Application of Machine Learning to Auto-Code Injury Data in the e-CHIRPP System: Development and Evaluation Study
Bibliographic record
Abstract
Background: The Canadian Hospitals Injury Reporting and Prevention Program (CHIRPP), a Public Health Agency of Canada program established in 1990, is an injury and poisoning sentinel surveillance system that collects and analyzes data on injuries to individuals who are seen at the emergency departments of numerous pediatric and general hospitals in Canada. Since its inception, the program has collected over 4 million records. The program's surveillance activities have contributed substantially to evidence-based decision-making to reduce injuries, support research, and establish preventive safeguards to protect the health and safety of Canadians. Patients presenting at participating hospitals are asked to complete a data collection form capturing the causes and circumstances contributing to the injury or poisoning event. Using this text, hospital and program staff have traditionally coded numerous surveillance variable codes manually for subsequent analysis within e-CHIRPP, the program's purpose-built analytical application on the Canadian Network for Public Health Intelligence public health informatics platform. Manual coding of this complex data is administratively burdensome and results in a significant time lag in the availability of important surveillance findings. Objective: With the initial goal of achieving a preliminary stage of implementation, the objective was to establish the capability to achieve enhanced timeliness of surveillance findings within a process of adaptability and continuous improvement by applying machine learning to auto-code injury data based on patient narratives. Methods: The research, development, and implementation of machine learning and auto-coding within the e-CHIRPP system were led by the Canadian Network for Public Health Intelligence team in collaboration with the CHIRPP program team. Data were extracted from e-CHIRPP and prepared for training, and candidate algorithms well suited for classification and supervised learning were initially assessed. Subsequently, 1 algorithm was chosen for further assessment based on initial accuracy, prediction confidence, and training time. The chosen algorithm was then further assessed in 2 stages, again using e-CHIRPP extracts: first, for a 2-year data set and then again for a 7-year data set. The sources of inaccuracies were investigated with a view to informing the refinement of the overall process and establishing ongoing adaptability and continuous improvement. Results: Auto-coding of injury variables showed a high level of accuracy in most cases when compared to variables previously coded manually. Importantly, insights were also gained into the sources of observed inaccuracies and measures to foster ongoing refinement of the process. Conclusions: The application of machine learning and auto-coding shows strong potential to benefit surveillance activities across various public health disciplines, yielding near real-time availability of intelligence, reduced administrative workload, continuous improvement, and adaptability to database growth and change.
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 | 0.016 | 0.002 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| 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".