From the 2018 HVPAA National Conference
Edward Herskovits (The University of Maryland), Jean Jeudy (The University of Maryland)
Background
Pulmonary nodules (PNs) are commonly encountered in clinical practice. The Fleischner Society first published recommendations for follow-up of PNs in 2006, and published revised Guidelines in 2017. Although awareness of the Fleischner Society guidelines is widespread, compliance is variable.
Objectives
To facilitate compliance, we developed and completed preliminary evaluation of machine-readable natural-language-processing (NLP) rules that implement the 2017 Fleischner Society guidelines for solitary PNs. We evaluated these rules using a corpus of chest-CT reports describing a solitary PN.
Methods
One of us developed an application that analyzes radiology reports during dictation; it has 3 major components. A communication module obtains information from PowerScribe 360; a NLP module evaluates the current report; and a user interface changes color to alert the radiologist to potential errors in the report, or to suggest follow-up recommendations. Working from sample reports describing PNs, we constructed 6 rules that detect solitary-PN statements in free text, and 3 rules that extract PN size and compare it with the Fleischner solitary-PN thresholds: < 6 mm, 6-8 mm, and > 8 mm diameter.
Results
Of 276 outpatient chest-CT reports during the last 3 years that contained the word “nodule”, 202 described a solitary PN. We evaluated these reports in batch mode using our application and the 9 rules described above. Our rules made the correct recommendation in 107/112, 56/57, and 29/33 reports describing PNs < 6 mm, 6-8 mm, and > 8 mm, respectively. Overall accuracy was 192/202 (95%).
Conclusion
Our preliminary results indicate that machine-readable NLP rules effectively implement the 2017 Fleischner criteria for solitary PNs.
Implications for the Patient
The 2017 Fleischner criteria for solitary PNs were designed with two primary goals. First, to ensure that nodules are followed, to minimize the chance that an early lung cancer is missed. Second, to ensure that CT is not overutilized, and to minimize radiation dose, by preventing overzealous follow-up.
