Automated Extraction of Grade, Stage, and Quality Information From Transurethral Resection of Bladder Tumor Pathology Reports Using Natural Language Processing

Abstract

Purpose Bladder cancer is initially diagnosed and staged with a transurethral resection of bladder tumor (TURBT). Patient survival is dependent on appropriate sampling of layers of the bladder, but pathology reports are dictated as free text, making large-scale data extraction for quality improvement challenging. We sought to automate extraction of stage, grade, and quality information from TURBT pathology reports using natural language processing (NLP). Methods Patients undergoing TURBT were retrospectively identified using the Northwestern Enterprise Data Warehouse. An NLP algorithm was then created to extract information from free-text pathology reports and was iteratively improved using a training set of manually reviewed TURBTs. NLP accuracy was then validated using another set of manually reviewed TURBTs, and reliability was calculated using Cohen’s κ. Results Of 3,042 TURBTs identified from 2006 to 2016, 39% were classified as benign, 35% as Ta, 11% as T1, 4% as T2, and 10% as isolated carcinoma in situ. Of 500 randomly selected manually reviewed TURBTs, NLP correctly staged 88% of specimens (κ = 0.82; 95% CI, 0.78 to 0.86). Of 272 manually reviewed T1 tumors, NLP correctly categorized grade in 100% of tumors (κ = 1), correctly categorized if muscularis propria was reported by the pathologist in 98% of tumors (κ = 0.81; 95% CI, 0.62 to 0.99), and correctly categorized if muscularis propria was present or absent in the resection specimen in 82% of tumors (κ = 0.62; 95% CI, 0.55 to 0.73). Discrepancy analysis revealed pathologist notes and deeper resection specimens as frequent reasons for NLP misclassifications. Conclusion We developed an NLP algorithm that demonstrates a high degree of reliability in extracting stage, grade, and presence of muscularis propria from TURBT pathology reports. Future iterations can continue to improve performance, but automated extraction of oncologic information is promising in improving quality and assisting physicians in delivery of care.

Natural Language Processing for Automated Quantification of Brain Metastases Reported in Free-Text Radiology Reports

“…NLP for the analysis of radiology reports has also been explored in the context of other cancer types, including hepatocellular carcinomas, [13][14][15][16] breast cancer, [17][18][19][20] lung cancer, [21][22][23] and other abdominal or pelvic tumors. 11,[24][25][26][27] All studies that provided sufficient insight into their modeling approach used a bag-of-words approach. To our knowledge, this study presents the first sequence-based NLP approach for analyzing free-text radiology reports in oncology patients as well as the first head-to-head comparison of sequence-based and bag-of-words models for medical text analysis.…”

Automating the Capture of Structured Pathology Data for Prostate Cancer Clinical Care and Research

“…24 Glaser and colleagues report an NLP approach to bladder cancer pathology showing concordance of 0.82 for stage, 1.00 for grade, and 0.81 for muscularis propria invasion. 25 Thomas et al 26 reported the use of a rule-based NLP system to identify patients with prostate cancer from prostate biopsy reports and extract Gleason scores, percentage of tumor involved, and presence of atypical small acinar proliferation (ASAP), high-grade prostatic intraepithelial neoplasia (HGPIN), and perineural invasion, with accuracy between 94% and 100% in a sample of 100 reports. They also extract information from 100 radical prostatectomy pathology reports to identify detailed pathologic features with 95% to 100% accuracy.…”

“…28 Our work, using a rule-based NLP approach, achieved similar results to other nvestigators. 24,25,29 However, instead of comparing results to only manual abstraction in a test environment, we report results from a real-world clinical implementation and compare with both manual abstraction and clinician-entered SDEs at the point of care. Our parallel approach has several advantages.…”

A frame semantic overview of NLP-based information extraction for cancer-related EHR notes

“…GRADE: appearance of the cancerous cells [8], [52], [9], [48], [87], [26] INVASION: whether or not more than 50% of an organ is invaded [52], [88] SIZE: quantitative size of tumor (e.g., 2.2 x 2.0 cm), diameter/volume of the tumor, including unit (e.g., 1 cm, 0.3 x 0.5 x 0.7 cm) [13], [52], [88], [85], [42], [25], [48], [41] SIZE TYPE: radiological/pathological [25] NEGATION: indicator to some negation of a tumor reference (e.g., no) [85], [86], [41] COUNT: number of tumor/nodule references (e.g., two or multiple) [13], [88], [85] TUMOR REFERENCE: a radiologic artifact that may reference a tumor (e.g., lesion or focal density) [88], [85] MENTION: tumor major object (e.g., tumor, lesion, mass, and nodule) [13], [42], [47], [49], [41] QUANTIFIER: one, two, three, several [42] TEMPORAL INFORMATION: refers to information about time (e.g., year, month, and date, 2007-08-04) [52], [42], [86] NON-TUMOR SIZE ITEMS: LeVeen needle, which is used in RFA treatment [42] STATUS: this indicates the final overall tumor status (e.g., regression, stable, progression, irrelevant) [89], [41] METASTATIC STATUS INDICATORS: phrases denoting a metastatic tumor [9], [90] MAGNI...…”