Predictive Modeling of Call Outcomes to Poison Control Center Recommendations

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Over 4 million poisoning episodes occur in the US annually with hospitalization occurring in 300,000. During the 90's, the death rate by poisoning increased by 56% and is now the second leading cause of injury-related deaths. The nation's 61 Poison Control Centers (PCCs) handle approximately 60% of all annual poisoning cases via telephone services. In responding to these phone calls, PCC staff assess the likelihood of adverse medical outcomes secondary to poisonings. Their role is critical in making efficient use of emergency health care services-triaging those individuals who can be managed on site and referring those who may need emergency medical care. Furthermore, HRSA recommended that PCCs serve as the nation's first response system to bioterrorism events. PCC services are dependent on the accurate, rapid, efficient telephone consultation provided by poison control specialists. This focus of this application is the development of an evidence base for PCCs and their staff to use in responding to the increasing national problem of poisoning. Using data from a regional PCC, we propose to develop and test multivariate models of call outcomes to PCC recommendations using behavioral science and informatics-based methodology. The first arm of the study focuses on a modifiable factor-the communication process that occurs during calls at a regional PCC. One thousand calls will be coded with a widely used medical communication coding system. These calls will be stratified based on exposee age and surge (i.e., incidence of high call volume). Guided by a relationship-centered care framework, we will conduct path analyses to test the mediational role specific communications strategies play between a priori selected, nonmodifiable factors (e.g., surge, severity,) and call outcomes. In the second arm of the project, predictive models of call outcomes, based on routinely collected clinical data for one year will be created and evaluated as a potential basis for clinical decision support applications to promote optimal PCC call outcomes. Data mining methods will be used to identify patterns of both coded and textual data and then used to create predictive models. Finally, we will synthesize the findings from Arms 1 and 2 into an exploratory hybrid model. Unique nonmodifiable clinical features identified from Arm 2 will be assessed for their predictive relationship to communication patterns and to call outcomes within the 1000 recorded calls. These Arm 2 features are likely to include and expand upon the nonmodifiable, a priori variables used in Arm 1. This hybrid model-testing will potentially allow us expand the application of communication intervention strategies (resulting from time-intensive quantitative coding) by the use of information derived from large scale predictive modeling with the ultimate goal of promoting optimal PCC call outcomes, and thus reducing adverse health effects.