Study design
Following the methodology developed in Sawyer et al. [23], a systematic review was conducted in order to obtain an evidence base from which to derive the CLDs. The protocol for the review was registered with PROSPERO prior to conducting the review (CRD42021244927). Using the selected evidence base, a conceptual systems map was constructed as a series of sub-system CLDs to explicate non-linear system mechanisms, such as reinforcing and balancing feedback loops. Unlike a traditional systematic review, the aim of this review was not to include, appraise and report all evidence published on this topic. Rather, it was to draw on a comprehensive, non-biased set of peer-reviewed publications to develop a conceptual model of the complex influences on physical activity in a specific population sub-group of people from ethnic minority groups.
Systematic review of the literature
The systematic review of the literature was an update of Langøien, Terragni, Rugseth et al.’s 2017 review of the determinants of physical activity in ethnic minority groups living in Europe [24]. The same search protocol was used for the updated search. A pluralist approach to evidence selection (i.e. inclusion of different study designs) suited the current review’s exploratory research question and is useful in building an evidence base from which to understand complex relationships between multiple factors [23, 25]. The approach used in the current review draws on: conventional methods for systematically reviewing peer-reviewed literature; methods for reviewing the literature from a realist perspective to understand mechanisms underlying relationships [26]; and emerging methods and approaches used to conduct systems-based literature reviews [23, 27].
The boundaries of the system under study (i.e. the scope of the research question) as well as the choice for Langøien et al.’s systematic review informed the inclusion and exclusion criteria and the population, outcomes and determinants of interest [24]. Given the central aim of synthesising the current evidence of underlying mechanisms of physical activity in ethnic minority groups living in Europe, we used the following inclusion and exclusion criteria:
Population
Our targeted population included children and adolescents (2–18 years), adults and older adults (18 + years) living in Europe, with a familial migration background from any low- and middle-income countries (including the former Eastern European Bloc countries) or from minority indigenous populations in Europe.
Outcomes
All outcomes pertaining to non-work related physical activity of light, moderate or vigorous intensity performed in any setting were included. Physical activity is defined as energy expenditure resulting from bodily movement created by skeletal muscles [2].
Determinants
Results from Langøien et al.’s [24] systematic review were used in a concept mapping exercise by Holdsworth et al. [28], resulting in 8 clusters of factors of physical activity: psychosocial, institutional environment, political environment, social and cultural environment, physical environment and opportunity, social and material resources, health and health communication, migration context. Determinants related to these clusters were considered relevant to the research question.
Inclusion criteria
-
• Studies with a target population of (an) ethnic minority group(s): “immigrants and their offspring/populations of immigrant background (not differentiating on their migration status) from low and middle income countries, population groups from the former Eastern European Bloc countries who migrate to other parts of Europe and minority indigenous populations in Europe” [24].
-
• Studies with a target population of a ‘majority’ ethnic group and (an) ethnic minority group(s) were included where sub-analyses were reported for (an) ethnic minority group(s).
-
• Children, adolescents and young people, adults, and older adults (age ranges as characterised by study authors);
-
• Qualitative or quantitative studies using non-experimental study designs or quasi-, controlled or natural experimental study designs; the combination of different study designs enables us to build CLDs and take a systems perspective;
-
• European setting;
-
• Studies exploring determinants (independent variables) in relation to migration background and/or ethnicity; i.e. studies must study determinants or mechanisms of physical activity by migration background and/or ethnicity.
-
• No restrictions on language or date of publication.
Exclusion criteria
-
• Studies that only describe (differences in) physical activity levels by migration background and/or ethnicity;
-
• Studies that examine physical activity in the relationship between disease and migration background and/or ethnicity, without examining determinants of physical activity;
-
• Studies that examine interaction between determinants of physical activity but not interaction between contextual factors and conversion factors (as described in Fig. 1);
-
• Literature reviews, position or conceptual papers were excluded from data extraction but included in search in order to conduct reference search;
-
• Grey literature.
Search strategy
Six databases (MEDLINE, EMBASE [Ovid], Web of Science, Cochrane Library, CINAHL, PsycINFO [Ovid]) were searched in March 2021 using predefined free-text words and MESH terms, modified for each database. Search terms were taken from Langøien et al. [24]; the only change was the removal of terms relating to sedentary behaviour and the addition of the term ‘physical inactivity’. The search terms for each database are reported in Supplementary File 1. AS performed the search and stored retrieved records using Rayyan QCRI software. Duplicated records were removed prior to screening.
Title, abstract and full-text screening was conducted by AS, MS, SF, KV, ANP and KWT. At each stage of screening, records were screened independently by two researchers and any disagreements between reviewers were assessed by a third researcher; further disagreements were resolved through discussion.
Data extraction
Included studies were stored in EndNote. Data were extracted by AS, CBMK, SF and MS. Data extraction was independently completed for 10% of included studies to check consistency and accuracy in the extraction process. Study characteristics were extracted for the studies which were identified through the updated search (i.e. were not included in Langøien et al. [24]) and checked for the studies which were included in the original review by Langøien et al. [24].
For each study, factors that were examined in direct relation or indirect relation (e.g. effect modification) to our outcomes of interest were extracted using their original labelling. Factors were sorted using the categories from Holdsworth et al. [28]. A new category was added by the reviewer if no suitable category was available; an example of a new category is ‘use of motorised private transport’. Where appropriate, new categories added by different reviewers were conflated by AS, retaining the original factor labels to enable double-review.
Following the categorisation of all factors, a selection of factors originally drawn from Holdsworth et al. [28] were relabelled to better reflect the included factors (e.g. ‘opportunities in life’ was relabelled ‘skilled work opportunities’ as factors within this category all related to employment opportunities). KS conducted a double-review of: variable names included in each category; creation and conflation of new categories; and the definition or relabelling of categories. Any disagreements were resolved in discussion with AS.
