Optimizing the implementation of personalized risk-prediction models for venous thromboembolism among hospitalized adults

点击翻译按钮获取中文摘要

In the last 30 years, there has been no significant improvement in rates of venous thromboembolism (VTE). These blood clots develop in the limbs and can travel to the lungs and form pulmonary emboli, which are the most common cause of preventable deaths in the hospital. Currently available tools for predicting and preventing hospital-acquired VTE (HA-VTE) were developed without sufficient input from frontline clinicians, add to clinician workload, are too cumbersome to implement in daily clinical practice, exhibit poor-to-fair prediction accuracy, and do not consider the risk of bleeding complications. Importantly, use of these tools has not been shown to improve patient outcomes. A significant gap therefore exists between the current system of variable practice patterns in VTE risk assessment and the goal of driving down rates of HA-VTE and reducing preventable deaths. Our objective is to refine, implement, and test a real-time prognostic model for HA-VTE among hospitalized adults to facilitate appropriate and timely initiation of thromboprophylaxis by busy clinicians. Our multidisciplinary team has developed a model that predicts the probability of HA-VTE among all adult inpatients based on clinical factors and medical history. The model updates as the clinical scenario evolves, discriminates well between high- and low-risk patients, and exhibits superior prediction performance compared with extant risk-stratification tools. It is unknown whether use of a prognostic model for HA-VTE in clinical practice improves patient outcomes. To achieve this important objective, we will: conduct observations and interviews with clinicians to elucidate their challenges with the current risk-assessment workflow and preferences for timing, content, and visualization of a prognostic model (Aim 1); create user-friendly clinical decision support (CDS) tools—based on an accurate and validated prognostic model for HA-VTE—that can be seamlessly integrated into existing clinical workflows, simultaneously consider the risk of bleeding complications, and maximize use of electronic health record data in real time (Aim 2); and conduct a pragmatic randomized trial and implementation evaluation of the prognostic model plus CDS for prophylaxis compared with usual care for the prevention of HA-VTE. In an adaptive platform trial, we will evaluate on a prospective basis the effectiveness of model-guided CDS to reduce HA-VTE, both overall and among key patient subgroups, and study through randomization the implementation strategies that work best for clinicians and improve patient outcomes (Aim 3). We will broadly disseminate the generalizable knowledge and implementation tools that are urgently needed to prevent HA-VTE and avoid deaths in the hospital, including an implementation manual, CDS knowledge artifacts, and open-source statistical software. Relevance: Our proposal closely aligns with NHLBI objectives, namely: developing and optimizing a real-time prognostic model to prevent HA-VTE, a HLBS disease; creating sustainable, adaptive implementation strategies to reduce rates of HA-VTE; and leveraging emerging opportunities in data science through integration of multiple types of data, innovative statistical methods, and informatics methodology to facilitate broad dissemination.