Criteria for considering studies for the review
We will include studies examining people with multi-morbidity, of which diabetes and/or hypertension is one, living in LMICs, irrespective of age. We define multi-morbidity as the presence of two or more chronic medical conditions in an individual [
8]. Studies addressing both adults and children will be considered. LMICs will be defined according to the classification of the World Bank [
33].
This review will only consider studies that describe integration of service delivery at PHC and community level. We will consider models of partial integration and full integration of service delivery (Fig.
1). Partial integration of service delivery will be defined as models where patients treated for diabetes, hypertension or any other chronic disease receive part of the package of care (prevention, diagnosis, treatment) for another disease. Full integration of service delivery will be defined as models where patients (primarily treated for diabetes, hypertension or any other disease) receive the full package of care (prevention, diagnosis and treatment) for diabetes/hypertension and any other chronic disease at the same point of care by one or more healthcare professionals.
The main comparison will be stand-alone models of care, defined as models of care that are limited to one disease. We will include the following comparisons:
-
◦ Fully integrated models of care vs. stand-alone care
-
◦ Partially integrated models of care vs. stand-alone care
-
◦ Fully integrated models of care vs. partially integrated models of care
We will include studies that report on either primary health outcomes or secondary outcomes. However, the absence of reporting the pre-specified health and process outcomes will not be a deciding factor for inclusion of studies. Types of outcomes to consider are the following:
Secondary outcomes (process outcomes)
-
▪ Access to care as reported in the included studies
-
▪ Retention in care and adherence as reported in the included studies
-
▪ Continuity of care as reported in the included studies
-
▪ Quality of care as reported in the included studies
-
▪ Cost of care as reported in the included studies
We will consider randomised controlled trials (RCTs), including cluster RCTs, controlled (non-randomised) clinical trials (CCTs) or cluster trials, interrupted time series (ITS) studies with at least three data points before and after the intervention, and controlled before-and-after (CBA) studies for inclusion. Cluster randomised, cluster non-randomised or CBA studies will be included only if there are at least two intervention sites and two control sites. Cross-sectional studies, case series and case reports will be excluded.
Search methods for identifying studies
We will search the following electronic databases: MEDLINE (PubMed), EMBASE (Ovid), the Cochrane Central Register of Control Trials (CENTRAL), LILACS, Africa-Wide Information (via EBSCO host), CINAHL, Web of Science (Core collection). For ongoing studies, we will search the following trial registries: WHO ICTRP and Clinicaltrials.gov. We will search conference abstracts from the International AIDS Society Online Resource Library, the HIV/AIDS Implementers’ Meetings and the NCDs Alliance meetings.
Search terms will include ‘diabetes’, ‘hypertension’, ‘comorbidities , ‘integrated healthcare delivery’, ‘low- and middle-income countries’ and their synonyms. The full search strategy for MEDLINE (PubMed) is provided in Additional file
2. We will adapt it for other electronic databases. We will report all search strategies in full in the final version of the review. In addition, we will screen reference lists of included studies and reference lists of relevant systematic reviews, and contact experts in the field and relevant organisations (e.g. NCD Alliance) for unpublished studies. All languages will be included.
Study selection and data extraction
Two authors will independently screen titles and abstracts of studies identified by the search using Covidence software, and we will retrieve full-text of all potentially eligible studies. Two authors will independently screen full texts for eligibility. Discrepancies in the selection process will be resolved through discussion or by consulting a third author. Studies will be classified as included, excluded or awaiting assessment. We will provide reasons for excluding studies. Studies only available as abstracts will be included in the general results of the review, but not in the analysis.
We will extract data based on the description of various models of integrated care as illustrated in Fig.
1 with consideration of possible scenarios of various packages of care provided in terms of partial integration. A pre-specified, standardised and piloted data extraction form will be used. Two authors will independently extract data and compare the results. Discrepancies will be resolved by discussion or by consulting a third reviewer. We will contact study authors in case of missing data. We will extract data on the participants, intervention, comparisons, outcomes, setting, context and funding sources.
Data items will be in line with recommendations from the template for intervention description and replication (TIDieR) [
34] and the PRISMA-Complex Interventions (CI) extension checklist [
35]. We will look at the following items:
-
▪ Provide name or a phrase to describe the intervention
-
▪ Rationale, theory or goal of the element essential to the intervention
-
▪ Materials used
-
▪ Procedures or processes used in the intervention
-
▪ Who provided—each category of the intervention provider
-
▪ Describe modes of delivery of the intervention
-
▪ Types of location where the intervention took place
-
▪ When and how much—number of times the intervention was delivered
-
▪ Tailoring of the intervention
-
▪ How well the intervention was planned
-
▪ Replicability
For the ‘implementation components’, we will look at the following items:
Integration strategies may include facilitators (distinct from intervention elements) such as attestations, financial incentives, periodic reports of findings, reminders, supplemental trainings or physical environmental changes.
Where there is insufficient detail reported in the study, we will contact the study investigators for clarity or more information on the study. Disagreements will be resolved first by discussion and then by consulting at third author for consensus.
