Oral Presentation Society of Obstetric Medicine of Australia and New Zealand and Australasian Diabetes in Pregnancy Society Joint Scientific Meeting 2017

Could we obtain better “value” from the pregnancy oral glucose tolerance test (OGTT) in diagnosis of gestational diabetes mellitus (GDM)?  A secondary analysis using the Australian HAPO data. (#40)

David McIntyre 1 , Kristen Gibbons 2 , Jeremy JN Oats 3 , Julia Lowe 4
  1. Mater Health, Mater Research and University of Queensland, South Brisbane, QLD
  2. Mater Research, University of Queensland, South Brisbane, Queensland, Australia
  3. University of Melbourne, Melbourne, Victoria, Australia
  4. University of Toronto, Toronto, Ontario, Canada

The 75 gram OGTT remains the standard for GDM diagnosis in Australia, with one value > threshold considered diagnostic.  IADPSG thresholds predominate, but “old ADIPS” thresholds persist in some centres.  Both sets of thresholds originate from expert consensus.  We hypothesised that using receiver operator curve (ROC) analysis and derived objective parameters would increase sensitivity and specificity of the OGTT results related to outcomes, specifically large for gestational age (LGA), fetal adiposity (FA) and shoulder dystocia (SD).

We used data from 1248 women in the HAPO Brisbane cohort to develop the model, subsequently validated using 619 women from the HAPO Newcastle cohort.  “Old ADIPS” (9.1% GDM) and IADPSG criteria (8.9% GDM), presented in that order, showed specificities of 90 - 92% but very low sensitivities for LGA (14%; 15%), FA (9%; 13%) and SD (20%, 30%).  Areas under ROCs (AUROCs) were low, varying between 0.50 - 0.61.

Using ROC analysis and derived “optimal” Youden cut-off points in a model including all standardised OGTT values (0, 1, 2 hours) markedly improved sensitivity: LGA 55%; FA 51%; SD 70%, but at the cost of lower specificity: LGA 68%; FA 73%; SD 67%.  AUROCs improved to 0.62 – 0.79 and Youden indices were:- LGA 23%; FA 25%; SD 47%.  However, this approach classified between 28 and 35% of women (depending on the outcome chosen) as at risk meriting intervention (ARMI).

“Diagnostic” ROC thresholds, calibrated to a specificity: sensitivity threshold of 3: 1 in the same model, gave sensitivities of:- LGA 28%; FA 28%; SD 30% with specificities of: LGA 85%; FA 85%; SD 90%) at ARMI prevalences between 10 and 17%.

Current GDM diagnostic algorithms provide poor risk stratification.  Novel ROC based models still involve inevitable compromises between sensitivity, specificity and pragmatically manageable prevalences of women identified as at risk and meriting intervention.