One hundred and twenty-eight patients (45.7%, 95% CI: 39.8-51.7) positively screened for depression among the 280 geriatric ED patients with GDS-15 using a cut off value of 4/5. The prevalence of previously unrecognized depression was 42.2 % (95% CI: 32.8-52) in males and 48% (95% CI: 40.3-55.7) in females, but the difference was not statistically significant (P-value=0.35). The mean GDS-15 score was 4.43 (95% CI :4.11-4.78, SD ±2.87, range 0-13). Among all the positively screened patients (n=128), 104 (81.3%, 95% CI:73.4-87.6) screened positive for mild depression (score 5-8) and 24 (18.8%, 95% CI:12.4-26.6) were screened positive for severe depression (score >8) (Table 1). When GDS-5 was used as the screening tool with a cutoff score of 2/3, 103 (36.8%, 95% CI:31.1-42.7) screened positive for depression with mean score of 1.3 (95% CI:1.16-1.43, SD ±1.146, Range 0-5) (Table2). There was a high degree of agreement between one of the authors and the research assistant on 20 randomly selected geriatric patients (Cohen’s Kappa k=.841(95% CI:.633-1.05), P-value<.001).
Table 1. Prevalence of geriatric depression according to GDS-15 (n=280)
Category
|
GDS-15 score
|
n (%)
|
Subtotal n (%)
|
Negatively screened for depression
|
0
|
12 (4.3)
|
152 (54.3)
|
1
|
38 (13.6)
|
2
|
38 (13.6)
|
3
|
31 (11.1)
|
4
|
33 (11.8)
|
Positively screened for mild depression
|
5
|
26 (9.3)
|
104 (37.1)
|
6
|
35 (12.5)
|
7
|
21 (7.5)
|
8
|
22 (7.9)
|
Positively screened for severe depression
|
9
|
11 (3.9)
|
24 (8.6)
|
10
|
5 (1.8)
|
11
|
4 (1.4)
|
12
|
2 (.7)
|
13
|
2 (.7)
|
Table 2. Prevalence of geriatric depression according to GDS-5 (n=280)
Category
|
GDS-5 score
|
n (%)
|
Subtotal n (%)
|
Negatively screened for depression
|
0
|
78 (27.9)
|
177 (63.2)
|
1
|
99 (35.4)
|
Positively screened for depression
|
2
|
60 (21.4)
|
103 (36.8)
|
3
|
31 (11.1)
|
4
|
9 (3.2)
|
5
|
3 (1.1)
|
Associations of geriatric depression
Mean age of the total sample was 71.36 years (95% CI:70.42-72.31, SD± 8.06, range: 60-94) with no significant difference by gender. The majority of patients were females (61%) largely single (78.9%) (P-value<.001). Among them, 82.5% had no formal education and females were more illiterate (94% vs 64%, P-value<.001). Male patients consumed significantly more alcohol than their female counterpart (24% vs 14%, P-value=0.036). Two hundred and eight (74.3%) patients had previous visits to ED in the last one year, and the proportion was higher in females (78%) in comparison to males (68%) (P-value=0.051). The comorbidities like COPD, cardiovascular disease (CVD), musculoskeletal (MSK) disease and DM were present in 59.3%, 37.1%, 16.8% and 7.9% of the study population respectively without significant difference by gender. Only one patient among the study group had no family support.
