A Natural Language Processing–Assisted Extraction System for Gleason Scores: Development and Usability Study
Bibliographic record
Abstract
BACKGROUND: Natural language processing (NLP) offers significantly faster variable extraction compared to traditional human extraction but cannot interpret complicated notes as well as humans can. Thus, we hypothesized that an "NLP-assisted" extraction system, which uses humans for complicated notes and NLP for uncomplicated notes, could produce faster extraction without compromising accuracy. OBJECTIVE: The aim of this study was to develop and pilot an NLP-assisted extraction system to leverage the strengths of both human and NLP extraction of prostate cancer Gleason scores. METHODS: We collected all available clinical and pathology notes for prostate cancer patients in an unselected academic biobank cohort. We developed an NLP system to extract prostate cancer Gleason scores from both clinical and pathology notes. Next, we designed and implemented the NLP-assisted extraction system algorithm to categorize notes into "uncomplicated" and "complicated" notes. Uncomplicated notes were assigned to NLP extraction and complicated notes were assigned to human extraction. We randomly reviewed 200 patients to assess the accuracy and speed of our NLP-assisted extraction system and compared it to NLP extraction alone and human extraction alone. RESULTS: Of the 2051 patients in our cohort, the NLP system extracted a prostate surgery Gleason score from 1147 (55.92%) patients and a prostate biopsy Gleason score from 1624 (79.18%) patients. Our NLP-assisted extraction system had an overall accuracy rate of 98.7%, which was similar to the accuracy of human extraction alone (97.5%; P=.17) and significantly higher than the accuracy of NLP extraction alone (95.3%; P<.001). Moreover, our NLP-assisted extraction system reduced the workload of human extractors by approximately 95%, resulting in an average extraction time of 12.7 seconds per patient (vs 256.1 seconds per patient for human extraction alone). CONCLUSIONS: We demonstrated that an NLP-assisted extraction system was able to achieve much faster Gleason score extraction compared to traditional human extraction without sacrificing accuracy.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".