Response to Comments on Zhang et al. Second-Line Agents for Glycemic Control for Type 2 Diabetes: Are Newer Agents Better? Diabetes Care 2014;37:1338–1345

Abstract

We are grateful for the comments from Messori and Bellia (1) and those of Bae et al. (2) regarding our article (3). We hope the following response will clarify the questions they raised.Messori and Bellia (1) pointed out that the incremental cost-effectiveness ratio (ICER) is preferred over the absolute cost-effectiveness ratio. We agree, and it is worth noting that Fig. 1 in our article provides complete details about the expected quality-adjusted life-years (QALYs) prior to first events and the expected medication costs under all treatment regiments. As such, it subsumes ICER estimates. Admittedly, the readers must calculate them from the graph and therefore we provide the estimates for the glycemic control goal of HbA 1c ,7% here. For males, for regimens T1-T4 the ICERs in units of $/QALY are 10, 369, 11,062, 11,277, 11,144, respectively. For females, for regimens T1-T4 the ICERs are 9, 310, 9,865, 10,043, 9,933, respectively. Messori and Bellia also express confusion over the computation of life-years to first event. To clarify, our reported results are based on a median diagnosis age between 54 and 55 years. The lifeyears to first event are the sum of time to diagnosis and the time from diagnosis to first event, where the latter is computed using our model. Bae et al. question an apparent assumption in our model that "underlying HbA 1c fluctuation at baseline remains constant throughout the patient's life" (2). However, we did not assume a constant baseline; we assumed a linear increasing trend in HbA 1c , consistent with other published glycemic control models (4,5).Bae et al. suggested the use of treatment effect estimates, based on realworld data, is a major limitation. We take the contrary perspective, i.e., that this is a strength of our study, as realworld data accounts for potential lack of adherence and other behavioral factors that may not be present in the ideal setting of a randomized controlled trial (RCT). Of course, we also recognize the merit of RCTs in limiting bias, and for this reason we reported results based on treatment effects reported in RCTs. As readers can see from our results, these do not change our main conclusions.Bae et al. propose the use of a longer time window for measuring the effect of medications. In our analysis, we found no evidence that a longer time window would make a significant difference. Compared with the 3-month time window used in our article, the use of 6-and 12-month time windows before and after treatment initiation resulted in similar or decreased treatment effects for the newer medications. Bae et al. also suggested the use of propensity score methods to account for possible confounding. This would be interesting to explore in theory, but in practice there are likely a number of unmeasured confounders. Fortunately, our sensitivity analysis based on RCT estimates obviates all of these concerns.Bae et al. fairly criticize the fact that the cost of hypoglycemia was not included in the model. As they pointed out, we adjusted for hypoglycemia outcomes, but n...