Validation of an Improved Computer-Assisted Technique for Mining Free-Text Electronic Medical Records
Bibliographic record
Abstract
BACKGROUND: The use of electronic medical records (EMRs) offers opportunity for clinical epidemiological research. With large EMR databases, automated analysis processes are necessary but require thorough validation before they can be routinely used. OBJECTIVE: The aim of this study was to validate a computer-assisted technique using commercially available content analysis software (SimStat-WordStat v.6 (SS/WS), Provalis Research) for mining free-text EMRs. METHODS: The dataset used for the validation process included life-long EMRs from 335 patients (17,563 rows of data), selected at random from a larger dataset (141,543 patients, ~2.6 million rows of data) and obtained from 10 equine veterinary practices in the United Kingdom. The ability of the computer-assisted technique to detect rows of data (cases) of colic, renal failure, right dorsal colitis, and non-steroidal anti-inflammatory drug (NSAID) use in the population was compared with manual classification. The first step of the computer-assisted analysis process was the definition of inclusion dictionaries to identify cases, including terms identifying a condition of interest. Words in inclusion dictionaries were selected from the list of all words in the dataset obtained in SS/WS. The second step consisted of defining an exclusion dictionary, including combinations of words to remove cases erroneously classified by the inclusion dictionary alone. The third step was the definition of a reinclusion dictionary to reinclude cases that had been erroneously classified by the exclusion dictionary. Finally, cases obtained by the exclusion dictionary were removed from cases obtained by the inclusion dictionary, and cases from the reinclusion dictionary were subsequently reincluded using Rv3.0.2 (R Foundation for Statistical Computing, Vienna, Austria). Manual analysis was performed as a separate process by a single experienced clinician reading through the dataset once and classifying each row of data based on the interpretation of the free-text notes. Validation was performed by comparison of the computer-assisted method with manual analysis, which was used as the gold standard. Sensitivity, specificity, negative predictive values (NPVs), positive predictive values (PPVs), and F values of the computer-assisted process were calculated by comparing them with the manual classification. RESULTS: Lowest sensitivity, specificity, PPVs, NPVs, and F values were 99.82% (1128/1130), 99.88% (16410/16429), 94.6% (223/239), 100.00% (16410/16412), and 99.0% (100×2×0.983×0.998/[0.983+0.998]), respectively. The computer-assisted process required few seconds to run, although an estimated 30 h were required for dictionary creation. Manual classification required approximately 80 man-hours. CONCLUSIONS: The critical step in this work is the creation of accurate and inclusive dictionaries to ensure that no potential cases are missed. It is significantly easier to remove false positive terms from a SS/WS selected subset of a large database than search that original database for potential false negatives. The benefits of using this method are proportional to the size of the dataset to be analyzed.
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 machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.014 | 0.058 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.007 | 0.004 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
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 source (direct Gemma or distilled Codex), 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".