Mortality in Type 2 Diabetes Mellitus
Mortality in Type 2 Diabetes Mellitus
A systematic review of published and unpublished literature was undertaken including grey literature. The review followed the guidance of the CRD (March 2001) report number 4. An electronic database search was conducted using MeSH and search terms as shown in Table 1. References of all retrieved articles were checked for relevant studies and if needed experts were contacted for advice, which identified additional published and unpublished references. The process of conducting the search was documented as it developed to ensure transparency.
Inclusion and exclusion criteria
Studies were included if they fulfilled the following criteria:
Participants: people with type 2 diabetes
Studies only available as an abstract or studies that included both type 1 and type 2 diabetes were excluded unless specific analysis for each type was published and the sample size for each was >100. Studies that did not specify the type of diabetes were excluded. Studies that did not have a comparison population were excluded. Poor quality studies as decided by quality appraisal and studies that did not have adequate data or lacked information necessary for the synthesis of the data were excluded.
The search was limited to studies published between 1990 and 2010 because two systematic reviews had already been published, one in 1983 and the other in 1999; also diabetes was undifferentiated into type 1 and type 2 diabetes in Medline-PubMed. In addition changes in the clinical diagnostic criteria (as defined by the World Health Organisation and the American Diabetes Association) occurred in 1999 making standardisation of methods and reporting comparable in different settings. All these affected the reporting of diabetes mortality outcome data before 1990. Where there was an overlap in the study population the relevant paper was included and the others were used to provide context. Titles and abstracts of articles were checked by two reviewers (C.N. and H.C.). Full texts of selected studies were assessed for inclusion by one reviewer (C.N.) and triangulated by the other reviewers (H.C. and D.B.J.). Any variations in comments were resolved through discussion.
The internal validity of individual studies was assessed independently (C.N. and H.C.), using CRD's guidance checklist for quantitative research, and was evaluated in accordance with the inclusion/exclusion criteria, methodological quality (that met the critical appraisal framework), and relevance to the research questions. Figure 1 shows the flow chart of the study selection based on the inclusion and exclusion criteria. To assess the quality of the included studies we used the scoring criteria outlined by CRD and followed the PRISMA Statement, through the course of the study.
(Enlarge Image)
Figure 1.
Flow chart of study selection for inclusion in the systematic review.
Data extraction was undertaken by one reviewer (C.N.), triangulated by the second reviewer (H.C.), and where there was uncertainty by the third reviewer (D.B.J.). Data from individually selected studies were extracted as follows: author, year of publication, study design, setting/location, year of study, comparison population, study size, number of deaths, patients' characteristics, follow-up years, outcome measure of mortality, and degree of mortality in RRs or odds ratio. Adjustments for confounding in the analyses were factored in the selected papers including age, smoking, sex, hypertension, body mass index, previous acute MI and hypercholesterolaemia.
Meta-analysis was carried out to show precise estimate of mortality risk. Sub-analyses of the effects of sex, drugs, CVD, smoking and alcohol, age at diagnosis and cancer on mortality risks were also assessed. Evidence from data on mortality in type 2 diabetes was synthesised through a descriptive epidemiological review from included studies.
Using the MIX 2.0 Pro software package, we analysed individual effects using RRs, displayed in the forest plot in Figure 2. The point estimate, lower and upper limits of 95% confidence intervals were log scaled and the effects observed in the studies were pooled to produce a weighted average effect of all the studies (meta-analysis). Larger studies were weighted preferentially by fixed effects statistical models; smaller studies were weighted preferentially by random effects models thereby enhancing the robustness of the summary effects according to variations in statistical methods. The test for heterogeneity of the various studies combined was computed using the statistical heterogeneity funnel plots, as shown in Figure 3, Cochran's Q score, inconsistency index (I2), Tau-square estimate (t2) and evidence of p-value <0.01. I2 is the measure of quantity of amount of heterogeneity and is deemed more reliable in assessing inconsistency between studies, and t2 is the measure of study variance in random effect which will be followed if the I2 is greater than 50%. Cochran's Q score is the summation of all the z-scores of the studies. A sensitivity analysis plot and meta-regression were used to assess the subgroups on the effects of RR of combined studies.
(Enlarge Image)
Figure 2.
Relative risk for overall mortality in type 2 diabetes population selected. The size of the box represents the weight assigned to individual effects (point estimate of effect). The horizontal lines indicate the lower and upper limits of 95% confidence intervals.
(Enlarge Image)
Figure 3.
Heterogeneity funnel plot for the included studies to detect publication bias in the meta-analysis. The individual studies are symmetrically aligned demonstrating limited publication bias.
Methods
Sources of Data
A systematic review of published and unpublished literature was undertaken including grey literature. The review followed the guidance of the CRD (March 2001) report number 4. An electronic database search was conducted using MeSH and search terms as shown in Table 1. References of all retrieved articles were checked for relevant studies and if needed experts were contacted for advice, which identified additional published and unpublished references. The process of conducting the search was documented as it developed to ensure transparency.
