A comparative study of machine learning models predicting post-hepatectomy liver failure: Enhancing risk estimation in over 25,000 National Surgical Quality Improvement Program patients
Bibliographic record
Abstract
Backgrounds/Aims: Post-hepatectomy liver failure (PHLF) is a significant complication with an incidence rate between 8% and 12%. Machine learning (ML) can analyze large datasets to uncover patterns not apparent through traditional methods, enhancing PHLF prediction and potentially mitigate complications. Methods: Using the National Surgical Quality Improvement Program (NSQIP) database, patients who underwent hepatectomy were randomized into training and testing sets. ML algorithms, including LightGBM, Random Forest, XGBoost, and Deep Neural Networks, were evaluated against logistic regression. Performance metrics included receiver operating characteristic area under the curve (ROC AUC) and Brier score loss. Shapley Additive exPlanations was used to identify individual variable relevance. Results: 28,192 patients from 2013 to 2021 who underwent hepatectomy were included; PHLF occurred in 1,305 patients (4.6%). Preoperative and intraoperative factors most contributed to PHLF. Preoperative factors were international normalized ratio > 1.0, sodium < 139 mEq/L, albumin < 3.9 g/dL, American Society of Anesthesiologists score > 2, total bilirubin > 0.65 mg/dL. Intraoperative risks include transfusion requirements, trisectionectomy, operative time > 266.5 minutes, open surgical approach. The LightGBM model performed best with an ROC AUC of 0.8349 and a Brier Score loss of 0.0834. Conclusions: While topical, the role of ML models in surgical risk stratification is evolving. This paper shows the potential of ML algorithms in identifying important subclinical changes that could affect surgical outcomes. Thresholds explored should not be taken as clinical cutoffs but as a proof of concept of how ML models could provide clinicians more information. Such integration could lead to improved clinical outcomes and efficiency in patient care.
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 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.002 | 0.000 |
| Bibliometrics | 0.001 | 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".