Development and validation of an ensemble machine learning framework for detection of all-cause advanced hepatic fibrosis: a retrospective cohort study
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
BACKGROUND: Cirrhosis is the result of advanced scarring (or fibrosis) of the liver, and is often diagnosed once decompensation with associated complications has occurred. Current non-invasive tests to detect advanced liver fibrosis have limited performance, with many indeterminate classifications. We aimed to identify patients with advanced liver fibrosis of all-causes using machine learning algorithms (MLAs). METHODS: In this retrospective study of routinely collected laboratory, clinical, and demographic data, we trained six MLAs (support vector machine, random forest classifier, gradient boosting classifier, logistic regression, artificial neural network, and an ensemble of all these algorithms) to detect advanced fibrosis using 1703 liver biopsies from patients seen at the Toronto Liver Clinic (TLC) between Jan 1, 2000, and Dec 20, 2014. Performance was validated using five datasets derived from patient data provided by the TLC (n=104 patients with a biopsy sample taken between March 24, 2014, and Dec 31, 2017) and McGill University Health Centre (MUHC; n=404). Patients with decompensated cirrhosis were excluded. Performance was benchmarked against aspartate aminotransferase-to-platelet ratio index (APRI), fibrosis-4 index (FIB-4), non-alcoholic fatty liver disease fibrosis score (NFS), transient elastography, and an independent panel of five hepatology experts (MB, GS, HK, KP, and RSK). MLA performance was evaluated using the area under the receiver operating characteristic curve (AUROC) and the percentage of determinate classifications. FINDINGS: The best MLA was an ensemble algorithm of support vector machine, random forest classifier, gradient boosting classifier, logistic regression, and neural network algorithms, which achieved 100% determinate classifications (95% CI 100·0-100·0), an AUROC score of 0·870 (95% CI 0·797-0·931) on the TLC validation set (fibrosis stages F0 and F1 vs F4), and an AUROC of 0·716 (95% CI 0·664-0·766) on the MUHC validation set (fibrosis stages F0, F1, and F2 vs F3 and F4). The ensemble MLA outperformed all routinely used biomarkers and achieved comparable performance to hepatologists as measured by AUROC and percentage of indeterminate classifications in both the TLC validation dataset (APRI AUROC score 0·719 [95% CI 0·611-0·820], 83·7% determinate [95% CI 76·0-90·4]; FIB-4 AUROC score 0·825 [95% CI 0·730-0·912], 72·1% determinate [95% CI 63·5-80·8]) and the MUHC validation dataset (APRI AUROC score 0·618 [95% CI 0·548-0·691], 75·5% determinate [95% CI 71·5-79·2]; FIB-4 AUROC score 0·717 (95% CI 0·652-0·776), 75·5% determinate [95% CI 0·713-0·797]), and achieving only slightly lower AUROC than transient elastography (0·773 [95% CI 0·699-0·834] vs 0·826 [95% CI 0·758-0·889]). INTERPRETATION: We have shown that an ensemble MLA outperforms non-imaging-based methods in detecting advanced fibrosis across different causes of liver disease. Our MLA was superior to APRI, FIB-4, and NFS with no indeterminate classifications, while achieving performance comparable to an independent panel of experts. MLAs using routinely collected data could identify patients at high-risk of advanced hepatic fibrosis and cirrhosis among patients with chronic liver disease, allowing intervention before onset of decompensation. FUNDING: Toronto General Hospital Foundation.
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) | 0.000 | 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