Response to Letter to the Editor: “Cardiovascular Effects of Pioglitazone or Sulfonylureas According to Pretreatment Risk: Moving Toward Personalized Care” Free

Defining subgroups of patients with a different likelihood of benefitting from a specific treatment represents an important clinical need, considering the large heterogeneity of patients with type 2 diabetes and the need for a personalized approach.

In this respect, recursive partitioning techniques can help in identifying distinct and homogeneous subgroups, based on the combination of different characteristics. By definition, this is an exploratory, data-dependent approach, and any result needs to be validated in independent samples. This is obviously difficult for many trials because of their cost and duration. However, we believe that important information can be derived from statistical approaches focusing on the interaction between different variables, rather than on the independent contribution of individual characteristics, as usually happens by the application of standard multivariate regression models.

Recursive partitioning techniques also allow the identification of the best cut-off for continuous variables. The cut-off values chosen by the model, as well as the combination of the variables within each cluster, are not defined a priori, but they represent the split effect that maximizes the difference in the risk of the outcome of interest. As such, they do not necessarily coincide with standard cut-offs, often established for different purposes.

Results of recursive partitioning analysis need to be interpreted in the light of their clinical plausibility. In our view, the findings of our study have a strong biological plausibility and are fully coherent with the existing literature on this topic. Pioglitazone transacts its effects through activation of the nuclear hormone receptor peroxisome proliferator–activated receptor-gamma (PPARγ). PPARγ receptors are expressed in endothelial cells, arterial smooth muscle cells, and monocytes/macrophages, providing a pathway for direct anti-inflammatory, antioxidant, and other protective actions of pioglitazone. Pioglitazone is the only true insulin-sensitizing antidiabetic agent and insulin resistance has been independently associated with atherosclerotic cardiovascular disease (CVD) in many cross-sectional and prospective studies. Pioglitazone treatment is associated with an improvement in major cardiovascular risk factors such as plasma lipids profile, blood pressure, and C-reactive protein (1–3). The PERISCOPE trial showed a reduction in the progression of coronary atherosclerosis (atheroma volume) in patients treated with pioglitazone versus glimepiride (4). Furthermore, the cardiovascular benefits of pioglitazone have been demonstrated in a trial conducted in nondiabetic, highly insulin-resistant subjects (1).

We agree that randomized trials remain the cornerstone for the generation of unbiased evidence. However, they provide an estimate of the average effect of an intervention, when applied to a specific study population. They are of little value for an in-depth exploration of the heterogeneity in the response, determined by a series of patient characteristics. The increasing number of therapeutic options available makes more and more relevant the identification of patient profiles that could help clinicians make the best choice for the individual patient. The application of machine learning models can help meeting these clinical needs.

Acknowledgments

Financial Support: The study is supported by the Agenzia Italiana del Farmaco (AIFA) within the Independent Drug Research Program (contract No. FARM6T9CET) and by Diabete Ricerca, the nonprofit Research Foundation of the Italian Diabetes Society. The funding agency played no role in the study design; in the data collection, analysis, and interpretation; in the writing of the manuscript; or in the decision to submit the manuscript for publication.

Disclosure Summary: All authors declare that no conflicts of interest exist with regard to this manuscript.

References