Eligibility criteria
We will include randomised controlled trials (RCTs) with a comparator group examining the impact of a KT intervention targeting CDM in older adults after more than 1 year of implementation or after the termination of research/project funding is described. To examine cost, all economic studies (e.g. cost studies with a relevant comparator, cost-effectiveness analyses) will be included.
The target population for the CDM intervention includes patients (aged 65 years and older with one or more chronic disease including noncommunicable diseases) [
17] or their caregiver. End-users of the KT intervention will include patients aged 65 years and older with one or more chronic disease, their caregivers, clinicians (from all disciplines), public health officials (including medical officers of health, department chairs, programme managers), health care managers and policy-makers (including regional, state/provincial, federal).
The CDM intervention is the clinical intervention such as use of exercise in a patient with type 2 diabetes; the KT intervention is the strategy used to support implementation of the CDM intervention, such as a reminder system for patients to exercise or motivational interviewing for clinicians to promote patient exercise. CDM interventions may target any chronic condition and may target the patient, caregiver, clinician or health system, with a goal to optimise health and well-being of the patient. All comparators are eligible for inclusion including other KT interventions or usual care.
Our primary outcome will be sustained implementation of the KT intervention beyond 1 year after implementation or the termination of funding as this was determined by our knowledge users to be relevant for decision-making in their contexts. Additionally, as secondary outcomes, we will consider sustained delivery of the CDM interventions as well of any patient (e.g. quality of life), clinician (e.g. eye examination frequency) or health care system (e.g. hospital admission) outcomes resulting from sustained behaviour change. Studies in all settings will be eligible, including primary and specialist care; acute and long-term care; inpatient and outpatient care; and regional, national, and international settings. Both published and unpublished material will be included, as well as those disseminated in any language.
Charting and outcome selection
The outcomes and outcome measures reported in all identified studies will be charted in Excel. All knowledge users (patients, clinicians, health care managers, policy-makers) on the team will then review this list and select the outcomes that are most relevant for their decision-making. We will aim to identify equal numbers of participants from each KU group, and in the case of non-response KUs will be asked to nominate an individual to replace them on the panel. Specifically, a modified Delphi approach [
18] commonly used in quality improvement research [
19] with two rounds will be used to achieve consensus on what KT and patient, clinician and health system outcomes should be considered for inclusion. First, an online survey using Qualtrics [
20] will be sent to the knowledge users, which will include the list of outcomes, their definitions, and the frequency with which they appeared in the studies. The knowledge users will be invited to rate the importance of each outcome on a 7-point Likert scale ranging from ‘not at all important’ to ‘extremely important’. Second, their ratings will be aggregated and median ratings will be calculated. Outcomes with median ratings of ≥ 5 and appropriate levels of agreement among KUs (e.g. ≥ 75% of ratings must fall within three points of the median score) will be considered important outcomes and will be included in the second round. Third, in the second round, the knowledge users will assess the shortened list of preferred outcomes and will be asked to rank their top 3 outcomes. This process will be used to identify the primary and secondary outcomes by selecting the three outcomes with the highest ranking.
We will conduct a systematic search of the published and difficult to locate or unpublished (i.e. grey) literature within health. Health will be defined using the World Health Organization (WHO) definition [
21]. The following electronic databases will be searched from inception onwards: MEDLINE, EMBASE, Cochrane Central Register of Controlled Trials, Cumulative Index to Nursing and Allied Health Literature (CINAHL) and Campbell databases. A search of the grey literature will be conducted using a strategically developed Google search strategy, using guidance from the Canadian Agency for Drugs and Technologies in Health Grey Matters tool [
22]. We will search the websites of key funding agencies (e.g. Canadian Institutes of Health Research) and health care provider organisations from Australia, Canada, the UK and the USA who have similar health care systems or similar challenges related to CDM. References from included studies and relevant articles will be scanned to ensure literature saturation. Team members will use their linkages with experts in the field to identify additional articles.
