The Development of a Rapid Classification Scale for Sleep Quality in Community-Dwelling Older Adults – The Yilan Study, Taiwan

Figures & data

Figure 1 Radar charts comparing the distributions of the seven component scores of the Pittsburgh Sleep Quality Index (PSQI) between subgroups across datasets. (A) shows the original empirically-derived subgroups in the development dataset. (B) shows the three subgroups specified by the Four-item Rapid Classification Scale for Sleep Quality (RCSSQ-4) in the validation dataset. (A) is adapted from Chen HC, Hsu NW, Chou P. Subgrouping Poor Sleep Quality in Community-Dwelling Older Adults with Latent Class Analysis - The Yilan Study, Taiwan. Sci Rep. 2020;10(1):5432.³⁹

Figure 2 Radar charts comparing the clinical features between subgroups across datasets. (A) shows the original empirically-derived subgroups in the development dataset. (B) shows the three subgroups specified by the Four-item Rapid Classification Scale for Sleep Quality (RCSSQ-4) in the validation dataset. (A) is adapted from Chen HC, Hsu NW, Chou P. Subgrouping Poor Sleep Quality in Community-Dwelling Older Adults with Latent Class Analysis - The Yilan Study, Taiwan. Sci Rep. 2020;10(1):5432.³⁹

Table 1 The Comparison of Demographic and Clinical Characteristics of Participants by Datasets*

	Development Dataset	Validation Dataset		p-value for χ^2/t/ANOVA Test
	Yilan City (n=2622)	Yilan City (n=964)	Taipei City (n=193)
Age (years, mean± SD)	76.6 ± 6.4	73.7 ± 7.5	77.8 ± 7.3	<0.001
Sex (n, %)
Female	1548 (59.0)	547 (56.7)	124 (64.2)	0.13
Education status (n, %)
Illiterate	593 (22.6)	62 (6.4)	–	<0.001
Body mass index (n, %)
< 18.5 kg/m^2	105 (4.0)	35 (3.6)	6 (3.1)	0.01
18.5–23.9 kg/m^2	961 (36.7)	390 (40.5)	82 (42.5)
≥ 24 kg/m^2	1404 (53.5)	486 (50.4)	105 (54.4)
Disabled	152 (5.8)	53 (5.5)	0 (0)
Living status (n, %)
Alone	197 (7.5)	97 (10.1)	–	0.01
Smoking status (n, %)
Current smoker	238 (9.1)	75 (7.8)	–	0.003
Ex-smoker	421 (16.1)	115 (12.0)
Non-smoker	1963 (74.9)	771 (80.2)
Drinking status (n, %)
Current drinker	331 (12.6)	116 (12.0)	–	0.78
Ex-drinker	137 (5.2)	55 (5.7)
Non-drinker	2154 (82.2)	793 (82.3)
Frequency of exercise (n, %)
< 3/ week	1174 (44.8)	221 (22.9)	–	<0.001
Falls in the past one year (n, %)	454 (17.3)	159 (16.6)		0.61
Pittsburgh Sleep Quality Index (mean± SD)	5.1 ± 4.0	4.3 ± 4.0	6.6 ± 4.1	<0.001
Athens Insomnia Scale (mean± SD)	2.1 ± 3.9	1.7 ± 3.2	–	0.01
Epworth Sleepiness Scale (mean ± SD)	4.1 ± 4.8	3.0 ± 4.2	6.1 ± 5.1	<0.001
Hypnotics use (days, mean± SD)	6.1 ± 11.7	5.5 ± 11.2	–	0.19
Number of physical diseases (mean± SD)	1.6 ± 1.3	1.5 ± 1.3	–	0.14
Hospital Anxiety and Depression Scale (mean± SD)
Depression subscale	1.9 ± 2.6	3.0 ± 3.4	–	<0.001
Anxiety subscale	2.3 ± 2.9	2.7 ± 3.1	–	<0.001
Short-form 12 (mean ± SD)
Mental health composite scores	58.4 ± 7.5	57.2 ± 7.8	–	<0.001
Physical health composite scores	46.6 ± 9.3	48.6 ± 9.4	–	<0.001
Groningen Activity Restriction Scale (mean± SD)	23.0 ± 11.3	22.2 ± 10.9	21.6 ± 5.4	0.04
Hand grip strength (kg, mean± SD)	19.5 ± 8.5	22.4 ± 8.8	–	<0.001

Note: *Some of the original variables were not available in the Taipei City cohort because the original study design differed.

