# Evaluating predictive models for hemorrhagic transformation post-mechanical thrombectomy in acute ischemic stroke

## TL;DR

This paper evaluates predictive models for bleeding after stroke treatment to find the most accurate one for improving patient outcomes.

## Contribution

The study compares predictive models for post-MT hemorrhagic transformation using ROC curves and C-index for accuracy.

## Key findings

- Receiver operating characteristic curves and C-index were used to assess model accuracy.
- The study highlights the need for consensus on the most effective predictive model for clinical use.
- Improved prediction models can enhance tailored treatment plans and patient outcomes.

## Abstract

Acute ischemic stroke (AIS), a condition defined by a decrease in cerebral blood flow, is primarily treated through mechanical thrombectomy (MT) for blockages in major anterior circulation arteries. Approaches encompass stent retrieval, suction thrombectomy, or a combination of both. MT is increasingly recognized for its rapid revascularization, low hemorrhagic transformation (HT) rate, and extended therapeutic time window. Nonetheless, multiple risk factors lead to post-MT HT through different mechanisms, resulting in adverse outcomes such as increased mortality and morbidity. Therefore, assessing the relevant risks based on predictive models for post-MT HT is necessary. These predictive models incorporate a series of risk factors and conduct statistical analyses to generate corresponding assessment scales, which are then used to evaluate the risk of postoperative bleeding. As this is a rapidly developing field, there is still controversy over which model is more effective than another in improving clinical efficacy, and there is a lack of consensus on the comparison of these data. In this paper, we assess the accuracy of these predictive models using receiver operating characteristic (ROC) curves and the concordance C-index. Determining the most accurate predictive model for post-MT HT is crucial for improving the prediction of patient outcomes and for the development of tailored treatment plans, thereby enhancing clinical relevance and applicability.