Can We Do More With Less While Building Predictive Models? A Study in Parsimony of Risk Models for Predicting Heart Failure Readmissions
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Hospital readmission due to heart failure is a topic of concern for patients and hospitals alike: it is both the most frequent and expensive diagnosis for hospitalization. Therefore, accurate prediction of readmission risk while patients are still in the hospital helps to guide appropriate postdischarge interventions. As our understanding of the disease and the volume of electronic health record data both increase, the number of predictors and model-building time for predicting risk grow rapidly. This suggests a need to use methods for reducing the number of predictors without losing predictive performance. We explored and described three such methods and demonstrated their use by applying them to a real-world dataset consisting of 57 variables from health data of 1210 patients from one hospital system. We compared all models generated from predictor reduction methods against the full, 57-predictor model for predicting risk of 30-day readmissions for patients with heart failure. Our predictive performance, measured by the C-statistic, ranged from 0.630 to 0.840, while model-building time ranged from 10 minutes to 10 hours. Our final model achieved a C-statistic (0.832) comparable to the full model (0.840) in the validation cohort while using only 16 predictors and providing a 66-fold improvement in model-building time.
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.
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 | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| 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 it