The association of the demographic data, comorbidities and health seeking behavior with depression was evaluated (Tables 3). The negative, mild depression and severe depression groups were similar with regard to age, gender, marital status, educational level, employment, financial condition, presence of pain, presence of comorbid conditions and previous ED visits. Females consisted 83% of those who were severely depressed, however the difference was not statistically significant (P-value=0.064) (Figure 2). Those who self-reported having the "heart and mind disease" had higher GDS-15 scores (P-value<.001). Furthermore, the depressed and severely depressed subjects, compared with non-depressed subjects, had more disturbed sleep (P-value<.001) and visited the local healers more frequently(P-value<.001). Similarly, those who had more pain, visited the ED more often and who had MSK diseases showed statistically significant differences among the three categories of depressive status (P-value=0.014, 0.047 and0.016 respectively)
Table 3. Association of socio-demographic characteristics, comorbidities and health seeking behavior with depression categories (n=280)
Variables
|
Total (n=280)
|
GDS-15 score <5 (n=152)
|
GDS-15 score 5-8 (n=104)
|
GDS-15 score 9-15
(n=24)
|
P- value
|
GDS-15 Score, Median (IQR)
|
4(2-6)
|
2(1-3)
|
6(5.5-7)
|
10(9-11)
|
NA
|
GDS-5 Score, Median (IQR)
|
1 (0-2)
|
0.5(0-1)
|
2 (1-2)
|
3(3-4)
|
<.001a
|
Age, median (IQR)
|
70 (65-70)
|
70(65-76)
|
71(65-78)
|
68(64.5-75)
|
0.458a
|
Female, n (%)
|
171 (61)
|
89 (58.6)
|
62(59.6)
|
20(83.3)
|
0.064b
|
Marital status-single, n (%)
|
109(38.9)
|
59 (38.8)
|
43(41.3)
|
7(29.2)
|
0.544b
|
No formal education, n (%)
|
231 (82.5)
|
126(82.9)
|
84(80.8)
|
21(87.5)
|
0.723b
|
unemployed, n (%)
|
195 (69.6)
|
101(66.4)
|
79(76)
|
15(62.5)
|
0.194b
|
Finance resource (not enough and just enough), n (%)
|
36 (12.9)
|
19(12.5)
|
11(10.6)
|
6(25)
|
0.161b
|
Alcohol, n (%)
|
50 (17.9)
|
25(16.4)
|
19(18.3)
|
6(25)
|
0.591b
|
Pain, n (%)
|
217(77.5)
|
113(74.3)
|
83(79.8)
|
21(87.5)
|
0.278b
|
Frequency of pain in 1 week, median (IQR)
|
3(2-4)
|
3(0-4)
|
3(2-4)
|
4(2.5-5.5)
|
0.014a
|
Disturbed sleep, n (%)
|
159(56.8)
|
68(44.7)
|
76(73.1)
|
15(62.5)
|
<.001b
|
Visit to local healer, n (%)
|
85(30.4)
|
21 (13.8)
|
52(50)
|
12(50)
|
<.001b
|
Self-perceived presence of heart and mind disease, n(%)
|
94(33.6)
|
30(19.7)
|
52(50)
|
12(50)
|
<.001b
|
Mobility, needs support, n (%)
|
89 (31.8)
|
43(28.3)
|
42(40.4)
|
4(16.7)
|
0.031b
|
Previous ER visit, n (%)
|
208(74.3)
|
109(71.5)
|
82(78.8)
|
17(70.8)
|
0.405b
|
Frequency of ED visits in last one year, median (IQR)
|
2 (0-3)
|
1.5(0-3)
|
2(1-3)
|
3(0-5)
|
0.047a
|
COPD, n (%)
|
166(59.3)
|
88(57.9)
|
62(59.6)
|
16(66.7)
|
0.716b
|
CVD, n (%)
|
104(37.1)
|
60(39.5)
|
36(34.6)
|
8(33.3)
|
0.675b
|
DM, n (%)
|
22(7.9)
|
14(9.2)
|
6(5.8)
|
2(8.3)
|
0.601b
|
MSK, n (%)
|
47(16.8)
|
24(15.8)
|
14(13.5)
|
9(37.5)
|
0.016b
|
aIndependent samples Kruskal Wallis test, bChi square test
Table 4 illustrates the estimates of multinomial logistic regression coefficients, their p value and odds ratios for each variable after controlling for confounders. Those variables which showed P-value of <0.2 in the bivariate analysis of non-depressed vs mild and severe depression (Tables 3) (gender, occupation, financial resource, mobility, sleep, frequency of pain, frequency of ED visit, self perceived “heart and mind disease”, visit to the local healer and presence of MSK disease) were chosen for the multinomial logistic regression analysis. The “no depression” category was taken as reference. The Hosmer-Lemeshow goodness-of-fit test indicated that the model described the data well (P-value= 0.305). Visits to local healers were associated with both mild depression and severe depression, independently of other factors (P-value=0.003 and 0.019 respectively). Those who reported to have disturbed sleep were 2.3 times more likely to screen positive for mild depression as opposed to no depression (P-value=0.006). Patients who experienced more frequent pain were 1.3 times more likely to screen positive if they were severely depressed versus people with no depression (P-value=0.05). No other variable was associated with either mild or severe depression in multinomial regression analysis.
