What is the coverage of retina screening services for people with diabetes? Protocol for a systematic review and meta-analysis

Objectives

The objectives of this review are to:

1. Summarise the different ways retina screening coverage for people with diabetes has been defined in published reports.

2. Summarise retina screening coverage for people with diabetes by World Bank country income level and WHO regions.

3. Compare disparities in retina screening coverage for people with diabetes by factors in the PROGRESS framework (ie, between subgroups within any of: Place of residence, Race/ethnicity/culture/language, Occupation, Gender/sex, Religion, Education, Socioeconomic status and Social capital).17

Eligibility criteria

Information sources and search strategy

Using a search strategy developed by an experienced information specialist, we will search MEDLINE (Ovid), Cochrane Central Register of Controlled Trials (CENTRAL), Embase, Global Health (Ovid), Emerging Sources Citation Index (ESCI), Science Citation Index Expanded (SCI-EXPANDED) and Social Sciences Citation Index (SSCI) on Web of Science (Clarivate) for reports using data collected since 1 January 2000. Our search strategy is included in online supplemental annex 2. We will download and de-duplicate the results in EndNote, and then export the results into Covidence (Veritas Health Innovation, Melbourne, Australia; available at www.covidence.org) for screening. We will examine reference lists of all included reports to identify further potentially relevant reports.

We will search for grey literature in the form of government reports of retina screening coverage at the national or subnational level. Grey literature reports will be searched for by one investigator (NC); another investigator (LG) will repeat the searches for 20 countries to ensure the consistency of relevant results. Government websites, including that of national and provincial government health departments, of the 194 WHO member states and eight associate countries and territories (American Samoa, Bermuda, French Polynesia, Greenland, Hong Kong, Palestine, Puerto Rico and Tokelau) will be identified and searched for relevant reports and coverage statistics. The process for systematically collecting government reports will be based on the grey literature search framework outlined by Godin et al.20 First, to identify the relevant government websites, a Google search will be conducted using the search strategy in online supplemental annex 3. The first ten pages of each search (representing 100 results) will be reviewed, using the title and short text underneath. The relevant websites’ name/organisation, country and URL will be entered into an Excel spreadsheet. Then, each of these websites will be searched using the search feature and a combination of keywords included in online supplemental annex 4; when the search feature is not available, the website will be scanned for relevant information. When the search feature is available, results will be reviewed using the title and short text underneath and until no more relevant items are identified on a following page. All potentially relevant reports will be exported into Covidence and de-duplicated for title and abstract screening.

Selection

For published literature and grey literature, all title and abstracts (for published literature) or short text (for grey literature) will be screened by two investigators independently using Covidence. Full-texts of all potentially relevant publications will then be acquired and assessed by two investigators independently to establish eligibility for inclusion into the review. Any disagreements in the screening of search results will be resolved by discussion and consultation with a third investigator as needed.

Following the selection process of published and grey literature, field experts—including clinicians, academics, non-governmental staff, the Diabetic Retinopathy Work Group of the International Agency for the Prevention of Blindness and WHO representatives—will be provided with a list of the included reports and requested to nominate further potentially relevant reports. An Excel spreadsheet will be used to track who has been contacted, who has responded, the need for a follow-up message, and identified reports.

Data collection and data items

Data extraction will be performed in Covidence. The data extraction form will be piloted by two investigators on five studies and modifications undertaken as required. The following details of each report will be extracted by two investigators independently. Any discrepancies will be resolved by discussion, and with a third reviewer if necessary. In the event of unclear outcome data, two attempts will be made to contact the corresponding author by email.

1. Study characteristics: country/countries of study, type of report, source of data, study design, administrative region of study (when subnational eg, province, district, state), year(s) of data collection, number of participants, overall participation rate, extent of missing data.

2. Participant characteristics: inclusion criteria (age, diabetes type), age (median/mean and range), diabetes type, diabetic retinopathy status, vision impairment status,21 available information across any PROGRESS dimension (outlined above).17

3. Outcome(s): definition of screening coverage (numerator, denominator, screening interval), period of follow-up (if applicable), retina screening coverage result (aggregated and if disaggregated by diabetes type, diabetic retinopathy status or any equity/PROGRESS dimensions (outlined above)17).

Outcomes

The primary outcome measure is retina screening coverage for people with diabetes. This is defined in a range of ways. We will include the broad categories of outcomes outlined in Table 1 and will expand this list based on what is identified during the search.