Search strategy
The main (i.e. MEDLINE) search strategy was developed by an experienced librarian, circulated to the team and revised, as necessary. The search used for our previous scoping review on this same topic was used as the basis for this search [
23]; however, it was substantially modified to include various categorisations of potential KT interventions including the Effective Practice and Organisation of Care (EPOC) taxonomy of health systems interventions [
24]. The search was organised according to the research question and PICO and included both MeSH and keyword terms for chronic medical diseases, specific MeSH and keyword terms for EPOC-related concepts and concepts related to knowledge translation and sustainability. The search was limited to humans and validated filters were used for older adults and randomised trials [
25].
The search for the current review was then peer reviewed by a second experienced librarian using the Peer Review of Electronic Search Strategies (PRESS) Checklist [
26]. The final search strategy can be found in Additional file
2. The search was adjusted and translated for other databases, as necessary, and these other search strategies are available from the authors upon reasonable request.
Study selection
The eligibility criteria will be pilot-tested on a random sample of 50 titles and abstracts from the literature search. All team members will screen these citations using the eligibility criteria independently and conflicts will be discussed. The eligibility criteria will be revised if deemed necessary by the team or if low agreement (< 90%) is observed. Two team members will then screen each citation independently using Synthesi.SR [
27]. Similarly, we will conduct a calibration exercise of the eligibility criteria prior to screening potentially relevant full-text articles, which will then be screened by two team members independently. At both the citation and full-text levels of screening, conflicts will be resolved via discussion among pairs or reviewers or with a third member, if required.
Data items and data collection process
Two investigators will independently read each article and extract relevant data when available. We will abstract data at the following levels:
1.
Study level: study design, year of study conduct, sample size, setting, country of study conduct, study funding source and KT theory/model/framework used to inform study
2.
Patient level: type and number of patients, age mean and standard deviation, proportion of female participants, type of chronic condition(s) and number of clusters
3.
Intervention level: type (e.g. KT interventions will be classified using a behaviour change techniques taxonomy) [
28], frequency and duration of CDM and KT interventions, provider and target
4.
Outcome level: patient, clinical and health system including cost-effectiveness (e.g. incremental cost-effectiveness ratios (ICERs) cost per quality adjusted life year (QALY))
Prior to data abstraction, we will calibrate our data collection form on a random sample of five full-text articles. Each team member will extract the data, and the team will meet to discuss conflicts. The data collection form will be revised for clarity, as needed. Subsequently, two team members will conduct all data collection for each study independently. Study authors will be contacted for further information as needed when considering studies for inclusion and when conducting data abstraction. When multiple studies report data from the same study population, the study with the longest follow-up and available data will be considered the main publication and the others will be retained for supplementary material only.
Two reviewers with expertise in KT and research methods will independently code each KT intervention first using the pre-existing taxonomy developed by the Cochrane EPOC group and then a behaviour change technique taxonomy to identify the specific active components in each intervention [
24,
28]. Conflicts in KT intervention coding will be resolved through discussion. We will use this coding structure to create the nodes for the NMA by ‘lumping’ interventions according to the taxonomy categories and have used this approach in other NMAs of complex interventions [
29].
Methodological quality/risk of bias appraisal
Two reviewers will conduct risk of bias independently on each included study. If there is disagreement, a third reviewer will be available. The risk of bias of RCTs will be done using the Cochrane EPOC Risk of Bias tool [
30], and cost-effectiveness analysis studies will be evaluated against the reporting checklist developed by Drummond and colleagues [
31]. For outcomes reported in 10 or more studies, small-study effects (e.g., publication bias) will be assessed using comparison-adjusted funnel plots [
32].
Synthesis of included studies
We will describe the study characteristics, patient characteristics, outcome results, the methodological quality and risk of bias. We will report the results using the PRISMA extension for NMAs [
33].