Table 4. Multinomial logistic regression analysis of the relationship between different variables and different severity of depression as compared to not depressed. (n=280)
Variables
|
Mild depression vs no depression
OR (95% CI)
|
P-value
|
Severe depression vs no depression
OR (95% CI)
|
P-value
|
Gender
|
Female
|
1
|
0.756
|
1
|
0.087
|
Male
|
1.09 (.617-1.943)
|
0.358 (.110-1.161)
|
Occupation
|
Not employed
|
1
|
0.149
|
1
|
0.768
|
Employed
|
0.606 (.306-1.197)
|
0.854 (.299-2.437)
|
Disability
|
Needs support
|
1
|
0.699
|
1
|
0.183
|
On their own
|
1.13 (.608-2.099)
|
0.426 (.121-1.496)
|
Self-perceived financial resource
|
Not enough
|
1
|
0.853
|
1
|
0.248
|
enough
|
1.091 (.432-2.757)
|
.493(.149-1.635)
|
Self-perceived mind and heart disease
|
Absent
|
1
|
0.295
|
1
|
0.217
|
Present
|
2.455 (.458-13.170)
|
5.316 (.375-75.301)
|
Disturbed sleep
|
No
|
1
|
0.006
|
1
|
0.504
|
Yes
|
2.326 (1.275-4.244)
|
1.411 (.514-3.877)
|
Visit to local healer
|
Yes
|
1
|
0.003
|
1
|
0.019
|
no
|
0.072 (2.326-1.275)
|
0.04(.003-.595)
|
MSK disease
|
Present
|
1
|
0.594
|
1
|
0.609
|
Absent
|
1.266(.532-3.009)
|
.734 (.224-2.402)
|
Frequency of pain during last week
|
1.019 (.864-1.201)
|
0.827
|
1.318(1-1.737)
|
0.05
|
Frequency of ED visits in last year
|
1.062 (.948-1.191)
|
0.3
|
1.111 (.942-1.310)
|
0.212
|
Significant P-values are bolded.
Performance of GDS-5 vs GDS-15 in our sample
The sensitivity of the brief original GDS-5 (items 1,4,8,9,12 from GDS-15) when compared to the standard GDS-15 was 75.78% (97/128; 95% CI: 67.42-82.91), with a specificity of 96.05% (146/152; 95% CI: 91.61-98.54) using a cutoff score of 2/3 (Youden index=0.718). Significant agreement was found between the 15-item GDS and the 5‐item GDS (kappa = 0.73). A reliability analysis was carried out on the GDS-15 scale items and the GDS-5 scale. The Cronbach’s alpha score reached acceptable reliability (alpha=0.714) for the GDS-15, however it was low for the GDS-5 (alpha=0.416). As illustrated in table 5, items 1, 3, 4, 7 and 14 in the GDS-15 appeared to be the worthiest of retention, resulting in a “decrease” in the alpha if deleted. The items 9, 11, 13 and 15 would “increase” the alpha if deleted. Therefore, a new version of the GDS-5 for our population was proposed selecting these 5 items from the GDS-15 with the highest corrected item- total correlation (items 1, 3, 4, 7 and 14).
