MASK: A Success Story for An International Collaboration
Bibliographic record
Abstract
IntroductionA significant amount of valuable information in Electronic Health Records (EHR) such as laboratory test results or echocardiogram interpretations is embedded in lengthy free-text fields. Often patients’ personal information is also included in these narratives. Privacy legislation in different jurisdictions requires de-identification of this information prior to making it available for research. This process can be challenging and time-consuming. In particular, rule-based algorithms may lead to over-masking of essential medical terms, conditions, or devices that are named after individuals. Objectives and ApproachWe aimed to enhance ICES’ existing rule-based application to make it contextually-driven by applying Artificial Intelligence (AI). The ICES team collaborated with computer scientists at the University of Manchester who had already published work in this area and Evenset, a Toronto-based software company. Based on the Manchester University de-identification framework for name entity recognition, three machine learning-based algorithms for name entity recognition were implemented: CRF, BiLSTM recurrent neural networks with GLoVe and ELMo word embeddings. The models were trained on three different types of ICES data: Laboratory results, Electronic Medical Record (EMR) and echocardiogram data. Evenset developed the user interface and the masking modules. ResultsPreliminary tests have generated very promising results. To improve accuracy of the models, additional data annotation to expand the training datasets is currently being undertaken at ICES. The final framework will be available as an open-source tool for public. Conclusion / ImplicationsA collaborative approach for solving complex problems like de-identification of text-based medical data is highly efficient, especially where there are unique sets of expertise, resources, data and clinical knowledge among stakeholders.
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 | 0.015 | 0.022 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.014 | 0.007 |
| Scholarly communication | 0.017 | 0.018 |
| Open science | 0.002 | 0.018 |
| Research integrity | 0.009 | 0.012 |
| Insufficient payload (model declined to judge) | 0.038 | 0.015 |
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".