The prospective association between frequency of contact with friends and relatives and quality of life in older adults from Central and Eastern Europe

The Health, Alcohol and Psychosocial factors in Eastern Europe is a population-based cohort study conducted in the Czech Republic (six towns), Poland (Krakow), Russia (Novosibirsk) and Lithuania (Kaunas). As Lithuania joined the project at wave two, we used Czech, Polish and Russian participant data from wave 1 (2002–2005) and wave 2 (2006–2008) for the analyses. Participants aged 45–69 were recruited at baseline from the population registers of the countries’ towns. The wave one response rate varied from 55 to 61% [21].

QoL was measured using the CASP (Control, Autonomy, Self-realisation and Pleasure) scale. CASP is a validated self-administered instrument used as a measure of eudemonic wellbeing in older populations based on the measurement of those four domains [22]. CASP asks participants to rate selected items on a Likert scale ranging from 0 (‘never’) to 3 (‘often’). QoL was measured by the 19-item version at baseline and the CASP-12 v.1 at follow-up developed by Borsch-Supan et al. [23]. Using wave 1 data from the twelve CASP-19 items comprising CASP-12 and CASP-12, v.1 scores were calculated by adding these item responses to derive scores ranging from 0 to 36 at both waves. Higher scores represent higher QoL.

Data on SN were collected at wave 1. Participants were asked ‘are you regularly in contact with your relatives who do not live in your household?’ and ‘how often do you visit friends?’ Responses ranged from 0 = ‘No relatives/No friends’, 1 = ‘Less than once a month’, 2 = ‘About once a month’, 3 = ‘Several times a month’, 4 = ‘About once a week’ to 5 = ‘Several times a week’. Responses to both questions were summed to create a score ranging from 0 to 10, where higher scores indicated more frequent contacts with both friends and relatives. Data on SN were analysed as scores, as well as tertiles denoting low (0–4), medium (5–6) and high (7–10) groups to capture both the change by unit and the gradient of change.

Depressive symptoms were measured using the Centre for Epidemiologic Studies Depression (CES-D) scale at waves 1 and 2, using the 20 and 10-item versions, respectively. CES-D has been shown to sufficiently screen for depression in older populations [24]. The CES-D 20 evaluated the frequency of 20 depressive symptoms during the past week on a four-point scale ranging from 0 (‘less than one a day’) to 3 (‘5–7 days’). Responses were added to obtain CES-D 20 scores (0–60). CES-D 10 asked whether 10 symptoms were felt ‘for much of the time’ in the previous week using yes (1) or no (0) response options, then summed to derive scores (0–10). Using suggested cut-offs, participants with CES-D 20 scores of ≥ 16 and CES-D 10 scores of ≥ 4 were identified as possible cases of depression.

Age, age squared, gender, country, marital status, educational attainment and material deprivation (based on how often participants lacked money for food, clothing or paying bills) were obtained at wave 1. Other baseline covariates included alcohol intake (grams per week), smoking status, self-rated health, physical functioning assessed with the SF-36 physical functioning subscale score [25] (0–100) and number of hours per week spent on doing sports or exercise.

The CASP scale at wave 1 was included in the retirement module given to retired individuals who received a pension. Of the total baseline sample (n = 28,945), wave 1 CASP data were available for 12,840 participants. Of these participants, 39.79% (n = 4981) were excluded due to missing CASP data at wave 2. The analytic sample was further reduced by 0.72% (n = 57) and 21.74% (n = 1696) due to missing data on SN and covariates, respectively. Therefore, the analytic sample of complete cases was N = 6106.

Linear regression was used to assess the prospective association between frequency of contact with friends and relatives at baseline, measured in a linear fashion and in tertiles, and QoL at follow-up. The analyses fitted nested models which were adjusted for age, age squared, gender, country and marital status (model 1), socioeconomic and health-related covariates (model 2) and further adjusted for CASP-12 score at wave 1 (model 3). We also ran multivariable regressions to determine the explanatory role of each covariate included in model 2 and 3, which adjusted for each of these variables in turn. Interactions between the exposure and age, gender and country on QoL were tested using two-way interaction terms and likelihood ratio tests. As no interaction effects reached statistical significance, the models were not stratified by these covariates.

Path modelling was applied to test whether depressive symptoms at wave 1 and wave 2 mediated the longitudinal association between frequency of contact with friends and relatives at wave 1 and QoL at wave 2 [26]. The path model is based on a half-longitudinal design [27] because SN and depressive symptoms at wave 1 on one hand, and depressive symptoms and CASP-12 score at wave 2 on the other hand, were measured concurrently. The potential pathway via depression was hypothesised in a path model, which comprised four paths (Fig. 1). Path A, the ‘direct effect’ of frequency of contact with friends and relatives at wave 1 on QoL at wave 2, was measured using linear regression. Path B measured the effect of frequency of contact with friends and relatives on possible depression (both at wave 1) using logistic regression. Path C estimates the effect of being a possible case of depression at wave 1 on being a possible case of depression at wave 2, also using logistic regression. The last path, D, measured the effect of having possible depression on QoL (both at wave 2) using linear regression. Therefore, paths A and D are reported as linear regression coefficients, and paths B and C are reported as ORs. After estimating the path model, the indirect effect was calculated by multiplying the coefficients of paths B, C and D. The total effect was the sum of both the direct and indirect effects. The proportion of the total effect that was mediated through depressive symptoms was calculated by dividing the indirect effect by the total effect. The path model adjusted for the covariates used in the linear regression model by including them as predictors of QoL at wave 2 and frequency of contacts with friends and relatives at wave 1. For ease of interpretation, the path model was fitted using the continuous variable of frequency of contact with friends and relatives. All statistical analyses were carried out with Stata version 13 and with a 0.05 significance level.

