The impact of physical fitness, social life, and cognitive functions on work ability in middle-aged and older adults

Descriptive statistics

Table 1 shows the descriptive statistics of sociodemographic and health-related variables, and the variables used in the structural equation modelling for middle-aged and older adults. Middle-aged adults showed overall higher levels in health-related variables (body mass index, waist–hip ratio, HRmax) and global cognition (MMSE). They also scored higher in education than older adults, which might be explained by more access to higher education in younger generations. Participants of both groups performed predominantly mental work as compared to physical, or mixed mental–physical work.

Table 1 Descriptive statistics of sociodemographic and health-related variables, and of the variables used in structural equation model for middle-aged and older adults

Concerning the variables used in the structural equation model, older adults informed of more current diseases (WAI3) and higher work impairment due to diseases (WAI4). They also rated their WA in relation to the job demands (WAI2) lower than their middle-aged counterparts. On the other hand, older adults informed of having more friends (FriendsN), while middle-aged adults had more frequent social contacts per week (FriendsFreq). Furthermore, older adults scored lower than middle-aged adults in all cognitive variables (Stroop, Digit-Symbol, d2-R, Digit-Span Backward, LPS3, and LPS7), but achieved a higher maximum power output (Pmax) on the bicycle ergometer test than the younger group.

Correlations

Figure 2 depicts the correlational structures for each age group in a graphic network. Visual network analysis is a useful tool in exploratory data analysis for visualizing correlational structures within the data. In both groups, the indicator variables formed similar patterns, corresponding to the four hypothesized latent factors (Work Ability, Social Life, Fitness, and Cognitive Functions).

Fig. 2figure 2

Network of associations based on polychoric, polyserial, and Pearson’s correlations for (a) middle-aged, and (b) older adults. Each node represents one of the indicator variables and the edges represent the strength of the correlations between the variables. Green edges represent positive correlations and red edges represent negative correlations. The width and color of the edges correspond to the absolute value of the correlations. Nodes are placed according to a forced-embedded algorithm (Fruchterman and Reingold 1991) in which the length of edges depends on the absolute weight of the edges. d2  d2-R Test of attentional endurance, DS  Digit-Span Backward, DST  Digit Substitution Test, FrF  frequency of social contacts, FrN  number of friends, LPS3  logical reasoning, LPS7  spatial rotation, LPA  lüdenscheid physical activity Questionnaire, Str  stroop test, Pmx  maximum power output in watt of the physical work capacity test; PWR  power-to-weight ratio of the physical work capacity test, QLs  WHOQoL social dimension, WAI  work ability index

Social Life vs. WA. In both groups, the WHOQoL social dimension (QoLsoc) showed a positive relationship with current WA (WAI1), WA in relation to job demands (WAI2), fewer diseases (WAI3), and mental resources (WAI7). In middle-aged adults, more frequent social contacts (FriendsFreq) were associated with fewer diseases (WAI3), less work impairment due to diseases (WAI4), higher number of friends (FriendsN), and a more optimistic prognosis of WA (WAI6). In older adults, FriendsN was positively associated with current WA (WAI1), WA in relation to job demands (WAI2), the prognosis of WA (WAI6), and mental resources (WAI7). Furthermore, in this age group, FriendsFreq correlated with job demands (WAI2) and mental resources (WAI7), and QoLsoc with the prognosis of WA (WAI6).

Cognitive Functions vs. WA. In middle-aged adults, WA in relation to job demands (WAI2) was positively associated with processing speed (DST), logical reasoning (LPS3), selective attention (d2), and working memory (DS). Furthermore, in this age group, DS was related to fewer sick leaves (WAI5), and DST to more mental resources (WAI7). In older adults, LPS3 was associated with WA in relation to job demands (WAI2) and mental resources (WAI7), and no other significant correlations between cognitive and WAI variables were found in this age group.

Physical Fitness vs. WA. Overall, we found more associations between the fitness-related and the WAI variables in the group of middle-aged than in their older counterparts. In this sense, in middle-aged adults, the maximum power output (Pmax), the power-to-weight ratio (PWR), and the weekly activity level (LPAQ) were associated with more physical health (WAI3, WAI4, and WAI5). Furthermore, Pmax correlated in middle-aged adults positively with job demands (WAI2), and LPAQ with mental resources (WAI7). In older adults, only sick leaves (WAI5) were related to Pmax and PWR, and no other WAI variables showed a significant correlation with physical fitness in this age group.

