The oral glucose tolerance test for the diagnosis of diabetes mellitus in patients during acute coronary syndrome hospitalization: a meta-analysis of diagnostic test accuracy

A search of MEDLINE (1985 to March 2012) and EMBASE (1985 to March 2012) via EMBASE.com was conducted to identify all studies involving diagnostic tests assessing the value of the OGTT in the diagnosis of diabetes mellitus in subjects with or without ACS (Additional file1) (Yicong Ye and Hongzhi Xie). In addition, a manual search of the literature using the references of original manuscripts, reviews, and meta-analyses was performed. Finally, a search of the Cochrane database of ongoing systemic reviews was conducted. No language restriction was imposed.

The study eligibility was independently determined by two reviewers (Yicong Ye and Hongzhi Xie). Disagreement was resolved by consensus. The study eligibility criteria included: 1) published prospective cohort studies, 2) performing the first OGTT during ACS hospitalization (only for ACS studies), 3) repeating an OGTT more than 1 week and less than 3 years after the first one, and 4) use of a 75 g OGTT.

The following were criteria for exclusion: 1) pregnant and pediatric individuals, 2) patients with other chronic diseases, other than coronary heart disease, such as cystic fibrosis, polycystic ovary syndrome, acromegaly, liver disease, and renal disease, 3) studies in which the OGTT was repeated only in subjects with abnormal or normal OGTT results in the first test, and 4) studies with intended medical intervention (life-style change or medication) between administration of the two OGTTs.

Attempts were made to contact the author for further information on studies that fulfilled the above criteria but did not have sufficient data to build a two-by-two table before they were excluded from the final analysis.

Data extraction was carried out independently by two authors (Yicong Ye and Hongzhi Xie). Disagreements were resolved by discussion between the two reviewing authors. From each included trial, information was extracted on: 1) the study population, 2) the published language, 3) the time interval between administration of the two OGTTs, 4) the reference standard of diabetes, 5) mean age of study population and 6) the true positive value, false positive value, false negative value, and true negative value of each included study.

Study quality was assessed using the QUADAS list, with each item scored as “yes”, “no”, or “unclear”[10]. The results are presented in the text and in a graph. A summary score was not calculated when estimating the overall quality of an article since the interpretation of such summary scores is problematic and potentially misleading. The items in the QUADAS tool and their interpretation are presented in Additional file1.

The WHO (World Health Organization) 1985[11], ADA (American Diabetes Association) 2003[12], or WHO 1999[13] criteria was used as the reference standard depending on which criteria had been used in each study. All the included patients can be classified into two groups: diabetes and non- diabetes. The diagnostic threshold was the same as the reference standard in each included study. Accordingly, the two-by-two tables were constructed. The data in the two-by-two tables were used to calculate the sensitivity and specificity for each study. The individual study results were presented graphically by plotting the estimates of sensitivity (SEN) and specificity (SPE), and their 95% confidence intervals (95% CI), in paired forest plots.

A bivariate model was used for the meta-analysis of the pairs of SEN and SPE and for the construction of a summary receiver operating characteristic (ROC) curve[14]. This summary ROC curve represents the change in diagnostic accuracy according to changes in the cutoff value. The bivariate random effects approach enabled the calculation of summary estimates of SEN, SPE, the positive likelihood ratio (PLR), the negative likelihood ratio (NLR), and the diagnostic odds ratio (DOR) while correctly dealing with the different sources of variation: (1) the imprecision by which SEN and SPE were measured within each study, (2) variation beyond chance in SEN and SPE between studies, and (3) any correlation that might exist between SEN and SPE.

The heterogeneity (or absence of homogeneity) of the results between studies was assessed statistically using the quantity I ² , which describes the percentage of total variation across studies that is attributable to heterogeneity rather than chance[15]. Covariates were incorporated in the bivariate model to examine the effect of potential sources of heterogeneity across subgroups of studies. Due to the bivariate nature of the model, the effects of covariates on SEN and SPE can be modeled separately. The following covariates were analyzed to explore the heterogeneity: 1) Study population: the major objective of the studies was to compare the diagnostic accuracy in patients with and without ACS; 2) Threshold: although different criteria were used in the studies (including WHO 1985, ADA 2003, and WHO 1999), all studies used the same cutoff value of a 2-hour OGTT (11.1mmol/L) to diagnose diabetes. The major difference is that some of the studies included FBG in the diagnosis criteria, while others did not. Thus, the studies were divided into 2 subgroups (with or without FBG). 3) Blood sample: different reproducibility values for plasma and capillary glucose have been reported. 4) Interval between repeated tests (within 2 months versus more than 2 months): a prolonged interval between two OGTTs may also lead to poor reproducibility even though the optimal interval is still unknown. 5) Age: aging is associated with degradation of the pancreatic β-cells which may affect the accuracy of OGTT.

