To the Editor: We read with great interest the study by Young et al [1] on the impact of population-level HbA1c screening on reducing diabetes diagnostic delay in middle-aged adults. Based on HbA1c levels among UK Biobank participants aged 40–70 years, 1% of 166,846 participants were found to have undiagnosed diabetes. The median time to clinical diagnosis for those with undiagnosed diabetes was 2.2 years.

As a tool for the diagnosis of diabetes and prediabetes (impaired glucose tolerance and/or impaired fasting glucose), HbA1c screening is not yet universally accepted; nevertheless, it provides an interesting picture of the state of dysglycaemic disorders among the population. In our cross-sectional study conducted in the Czech Republic among a population aged 25–64 years (n=1189), we found that, based on HbA1c levels, 16.7% of participants with diabetes were undiagnosed and 27.8% of participants had prediabetes [2].

Young et al [1] should be congratulated on the results of their study. Nevertheless, we found several points that we believe should be clarified. First, diabetes status was defined using several criteria, one of which was evidence of taking or being prescribed glucose-lowering medication. As metformin is also widely prescribed to individuals with prediabetes, was a diagnosis of diabetes (rather than merely prediabetes) confirmed using another study criterion? As a large group with prediabetes would be present among the particular age group included in the study, this may have influenced the results substantially.

Another issue deserving some discussion is the possibility that Biobank study baseline measurements are not comparable to primary care measurements (e.g. because of some systematic measurement errors). If such a discrepancy exists, the situation might be substantially more complicated than the authors and their Cox model anticipate. Specifically, study-censored individuals might arise as a mixture of standard (uniformative) censoring and censoring of individuals who are diabetes negative and who will never be diagnosed with diabetes. Modelling such a process is complicated, but it can be said with some confidence that the standard Cox model does not allow for it (its standard application in the mixture case will probably lead to increased estimates of time until diagnosis). The need for calibration of the HbA1c values from Biobank (which the authors acknowledge [1]) seems to support the previous concerns. If interlaboratory baseline variability among different laboratories is a common occurrence, the idea of prescreening, as suggested in the paper, will be jeopardised, even in the future.

An obligatory statistical question in the context of the Cox regression relates to whether or not Young et al [1] checked the validity of the risk proportionality assumption. More importantly, is it possible that those who left their primary care practice during the study (and who were then censored as stated in the paper) have a larger (or smaller) risk of diabetes on average than those who stayed in the same practice? They may have left, for example, because they were dissatisfied with their practitioner (or because they are very healthy in general and do not care about primary care, except for the convenience, and hence change their primary care practice haphazardly and often).

Another important issue is that artificial discretisation of originally continuous explanatory variables can be avoided by employing, for example, complexity-penalised splines.

We also believe that time until diagnosis should be studied in relation to age [3,4,5]. One can anticipate that it will be strongly related to age in general. Moreover, with a fixed schedule of preventive check-ups, the shortening of time until diagnosis will be small or zero when diagnosis occurs close to the age at which a regular check-up takes place.

Finally, is it possible, at least in principle, that some individuals with a suspicious HbA1c level in UK Biobank were not checked for fasting blood glucose levels or HbA1c in their primary care practice? Such individuals would never receive a formal diabetes diagnosis (with effectively an ‘almost infinite’ time until diagnosis, increasing the mean time until diagnosis). We believe that this scenario would not occur frequently but would be possible, and so should have been addressed in the discussion.

We respectfully suggest that the points raised in this letter are taken into consideration, especially if a follow-up study is planned.