Methodological quality of the review: Medium confidence
Author: Zhou M, Wang W, Huang W, Zhang X
Geographical coverage: United States (US), Netherlands, Barbados, United Kingdom (UK), Korea, Congo, France and Denmark
Sub-sector: Primary open angle glaucoma, diabetes mellitus
Equity focus: None specified
Review type: Effectiveness review
Quantitative synthesis method: Meta-analysis
Qualitative synthesis method: Not applicable
Background: Diabetes mellitus (DM) is a serious and increasingly prevalent health problem worldwide due to lifestyle changes and an ageing population. DM is associated with severe, acute and chronic complications, and some studies have also found it to be another possible risk factor for primary open angle glaucoma (POAG). However, the relationship between DM and POAG is controversial.
Objectives: To determine the association between diabetes mellitus (DM) and primary open angle glaucoma (POAG).
Main findings: Thirteen studies – seven case-control studies and six population-based cohort studies – were included in this meta-analysis. Of the seven case-control studies, three originated from the US, one from Korea, one from the Congo, and two from Europe (France and Denmark). Of the six population-based cohort studies, three originated from the US, one from the Netherlands, one from the Barbados and one from the UK. Overall, all the studies were of high quality. The most common bias was ascertainment of exposure, with four studies assessing DM by self-reporting or medical records.
The pooled risk ratio (RR) of the association between DM and POAG based on the risk estimates of the six cohort studies was 1.40 (95% CI, 1.25–1.57). The pooled odds ratio (OR) of the association between DM and POAG based on the risk estimates of the seven case-control studies was 1.49 (95% CI, 1.17–1.88). There was considerable heterogeneity among the case-control studies that reported an association between DM and POAG (P<0.001) and no significant heterogeneity among the cohort studies (P = 0.377). After omitting the case-control study that contributed significantly to the heterogeneity, the pooled OR for the association between DM and POAG was 1.35 (95% CI, 1.06–1.74). Authors note that these findings have important public health implications pointing to a significant association between DM and the risk of POAG. Additionally, authors also recognise that further studies are needed to elucidate the exact underlying mechanisms linking DM and POAG.
Methodology: The full-length articles were required to meet the following inclusion and exclusion criteria for the meta-analysis. The inclusion criteria were: (a) they had a case-control or a cohort design (b) the exposure of interest was DM (c) the outcome of interest was POAG and (d) the odds ratios (ORs) or the relative risk (RR) estimates with their 95% confidence intervals (CIs) (or data to calculate them: raw data, P value, and/or variance estimates) were reported.
Authors conducted a search of PUBMED and EMBASE up to February 2014. The following terms, adapted for each database, were used for the searches: diabetes mellitus, DM, impaired glucose tolerance, hyperglycemia, insulin resistance, insulin secrete dysfunction, glaucoma, intraocular hypertension, intra-ocular pressure, ocular hypertension. To achieve maximum sensitivity, limits or filters were not placed on the searches. No language restrictions were imposed. A manual search was performed by checking the reference lists of the original reports. Although methods used to screen studies for inclusion were not reported, authors used appropriate methods to extract data and quality assess included studies.
Data from the cohort studies and the case-control studies was analysed separately. The risk ratio (RR) was used as a common measure of the association between DM and the risk of POAG in the cohort studies. The incidence of rate ratio (IRR) and the hazard ratio (HR) were considered as RRs, and the pooled adjusted RRs with the corresponding 95% confidence interval (CI) were calculated. The maximally adjusted RRs or ORs were used to assess the association between DM and POAG.
Applicability/external validity: Authors note that unlike previous studies, this review included several newly published case-control and cohort studies and excluded cross-sectional studies in the updated meta-analysis. This allowed for a greater number of subjects and, hence, a more detailed and accurate risk estimation than in prior meta-analyses. Additionally, authors state that findings from this meta-analysis have important public health implications pointing to a significant association between DM and the risk of POAG.
Geographic focus: Review authors did not restrict literature searches to specific income settings, however studies eligible for inclusion were conducted only in high-income countries. Authors did not describe particularly how applicable the results may be to low- and middle-income settings.
Summary of quality assessment: There is medium confidence in the conclusions about the effects of this review as limitations were identified. Although authors did not contact experts for further potentially relevant studies as part of the search strategy, results obtained from the funnel plot analysis and formal statistical tests did not provide evidence of publication bias present in the review.
Authors used appropriate methods to extract data and assess the quality assessment of each included. Authors used a validated tool to assess the quality of each included study and provided an overall assessment of the case-control and cohort studies, however assessment for each criterion for each individual study was not provided in the review. Additionally, it is not clear if selection bias was avoided, as reviewers do not mention if study selection was conducted by two reviewers independently. Appropriate methods were used to synthesise data, and authors ensured that studies were similar enough that it made sense to combine them.
Where heterogeneity was present, a sensitivity analysis was conducted. As with any meta-analysis of observational studies, there are several potential limitations with regards to the results. Limitations were acknowledged by the authors including heterogeneity across included studies, variation in effect sizes between the cohort studies and potential recall bias within included studies.