Timely identification of relevant or ?need to know? clinical information about a patient?s history in the acute care setting can be critical for patient safety and medical decision-making. Often the relevant information is buried in unstructured or free-text narratives within the Electronic Health Record (EHR), making it difficult to access in a timely fashion. Currently, it is estimated that over 50% of the EHR is free-text. EHR search tools today are often inefficient, simplistic, and unable to rank or evoke the relevance of information for a particular problem or complaint. This is compounded by the fact that EHRs are amassing clinical information at an exponential rate. While the benefits of having a wealth of information at a provider?s fingertips seem obvious, the time and energy cost of culling through enormous amounts of data creates new issues of decreased efficiency and information overload for providers seeking to identify the most pertinent and relevant information about their patients. In the emergency department, where patients can present with life threatening conditions, timely unlocking of clinically relevant information for a patient?s problem or complaint at the point of care can be critical to medical decision-making and patient safety. In this study, we plan to address this challenge through the development of a sophisticated natural language processing (NLP) search tool to automatically identify and rank clinically relevant information based on the patients presenting complaint. We will accomplish this through the following specific aims: 1) identify and define complaint- specific information elements within a patient?s history and 2) develop and test an NLP-based information retrieval tool.
Patient safety hinges on having right information about the right patient and the right time. Often the relevant information is buried in unstructured or free-text narratives within the Electronic Health Record (EHR), making it difficult to access in a timely fashion. The purpose of this study is to develop and evaluate a sophisticated natural language processing (NLP) search tool to automatically identify and rank clinically relevant information from EHRs that providers rely upon to make medical decisions for their patients. This study comes at an important time where data in the EHRs is increasing at an exponential rate, creating a new problem for clinicians, that of finding all the relevant information for patient?s particular problem. This is particularly true in the emergency department setting where providers have limited if any prior relationship and often have to make quick decisions for patients with life threatening conditions. This search tool will provide a snapshot of clinically relevant information that the providers can view alongside the structured information already in the EHR. We believe this approach has the potential to increase clinician efficiency, decrease healthcare costs by avoiding duplicate diagnostic tests, and provide clinicians with the tools they need to make well-informed medical decisions, thereby improving patient safety and reducing suffering.
| Cohen, K Bretonnel; Xia, Jingbo; Zweigenbaum, Pierre et al. (2018) Three Dimensions of Reproducibility in Natural Language Processing. LREC Int Conf Lang Resour Eval 2018:156-165 |