Structural equation modellingConfigural model and multi-group model fit

The posited model with the pooled data of both groups presented a regular fit \((^ = 347.435, df=146,p < 0.05, \mathrm= 0.905,\mathrm= .054, \mathrm= .07)\). An inspection of the modification indices indicated that linking the errors of WAI3 (number of current diseases diagnosed by a physician) and WAI4 (estimated work impairment due to diseases) would improve the model, reducing the standard \(^\) by 50.815. Given that the estimation of work impairment is conditioned by the number of current diseases, it seemed reasonable to us to include this parameter in the model. After this adjustment, the model showed an acceptable fit to the pooled data (\(^ = 300.529, df = 145, p < 0.05, \mathrm= 0.927, \mathrm= .047, \mathrm= .066)\). Factor loadings of all indicator variables were statistically significant (all \(p\mathrm< 0.001\)) and no variable presented a negative error variance. The standardized parameter estimates indicated that WA was significantly explained by Social Life (\(\gamma =.457, p<.001\)) and Cognitive Functions (\(\gamma = 0.172, p < 0.01\)), while the path between WA and Physical Fitness was not significant (\(\gamma = 0.088, p > 0.05\)). Social Life was associated with Cognitive Functions (\(r = 0.181, p < 0.05\)), whereas Cognitive Functions and Physical Fitness, and Physical Fitness and Social Life were unrelated to each other (\(p > 0 .05\)).

In a second step, we tested for configural invariance in a multi-group approach. Therefore, we fitted the data of middle-aged and older adults simultaneously in the same model with Age Group as the grouping variable. As showed in Table 2, fit indices were adequate (\(^ = 390.091, df = 290, p < 0.05, \mathrm= 0.945, \mathrm= 0 .038, \mathrm= 0 .079\)) and no variable presented a negative error variance.

Table 2 Model fits and invariance tests across middle-aged and older adults

In a third step, we tested for metric invariance, constraining the factor loadings to be equal across groups. Also, this model showed an adequate fit to the data of both groups (\(^ = 404.972, df = 305,p < 0.05,\mathrm= 0.945,\mathrm= .037, \mathrm= 0 .081, \Delta ^ = 14.88,\) n.s.), indicating that latent constructs were defined similarly by both groups. Detailed information on factor loadings and parameter estimates of the different models are given in Online Resource Tables S2, S3, and S4.

Regressions and correlations across groups

Figure 3 shows the partial regression coefficients and correlations between latent factors of the metric model for each age group. As indicated by the unstandardized parameter estimates, the path from Social Life to WA constituted the strongest predictor for WA in middle-aged adults (\(\gamma = 0.556, \mathrm = 0.196, p < 0.01)\), as well as in older adults (\(\gamma =.558, SE=.16,p < 0.001\)). Contrary to the analysis with the pooled data, the path from Cognitive Functions to WA did not reach statistical significance neither in the group of middle-aged (\(\gamma = 0.196, \mathrm = 0.127, p > 0.05),\) nor in the group of older adults \((\gamma = 0176,\mathrm = 0.112, p > 0.05)\). In middle-aged adults, Physical Fitness positively predicted WA (\(\gamma = 0.223,\mathrm= 0.084, p < 0.01)\), whereas in older adults, Fitness had no influence on WA (\(\gamma = -0.032,\mathrm= 0.073, p > 0.05)\). Regarding the correlations between the latent factors, in older adults, Physical Fitness correlated significantly with Cognitive Functions (\(r = 0.259, p < 0.01)\), whereas in middle-aged adults, Cognitive Functions were marginally associated with Social Life (\(r = 0.253, p = 0.05)\).

Fig. 3figure 3

Path diagram for (a) middle-aged, and (b) older adults. Social, CF, and Fitness constitute the independent (exogenous) factors, and WAI, the dependent (endogenous factor). One-headed lines represent unstandardized regression coefficients, and two-headed curved lines represent correlations. Social  social life, CF  cognitive functions, WAI work ability index. Observed variables are omitted. * < .05; ** < .01; *** < .001

In a last step, we tested for the statistical significance of group differences in these parameters. First, we set latent covariances and regression coefficients to be equal across groups, while maintaining the constraints on loadings. This model showed a significantly poorer fit in comparison to the unconstrained model (\(\Delta ^ = 28.88 , p < .05\)), confirming the existence of substantial differences in structural parameters across groups. Several t tests on the partial regression coefficients and on the correlations between latent factors revealed that the influence of Physical Fitness on WA was significantly larger in middle-aged than in older adults \((t = 2.304, p < .05)\). Social Life showed a stronger relationship with Cognitive Functions in middle-aged than in older adults \((t = 4.321, p < 0.01),\) whereas the correlation between Cognitive Functions and Fitness was significantly larger in older than in middle-aged adults \((t = 6.26, p < 0.01)\). None of the remaining structural parameters resulted statistically different across groups (all ps \(< 0.05\)). In a last step, to analyze how these differences influenced the model fit, we generated one more model, with latent covariances and regression coefficients constrained across groups, except for the path from Fitness to WA and the covariance between Fitness and Cognitive Functions. This model did not differ statistically from the metric model\((\Delta ^ = 0.47, n.s.)\), confirming that these were the only parameters whose group differences negatively affected the model fit.

留言 (0)

沒有登入
gif