Development of an Automated Follow-up Tracking and Feedback System for Radiologic, Clinical, and Laboratory Studies
 
Authors:
Tessa S. Cook, MD, PhD, Hospital of the University of Pennsylvania; Jason Itri, MD, PhD; William Boonn, MD; Woojin Kim, MD
 
Hypothesis:
By using a natural language processing algorithm, is it possible to effectively and automatically mine the radiology information system (RIS) to determine when a patient has received a follow-up study?
 
Introduction:
Radiologists often recommend further imaging, laboratory, or clinical follow-up as part of study interpretation, but rarely receive feedback as to the results of these additional tests they recommend[1]. In most cases, the radiologist must actively seek this information by searching through multiple different electronic medical records. There has been previous work examining whether self-referral in radiology contributes significantly to imaging-related health care costs[3], as well as determining how radiologists’ levels of experience affect their likelihood to order follow-up imaging[4]. Additional work has been done to evaluate how practice habits of primary care physicians are altered by analysis of the laboratory and radiologic studies they order[5-6]. However, none of these studies directly address how radiologists recommend follow-up studies based on imaging. In earlier work, we have developed a set of key phrases that radiologists use to recommend feedback in their reports. In this study, we determine if this set of phrases can be used to implement an automated follow-up tracking system that alerts radiologists when patients receive radiologist-recommended follow-up studies.
 
Methods:
Using a newly developed search engine at our institution that mines radiology reports (PRESTO), we formulated an algorithm to search radiology reports for phrases commonly used by radiologists to recommend follow-up testing and to determine if patients actually received the suggested test[2]. The algorithm searches through all the reports generated on a given day, then flags those patients whose reports include one or more of the phrases in the set. Each subsequent day’s reports are then matched against the list of patients thought to be awaiting follow-up, to determine if the recommended study was, in fact, obtained.
 
Results:
We have developed an automated online follow-up and feedback application for radiologists that analyzes reports generated each day in search of follow-up recommendations. Reports are simultaneously parsed to determine if they meet the criteria to represent a follow-up study. Patients whose studies recommend follow-up are added to a tracking database, while patients who have received follow-up are summarized in a report that outputs the text of the initial and follow-up studies.
 
Discussion:
Providing radiologists with feedback regarding follow-up imaging studies they have recommended can be valuable in many respects and, ultimately, improve patient care. For example, if a patient has been imaged frequently and repeatedly for a stable lesion, analysis of the imaging timeline may be helpful in clinical decision-making or in avoiding unnecessary subsequent follow-up. Furthermore, if follow-up after a specific interval had been suggested (e.g., repeat CT scan in 6-9 months), temporal analysis may reveal whether or not the patient received attention within the recommended period of time. Having an automated feedback system will provide radiologists with an additional perspective, as the feedback can be incorporated into the RIS and summarized for easy access by the radiologist tasked with interpreting a follow-up study. Understanding the ramifications of recommending follow-up can enable radiologists to adjust and improve their practice habits, and can provide valuable information to trainees, newly board-certified radiologists, or experienced radiologists who have recently undertaken a new subspecialty. Analysis of the impact of follow-up recommendations can also lead to increased workflow efficiency and decreased healthcare costs, by resulting in fewer unnecessary imaging studies.
 
Conclusion:
The algorithm developed in this work is the first step toward an automated feedback system that will keep track of follow-up recommendations made by radiologists in everyday practice, and alert them if and when patients receive the suggested further testing. The analysis of follow-up recommendations and their consequences has great potential to enable radiologists to adjust their practice and decision-making, ultimately improving patient care, increasing workflow efficiency and decreasing healthcare costs.
 
References:
1. Boonn WW, et al. “Radiologist Use of and Perceived Need for Patient Data Access.” Journal of Digital Imaging. 2009;22(4):357-62.

2. Cook TS, et al. “Analyzing How Radiologists Recommend Follow-up: Towards Development of an Automated Tracking and Feedback System for Clinical, Laboratory, and Radiologic Studies.” SPIE 2010, submitted.

3. Lee SI, et al. “Does Radiologist Recommendation for Follow-up with the Same Imaging Modality Contribute Substantially to High-Cost Imaging Volume?” Radiology. 2007;242:857-864.

4. Molins E, et al. “Association between Radiologists’ Experience and Accuracy in Interpreting Screening Mammograms.” BMC Health Services Research. 2008;8:91+.

5. Ramsay CR, et al. “Assessing the long-term effect of educational reminder messages on primary care radiology referrals.” Clinical Radiology. 2003;58(4):319-21.

6. Thomas RE, et al. “Effect of enhanced feedback and brief educational reminder messages on laboratory test requesting in primary care: a cluster randomised trial.” Lancet. 2006;367(9527):1990-6.