Regression analysis
Findings of conditional FE logistic regression analysis are shown in Table
2 (with problems in the EQ-5D dimensions as outcome measures) and findings of linear FE regression analysis are shown in Table
3 (with EQ-VAS as outcome measure). In the first case, odds ratios (OR) were reported and in the second case, beta-coefficients were reported. For non-continuous variables—e.g., the nominal time-varying covariate dementia -, the coefficients in Table
2 and Table
3 refer to the association between the onset of dementia and the outcome measures.
Table 2
Correlates of health-related quality of life (problems in the EQ-5D dimensions: 0 = absence of problems in the respective dimension; 1 = presence of problems in the respective dimensions). Findings of conditional FE logistic regressions
Social support (Lubben Social Network Scale) | 0.96 (0.90—1.03) | 0.91* (0.85—0.98) | 0.95 + (0.89—1.01) | 0.93* (0.88—0.99) | 0.94 + (0.89—1.00) |
Age | 1.82*** (1.44—2.31) | 1.84*** (1.40—2.41) | 1.66*** (1.34—2.06) | 1.23* (1.01—1.51) | 1.71*** (1.37—2.15) |
Married (Ref.: single/divorced/widowed) | 0.52 (0.08—3.42) | 0.71 (0.10—5.09) | 1.44 (0.24—8.84) | 0.77 (0.15—3.85) | 3.00 (0.38—23.70) |
Dementia (Global Deterioration Scale ≥ 4) | 0.56 (0.13—2.39) | 0.96 (0.26—3.61) | 2.05 (0.39—10.89) | 0.54 (0.16—1.86) | 0.39 (0.11—1.40) |
Visual impairment (Ref.: absence of visual impairment) | 3.87* (1.26—11.92) | 1.67 (0.79—3.52) | 1.75 (0.89—3.42) | 0.50* (0.27—0.95) | 1.02 (0.53—1.96) |
Hearing impairment (Ref.: absence of hearing impairment) | 1.48 (0.70—3.11) | 1.48 (0.67—3.26) | 2.04* (1.01—4.14) | 0.85 (0.43—1.67) | 1.04 (0.48—2.28) |
Observations | 511 | 435 | 602 | 614 | 537 |
Number of Individuals | 178 | 155 | 211 | 216 | 189 |
Pseudo R2 | .12 | .14 | .10 | .04 | .08 |
Table 3
Correlates of health-related quality of life (EQ-VAS, ranging from 0 (worst) to 100 (best)). Findings of linear FE regressions
Social support (Lubben Social Network Scale) | 0.02 |
| (0.14) |
Age | −0.97* |
| (0.46) |
Married (Ref.: single/divorced/widowed) | −2.97 |
| (3.28) |
Dementia (Global Deterioration Scale ≥ 4) | −7.20* |
| (3.11) |
Visual impairment (Ref.: absence of visual impairment) | −1.40 |
| (1.71) |
Hearing impairment (Ref.: absence of hearing impairment) | −3.14* |
| (1.50) |
Constant | 153.38*** |
| (40.86) |
Observations | 1792 |
Individuals | 648 |
R2 | 0.02 |
Adjusting for several time-varying covariates, conditional FE logistic regressions revealed that intraindividual decreases in social support were associated with an increased likelihood of developing problems in ‘self-care’ (OR = 0.91, p < 0.05), ‘usual activities’ (OR = 0.95, p < 0.10), ‘pain/discomfort’ (OR = 0.93, p < 0.05) and ‘anxiety/depression’ (OR = 0.94, p < 0.10) within individuals over time. Beyond that, only increasing age (intraindividual) was consistently associated with an increased likelihood of problems in all five dimensions within individuals over time.
Linear FE regressions did not show an association between intraindividual changes in social support and intraindividual changes in the EQ-VAS score. Intraindividual decreases in the EQ-VAS score were associated with intraindividual increases in age (β = −0.97, p < 0.05), the onset of dementia within individuals over time (β = −7.20, p < 0.05), and the onset of hearing impairment within individuals over time (β = −3.14, p < 0.05).
In additional analysis, full information maximum likelihood (FIML) was used to tackle missing data in linear FE regressions (please see Supplementary file 1: Supplementary Table 1). Similarly, in this model, intraindividual decreases in the EQ-VAS score were associated with the onset of dementia within individuals over time (β = −6.71, p < 0.05), and the onset of hearing impairment within individuals over time (β = −3.66, p < 0.05), whereas the association between intraindividual increases in age and intraindividual decreases in the EQ-VAS score vanished.
Moreover, in another sensitivity analysis, we restricted FE regressions to those individuals who continuously replied to the outcome measures from wave 7 to wave 9 (please see Supplementary file 1: Supplementary Table 2). While the marginal significant associations between intraindividual decreases in social support and intraindividual increases in the likelihood of developing problems in ‘usual activities’ within individuals over time (OR = 0.95, p = 0.15) and ‘anxiety/depression’ within individuals over time (OR = 0.97, p = 0.38) vanished, the significant associations between intraindividual decreases in social support and increases in the likelihood of developing problems in ‘self-care’ within individuals over time (OR = 0.91, p < 0.05) and ‘pain/discomfort’ within individuals over time remained nearly the same (OR = 0.92, p < 0.01).
In further sensitivity analysis, chronic conditions were added as a time-varying covariate to our main model (please see Supplementary file 1: Supplementary Table 3). Some differences (compared to our main model) are worth noting: While a marginal significant association between intraindividual decreases in social support and increases in the likelihood of developing problems in ‘mobility’ within individuals over time (OR: 0.93, p = 0.06) appeared, the marginal significant associations with ‘usual activities’ within individuals over time (OR: 0.96, p = 0.24) and ‘anxiety/depression’ within individuals over time (OR = 0.95, p = 0.14) disappeared. Furthermore, while the association between intraindividual decreases in social support and increases in the likelihood of developing problems in ‘self-care’ within individuals over time (OR = 0.93, p = 0.11) disappeared, the association with ‘pain/discomfort’ within individuals over time remained nearly the same (OR = 0.91, p < 0.01).
In further sensitivity analysis, the LSNS total score was replaced by the Family subscale and the Friend Subscale (please see Supplementary file 1: Supplementary Table 4). While intraindividual decreases in the family subscale (i.e., lower support from family) were associated with increases in the likelihood of developing problems in ‘self-care’ within individuals over time (OR: 0.88, p < 0.10) and ‘usual activities’ within individuals over time (OR: 0.85, p < 0.01), intraindividual decreases in the friend subscale (i.e., lower support from friends) were associated with increases in the likelihood of developing problems in ‘pain/discomfort’ within individuals over time (OR: 0.93, p < 0.10).
In our last sensitivity analyses, the LSNS total score was replaced by social isolation (dichotomized LSNS-6; please see Supplementary File 1: Supplementary Table 5). Our key findings remained similar. More precisely, while the marginal significant association between the presence of social isolation within individuals over time and increases in the likelihood of developing problems in ‘usual activities’ within individuals over time (OR: 1.04, p = 0.89) disappeared, the presence of social isolation within individuals over time was still associated with intraindividual increases in the likelihood of developing problems in ‘self-care’ (OR: 1.92, p < 0.05), ‘pain/discomfort’ (OR: 2.01, p < 0.05) and ‘anxiety/depression’ (OR: 1.60, p < 0.10).