Using medical language processing to support continuous quality improvement in radiology.

摘要

A medical language processing (MLP) system can be used to extract clinical concepts from radiology narrative reports that are useful for specific quality assessment and improvement applications in radiology. Three studies are presented in this dissertation to support this hypothesis. In all three studies an MLP system called SymText, postprocessed by algorithmic rules, was used to extract clinical concepts that could support different quality improvement projects in a radiology department.; In the first study, SymText was used to capture American College of Radiology utilization review codes from head CT scan reports. The system performed similarly to physicians on extracting those codes. The encoded information has the potential to support clinical radiology quality improvement applications such as utilization review studies and physician profiling.; In the second study, SymText was used to automatically classify Ventilation/Perfusion (V/Q lung scan and pulmonary angiogram reports into the criteria proposed by the PIOPED (Prospective Investigation of Pulmonary Embolism Diagnosis) study. The classification proposed by SymText achieved accuracy above 95%. The automatic classification of those reports can support quality initiatives that require correlation between V/Q lung scans and pulmonary angiograms. The correlation between radiology procedure and outcome data have been proposed as a quality indicator in a radiology department. In the third study, SymText was used to extract evidence of acute bacterial pneumonia described in chest x-ray reports. The performance of the system was indistinguishable from that of four board certified physicians. The automatic extraction of pneumonia information form chests x-ray reports can be used for different quality initiatives. A quality study on diagnostic interpretations of radiologists could compare pneumonia interpretations with outcome data from other sources such as the discharge diagnosis. It could also support peer review quality studies in diagnostic interpretations of the radiologists that usually require encoded data from chest x-rays reports.; In summary, an MLP system can be used to extract clinical concepts that are useful for quality improvement applications in radiology. It remains to be determined how the data extracted by the system will be used by radiology to implement the full cycle prescribed in continuous quality improvement theory.