We anticipate clinical and methodological heterogeneity and thus will conduct random-effects meta-analysis in a Bayesian framework. We will assess heterogeneity of the included studies in terms of clinical (e.g. patient characteristics), methodological (e.g. study design) and statistical (e.g. heterogeneity in study outcomes between studies) characteristics. For example, clinicians on the team will assess clinical heterogeneity and methodologists will assess study heterogeneity. Statistical heterogeneity will be assessed by visual inspection of each meta-analysis forest plot, by estimating the between-study variance and by using the
I2 statistic [
25,
34]. If extensive heterogeneity is observed, we will try to explain this via sub-group analysis and meta-regression analysis [
35].We will use vague priors for all model parameters aside from the between-study variance for which we will use the informative priors suggested by Turner and colleagues for dichotomous data and by Rhodes and colleagues for continuous data [
36,
37]. In each NMA, we will assume a common within-network between-study variance across treatment comparisons and we will use an informative prior as suggested in Turner et al. [
36], Rhodes et al. [
37] and Nikolakopoulou et al. [
38]. Multi-arm studies will be included in a head-to-head meta-analysis using the exact adjustment method as described by Rucker et al. [
39]. The meta-regression analyses will be conducted when 10 or more studies are available for the underlying outcome and intervention comparison and will examine the influence of factors such as age and severity of chronic disease and risk of bias on the meta-analysis results.
For continuous outcomes, to make use of all data, we will impute missing measures of variance using established methods [
40]. To ensure our imputations do not bias our results, we will conduct a sensitivity analysis to examine the missing data under random, completely at random and non-random assumptions.
The transitivity assumption will be assessed visually to ensure that potential effect modifiers (e.g. patient age, comorbidities) are balanced on average across comparisons [
41]. We expect that transitivity will be valid if the common intervention used to compare different KT interventions indirectly is similar when it appears in different trials. Consistency of the entire network will be assessed with the design-by-treatment interaction model [
42]. If inconsistency is found within the network, the loop-specific approach will be used to assess local inconsistency of the loops within each network [
43]. If the assumption of transitivity is valid and the evidence forms a connected network, we will conduct a NMA using a Bayesian random-effects model.
Across all analyses, the summary intervention effects will be presented using the odds ratio or mean difference with the corresponding 95% credible intervals. Predictive intervals will also be reported for meta-analyses and NMA point estimates. We will assess model convergence using the Gelman-Rubin diagnostic and goodness of model fit [
44] will be assessed using the posterior residual deviance, the degree of heterogeneity, and the Deviance Information Criterion (DIC) [
45]. A well-fitting model has a residual deviance close to the number of data points. A difference of ≥ 3 units in DIC is considered important and the lowest DIC value implies the best fitting model. Mean ranks and the surface under the cumulative ranking curves (SUCRAs) will be used to derive a relative ranking of the KT interventions based on the NMA results [
46]. Rank-heat plots will be used to display the intervention rankings across multiple outcomes [
47].
Network subgroup and meta-regression analyses will be done to explore potential effect modifiers that impact transitivity or consistency assumptions when ten or more studies are available in the underlying outcome. If there are a sufficient number of studies, we will explore the following potential effect modifiers: sex, type of chronic disease, cluster of diseases (as reported in the studies such as diabetes and depression), severity of chronic diseases, care setting, and duration of follow-up. Sensitivity analysis will be done whereby we use data from the following studies in the NMA: (1) studies at low risk of bias based on the two components of our risk of bias assessment found to be of the greatest threat to study validity and (2) studies with high participant retention. Since the NMA is dependent on different priors for variance parameters included in the Bayesian approach, we will conduct a sensitivity analysis using vague priors [
48]. All analyses will be performed within OpenBUGS, except for the consistency assessment that will be conducted in Stata using the
network command [
49,
50]. To ensure reproducibility, code listings or citations providing guidance on the methods employed in this NMA will be available upon request.