Table 5. Item-Total Statistics of GDS-15
Items in GDS15
|
Scale Mean if Item Deleted
|
Scale Variance if Item Deleted
|
Corrected Item- Total Correlation
|
Cronbach's Alpha if Item Deleted
|
Item 1a
|
4.2652
|
7.239
|
.427
|
.689
|
Item 2
|
4.2545
|
7.478
|
.292
|
.703
|
Item 3a
|
4.1004
|
6.889
|
.453
|
.682
|
Item 4a
|
3.9283
|
6.973
|
.381
|
.692
|
Item 5
|
4.2867
|
7.421
|
.359
|
.696
|
Item 6
|
4.0215
|
7.000
|
.380
|
.692
|
Item 7a
|
4.1685
|
7.112
|
.394
|
.691
|
Item 8
|
4.3011
|
7.449
|
.366
|
.696
|
Item 9b
|
4.2222
|
7.735
|
.147
|
.718
|
Item 10
|
3.8674
|
7.274
|
.265
|
.707
|
Item 11b
|
4.3978
|
8.053
|
.146
|
.714
|
Item 12
|
4.1470
|
7.183
|
.349
|
.696
|
Item 13b
|
3.8387
|
7.481
|
.189
|
.717
|
Item 14a
|
4.0430
|
6.905
|
.425
|
.686
|
Item 15b
|
4.1792
|
7.565
|
.200
|
.713
|
aItems of GDS-15 resulting in a “decrease” in the alpha if deleted (items for new GDS-5),
bItems of GDS-15 resulting in an “increase” in the alpha if deleted.
Sensitivity, specificity, diagnostic accuracy, and positive and negative predictive values were calculated for the proposed 5‐item scale and compared with the original one (Table 6 and 7). The validity was recalculated which improved the sensitivity to 90.62 (95%CI: 84.20% to 95.06%) (Youden index=0.755) and the Cronbach’s alpha increased to 0.623 from 0.416. Thus, the detection rate of geriatric depression with the new scale was higher (139 vs 103) (table 6). The agreement between the 15-item GDS and the proposed 5‐item GDS was also higher than with the original one (kappa = 0.75 vs 0.73) (Table 7). Other measures for validity like Spearman’s correlations (strong correlation for both) and AUC (excellent discriminant validity for both) are depicted in Table 7.
Table 6: Assessment of original and proposed GDS-5 accuracy in relation to the GDS-15.
|
GDS-15 n (%)
|
Depressed (n=128)
|
Not depressed (n=152)
|
Original GDS-5
|
Depressed (n=103)
|
97 (75.8)
|
6 (3.9)
|
Not depressed (n=177)
|
31 (24.2)
|
146 (96.1)
|
Proposed GDS-5
|
Depressed (n=139)
|
116 (90.6)
|
23 (15.1)
|
Not depressed (n=141)
|
12 (9.4)
|
129 (84.9)
|
Table 7: Performance of original and proposed GDS-5 against GDS-15
|
Original GDS-5
|
Proposed GDS-5
|
Sensitivity, % (95% CI)
|
75.8 (67.42-82.91)
|
90.6 (84.2-95.06)
|
Specificity, % (95%CI)
|
96 (91.61-98.54)
|
84.9 (78.17-90.16)
|
Positive likelihood ratio, (95% CI)
|
19.2 (8.71 -42.31)
|
5.99 (4.09-8.76)
|
Negative likelihood ratio, (95%CI)
|
0.25 (0.19 -0.34)
|
0.11 (0.06- 0.19)
|
Positive predictive value, % (95%CI)
|
94.2 (88.00-97.27)
|
83.45 (77.50-88.06)
|
Negative predictive value, % (95% CI)
|
82.5 (77.59-86.51)
|
91.49 (86.21-94.87)
|
Accuracy, % (95% CI)
|
86.8 (82.25-90.52)
|
87.5 (83.05-91.14)
|
AUC (area under curve) 95% CI)
|
0.919 (0.887-0.951)
|
0.929 (0.900-0.959)
|
Cronbach’s Alpha
|
0.416
|
0.623
|
kappa
|
0.73
|
0.75
|
Spearman’s correlation coefficient, rs
|
0.827
|
0.860
|