaction). No evidence of significant interaction in the ANOVA of residuals; the authors conclude that non-uniform DIF is not detected for any comparisons.
Magnitude of DIF	Not provided
Impact of DIF	The four items exhibiting DIF were deleted in order to create a new version of the scale. There was no significant difference between the area under the ROC curves (AUC) of the original and revised versions of the GDS.