Study Selection
Inclusion and exclusion criteria
Studies were included if they fulfilled the following criteria:
Participants: people with type 2 diabetes
Outcomes: all-cause or overall-cause or total mortality expressed as RRs or risk ratios
Evaluation of outcomes: mortality after diagnosis of type 2 diabetes; coronary heart disease, MI, nephropathy, cerebrovascular disease (stroke), or ischaemic heart disease.
Designs: existing systematic reviews, randomised controlled trials, cohort studies and epidemiological studies
Reporting: English language studies only were included as evidence shows that excluding non-English language studies does not change outcome results. Studies also had to provide sufficient detail including: a review of mortality indices in type 2 diabetes and partial or complete review of mortality in type 2 diabetes.
Studies only available as an abstract or studies that included both type 1 and type 2 diabetes were excluded unless specific analysis for each type was published and the sample size for each was >100. Studies that did not specify the type of diabetes were excluded. Studies that did not have a comparison population were excluded. Poor quality studies as decided by quality appraisal and studies that did not have adequate data or lacked information necessary for the synthesis of the data were excluded.
The search was limited to studies published between 1990 and 2010 because two systematic reviews had already been published, one in 1983 and the other in 1999; also diabetes was undifferentiated into type 1 and type 2 diabetes in Medline-PubMed. In addition changes in the clinical diagnostic criteria (as defined by the World Health Organisation and the American Diabetes Association) occurred in 1999 making standardisation of methods and reporting comparable in different settings. All these affected the reporting of diabetes mortality outcome data before 1990. Where there was an overlap in the study population the relevant paper was included and the others were used to provide context. Titles and abstracts of articles were checked by two reviewers (C.N. and H.C.). Full texts of selected studies were assessed for inclusion by one reviewer (C.N.) and triangulated by the other reviewers (H.C. and D.B.J.). Any variations in comments were resolved through discussion.
Quality Appraisal
The internal validity of individual studies was assessed independently (C.N. and H.C.), using CRD's guidance checklist for quantitative research, and was evaluated in accordance with the inclusion/exclusion criteria, methodological quality (that met the critical appraisal framework), and relevance to the research questions. Figure 1 shows the flow chart of the study selection based on the inclusion and exclusion criteria. To assess the quality of the included studies we used the scoring criteria outlined by CRD and followed the PRISMA Statement, through the course of the study.
(Enlarge Image)
Figure 1.
Flow chart of study selection for inclusion in the systematic review.
Data Extraction and Synthesis of Evidence
Data extraction was undertaken by one reviewer (C.N.), triangulated by the second reviewer (H.C.), and where there was uncertainty by the third reviewer (D.B.J.). Data from individually selected studies were extracted as follows: author, year of publication, study design, setting/location, year of study, comparison population, study size, number of deaths, patients' characteristics, follow-up years, outcome measure of mortality, and degree of mortality in RRs or odds ratio. Adjustments for confounding in the analyses were factored in the selected papers including age, smoking, sex, hypertension, body mass index, previous acute MI and hypercholesterolaemia.
Meta-analysis was carried out to show precise estimate of mortality risk. Sub-analyses of the effects of sex, drugs, CVD, smoking and alcohol, age at diagnosis and cancer on mortality risks were also assessed. Evidence from data on mortality in type 2 diabetes was synthesised through a descriptive epidemiological review from included studies.
Statistical Analysis
Using the MIX 2.0 Pro software package, we analysed individual effects using RRs, displayed in the forest plot in Figure 2. The point estimate, lower and upper limits of 95% confidence intervals were log scaled and the effects observed in the studies were pooled to produce a weighted average effect of all the studies (meta-analysis). Larger studies were weighted preferentially by fixed effects statistical models; smaller studies were weighted preferentially by random effects models thereby enhancing the robustness of the summary effects according to variations in statistical methods. The test for heterogeneity of the various studies combined was computed using the statistical heterogeneity funnel plots, as shown in Figure 3, Cochran's Q score, inconsistency index (I2), Tau-square estimate (t2) and evidence of p-value <0.01. I2 is the measure of quantity of amount of heterogeneity and is deemed more reliable in assessing inconsistency between studies, and t2 is the measure of study variance in random effect which will be followed if the I2 is greater than 50%. Cochran's Q score is the summation of all the z-scores of the studies. A sensitivity analysis plot and meta-regression were used to assess the subgroups on the effects of RR of combined studies.
(Enlarge Image)
Figure 2.
Relative risk for overall mortality in type 2 diabetes population selected. The size of the box represents the weight assigned to individual effects (point estimate of effect). The horizontal lines indicate the lower and upper limits of 95% confidence intervals.
(Enlarge Image)
Figure 3.
Heterogeneity funnel plot for the included studies to detect publication bias in the meta-analysis. The individual studies are symmetrically aligned demonstrating limited publication bias.