For randomised controlled trials, non-randomised trials and controlled before-after studies, we will use the tool proposed by the Effective Practice and Organisation of Care (EPOC) group [
36]. We will assess the following nine domains as having ‘low risk’, ‘high risk’ or ‘unclear risk’ of bias: (1) random sequence generation, (2) allocation concealment, (3) baseline outcome measurements, (4) baseline characteristics, (5) incomplete outcome data, (6) knowledge of allocated intervention (blinding), (7) protection against contamination, (8) selective outcome reporting and (9) other risks of bias. For cluster RCTs, we will assess additional risk of bias linked to recruitment, baseline differences, loss of clusters, incorrect analysis and compatibility with RCTs randomised by individuals. For ITS studies, we will also use the tool proposed by EPOC to assess whether: (1) the intervention was independent of other changes, (2) the shape of the intervention effect was pre-specified, (3) the intervention was unlikely to affect data collections, (4) knowledge of the allocated intervention was adequately prevented during the study, (5) incomplete outcome data was likely to bias results, (6) outcomes were reported selectively and (7) there were any other risks of bias. For each domain, risk of bias will be assessed as low, high or unclear. Two authors will independently assess risk of bias of included studies. We will resolve discrepancies through discussion or consulting a third author.
Data analysis
We will extract relevant outcome data for each study and enter it into Review Manager 5. For dichotomous outcomes, we will calculate risk ratios (RR) and report pooled effects with 95% confidence intervals. For continuous outcomes, we will calculate the mean differences (MD) if outcomes were measured in the same way across studies or standardised mean differences (SMD) where outcomes were measured differently across studies. We will report pooled effects with 95% confidence intervals.
Where cluster RCTs have appropriately adjusted for the effects of clustering in their analysis, we will use these adjusted effect estimates and standard errors in our meta-analysis using the generic inverse-variance method in Review Manager 5 [
37]. Where the included cluster RCTs did not perform any adjustment for clustering, we will adjust the raw data ourselves using the intra-class correlation coefficient (ICC). If the study authors do not report an ICC value in the published article, either we will obtain this value from similar studies or we will estimate the ICC value. We will not present results from cluster RCTs that were not adjusted for clustering. If we estimate the ICC value, we will perform sensitivity analyses to investigate the robustness of our analyses. Where multi-arm studies (e.g. two intervention arms and one control arm) contribute multiple comparisons to a specific analysis, we will split the ‘shared group’ to avoid including data from the same participant more than once.
We will contact the study authors to request missing data if needed. If after contacting the authors, there are still missing data and we consider the data to be missing at random, we will include only the data available in the analysis. Otherwise, we will impute the missing data and account for the data imputed with uncertainty in line with the Cochrane Handbook of Systematic Reviews of Interventions [
38]. We will then conduct a sensitivity analysis to analyse how sensitive the results are to the assumptions we made when imputing missing data and analyse all data as intention-to-treat.
We will explore clinical heterogeneity by clearly documenting study characteristics related to the population, intervention, outcomes and context in table format. We will assess the statistical heterogeneity in each meta-analysis by inspecting forest plots and calculating
χ2 test values and
I2 statistics. We will consider significant heterogeneity present if the
P value of the
χ2 test is < 0.10. We will interpret the
I2 statistic according to the thresholds recommended in the Cochrane Handbook of Systematic Reviews of Interventions [
39]. Therefore, we will consider an
I2 value above 30% to indicate important heterogeneity. In addition, we will explore the causes of statistical heterogeneity by conducting subgroup analyses.
We will examine reporting biases by means of funnel plots, if we are able to pool more than 10 studies per outcome in a meta-analysis.
We will organise the review findings according to the proposed models of integrated care as depicted in Fig.
1. We will pool data from individual studies if they are sufficiently homogeneous in terms of design, population, intervention and comparator. As we anticipate some degree of heterogeneity, we will perform random-effects meta-analysis. We will not pool data from RCTs and non-randomised studies in a single meta-analysis. If we judge included studies to be too heterogeneous to pool, we will make use of narrative synthesis and present data in table format.
We will carry out the following subgroup analyses on primary outcomes to explore heterogeneity: various co-morbidities (e.g. patients with diabetes and HIV vs patients with diabetes and depression), clinic vs. community level and age category (children, age 1–10 years vs. adolescents, age 10 to 19 years vs. adults, age > 19 years).
We will carry out the following sensitivity analyses on primary outcomes:
-
To examine the effect of excluding studies of high risk of bias
-
To examine the effect of various ICCs in case of adjusting outcomes for clustering ourselves
-
To examine the effect of imputed data
We will assess the certainty of the evidence for primary outcomes (all-cause mortality, disease-specific morbidity, HbA1c, BP, cholesterol levels) using GRADE. We will create a ‘summary of findings’ table using GRADEpro software [
40]. In the table, we will display the model of integrated care, primary outcomes (e.g. all-cause mortality, disease-specific morbidity) of the review, the comparative risks between intervention and control groups, the relative effects with 95% CI, the number of participants in the studies and the certainty of evidence.
The five domains that we will consider for our judgement to downgrade the certainty of evidence comprise study limitations, inconsistency, imprecision, indirectness and publication bias. We will consider upgrading the certainty of evidence if there is a large effect, a dose-response and cases where all plausible residual confounding would reduce a demonstrated effect or would suggest a spurious effect if no effect was observed. The quality of evidence for each outcome will be described as high, moderate, low or very low [
41].