The feasibility of using natural language processing to extract clinical information from breast pathology reports

Buckley, J. J. C.; Coopey, Suzanne B.; Sharko, John; Polubriaginof, Fernanda; Drohan, Brian; Belli, Aslı Akın; Kim, Elizabeth M. H.; Garber, Judy E.; Smith, Barbara L.; Gadd, Michele A.; Specht, Michelle C.; Roche, Constance A.; Gudewicz, Thomas M; Hughes, Kevin S.

doi:10.4103/2153-3539.97788

Cited by 95 publications

(42 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, an NLP engine developed to classify breast pathology histology achieved a PPV of 0.97 when compared to human annotators as the gold standard. 18 Similarly, NLP has been used to abstract histology from colon pathology reports and clinically relevant variables from prostatectomy pathology reports. 14,19 These prior results – as well as our own – demonstrate that NLP is highly accurate to retrieve certain components of pathology reports such as histology and grade.…”

Section: Discussionmentioning

confidence: 99%

Development of a Natural Language Processing Engine to Generate Bladder Cancer Pathology Data for Health Services Research

Schroeck

Patterson

Alba

et al. 2017

Urology

View full text Add to dashboard Cite

Objective To take a first step towards assembling population based cohorts of bladder cancer patients with longitudinal pathology data, we developed and validated a natural language processing (NLP) engine that abstracts pathology data from full text pathology reports. Methods Using 600 bladder pathology reports randomly selected from the Department of Veterans Affairs, we developed and validated an NLP engine to abstract data on histology, invasion (presence versus absence and depth), grade, presence of muscularis propria, and presence of carcinoma in situ. Our gold standard was based on independent review of reports by two urologists, followed by adjudication. We assessed NLP performance by calculating accuracy, positive predictive value (PPV), and sensitivity. We subsequently applied the NLP engine to pathology reports from 10,725 bladder cancer patients. Results When comparing the NLP output to the gold standard, NLP achieved the highest accuracy (0.98) for presence versus absence of carcinoma in situ. Accuracy for histology, invasion (presence versus absence), grade, and presence of muscularis propria ranged from 0.83 to 0.96. The most challenging variable was depth of invasion (accuracy 0.68), with acceptable PPV for lamina propria (0.82) and muscularis propria (0.87) invasion. The validated engine was capable of abstracting pathologic characteristics for 99% of bladder cancer patients. Conclusions NLP had high accuracy for five of six variables and abstracted data for the vast majority of patients. This now allows for assembly of population based cohorts with longitudinal pathology data.

show abstract

Section: Discussionmentioning

confidence: 99%

Development of a Natural Language Processing Engine to Generate Bladder Cancer Pathology Data for Health Services Research

Schroeck

Patterson

Alba

et al. 2017

Urology

View full text Add to dashboard Cite

show abstract

“…This result is in line with the experience of others. [37383940] In contrast, with our embedded structure of pipes, a single simple algorithm allowed extraction of 100% of data from all tumor sites. Thus, we believe that embedding an appropriate structure into the synoptic report can significantly improve the success of extracting a structured data file from that report using free text searches.…”

Section: Discussionmentioning

confidence: 99%

Performance of a Web-based Method for Generating Synoptic Reports

Renshaw

Mena-Allauca

et al. 2017

Journal of Pathology Informatics

View full text Add to dashboard Cite

Context:The College of American Pathologists (CAP) requires synoptic reporting of all tumor excisions.Objective:To compare the performance of different methods of generating synoptic reports.Methods:Completeness, amendment rates, rate of timely ordering of ancillary studies (KRAS in T4/N1 colon carcinoma), and structured data file extraction were compared for four different synoptic report generating methods.Results:Use of the printed tumor protocols directly from the CAP website had the lowest completeness (84%) and highest amendment (1.8%) rates. Reformatting these protocols was associated with higher completeness (94%, P < 0.001) and reduced amendment (1%, P = 0.20) rates. Extraction into a structured data file was successful 93% of the time. Word-based macros improved completeness (98% vs. 94%, P < 0.001) but not amendment rates (1.5%). KRAS was ordered before sign out 89% of the time. In contrast, a web-based product with a reminder flag when items were missing, an embedded flag for data extraction, and a reminder to order KRAS when appropriate resulted in improved completeness (100%, P = 0.005), amendment rates (0.3%, P = 0.03), KRAS ordering before sign out (100%, P = 0.23), and structured data extraction (100%, P < 0.001) without reducing the speed (P = 0.34) or accuracy (P = 1.00) of data extraction by the reader.Conclusion:Completeness, amendment rates, ancillary test ordering rates, and data extraction rates vary significantly with the method used to construct the synoptic report. A web-based method compares favorably with all other methods examined and does not reduce reader usability.

show abstract

“…The previously reported methods for data extraction used two-step approaches[789] including obtaining the report texts in step one and performing extraction and analysis in step two. The integrated one-step approach enables the second round of queries of database using the results of the first query.…”

Section: Discussionmentioning

confidence: 99%

“…Natural language processing (NLP) has been used to extract information from breast carcinoma pathology reports with variable degrees of success. [78] Recently, Boag described a simpler yet powerful approach: programing language R was used to extract and analyze data from discrete synoptic pathology reports (from the reports of prostate needle core biopsies). [9] First, all the reports with synoptic reports of prostate needle core biopsies were retrieved using a built-in report retrieving mechanism of their pathology information system.…”

Section: Introductionmentioning

confidence: 99%

Pathology report data extraction from relational database using R, with extraction from reports on melanoma of skin as an example

2016

Journal of Pathology Informatics

View full text Add to dashboard Cite

Background:Different methods have been described for data extraction from pathology reports with varying degrees of success. Here a technique for directly extracting data from relational database is described.Methods:Our department uses synoptic reports modified from College of American Pathologists (CAP) Cancer Protocol Templates to report most of our cancer diagnoses. Choosing the melanoma of skin synoptic report as an example, R scripting language extended with RODBC package was used to query the pathology information system database. Reports containing melanoma of skin synoptic report in the past 4 and a half years were retrieved and individual data elements were extracted. Using the retrieved list of the cases, the database was queried a second time to retrieve/extract the lymph node staging information in the subsequent reports from the same patients.Results:426 synoptic reports corresponding to unique lesions of melanoma of skin were retrieved, and data elements of interest were extracted into an R data frame. The distribution of Breslow depth of melanomas grouped by year is used as an example of intra-report data extraction and analysis. When the new pN staging information was present in the subsequent reports, 82% (77/94) was precisely retrieved (pN0, pN1, pN2 and pN3). Additional 15% (14/94) was retrieved with certain ambiguity (positive or knowing there was an update). The specificity was 100% for both. The relationship between Breslow depth and lymph node status was graphed as an example of lesion-specific multi-report data extraction and analysis.Conclusions:R extended with RODBC package is a simple and versatile approach well-suited for the above tasks. The success or failure of the retrieval and extraction depended largely on whether the reports were formatted and whether the contents of the elements were consistently phrased. This approach can be easily modified and adopted for other pathology information systems that use relational database for data management.

show abstract

The feasibility of using natural language processing to extract clinical information from breast pathology reports

Cited by 95 publications

References 7 publications

Development of a Natural Language Processing Engine to Generate Bladder Cancer Pathology Data for Health Services Research

Development of a Natural Language Processing Engine to Generate Bladder Cancer Pathology Data for Health Services Research

Performance of a Web-based Method for Generating Synoptic Reports

Pathology report data extraction from relational database using R, with extraction from reports on melanoma of skin as an example

Contact Info

Product

Resources

About