Risk of bias in individual studies

Two investigators will independently assess the risk of bias in each included study. Any conflict in relation to the appraisal will be discussed between the two investigators and resolved with a third investigator if necessary. Relevant Joanna Briggs Institute critical appraisal tools will be used to evaluate each included study .22

Assessment of heterogeneity

Clinical heterogeneity will be assessed by comparing key participant characteristics at the study level (eg, age, sex, diabetes type, diabetic retinopathy status). Methodological heterogeneity will also be considered, including variation in the specific definition of the numerator and denominator as well as the reporting period and method of data collection. If meta-analysis is possible, we will assess statistical heterogeneity by inspecting forest plots and the I² and τ² statistics to examine the proportion of heterogeneity across studies that is due to chance. If high levels of heterogeneity are detected (eg, I²>50%) and there are sufficient data available (eg, meta-analysis including more than 10 studies), we will explore likely sources of this heterogeneity via meta-regression.

Data synthesis

We will take an inclusive approach for objective 1 and will then determine the potential for synthesis of results reporting the WHO definition (table 1) for objective 2, before exploring disparities within all included studies in objective 3.

To achieve objective 1, we will outline the full range of definitions used across included studies and summarise the study characteristics of each definition, including geographic (country, national vs subnational), source of data, study design, inclusion criteria and extent of data missingness.

To achieve objective 2, in countries where estimates on retina screening coverage for people with diabetes using the WHO definition are available from multiple sources, we will establish the country estimate according to the decision tree included in online supplemental annex 5. We will first follow the process separately for estimates generated for the preferred and alternative data sources for the WHO definition (facility-based and survey-based, respectively, see table 1). The process outlined in the decision tree prioritises more recent studies and studies with nationally representative sampling frames. Where no nationally representative estimates are available, we will pool two or more subnational estimates if the definitions of the outcome are comparable, and the estimates are close enough in value that the pooled result provides a meaningful measure of retina screening coverage for people with diabetes. We will combine these subnational estimates from the same country using an inverse variance weighted average, using the metagen command from the meta package in R. If two sources of data are then available for a country, we will prioritise facility data in countries with organised retina screening programmes and accurate diabetes registries, and survey data in countries with opportunistic screening or with no centralised registries. We will document the extent to which these judgements affect the final country estimates by a sensitivity analysis.

We will use the country estimates to calculate the median, interquartile range and range of retina screening coverage for each WHO region and World Bank country income level,23 stratified by screening interval (eg, 1–2 years and 3 years). We will also compare retina screening coverage estimates over time where comparable sampling frames have been used in repeat studies.

To achieve our third objective, and depending on availability of data, we will calculate risk ratios comparing coverage in different groups, specifically:

1. Diabetes type (type 2 vs type 1).

2. Duration of diabetes (per year increase).

3. Diabetic retinopathy status (present vs absent).

and PROGRESS categories:

1. Place of residence (eg, rural vs urban)

2. Ethnicity (eg, non-dominant vs dominant)

3. Occupation status (eg, not-employed vs employed).

4. Gender/sex (eg, female vs male)

5. Religion (eg, non-dominant vs dominant).

6. Education level (eg, no educational qualification vs school level qualification vs further educational qualification).

7. Socioeconomic status (eg, high vs low area level deprivation).

8. Social capital (eg, low vs high social capital).

We will plot these risk ratios in separate forest plots and combine values where appropriate that is, where we judge that pooling will provide a meaningful measure of comparative risk (see the Assessment of heterogeneity section), using the metagen command from the meta package in R. If there are sufficient data available (eg, meta-analysis including 10 or more studies), we will also assess small study effects, one of which may be publication bias, by preparing a funnel plot,24 which is a scatter plot of effect size versus precision (SE).

Certainty assessment

Currently there is no formal guidance from GRADE regarding the assessment of certainty in systematic reviews and meta-analysis of prevalence studies.25 Therefore, we do not plan to assess certainty formally but we will consider the GRADE domains, that is, risk of bias, inconsistency, imprecision, indirectness and publication bias when drawing conclusions from the available evidence.26

Patient and public involvement

Patients and the public will not be involved in the conduct of the systematic review and meta-analysis, as we plan to review existing published literature and publicly available reports only.

Ethics and dissemination

This review will only include published data thus no ethical approval will be sought. The findings of this review will be published in a peer-reviewed journal and presented at relevant conferences.