Retrospective evaluation of a machine learning model to facilitate pharmacogenetic testing
Bibliographic record
Abstract
Objectives: To describe the epidemiology of nine medications with pharmacogenetic guidance (targeted medications) and frequency of pharmacogenetic testing; and to develop a retrospective machine learning (ML) model to predict prescription of targeted medications within 3 and 6 months of admission. Methods and Analysis: For the epidemiology aim, the cohort included patients prescribed at least one targeted medication. Pharmacogenetic testing rates were determined before and after first targeted medication prescription. For the ML aim, the cohort included all inpatient admissions. Outcome was receipt of a targeted medication within 3 or 6 months. Models were trained using L2-regularised logistic regression and two gradient boosting machine frameworks (LightGBM and XGBoost). Data were temporally split into training (80%), validation (10%) and test (10%) sets. Results: For the epidemiology cohort, 4520 patients were prescribed at least one targeted medication. Only 4.3% (n=194) had pharmacogenetic testing performed at any point, and only 1.0% (n=44) completed testing before the first prescription. For the ML cohort, 57 368 admissions were included. LightGBM was the best-performing ML model. For the prediction of prescription of a targeted medication at 3 and 6 months, area under the receiver operating characteristic curves was 0.926 (95% CI 0.911 to 0.939) and 0.922 (95% CI 0.911 to 0.932), respectively. Area under the precision recall curves was 0.477 (95% CI 0.462 to 0.495) and 0.450 (95% CI 0.411 to 0.494), respectively. Conclusion: Only 1% of patients receiving a targeted medication had pharmacogenetic testing before the medication order. We developed an ML model with acceptable performance to predict targeted medication administration with the goal of facilitating earlier pharmacogenomic testing.
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 | 0.012 | 0.025 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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".