Developing Machine Learning Algorithms on Routinely Collected Administrative Health Data - Lessons from Ontario, Canada.
Bibliographic record
Abstract
There has been considerable growth in the development of machine learning models for clinical applications; however, less attention has been paid to applications at the health systems level. Here, we survey recent models developed using provincial administrative health data holdings in Ontario, Canada to synthesize key learnings across use cases. We have developed four models in the areas of diabetes incidence and complications, hospitalization due to ambulatory care sensitive conditions, and hospitalization due to SARS-CoV-2 infection. Our team was highly multidisciplinary with expertise across clinical medicine, administrative health data, epidemiology, and computer science. We used a “sliding window” approach to aggregate healthcare events across multiple health administrative data sets chronologically and map them dynamically onto a patient timeline. Tree-based algorithms, specifically gradient boosted decision trees, are well suited for the underlying tabular structure of administrative data and were used for each prediction task. Our models achieved excellent discrimination, measured by the area under the receiver operating characteristic curve, between 0.77-0.85 at prediction windows between 30 days and 3 years in advance. They were also well-calibrated, both in-the-large and in population subgroups such as older adults, those living in rural areas, and the materially deprived. Measures of feature importance revealed that our models were leveraging predictors across administrative datasets (e.g. demographics, interactions with the healthcare system, medications) in intuitive and defensible ways. Finally, we demonstrated the utility of our models with “recall at top k” metrics - for example, the top 1% of patients predicted at risk of diabetes complications represented a cost of over $400 million to the healthcare system. We identify three key learnings needed for the successful application of machine learning methods to health administrative data: synergy between nature of training data and intended algorithm use, adherence to methodological best practices for rigour and transparency, and multidisciplinary teams with expertise across data provenance, methodological approach, and impact assessment.
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 | 0.003 | 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 | 0.003 | 0.000 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.010 | 0.004 |
| Research integrity | 0.000 | 0.001 |
| 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".