This article is more than 5 years old. Information may no longer be current.
Free text search of EHRs bested current automated method
Murff HJ. JAMA. 2011;306:848-855.
Click Here to Manage Email Alerts
A natural language processing search that extracts specific medical concepts via free text searches of electronic health records identified more postoperative surgical complications vs. the current automated method based upon discharge codes, according to new findings published in the Journal of the American Medical Association.
“Administrative codes are inexpensive to obtain and readily available; however, there are major concerns regarding how accurately administration codes reflect actual substandard quality of care,” Harvey J. Murff, MD, MPH, of the Veterans Affairs Medical Center and Vanderbilt University, in Nashville, Tenn., told Infectious Disease News. “While manual chart review is more accurate, it is costly and logistically difficult to implement on a large-scale basis. Using free text searches of electronic medical records represents a potentially cost-effective and accurate means of monitoring patient safety and should be extended to other quality-of-care measures.”
In the cross-sectional study, Murff and colleagues evaluated the sensitivity and specificity of the natural language processing approach at identifying postoperative complications among the electronic health records (EHRs) of 2,974 patients undergoing inpatient surgical procedures across six Veterans Health Administration medical centers between 1999 and 2006.
Natural language processing correctly identified 82% (95% CI, 67-91) of acute renal failure cases compared with 38% (95% CI, 25-54) with administrative data coding. This was followed by 59% (95% CI, 44-72) of venous thromboembolism cases vs. 46% (95% CI, 32-60); 64% (95% CI, 58-70) of pneumonia cases vs. 5% (95% CI, 3-9); 89% (95% CI, 78-94) of sepsis cases vs. 34% (95% CI, 24-47); and 91% (95% CI, 78-97) of postoperative myocardial infarction cases vs. 89% (95% CI, 74-96).
Further, natural language processing was associated with a twofold increased sensitivity for acute renal failure and sepsis and more than a 12-fold increased sensitivity for pneumonia.
“These findings demonstrate the feasibility and utility of using computer algorithms to perform automated chart reviews,” Murff said. “A major barrier to quality improvement in health care is the lack of an accurate, reliable and cost-efficient means to routinely identify patient safety and quality-of-care concerns. With more health care systems adopting electronic medical records, this methodology could be applicable to a wide range of patient care outcomes and allow institutions to perform accurate and comprehensive monitoring of their own performance on a routine basis.” – by Ashley DeNyse
Disclosure: This study was supported by grant SAF-03-223 of the Department of Veterans Affairs.
 |
Follow InfectiousDiseaseNews.com on Twitter. |