It is time for some deep learning: a statistical commentary on machine learning for clinical prediction models using imbalanced datasets
Bibliographic record
Abstract
Training in research and statistical techniques are increasingly a core part of modern medical training. From undergraduate and medical school courses to board exams and journal clubs, physicians and surgeons are generally well-equipped to interpret contemporary studies that inform evidence-based practice. Machine learning (ML), however, so far lacks this integration. ML is a field of study in artificial intelligence (AI) that seeks to use algorithms that are capable of independently ‘learning’ to optimize their own performance. ML techniques are being increasingly applied in medical and surgical research.1 Since 2010, the number of PubMed articles containing the term ‘machine learning’ in the title or abstract has accelerated from 560 to 28 000 articles per year. Yet for those without training in applied mathematics or computer science, these techniques remain unfamiliar and inaccessible. The most advanced ML models, which are highly flexible and capable of mapping complex relationships, are also the most difficult to interpret (the AI black box problem).2 Standards for reporting and evaluating model performance are also highly variable. Taken together, these challenges represent a major barrier to implementation of ML-based prediction tools in clinical practice.
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 | 0.044 | 0.187 |
| Meta-epidemiology (narrow) | 0.002 | 0.002 |
| Meta-epidemiology (broad) | 0.003 | 0.004 |
| Bibliometrics | 0.002 | 0.003 |
| Science and technology studies | 0.008 | 0.013 |
| Scholarly communication | 0.009 | 0.012 |
| Open science | 0.007 | 0.004 |
| Research integrity | 0.090 | 0.131 |
| Insufficient payload (model declined to judge) | 0.007 | 0.008 |
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".