Meta-machine learning for enhanced detection of atrial fibrillation after stroke: FIND-AFDAS
Bibliographic record
Abstract
Abstract Background Atrial fibrillation (AF)-related strokes have a high rate of recurrence and are associated with morbidity, healthcare expenditure, and mortality.(1-3) Accurately identifying patients with stroke at high risk for AF could enable targeted extended monitoring to diagnose AF and prevent recurrent stroke.(4) Purpose To derive a scalable and internationally generalisable prediction model for incident AF after stroke presentation through meta-machine learning. Methods Candidate variables were selected based on a previous systematic review and logistic regression analysis.(4) We trained and tested logistic regression, random forest, XGBoost, Neural Networks, linear discriminant analysis and Naïve Bayes models in data (partitioned 7:3) from: CPRD (United Kingdom), NTUH (Taiwan), and AF-ESUS (France, Greece). An ensemble learning technique (stacking) was applied to the best performing models from each cohort to develop a meta-model (FIND-AFDAS). External validation was conducted in three international routine EHR cohorts (JMDC Claims Database (Japan), CDARS (China), and PRECISE (Scotland)). Prediction performance and clinical impact was evaluated in the PER DIEM and ARCADIA randomised clinical trial (RCT) populations to estimate the optimum threshold for negative predictive value (NPV), positive predictive value (PPV) and number needed to screen (NNS). Results A prior candidate variables selection and logistic regression analysis led to a final parsimonious selection of variables: age, sex, ethnicity (white versus other) and five comorbidities. In CPRD-GOLD (n = 36160), NTUH (n = 2661), and AF-ESUS (n = 730), XGBoost models had the best performance and stacking the XGBoost models (FIND-AFDAS) led to excellent prediction performance in each of the cohorts (CPRD AUC 0.810, 95% CI 0.795-0.825; NTUH AUC 0.936, 95% CI 0.918-0.953; AF-ESUS AUC 0.967, 95% CI 0.923-0.985) (Table 1). The FIND-AFDAS meta-model had excellent prediction performance on external validation in JMDC (n = 23474, AUC = 0.770, 95% CI = 0.752-787, CDARS n = 3840, AUC = 0.979, 95% CI = 0.947-0.992), and PRECISE (n = 4037, AUC = 0.898, 95% CI 0.878-0.915) (Table 1). In the PER DIEM RCT population (n=300) of patients with ischaemic stroke or TIA who were randomized 1:1 to implantable loop recorder or external loop recorder, prediction performance of FIND-AFDAS was excellent (AUC 0.981, 0.927-0.995) (Table 1) and an optimised risk threshold of 0.11 led to sensitivity, specificity, PPV, and NPV of 100%, 88,1%, 48.4% and 100%, respectively, and an 80% reduction in NNS (10 to 2) (Figure 1). These excellent results were confirmed in the ARCADIA RCT (n=1005) (Table 1, Figure 1). Conclusions The internationally generalisable and scalable FIND-AFDAS meta-machine learning algorithm can accurately identify individuals for extended monitoring for AF after presentation with stroke. Clinical and cost-effectiveness evaluation in a prospective RCT is now required. Figure 1
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 | 0.061 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.007 | 0.019 |
| Bibliometrics | 0.006 | 0.004 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.003 | 0.002 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.003 | 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 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".