Mining for clinical expertise in (undocumented) order sets to power an order suggestion system.
Lists
Only you see this
Abstract
Physician orders, the concrete manifestation of clinical decision making, are enhanced by the distribution of clinical expertise in the form of order sets and corollary orders. Conventional order sets are top-down distributed by committees of experts, limited by the cost of manual development, maintenance, and limited end-user awareness. An alternative explored here applies statistical data-mining to physician order data (>330K order instances from >1.4K inpatient encounters) to extract clinical expertise from the bottom-up. This powers a corollary order suggestion engine using techniques analogous to commercial product recommendation systems (e.g., Amazon.com's "Customers who bought this…" feature). Compared to a simple benchmark, the item-based association method illustrated here improves order prediction precision from 13% to 18% and further to 28% by incorporating information on the temporal relationship between orders. Incorporating statistics on conditional order frequency ratios further refines recommendations beyond just "common" orders to those relevant to a specific clinical context.
BibTeX
@inproceedings{chen2013mining,
author = {Jonathan H. Chen and Russ B. Altman},
title = {{Mining for clinical expertise in (undocumented) order sets to power an order suggestion system.}},
booktitle = {PubMed},
year = {2013},
}Loading related content...