Pioglitazone associated with lower rate of stroke and MI in patients with prediabetes and recent stroke

Référence

Spence JD, Viscoli CM, Inzucchi SE, et al, for the IRIS Investigators. Pioglitazone therapy in patients with stroke and prediabetes: a post hoc analysis of the IRIS randomized clinical trial. JAMA Neurol 2019; Feb 7: doi: 10.1001/jamaneurol.2019.0079. [Epub ahead of print]

Sommaire

Insulin Resistance Intervention after Stroke (IRIS) was a randomized clinical trial that demonstrated that pioglitazone reduced the risk for recurrent stroke or MIs among 3876 patients with insulin resistance who also had a stroke or transient ischemic attack in the preceding 6 months. This reduction was at the cost of significant harms, such as fracture, edema, and weight gain (http://www.essentialevidenceplus.com/content/poem/180603). This current paper is a post hoc analysis of 2885 patients with prediabetes (glycohemoglobin between 5.7% and 6.4% or fasting plasma glucose level between 100mg/dL and 125 mg/dL). This analysis also emphasizes outcomes in 1454 adherent patients (ie, those who took more than 80% of the study medication) and pays lip service to the study's original intention-to-treat analysis. Since this was not a straight-forward analysis of clinical trial data, the authors used some fancy statistical gymnastics (Cox regression) to estimate the likelihood of various outcomes at the end of the study. After a median of 4.8 years of follow up, the intention-to-treat analysis indicated that the patients given pioglitazone had a lower rate of MI (number needed to treat [NNT] = 28); subsequent stroke (NNT = 39); acute coronary syndromes (NNT = 62); hospitalization for stroke, myocardial infarction, or heart failure (NNT = 34); and new onset diabetes (NNT = 19). Among the harms, the rate of fracture resulting in hospitalization, surgery, or other procedure was higher (number needed to treat to harm [NNTH] = 59). The rates of weight gain of 10% or more (NNTH = 7) and edema (NNTH = 8) were also higher. When analyzing only the adherent patients, the numbers were slightly better. Post hoc analyses are generally best at developing hypotheses to be tested in future studies, so stay tuned.