Sample profile and attrition. The mean age of the sample was 55.2 years (SD = 7.92). More than half of the sample was male (61%), married (60%), and attained at least post-secondary education (60%). More than half of the sample experienced ischemic stroke (57%) and the onset age of stroke was 51.4 years (SD = 8.43) for the participants with average stroke duration of 3.83 years (SD = 4.25). Over 90% of the participants reported comorbidities such as hypertension, hyperlipidemia, diabetes, and/or heart diseases. Half (51%) of the sample reported slight disability and one-forth of the sample (26%) reported moderate disability.
In the present sample, the mean score for anxiety and depressive symptoms was 6.73 (SD = 4.13) and 7.28 (SD = 3.89), respectively. Around half of the participants reported possible anxiety and depression (41% and 50%). The sample showed average levels of hope (M = 33.5, SD = 9.1) and self-esteem (M = 26.5, SD = 5.2) and a lower level of functioning in the physical domain (M = 36.6, SD = 8.5) than mental domain (M = 46.7, SD = 10.4). Attrition analysis did not reveal significant differences between the study completers (N = 148) and dropouts (N = 36) in the mental health variables (p = 0.06-0.85), quality of life (p = 0.19-0.79), and demographic and clinical characteristics (p = 0.26-0.89), except for a significantly lower degree of disability (Cohen d = 0.16, p = 0.028) for the dropouts (mRS score = 1.86) than the study completers (mRS score = 2.18).
Construct validity and reliability. At Time 0 and Time 1, the 2-factor CFA model provided a better model fit than the 1-factor model (Δχ2 = 8.90-13.0, Δdf = 1, p < 0.01). From Table 1, neither model provided an adequate fit to the data with CFI and TLI < 0.95 and RMSEA > 0.10. Examination of modification indices suggested four residual correlations between item 1 (“fatigue”) and item 2 (“interference with personal life”), item 3 (“language”) and item 10 (“upper extremity”), item 5 (“mood”) and item 6 (“personality”), and item 7 (“self-care”) and item 12 (“housework”). Addition of these residual correlations led to an acceptable approximate fit (CFI and TLI ≥ 0.95, RMSEA ~ 0.08, and SRMR < 0.06) for the modified 1-factor model. The modified 2-factor model did not provide a superior fit over the modified 1-factor model (Δχ2 = 0.12-2.24, Δdf = 1, p = 0.13-0.73). The two factors in the modified 2-factor model were extremely highly correlated (r = 0.95-0.99). The equivalent model fit and potential factor redundancy supported a unidimensional factor structure for the SSQOL-12.
[Insert Table 1 about here]
Figure 1 depicts the factor structure of the modified 1-factor model at both Time 0 and Time 1. All of the specified factor loadings were substantial (λ = 0.40-0.87) at Time 0 and Time 1. The four residual correlations among the SSQOL-12 items were significant and moderate (r = 0.25-0.45, p < 0.05). The SSQOL-12 factor showed good reliability at both Time 0 and Time 1 (Ω = 0.88-0.89).
[Insert Figure 1 about here]
Measurement invariance across time. Table 2 shows the fit indices of measurement invariance models of the SSQOL-12 across time. The configural invariance model provided an adequate fit (CFI ≥ 0.95, RMSEA = 0.06, and SRMR = 0.065) to the data. The metric invariance model, which constrained factor loadings to be equal across time, provided better model fit indices over the configural invariance model without significant difference (Δχ2 = 19.4, Δdf = 12, p = 0.08) in chi-square difference test. The scalar invariance model fixed the four thresholds for each of the 12 items to be equal across time. This invariance model showed improved RMSEA and TLI over the metric invariance model without significant difference (Δχ2 = 52.5, Δdf = 47, p = 0.27) in chi-square difference test. These findings supported scalar measurement invariance across time. The latent mean of the SSQOL-12 factor showed a non-significant increase from Time 0 to Time 1 (standardized mean difference = 0.16, SE = 0.08, p = 0.051). The SSQOL-12 factors were strongly correlated (r = 0.70, p < 0.01) across Time 0 and Time 1. Besides, the model with invariant residual covariances did not display significantly worse fit (Δχ2 = 4.44, Δdf = 4, p = 0.35) than the scalar invariance model.
[Insert Table 2 about here]
Convergent and discriminant validity. The path model that associated the SSQOL-12 factor with mental health outcomes and demographic and clinical covariates provided a good fit (χ2 = 24.2, df = 20, p = 0.24, CFI = 0.990, TLI = 0.982, RMSEA = 0.034, and SRMR = 0.030) to the Time 0 data. As shown in Table 3, gender, age, and onset time of stroke were not significantly associated (p = 0.25-0.52) with SSQOL. Participants with ischemic stroke showed significantly higher SSQOL (β = 0.34, SE = 0.15, p = 0.018) than those with hemorrhagic stroke. The degree of disability was negatively associated (β = -0.40, SE = 0.06, p < 0.01) with SSQOL. SSQOL was moderately linked with higher levels of hope and self-esteem (β = 0.42-0.51, SE = 0.05-0.06, p < 0.01) and lower levels of anxiety and depression (β = -0.37 to -0.62, SE = 0.05-0.07, p < 0.01). The SSQOL-12 factor explained 17.5%, 13.9%, 37.9%, and 26.2% of the total variance of hope, anxiety, depression, and self-esteem, respectively.
[Insert Table 3 about here]
The SSQOL-12 factor was significantly and moderately correlated (r = 0.42-0.48, p < 0.01) with the physical and mental SF-12. Males showed significantly higher levels of physical QoL (β = 0.27, SE = 0.13, p = 0.042) than females. Age, onset time of stroke, and type of stroke were not significantly associated (p = 0.42-0.80) with physical and mental QoLs. The degree of disability was significantly and negatively associated (β = -0.43, SE = 0.07, p < 0.01) with physical QoL but not with mental QoL (β = -0.02, SE = 0.09, p = 0.85). Mental QoL showed positive and moderate associations (β = 0.27-0.40, SE = 0.06-0.09, p < 0.01) with hope and self-esteem and negative and moderate associations (β = -0.58 to -0.42, SE = 0.05-0.06, p < 0.01) with anxiety and depression. Weaker associations (β = -0.19 to 0.18) were found between physical QoL and mental health outcomes.
Controlling for the effects of physical and mental QoLs, the association between SSQOL and anxiety became non-significant (β = -0.06, SE = 0.08, p = 0.45). SSQOL was still significantly and positively linked with hope and self-esteem (β = 0.24-0.26, SE = 0.07-0.09, p < 0.01) and negatively linked with depression (β = -0.35, SE = 0.07, p < 0.01). This model explained 23.0%, 40.4%, 51.8%, and 38.6% of the total variance of hope, anxiety, depression, and self-esteem, respectively.