To document the reported relationships between factors, all relevant examined associations between factors were reported in a spreadsheet which had all factor categories listed in the first column and first row to create a matrix. In the appropriate cell (e.g. cell X by Y to report the relationship between a factor in category X and a factor in category Y), the reviewer noted: the study reference, the original factor name (rather than the category label), the original framing of the factor (e.g. ‘low levels of social support’ or ‘high levels of social support’), the direction of the studied relationship (the effect of X on Y, Y on X, a correlation, or a modifying effect), the reported significance of the relationship according to the measure of significance used in the original study (e.g. p-value, qualitative interpretation; reported as: significant, non-significant or a trend) and the physical activity outcome assessed in relation to these factors. Relationships between a single factor and a physical activity outcome was not recorded as the objective of this study was to examine the relationships between determinants of physical activity. A hypothetical entry could read: “ [19] low levels of parental support for physical activity ➔ (significant) low levels of motivation for physical activity ➔ (significant) reduced interest in physical activity”. Each reviewer received detailed guidance on data extraction and populated their own spreadsheet. Spreadsheets were collated by AS at the end of data extraction.
Quality assessment
As the aim of this literature review was to identify and elucidate mechanisms rather than make conclusive judgements (e.g. on clinical effectiveness), we took a pluralistic approach, synthesising qualitative and quantitative evidence from multiple types of study design. The quality assessment checklist by Kmet, Lee & Cook [29] allowed assessments of risk of bias in primary research using a range of study designs. Where physical activity in ethnic minority groups was a partial focus of a selected study, quality assessments were only conducted on the parts of the study which pertained to our research question. To ensure consistency, quality assessments were performed for all included studies, not just those obtained in the updated search. Quality assessments were independently conducted by AL, CW, EGB, AS and GR.
Data synthesis and CLD development
All factors and associations which, on balance across the selected evidence, were qualitatively or statistically significant were entered into a CLD using Kumu software. Factors (or CLD ‘nodes’) were colour-coded according to their overarching category [28]: ‘psychosocial’, ‘sociocultural’, ‘health and health communication’, ‘migration context’, ‘social and material resources’, ‘institutional environment’ and ‘physical environment and opportunity’. The direction and polarity of associations were recorded using connection arrows.
In line with the capability approach, factors within the categories of ‘institutional environment’, ‘migration context’ and ‘physical environment and opportunity’ were conceptualised as contextual factors. Factors within all other categories were conceived as characteristics of the target group and therefore conceptualised as conversion factors. It was theorised that the interaction between contextual factors and conversion factors produce an emergent set of capabilities which do or do not enable physical activity (Fig. 1). As previously outlined, this interaction was the primary interest for the current study.
Sub-system CLDs were created to understand the hypothetical relationship between the environmental context and conversion factors. Feedback loops and mechanisms within these CLDs were analysed in order to understand how this relationship could lead to the emergence of particular capabilities for physical activity. The choice to act on capabilities to achieve a certain functioning (i.e. physical activity outcome) was not in the scope of this study and therefore not within the boundaries of the CLD (light grey boxes in Fig. 1).
Separate sub-system CLDs were created for each category of conversion factors. Exogenous contextual factors, defined as factors that did not have incoming connections, were excluded. Because they were not connected to other factors, they cannot contribute to potential system dynamics of interest. By screening the results from the automatic detection of feedback loop using Kumu software, feedback loops involving a conversion factor and at least two but no more than 5 other (conversion or contextual) factors were identified for interpretation by co-authors with expertise in the relevant field. Feedback loops comprising more than 6 nodes were deemed too complex for qualitative interpretation.
CLD interpretation
An initial review session of all 4 sub-system CLDs was attended by AS, KS and FvL and used to check for possible human errors in the translation of the evidence to the CLD. This check involved inspecting factor labels and the direction and polarity of connection arrows. Subsequently, a separate review session for each sub-system CLD was held in June 2023, involving co-authors with demonstrable expertise in the field (e.g. expertise in psychosocial influences on physical activity). Sessions were held online and typically lasted 90 min. Prior to the review session, reviewers were asked to scrutinise a document including the sub-system CLD and each identified feedback loop presented separately. Reviewers were asked to inspect the CLD and feedback loops for errors and consider their initial interpretation of feedback loops.
During the review sessions, reviewers were first asked to report any identified errors or raise any queries about unexpected present or missing connection arrows. AS used the collated data extraction spreadsheet which noted linkage between connection arrows and the primary studies to check errors and queries in real time; where it was not possible to resolve the issue during the session, AS noted queries for further examination. Next, reviewers examined each feedback loop in turn, discussing interpretations and supporting evidence. Initially, AS presented each feedback loop as a set of connection arrows, drawing on the supporting evidence from the primary studies to explain the association between each node (to facilitate this, studies were cited for each connection arrow using the Notes feature in Kumu). The review groups then interpreted the feedback loops as presenting reinforcing or balancing mechanisms, where a reinforcing feedback loop indicates increasing growth or decline in the level of included nodes over time (e.g. increasing levels of social support) and a balancing feedback loop indicates stabilisation in the level of included nodes over time (e.g. stable levels of social support). Missing evidence was also discussed at reviewer sessions, identifying one systematic review; the primary studies included in this review were latterly screened and excluded by AS. The following co-authors took part in the review sessions: environmental, migration and psychosocial factors sub-system: AS, KV, EGB, ANP, CW; environmental, migration and sociocultural factors sub-system: AS, AL, GR, SF, EGB; environmental, migration and health and health communication factors sub-system: AS, LL, LT, KWT, EGB; environmental, migration and social and material resources sub-system: AS, EGB, CBMK, SF.
Add Comment