We read with great interest the article by Zhang et al. (1) in which they compared in a model four second-line treatment strategies following metformin monotherapy: sulfonylurea, dipeptidyl peptidase-4 inhibitor, GLP-1 agonist, and insulin. They concluded that all strategies produced similar glycemic benefits, but favored sulfonylurea from a treatment cost perspective. Diabetes models are used to simulate outcomes that cannot be directly observed during clinical trials. The results are often used by payers to inform formulary decisions and, as such, diabetes models play an increasingly important role on the availability of therapeutic options to both patients and prescribers.
Notable methodological achievement in the study by Zhang et al. was that the authors jointly estimated the underlying probability of the HbA1c changes and the initial treatment effects to support the Markov model design. However, the key assumption that underlying HbA1c fluctuation at baseline remains constant throughout the patient’s life is not supported (2). In addition, reliance on the first 3-month treatment effect as the main efficacy is a significant limitation. Longer-term studies have suggested that capacity of various agents to sustain glucose control varies (3). Moreover, due to heterogeneous treatment response, the efficacy of these drugs may not be sustained for some; but for others, third-line therapies may be delayed more than 5 years in clinical practice (4).
Using real-world patient data to estimate treatment effects also raises the concern that, in contrast to randomized control clinical trials, real-world clinical evidence will, in part, reflect the selection of therapies based on the individual needs of a patient. Adjusting for covariates would not adequately correct for this selection bias. Using propensity score methods prior to estimating treatment effects should have been included to better account for this confounding.
Another important limitation is the model’s ability to account for the effects of hypoglycemia. The model adjusted for hypoglycemia outcomes but did not account for the cost associated with management of hypoglycemia care. Research shows that the cost of hypoglycemia hospitalization among patients with type 2 diabetes treated with oral medication to be $17,564 per event in 2008 (5). In addition, more recent evidence suggests that nonsevere hypoglycemic events, which occur much more frequently, incur an average cost of $127 per person per event (6). Full accounting of all health care costs associated with therapy, and not solely medication acquisition costs, should be used to more accurately evaluate cost-effectiveness.
We encourage the further refinement of models that take into account the full range of important clinical and economic factors to better understand diabetes-related health care burden.
Article Information
Duality of Interest. J.B., B.H.C., D.M.K., and R.J.H. are employees and shareholders of Eli Lilly and Company, which is involved in the manufacture and marketing of a variety of antihyperglycemia agents that [at a therapeutic class level] would be considered to be involved in the analysis undertaken by Zhang et al. (e.g., insulin). No other potential conflicts of interest relevant to this article were reported.
