Sample Characteristic. Participants were well-educated, Caucasian, living alone, and unemployed (). In general, women had high levels of psychological well-being, low mean levels of depressive symptomatology and anxiety, “good” subjective health and low number of illnesses. The most common illnesses reported at baseline were arthritis, hypertension, glaucoma/cataracts, thyroid problems, and heart disease.
Baseline demographic characteristics of community-dwelling older women (N = 115)
Does the sleep quality of older women change over time? Overall, sleep quality declined over time from M = 5 (SD = 2.72) to 6.29 (SD = 3.65); nearly half of the women reported the use of sleep aids at Time 3. To identify the general form of change in sleep quality over time, a linear growth curve was tested. Fit statistics indicate that a linear growth curve model was a good fit (X2 = 0.46; p = 0.5; Comparative Fit Index (CFI) = 1; Tucker Lewis Index (TLI) = 1.02; Root Mean Square Error of Approximation (RMSEA) = 0.00, Standardized Room Mean Square Residual (SRMR) = 0.01). The overall growth curve of sleep quality in community-dwelling older women increased significantly over time, indicating reduced sleep quality over time (β0 = 4.97; SE = 0.24; β1 = 0.14; SE = 0.03; p < .001). Note that higher numbers reflect declining sleep quality.
Are there different patterns of change in sleep quality over time?
The most parsimonious model (a one-class model) was followed by sequentially increasing the number of growth model classes to three latent class models, with the choice of best fitting model based on the following criteria: the Akaike Information Criterion (AIC) statistic (Akaike, 1974
), the Bayesian Information Criterion (BIC), the adjusted BIC statistical fit index (Schwarz, 1978), and the CAIC (Consistent AIC). Lower AIC, BIC, Adjusted BIC, and CAIC values indicate better model fit and a significant Lo-Mendell Rubin Likelihood Ratio (LMR) demonstrating a significant improvement in fit for the inclusion of one more class (Li & Nyholt, 2001
). While Adjusted BIC has been found to identify the number of classes better than other fit statistics, it is not perfect. Since no single parameter demonstrates a best fit, examining across the parameters provides the best estimate of fit (Nylund, Muthen, & Asparouhov, 2007
The three-category model demonstrated the best fit of the three models tested (). Group 1 (n = 23) had an estimated mean sleep quality score that did not change significantly over time (β0 = 7.34; SD = 0.47; β1 = 0.24; SD = 0.13; p < .058). Group 2 (n = 4) had the worse mean sleep quality score at baseline that became significantly worse over time (β0 = 10.62; SD = 2.74; β1 = 0.61; SD = 0.21; p < .004). Group 3 (n = 88) had the best mean sleep quality score at baseline, but it significantly worsened over time (β0 = 4.06; SD = 0.29; β1 = 0.09; SD = 0.04; p < .009).
Growth Mixture Model Fit Statistics for One-, Two- and Three-Category Sleep Quality Models Original Sample (N=115) and Retested Sample (N = 111)
Preliminary to performing the analysis for question 3, GCM was conducted for each of the health and well-being variables. Fit statistics were examined to ensure that each predictor variable met the assumptions of a linear model. For physical health, number of illnesses was dropped from this step of the analysis because of poor fit (). For psychological well-being, positive relations, personal growth, and purpose in life did not change, but, autonomy, environmental mastery, and self-acceptance had significant increases. For psychological distress, anxiety and depression scores improved significantly over time. Subjective health declined significantly over time. (See ).
Linear Growth Curve Model Fit Statistics for Each Health and Well-Being Variable (N=115)
Linear Growth Curve Model for Each Health and Well-Being Variable (N = 111)
Do psychological well-being (positive relations with others, environmental mastery, personal growth, purpose in life, and self-acceptance), psychological distress (depression and anxiety) or physical health (subjective health) predict different patterns of change in sleep quality over time? The three-category model could not be used in these analyses because the smallest group (n = 4) was too small for statistical comparisons. The four women in this group were removed from analysis. GCM and GMM modeling were conducted with the remaining 111 women. A sequential testing of one-, two- and three-category models indicated that a two-category model was the best fit, and a better fit than the previous two- and three-category models using all 115 women (see ). In this model, group 1 (n = 87) had relatively good sleep quality scores at baseline, but worsened significantly over time (β0 = 3.95; SE = 0.39; β1 = 0.10; SE = 0.10; p < .01). Group 2 (n = 24) had worse sleep quality scores at baseline that remained poor (β0 = 7.38; SE = 0.42; β1 = 0.20; SE = 0.14; p > .05). A binary classification of sleep quality was created from these distinctly different groups. The larger group (n = 87) had low sleep quality scores at baseline and scores increased significantly over time indicating good but diminished sleep quality over time. This group was classified as the “good sleep” (GS) in that while sleep worsened significantly overtime, overall sleep scores remained within a good vs poor range. The smaller group (n = 24) had poor sleep quality scores at baseline and no change in the level of sleep quality over time, indicating disrupted sleep that persisted over time. This group was classified as “disrupted sleep” (DS).
GMM was conducted to identify whether any psychological well-being, psychological distress, or physical health variables at baseline predicted membership in the GS or DS group. The growth class membership, based on posterior probabilities, was modeled using logistic regression. Predictors were modeled one at a time. The results indicate that baseline levels of five dimensions of psychological well-being (positive relations with others, environmental mastery, personal growth, purpose in life, and self-acceptance) predicted membership in the DS group of women, that is, higher levels of PWB were associated with lower odds of disrupted sleep (see ). Higher depression and more illnesses at baseline were associated with greater odds of having disrupted sleep. Baseline autonomy, anxiety, and subjective health were not significant.
Baseline Health and Well-Being Variables Predicting Membership in the Disrupted Sleep Group (N = 111)