Meta-regression was conducted to further explore the heterogeneity quantitatively among the studies and to determine the diagnostic accuracy of the OGTT in different conditions. The lnDOR was used as the dependent variable. The standard error of the lnDOR was used to measure the within-study variability, and the residual maximum likelihood method to estimate the between-study variance. Factors associated with significantly different SEN and/or SPE in the subgroup analysis were included in the meta-regression model as covariates.

The potential publication bias was assessed using the Deeks funnel plots[16]. The studies by Eschwege et al. and De Vegt F et al. had much larger sample sizes and much longer intervals between the administrations of the two OGTTs, compared with the other studies included in the meta-analysis (Table1). In order to assess the effect of this study on the pooled result, a post-hoc sensitive analysis was performed to calculate DOR for ACS in the meta-regression model without the studies by Eschwege et al. and De Vegt F et al. In addition, the diagnostic criteria for DM have changed over time, thus the criteria used in the included studies has varied. Another post-hoc sensitive analysis was performed in order to calculate DOR for ACS in the meta-regression model without the study using the 1985 WHO criteria.

Of the 5,521 references identified in the initial search, only 18 reports fulfilled the criteria for inclusion in the meta-analysis. After communicating with the authors, 15 reports offered sufficient data to build a two-by-two table and thus were included in the final meta-analysis (Figure1)[17‐31]. Of the 15 included studies involving 8,027 subjects, 10 studies included patients with ACS, and the remaining consisted of non-ACS individuals. The characteristics of the included studies are detailed in Table1. The results on the methodological quality of the included studies are presented in text form in Additional file1 and in a graph in Figure2.

The SEN of the included studies ranged from 0.38 to 0.96, whereas the SPE ranged from 0.64 to 0.98 (forest plots in Figure3). The hierarchical summary ROC curve represents the relationship between sensitivity and specificity across the included studies with a 95% confidence ellipse and a 95% prediction ellipse (Figure4). The area under the summary ROC curve (AUC) was 0.87 (95% CI, 0.16-1.00). Using the bivariate model, the pooled results for SEN, SPE, PLR, NLR, and DOR were 0.70 (95% CI, 0.60-0.78), 0.91 (95% CI, 0.86-0.94), 7.6 (95% CI, 4.9-11.7), 0.33 (95% CI, 0.25-0.45), and 23 (95% CI, 12–41), respectively. The I² value of all measures was 99% (95% CI, 98-99%), indicating significant heterogeneity across the included studies.

The subgroup analyses demonstrated that the OGTT performed in ACS patients has similar SEN (0.71 [95%CI, 0.60-0.82] versus 0.67 [95% CI, 0.54-0.81], p=0.43) but a slightly lower SPE (0.86 [95% CI, 0.81-0.92] versus 0.95 [95% CI, 0.93-0.98], p<0.01) compared with non-ACS patients. A prolonged interval between repeated tests (more than 2 months) is also associated with lower SEN (0.62 [95% CI, 0.50-0.73] versus 0.77 [95% CI, 0.68-0.86], p<0.01) and SPE (0.90 [95% CI, 0.84-0.95] versus 0.92 [95% CI, 0.87-0.97], p<0.01). Compared with the younger age (< 60 years) group, advanced age (≥60 years) is associated with lower SPE (0.89 [95% CI, 0.82-0.96] versus 0.92 [95% CI, 0.87-0.97], p=0.01) while the SEN is similar (0.73 [95% CI, 0.62-0.84] versus 0.63 [96% CI, 0.48-0.77], p=0.81). However, using a different threshold (2-hour OGTT with or without FBG) or blood sample (plasma glucose or capillary glucose) did not lead to different diagnostic accuracy (all p>0.05) (Figure5).

Since ACS, interval between repeated tests and age were found to be associated with different SEN and/or SPE in the subgroup analysis, multiple meta-regressions were performed to further determine the effect of these factors on the DOR. However, none of these covariants was found to be associated with different diagnostic accuracy in the multiple meta-regression model (Table2).

The Deeks funnel plot asymmetry test showed insignificant publication bias (p=0.24, Figure6).

The post hoc sensitive analysis, which excluded the largest studies by Eschwege et al. and De Vegt et al., demonstrated a similar result (DOR 0.91, 95% CI 0.02-47.95, p=0.955), indicating the final conclusion of our meta-analysis was not markedly affected by these studies. Another post hoc sensitive analysis excluded studies using the 1985 WHO criteria also demonstrated a result similar to the final conclusion of the meta-analysis (DOR 0.53, 95% CI 0.01-19.36, p=0.687).