Table 2 Internal Validity of the Predictive Models by Various Predictor Combinations

	Number of Predictors	Validity for Pittsburgh Sleep Quality Index ≥ 6					Overall Accuracy for Predicting Four Subgroups (%)
		Accuracy (%)	SEN (%)	SPE (%)	PPV (%)	NPV (%)
Full model*	8	89.9	83.7	93.9	89.5	90.1	81.6
Reduced models
(I). Remove Nightmare from full model	7	89.7	82.9	93.9	89.5	89.7	81.3
(II). Remove Sleepy Activities from (I)	6	89.4	82.4	93.9	89.4	89.5	81.1
(III). Remove Pain from (II)	5	89.4	81.9	94.0	89.6	89.2	81.0
(IV). Remove Sleep Disruption from (III)	4	89.0	83.7	92.3	87.2	90.0	79.9
(V). Remove Lacking Enthusiasm from (IV)	3	88.7	82.5	92.7	87.6	89.4	78.9
(VI). Remove Sleep Onset from (V)	2	87.0	77.8	92.7	87.0	86.9	77.3
(VII). Remove Hypnotics Use from (VI)	1	82.4	61.0	95.8	90.2	79.6	74.7

Notes: *Predictors and their priority in the full model were determined by stepwise multinomial logistic regression. The full model includes predictors of sleep quality, hypnotics use, sleep onset, lacking enthusiasm, sleep disruption, pain, sleepy activities, and nightmares.

Abbreviations: SEN, sensitivity; SPE, specificity; PPV, positive predictive value; NPV, negative predictive value.

Table 3 Internal Validity of Predictive Models for the Pittsburgh Sleep Quality Index ≥6 by Various Combinations of Predictors

Number of Predictors	Subgroups	SEN (%)	SPE (%)	PPV (%)	NPV (%)
Full model
8	Predictors: sleep quality, hypnotics use, sleep onset, lacking enthusiasm, sleep disruption, pain, sleepy activities, nightmare
	Low insomnia	62.7	90.7	65.3	89.7
	High insomnia	64.4	98.8	81.5	97.2
	High hypnotics	58.9	95.9	59.7	95.7
Reduced models
5	Predictors: sleep quality, hypnotics use, sleep onset, lacking enthusiasm, sleep disruption
	Low insomnia	63.9	90.0	64.2	89.9
	High insomnia	62.3	98.9	82.1	97.1
	High hypnotics	49.6	96.4	59.0	94.9
4	Predictors: sleep quality, hypnotics use, sleep onset, lacking enthusiasm
	Low insomnia	61.3	89.4	61.8	89.2
	High insomnia	55.5	99.3	86.2	96.6
	High hypnotics	60.6	94.6	53.8	95.9
3	Predictors: sleep quality, hypnotics use, sleep onset
	Low insomnia	49.8	92.1	64.0	86.8
	High insomnia	51.8	99.5	90.0	96.3
	High hypnotics	77.2	91.4	48.2	97.5

Abbreviations: SEN, sensitivity; SPE, specificity; PPV, positive predictive value; NPV, negative predictive value.

