Recorded presentation

Gordon S. Smith, MD, MPH¹, Dheeresh K. Mamidi, MBBS, MPH², David A. Lombardi, PhD¹, Simon Matz, MS¹, Robert A Timmons, DO, MPH³, and Theodore K. Courtney, MS, CSP¹. (1) Quantitative Analysis Unit, Liberty Mutual Research Institute for Safety, 71 Frankland Rd, Hopkinton, MA 01748, (508) 497-0250, Gordon.Smith@LibertyMutual.com, (2) Department of Epidemiology and Biostatistics, University of Massachusetts, 715 North Pleasant Street, 408 Arnold House, Amherst, MA 01003-9304, (3) Harvard School of Public Health, Occupational & Environmental Medicine, 401 Park Dr, Boston, MA 02215

Purpose: To evaluate the ability of narrative text searches to identify specific injury causes such as ladder falls from data containing narrative text information but lacking detailed cause of injury codes. Methods: Multiple approaches, including using coded data and narrative text searches were used to identify ladder injuries from 535,605 workers’ compensation (WC) claims. Word frequency lists from “accident description” texts determined index terms (letter groups) useful for identifying ladder injury claims. Claim reviews determined if they were true ladder falls. Results: We identified 9,748 potential ladder injuries by searching “accident” narratives using index search terms “ladd”- 9458, “lader”-15, “_lander”-2 and “latter”-273, and 21 more from injury descriptions . Subsequent searches of specialized cause and agency codes identified 152 additional ladder cases among the 7389 claims identified. While 98% of potential cases were identified using narrative text searches only 50% were identified from specific available codes associated with ladders. Reviewing 589 potential ladder cases with fractures found that 87% were falls from ladders, 3.8% were injured on a ladder, 4.8% tripped on or were struck by/against a ladder and 1.2% were moving a ladder. Another 2.2% mentioned a ladder but their injury was unrelated to ladders and 0.5% did not involve ladders. Detailed analyses comparing the sensitivity and specificity of different methods for identifying cases will also be presented. Conclusions: Narrative text analysis greatly improves the usefulness of incompletely coded data, such as workers compensation claims, for identifying specific causes of injury for use in more detailed studies.

Learning Objectives:

• to improve identification of cases by text analysis
• To evaluate the ability of narrative text searches to identify specific injury causes

Keywords: Occupational Injury and Death, Methodology

Related Web page: www.libertymutual.com/research/

Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.

Recorded presentation