This is a longitudinal analysis of social participation, which was examined as a secondary outcome within the AgeWell.de trial, a cluster-randomized trial of a primary care multi-domain intervention against cognitive decline conducted in Germany between June 2018 and January 2022. The respective study protocol, baseline characteristics, and main results are described in detail in previous publications [5,6,7]. Study participants were recruited from general practitioner (GP) offices at five trial sites across Germany, with the GP offices serving as randomization clusters. The target population were community-dwelling adults aged 60–77 years, who were at increased risk for developing dementia based on the “Cardiovascular Risk Factors, Aging, and Dementia” (CAIDE) risk score. The multi-domain intervention was delivered by trained study nurses as three face-to-face and five telephone appointments. It targeted social, cognitive, and physical activity, as well as optimization of nutrition and medication. The social participation intervention consisted of setting and reviewing individual goals for social activities during the visits, and encouraging participants to engage in social activities such as community events or family gatherings using a motivational interviewing technique. Participants in the control group received GP treatment as usual and general health advice.

Out of 1,030 participants at baseline, 819 completed the follow-up assessment at 24 months. The variables shown in Table 1 were used for the analyses in this study. The 6-item version of the Lubben Social Network Scale (LSNS-6) [9, 10] and self-reported social activity represented the main outcomes, along with sociodemographic variables, the Montreal Cognitive Assessment (MoCA) and Geriatric Depression Scale (GDS) used as covariates [11,12,13].

Social participation can be defined as “engagement of individuals in social leisure activities (focusing on activities undertaken with other people), contact with social networks and their satisfaction with this participation” [3]. Social networks play a crucial role in facilitating social participation, which encompasses a broad range of activities, such as group recreational activities, social events, or volunteering [14]. For the purpose of this study, our operational definition of social participation focuses on contact with social networks and engagement in social activities based on the available data and instruments: The LSNS-6 measures the quantitative aspect by assessing the size and depth of social networks, while Levasseur’s taxonomy evaluates the qualitative aspect by categorizing social activities based on their nature and the goals they fulfill, thus reflecting the meaning and depth of social involvement.

The LSNS-6 is a validated and commonly used tool to assess social networks among older adults. It captures the self-reported number of social contacts, as well as support and trust with these connections. The LSNS-6 consists of six questions, one set of three for family and one for friends, respectively: How many relatives/friends do you see or hear from at least once a month? How many relatives/friends do you feel at ease with that you can talk about private matters? How many relatives/friends do you feel close to such that you could call on them for help? The questions can be answered on a 5-point Likert scale as follows: 0 (None), 1 (One), 2 (Two), 3 (Three or four), 4 (Five through eight), 5 (Nine or more). Out of a maximum score of 30 points, a value smaller than 12 points has been defined as a cut-off indicating risk for social isolation [9].

As previously described for the analysis of baseline data [15], we additionally assessed social participation according to the framework described by Levasseur et al. [16], which has also been used in other clinical trials [17, 18]. It is based on the goal of the activity and the interaction with others, with more engagement reflected in higher levels of social involvement. Participants were asked about their social participation based on a list of activities, which consisted of social hobbies and events, as well as engagement in a local church, clubs or other volunteer occupations. These self-reported social activities were classified based on the highest ranking answer as “high involvement” if they had a higher level of engagement with a common goal, “low involvement” if they included regular past-time social interactions, and “no involvement” if participants did not regularly engage in any social activities. For example, social hobbies like restaurant or theater visits represented low involvement, while engagement in organizations or other volunteer occupations represented a high level of involvement.

To account for the Covid-19 pandemic in Germany, an additional questionnaire was used to gather data between January and May 2021, assessing the participants’ attitudes and subjective restrictions due to pandemic containment measures. The questionnaire item on social restrictions was used for adjustments in this analysis. Furthermore, a variable indicating the phase during which a participants’ follow-up exam took place, based on the waves of Covid-19 variants of concern according to the German public health institute (RKI) [19], was introduced. Both variables are summarized in Table 2.

Missing data at the single item level amounted to 1–2% for social participation questionnaires, MoCA and GDS scores at baseline, 2–5% for social participation questionnaires at follow-up, and 15% for the Covid-19 questionnaire item. Missing items were completed using a hot deck imputation procedure conditioned on age, gender, and education [20]. The complete case sensitivity analysis, shown in Table A1, did not reveal any relevant differences between observations with complete or incomplete information for the primary outcomes.

Descriptive statistics were used to summarize the baseline characteristics of the study sample. The outcomes of interest for this study were defined as (1) the between-group difference in mean change of the LSNS-6 score at 24-months follow-up, assessed using a two-sample t-test, (2) being at risk for social isolation as a binary outcome, indicated by a LSNS-6 score < 12 points, and (3) the social involvement framework classification as a three-tiered categorical outcome, both compared between groups at 24-months follow-up using Fisher’s exact test. All assessments were done under the intention-to-treat principle as the primary analysis. The mobility of study participants within the social involvement framework from baseline to follow-up was visualized using a Sankey diagram (Fig. 1). [21]

Sankey diagram for the flow of study participants within the social participation framework from baseline (T1) to 24-month follow-up (T2). Social involvement levels based on the framework by Levasseur et al.: None (no regular social contact) - Low involvement (Restaurant visits, sports groups, etc.) - High involvement (Engagement in clubs or volunteer organizations). LTF: Loss to follow-up

For confirmatory analyses, we then calculated multi-level mixed effects generalized linear regression models, which were adjusted in multiple ways to address potential bias. For the LSNS-6 score, a Gaussian distribution and identity link function were applied, for social isolation, a binomial distribution with logit link was chosen, and an ordered logistic model was applied to the three-tiered framework classification of social involvement, with robust standard errors reported for all models. To account for the cluster-randomized study design, 117 clusters at the GP office level were introduced as random effects. All models were further adjusted for the respective baseline value of the dependent variable, sociodemographic factors (age, gender, education, work, relationship status), as well as depressive symptoms and mild cognitive impairment, all thought to be associated with social participation. A three-tiered factor variable of treatment allocation and protocol adherence, based on goal achievement in the main intervention components [7], was chosen as the main predictor to allow for a per-protocol analysis.

Additionally, the drop-out analysis revealed differences in key variables for those participants lost to follow-up (Appendix, Table A2). To account for potentially resulting informative missingness, a logistic model for study completion, adjusted for significant predictors, was used to derive stabilized inverse probability weights for regression analyses (Appendix, Table A3).

Regarding the Covid-19 pandemic in Germany, the association between the pandemic and social participation outcomes at follow-up was visualized, showing a negative trend (Fig. 2, Panel A). Consequently, regression models were adjusted for pandemic effects using the variables shown in Table 2. The fully adjusted models were also used to calculate average marginal effects for the pandemic phases as a factor variable, resulting in Panels B-D shown in Fig. 2.

Social participation during the Covid-19 pandemic. Panel A: Scatterplot of LSNS-6 scores at 24-month follow-up throughout the follow-up period between July 2020 and April 2022. ß=-0.11 (SE = 0.044, p = 0.011). Panels B – D: Average marginal effects on given outcomes at 24-month follow-up during the respective pandemic phases in Germany between May 2020 and May 2022 [19] as listed in Table 2, obtained from fully adjusted regression models shown in Table 4. Panel B shows the change in LSNS-6 scores from baseline, and Panels C and D show the probability of observing the denoted level of social involvement or risk for social isolation (LSNS-6 < 12) at the respective pandemic phase

All statistical analyses were performed using Stata 15.1 (Stata Corp, College Station, TX).