Table 4 Parameter Estimates of Multinomial Logistic Regression Analyses for Predicting Subgroups of Poor Sleep Quality in the Development Dataset

	Reference Group (Pittsburgh Sleep Quality Index < 6)
	Low Insomnia		High Insomnia		High Hypnotics
	b (se)	OR (95% CI)	b (se)	OR (95% CI)	b (se)	OR (95% CI)
Model I: 5 predictors
Intercept	−5.20 (0.23)		−16.45 (0.95)		−7.35 (0.37)
Sleep quality	1.94 (0.17)	6.93 (4.96–9.68)	5.39 (0.41)	219.15 (98.46–487.80)	1.79 (0.23)	5.96 (3.81–9.32)
Hypnotics use	0.75 (0.07)	2.12 (1.84–2.45)	0.92 (0.12)	2.51 (1.99–3.15)	1.84 (0.09)	6.32 (5.25–7.60)
Sleep onset	1.18 (0.07)	3.26 (2.86–3.72)	1.88 (0.16)	6.57 (4.79–9.01)	0.90 (0.10)	2.47 (2.05–2.97)
Lacking enthusiasm	0.78 (0.29)	2.18 (1.24–3.83)	2.15 (0.33)	8.54 (4.49–16.24)	2.17 (0.31)	8.78 (4.77–16.14)
Sleep disruption	0.68 (0.08)	1.98 (1.70–2.30)	1.09 (0.14)	2.98 (2.28–3.89)	0.48 (0.11)	1.62 (1.31–2.00)
Model II: 4 predictors
Intercept	−5.21 (0.23)		−16.10 (0.91)		−7.36 (0.37)
Sleep quality	2.35 (0.17)	10.46 (7.55–14.48)	5.93 (0.39)	376.14 (175.02–808.39)	2.06 (0.22)	7.86 (5.07–12.19)
Hypnotics use	0.73 (0.07)	2.07 (1.80–2.38)	0.89 (0.12)	2.43 (1.95–3.05)	1.82 (0.09)	6.18 (5.15–7.41)
Sleep onset	1.22 (0.07)	3.38 (2.97–3.84)	1.96 (0.16)	7.11 (5.17–9.77)	0.93 (0.09)	2.53 (2.11–3.04)
Lacking enthusiasm	0.85 (0.28)	2.34 (1.35–4.06)	2.32 (0.32)	10.22 (5.46–19.12)	2.21 (0.31)	9.09 (4.98–16.60)
Model III: 3 predictors
Intercept	−5.28 (0.23)		−16.01 (0.89)		−7.39 (0.36)
Sleep quality	2.46 (0.16)	11.67 (8.47–16.09)	6.29 (0.38)	541.52 (256.99–1141.07)	2.42 (0.22)	11.26 (7.34–17.25)
Hypnotics use	0.72 (0.07)	2.06 (1.79–2.36)	0.86 (0.11)	2.36 (1.90–2.93)	1.76 (0.08)	5.82 (4.89-6.91)
Sleep onset	1.21 (0.07)	3.36 (2.96–3.81)	1.90 (0.16)	6.72 (4.92–9.17)	0.93 (0.09)	2.51 (2.11–3.02)

Table 5 The Validity of the Four-Item Rapid Classification Scale for Sleep Quality for the Pittsburgh Sleep Quality Index ≥6

	Prevalence of PSQI≥ 6 (%)	Accuracy (%)	SEN (%)	SPE (%)	PPV (%)	NPV (%)
Development dataset
Yilan study (n=2622)	38.6	89.0	83.7	92.3	87.2	90.0
Validation dataset
Yilan city (n=964)	30.6	89.1	87.5	89.7	75.4	95.2
Taipei city (n=193)	53.9	90.7	87.5	94.4	94.8	86.6

Abbreviations: PSQI, the Pittsburgh Sleep Quality Index; SEN, sensitivity; SPE, specificity; PPV, positive predictive value; NPV, negative predictive value.

Chen HC, Hsu NW, Chou P. Subgrouping poor sleep quality in community-dwelling older adults with latent class analysis - the Yilan Study, Taiwan. Sci Rep. 2020;10(1):5432. doi:10.1038/s41598-020-62374-4

 Web of Science ®Google Scholar