*2.4. Other Study Variables*


with others in their home, helping others, doing something outside their home with others, and doing something outside their home alone.

• Fear of falling was assessed with a dichotomous yes/no question.

### *2.5. Analysis Design and Statistics*

Data on participant age showed a non-normal distribution and was therefore split into three age categories: category 1 (age < 75 years), category 2 (age 75–90 years) and category 3 (age > 90 years). Category 2 (75–90 years) was used as the reference category in the regression analysis.

Living conditions were reclassified into two response alternatives. "Living alone and having no close relationship with a partner", "Living alone and having a close relationship with a partner" and "Other" were classified as "Living alone". "Living together with partner/family" was classified as "Living together". The scores from MoCA, ADL-staircase and UIMH (all three aspects) were reclassified from continuous to categorical variables. This was done to describe and visualize differences between groups. MoCA scores were classified into moderate, mild, or normal, and ADL-staircase and UIMH scores were divided into quartiles.

Descriptive statistics included independent T-tests and ANOVA examining the variables associations to health-related quality of life as measured separately by the EQ-5D-5L and the EQ VAS. Variables potentially associated with the outcomes (*p*-value below 0.2 at bivariate level) were then included in the multiple regression analysis by stepwise-forward selection, starting with variables that significantly differed between groups in the bivariate analysis. The independent variables were gender, age, living conditions, cognitive impairment, ADL dependence and usability of the home. We also tested the influence of items related to frequency and satisfaction with participation, and fear of falling but since they showed no potential association with the outcomes of interest, these variables were not included in the final analysis.

Associations with health-related quality of life were analyzed with multiple linear regression analysis using robust standard errors [30]. We used the EQ-5D-5L and the EQ VAS separately as the dependent variables of the regressions. Level of significance was set to *p* < 0.05. Post-estimation diagnostics was conducted by checking normality of residuals, the link test and measuring the information criteria of the models. Statistical analysis was performed using SPSS Statistics 25.0 (IBM Corporation, Armonk, NY, USA) and Stata 15.0 (Statacorp, College Station, TX, USA, 2015).
