Advertisement

Creation of a quality-improvement database for transurethral resection of bladder tumors

Login to Access Video or Poster Abstract: MP10-18
Sources of Funding: Use of the Northwestern Medicine Enterprise Data Warehouse was in part supported by Northwestern University Clinical and Translational Sciences Institute (NUCATS) grant UL1RR025741.

Introduction

Presence of muscle in transurethral resection of bladder tumor (TURBT) specimens is an important indication of quality of endoscopic resection. The presence or absence of muscle should be noted by the pathologist, and a sample is usually only considered adequate if there is muscle present. The objective of this study is to create natural language processing to evaluate the quality of TURBT specimens amongst many surgeons at a large institution.

Methods

The Enterprise Data Warehouse at Northwestern University was used to perform a retrospective analysis of patients undergoing TURBT over 10 years. Natural language processing was used to extract stage, grade, and muscle presence information from TURBT pathology reports. Initial construction of programming language was performed using a manually-created training set of 867 TURBTs. Outcomes included (1) rates of pathology reports mentioning the presence or absence of muscle, and (2) for pathology reports that mentioned muscle, rates of muscle presence in the surgical specimen. Since tumors that were cT2 involved muscle, these were excluded from the analysis. Logistic regression analysis was performed to determine associations with muscle being mentioned and present.

Results

3042 TURBTs from 1324 patients performed by 20 surgeons were included in the database. Validation of 150 randomly-selected data points generated with our algorithm revealed accuracy of 98.7%. Muscle was mentioned in 72% of all 2918 TURBTs stage

Conclusions

Automated natural-language processing algorithm was used to create a TURBT database for quality improvement. Patients with T1 disease are more likely to have muscle mentioned and present in the report, and variations in muscle sampling exist amongst surgeons. This algorithm could be portable among medical systems and allow for large-scale quality initiatives between institutions.

Funding

Use of the Northwestern Medicine Enterprise Data Warehouse was in part supported by Northwestern University Clinical and Translational Sciences Institute (NUCATS) grant UL1RR025741.

Authors
Jason Cohen
Alex Glaser
Leslie Okorji
Daniel Oberlin
Joshua Meeks
back to top