Odds, risks and appropriate diagnosis of gestational diabetes: comment

In my opinion, McIntyre and colleagues have misunderstood the primary point I raised in my earlier article in the Journal.1 My concern was not the increased number of women diagnosed with gestational diabetes mellitus (GDM), but the accuracy of the diagnostic thresholds. Although I suggested that many women in the Hyperglycemia and Adverse Pregnancy Outcome (HAPO) study with one elevated blood glucose level may not be at risk, I also stated that additional women with two or more blood glucose levels marginally below the new thresholds will be at risk. The authors' comment that an increased odds ratio has the same effect confirms that they misunderstood my concern.

My concerns about the combined primary end point were valid and I acknowledge the response provided — but is cord C-peptide > 90th centile a true adverse neonatal outcome?

I agree that thresholds should be determined by fully adjusted models.

However, no data have been provided to refute the assertion that many women with only one elevated blood glucose level may not be at risk. Each blood glucose threshold has been determined independently and not adjusted for normal blood glucose levels.2 The thresholds should accurately define patients at risk in the HAPO study,3 so the number of women diagnosed should approximate the total number of cases of GDM (since a woman can only have one case of GDM per pregnancy). Of 23 316 participants, 3746 (16.1%) were diagnosed with GDM. There were 1935 women with an elevated fasting blood glucose level, 1928 with a raised 1-hour blood glucose level, and 1309 with a raised 2-hour blood glucose level — 5172 cases of GDM (22.2%) in total, or 38% more cases than women diagnosed with GDM.2,4 This occurs because women with two or more elevated blood glucose levels are double- or triple-counted as diagnoses. Women with two or more elevated levels have the highest rate of adverse outcomes.4 These women contribute disproportionately to the total number of adverse events for each parameter.

There is an expectation that the lower limit of the confidence interval should exceed the diagnostic odds ratio threshold, given that the minimal odds ratio of groups of these women should be 1.75. This raised additional concerns1 that many women with only one elevated blood glucose level will belong to blood glucose level categories at reduced risk, when the overall odds ratio of these cohorts is equivalent to the risk threshold.4

This is not a new problem. It confronted the Framingham statisticians who stated “analyses that fail to examine risk factors in combinations usually greatly overestimate the population-attributable risks associated with individual risk factors”,5 which is exactly the situation for women with one elevated blood glucose level only. They also stated that “multivariable risk assessment also avoids overlooking high-risk CVD [cardiovascular disease] candidates with multiple marginal risk factors and avoids needlessly alarming persons with only 1 isolated risk factor”,5 which is also the situation for women with two or more blood glucose levels just below the diagnostic thresholds in contrast to those with a single marginally elevated blood glucose level. It is reasonable to question the recommended implementation of the current thresholds.2 Subgroup analysis of the 22 152 women with two normal blood glucose levels is a valid suggestion.

The HAPO study has a wealth of data.3 Clinicians require tools that accurately diagnose women at risk. The interaction of blood glucose parameters is similar to the interaction of cardiovascular risk factors; a single cholesterol threshold does not define absolute risk for all people. We have a great opportunity to develop a risk model that would more accurately identify women at risk, based on the three blood glucose levels, age, weight, ethnicity and other parameters. This would make the most of the data from this excellent study. It has been done for cardiovascular disease, why not for GDM?