With interest we read the article by Fadl et al. about fasting capillary glucose as a screening test for gestational diabetes mellitus (GDM). The authors evaluated the accuracy of the fasting capillary glucose and repeated random glucose measurement as compared with a 75 g oral glucose tolerance test. Furthermore, they explored whether combination of the test results with presence of traditional risk indicators optimises screening for GDM. They conclude that in their population, fasting glucose measurement is an acceptable and useful screening test. Application of a combined screening model was not considered superior over fasting capillary glucose alone.1
We would like to comment on the way their combined screening model was constructed. The authors used univariate analysis to create their integrated screening model. Two by two contingency tables were constructed to calculate accuracy measures of the fasting glucose test and the repeated random glucose test. Traditional risk indicators were assessed as one entity, being either present or absent. Subsequently, it was evaluated if the presence of risk indicators had additive value, calculating sensitivity and specificity for the situation in which either the screening test result was positive or traditional risk indicators were present.
This approach would be appropriate if the presence of one or more traditional risk indicators have exactly the same diagnostic value as the capillary glucose measurement at the respective cut-off values. However, it is more likely that different risk indicators would contribute differently, and that if multiple risk indicators are present, the risk of GDM is even higher. An evaluation based on the approach used in this study would not detect such associations, and it may lead to incorrect statistical findings and conclusions, as important information in the study data is ignored. Therefore, the conclusion made by the authors that a combined screening model is not superior over using fasting capillary glucose alone, is in our opinion premature. Instead, we propose an approach in which individual and combined associations are evaluated systematically.2 Such an analysis has the potential to show the true value of the test(s) under study.3 Only when these analyses indicate that the risk indicators do not substantially improve the accuracy in a combined screening model, the conclusion of the authors would be confirmed.
In conclusion, we would like to point out that univariate analysis of the value of diagnostic testing, in general, can be misleading, and that multivariate modelling is needed to reveal the true value of the use of individual traditional risk indicators combined with screening test results. In our opinion, test accuracy measures could still be increased by the use of individual traditional risk indicators combined with screening test results if the appropriate multivariate statistical approach would be used.