Department of Medical Informatics and Clinical Epidemiology
Oregon Health & Science University
Introduction Health care providers place a variety of orders in many clinical contexts: ambulatory, inpatient, emergency department, and beyond. Laboratory tests are a ubiquitous and sometimes costly aspect of medical care. Providers may order laboratory tests to assist with diagnosis, based on antiquated training, reflexively to mitigate liability, as a pre-requisite to pharmacologic therapy, or automatically as a part of an order set. Providers have little, if any, assistance from electronic health records (EHRs) to guide their understanding of trends in their own normal or abnormal lab results. The project aim was to develop a framework to visualize analytics and disparities in laboratory test results ordered by providers. Laboratory tests are the focus of the investigation given the relative ease of data analysis of numerical values. The conceptual principle may, however, be extended to a wide array of orders (e.g., radiology results, cardiac testing, or pathology results). The visualization framework is intended to provide information to providers at the point of order entry to identify their historical rates of abnormal (or normal, depending on context) results. Such data visualization is intended to exist as a contextually-based infobutton rather than a mandatory clinical decision support (CDS) alert. Contextual information (e.g., diagnoses) is considered along with correlation to colleagues’ historical data and patient outcomes.
Methods Data utilized in the study were prospectively collected on patients admitted to the Cleveland Clinic main campus medical intensive care unit between April 2010 and December 2013. Data elements available included demographics (gender, age), orderspecific details (attending physician, order date and time, and order status), and lab results (result date and time, result, abnormally low or high, and normal laboratory values). Additional data were obtained from a business intelligence system, which included International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis codes. Results The resulting visualization framework consists of components allowing data analysis, as desired by the user, in a variety of fashions. Each order’s constituent components are analyzed separately. Histograms allow the provider to review data compared to colleagues when focusing on normal, abnormal, or all values. The limits of normal are adjustable, depending on the provider’s preferences and possibly clinical context. For each order, the diagnoses most frequently associated with the order and odds ratios are listed in descending order, by the provider along with colleagues for comparison. Finally, to avoid overwhelming providers in anticipation of using the framework in a clinical environment, brief summarized notes of diagnosis-based context specific data are provided. The notes alone could be used in an eventual production environment to help providers change ordering behavior.
Conclusion Health care providers, limited by time and accessibility of clinical data through EHRs, need opportunities to determine appropriateness of clinical orders. This work demonstrates the potential infobutton-type approach to allowing providers an opportunity to evaluate their own performance in comparison to colleagues in a patient-specific contextual fashion. While constant alerts and reminders provide limited benefit due to fatigue, personalized data should be made available to interested providers to help make meaningful and informed patient care decisions.
School of Medicine
Rosenbaum, Benjamin P., "Modeling disparities in provider-ordered test results" (2014). Scholar Archive. 3530.