To assess possible biases in this study, several sensitivity analyses were carried out. To assess whether the analytic sample was representative of the recruited sample in HAPIEE, an analysis comparing the participants’ characteristics between the analytical sample (n = 6106) and those excluded (n = 6734) was done. Of the 12,840 participants with data on QoL at baseline, 7791 participants had available data on CASP-12 at wave 2 and on covariates included in model 1. Therefore, this model was re-run on these participants and the results were compared with those from the main analysis for robustness. Additionally, the cross-sectional association between frequency of contact with friends and relatives and QoL was assessed using the data available at baseline for CASP-12. As the main exposure in this study was derived from two variables, the prospective association was tested in each separate exposure, with a score ranging from 0–5 where a higher score indicates a more frequent contact. Lastly, to address the risk of over-adjustment by depression in the path analysis, another path model was estimated using only depression at wave 1.

Study characteristics of the sample differentiated by country can be seen in Table 1. The average age of the participants was 62.5 years. There were approximately 20% more women than men in the sample and around three-quarters of the sample lived with a partner at baseline. Overall, the mean CASP-12 score was 23.3 and 24.9, respectively, at baseline and follow-up. At both time points, Russian participants had the lowest mean score, while Czech participants had the highest. Frequency of contact with friends and relatives was higher among Czech participants, followed by Russian participants and, lastly, Polish participants, with a median of 7, 6 and 5, respectively. The interquartile range (IQR), indicating the variation of this variable, was the same for all countries. At baseline, a quarter of the total sample was classified as having probable depression. This proportion increased to 30.6% at follow-up.

Table 2 reports the prospective association between frequency of contact with friends and relatives at baseline and CASP-12 score at follow-up. For every one unit increase in contact frequency score, there was a 0.31 point (95% CI 0.24, 0.37) increase in CASP-12 score, after controlling for demographic factors (Model 1). According to tertiles of contact frequency, people in the medium and high tertiles had a higher CASP-12 score by 1.12 (95% CI 0.76, 1.47) and 1.53 (95% CI 1.17, 1.90) units, compared to those in the low tertile. The addition of socioeconomic and health-related covariates in the model reduced these coefficients, but they remained statistically significant (Model 2). For every one unit increase in contact frequency score, CASP-12 score increased by 0.21 (95% CI 0.15, 0.27) points. CASP-12 scores for participants in the medium and high tertiles of frequency of contact with friends and relatives remained higher by 0.77 (95% CI 0.44, 1.10) and 1.01 (95% CI 0.67, 1.36) units, respectively, compared to the reference group. Finally, the inclusion of baseline CASP-12 in the model further attenuated the prospective associations (Model 3), as every one unit increase in frequency of contact score was significantly associated with a smaller increase of 0.12 (95% CI 0.06, 0.17) in CASP-12 score at follow-up. The fully adjusted results, however, showed a strong positive association as CASP-12 scores for participants in the medium and high groups were greater by 0.50 (95% CI 0.19, 0.82) and 0.59 (95% CI 0.26, 0.92) units, compared to the reference group.

Figure 2 displays the path model results. After accounting for all other paths in the model, the prospective effect of frequent contact with friends and relatives on QoL remained statistically significant. The direct effect showed that per every one unit increase in contact frequency score at baseline, the CASP-12 score increased by 0.11 (95% CI 0.053, 0.16) points at follow-up. The indirect paths were also statistically significant. For every one unit increase in contact frequency, the odds of being a possible case of depression at baseline were reduced by 9% (95% CI 0.89, 0.93). Subsequently, possible cases of depression at baseline were 3.32 (95% CI 2.94, 3.75) times more likely to be possible cases of depression at follow-up. Possible cases of depression at follow-up had, on average, a CASP-12 score that was 4.14 points lower (95% CI − 4.42, − 3.85) compared to non-cases at follow-up, accounting for the other paths and covariates in the model.

The total effect of a unit increase in frequency of contact with friends and relatives on CASP-12 score at follow-up was 0.57 (95% CI 0.42, 0.72). The indirect effect, via depression was calculated at 0.46 (95% CI 0.33, 0.60), increases the CASP-12 score at follow-up (per every one unit on exposure), suggesting that approximately 80% of the total effect was mediated through possible depression at baseline and follow-up.

Participants who were excluded from the analysis due to missing data on baseline covariates and QoL at follow-up showed some significant difference from the analytical sample (Supplementary Table 2). Excluded individuals tended to have less frequent contacts with friends and relatives and were more likely to have possible depression at both time points compared to the analytical sample. Excluded individuals also had a slightly lower QoL score at both baseline and follow-up, alongside lower levels of educational attainment and less time spent on physical activity and lower proportion of smokers.

Repeating the linear regression model 1 on a larger share of individuals with complete information on the model 1 covariates generated results that were consistent with the main model 1 results (Supplementary Table 3). We estimated the cross-sectional association using nested linear regression models on all participants with complete data on baseline QoL and covariates. While the cross-sectional effect sizes were slightly larger than the prospective effect sizes in the main analyses, the difference in magnitude was not large enough to indicate that the prospective association was appreciably underestimated (Supplementary Table 4). Additional prospective analyses using separate contact frequency scores for friends and relatives indicated that both contact with friends and contact with relatives have independent but similar beneficial effects on QoL (Supplementary Table 5).

Finally, the path analysis considering only possible cases of depression at baseline showed a smaller indirect effect compared to the analysis using possible depression at both waves. The indirect effect of baseline depression, however, was strong and significant at 0.20 (95% CI 0.11, 0.29), and attributed to 44% of the model’s total effect (0.45 95% CI 0.29, 0.61) (Supplementary Fig. 1 ).