Xuxi Zhang

frailty cannot be inferred. In Chapter 5 , the process components of the UHCE approach implementation were evaluated using the Steckler and Linnan framework. One reason to develop the process evaluation was to explain why certain effects were found in the effect evaluation that was published previously and pinpoint to components of interventions that were effective. 22 However, the complex interventions may include the interplay of multiple components. In Chapter 6 ‐ 7 , the design and conduct of RCTs could affect the meta ‐ analyses. For example, the non ‐ blinding of trial treatment may lead to biased assessment of some outcomes. 37 None of the RCTs included in the meta ‐ analyses was a double ‐ blinded study, with the exception of one RCT in Chapter 6. In Chapter 8 , we proposed an evaluation design for the Social Engagement Framework for Addressing the Chronic ‐ disease ‐ challenge (SEFAC) project. We suggested to apply a 6 ‐ month pre ‐ post design, using the baseline measurement as the ‘control group’. However, this pre ‐ post design is relatively weak in terms of internal validity, because it does not eliminate the possibility that the posttest results might have occurred regardless of the intervention. 38 Study population All studies in this thesis, with the exception of Chapter 6 and 7, relatively healthier participants may have enrolled into the studies which potentially caused selection bias. In Chapter 5 , we used qualitative data from focus groups with older persons and professionals. Older persons included in the focus groups might have been those who were most positive about the UHCE. Additionally, there were relatively many missing data for questions on satisfaction of the UHCE approach. Participants who did not reply to these questions could have thought these questions were not applicable to them or were the people who were less involved. Measurements In all studies of this thesis, people from different countries were included, and cultural differences in the interpretation of questions might have caused some variation between countries. Most data collected in all the studies, with the exception of Chapter 6 and 7, were based on self ‐ reported questionnaires which could have led to response bias to some extent. For example, participants may have been tempted to provide socially desirable responses. 39 Furthermore, participants may report inaccurate data because they cannot remember or omit details, such as the frequency of alcohol use, smoking and physical activity. This problem, known as recall bias, is a potential weakness in studies that use self ‐ reporting. 40 Frailty was measured by the TFI in Chapter 2 ‐ 5. In Chapter 2 , cut points of frailty and its three domains of the TFI were applied to explore the association between frailty and HRQoL (compared to using continuous scores), which might cause information loss. However, we performed analyses on the association between continuous frailty scores and HRQoL and found similar results. In Chapter 4 , the socio ‐ cultural and language differences in the interpretation of individual items of the TFI between countries were not assessed. Most of the alternative measures chosen to examine convergent and divergent validity and 9 227 General discussion