Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
February 6, 2026European Heart Journal

FIND-AFDAS machine learning model predicts post-stroke AF, cutting the number needed to screen by ~80%.

View Full Paper
Ask AI
Bookmark
Share

Why the study?

AF-related strokes carry high recurrence, morbidity, and mortality, and accurately identifying stroke patients at high risk could enable targeted extended monitoring to diagnose AF.

Does the FIND-AFDAS meta-machine learning algorithm accurately predict incident atrial fibrillation in patients after stroke presentation?

Population

Patients with stroke presentation from multiple international routine EHR cohorts and RCT populations

Comparison

FIND-AFDAS meta-machine learning prediction model evaluated across cohorts

Design

Prediction model derivation and external validation study across international EHR cohorts and RCT datasets

Key result

The FIND-AFDAS meta-machine learning model accurately predicted incident atrial fibrillation after stroke (AUC 0.981; 95% CI 0.927-0.995 in the PER DIEM cohort), reducing the number needed to screen from 10 to 2.

Authors

RNR NadarajahJWJ WuKRKeerthenan Raveendra

Discussion

Loading...

Member takes

Overview

May guide targeted post-stroke AF screening; extends prior models but hypothesis-generating pending prospective validation.

Key Points

  • This research aims to develop a robust prediction model for identifying atrial fibrillation in stroke patients using meta-machine learning techniques.
  • Selected candidate variables from a previous systematic review and logistic regression analysis.
  • Trained various models including XGBoost and Neural Networks on diverse international datasets.
  • Applied ensemble learning to enhance model prediction.
  • Conducted external validation with additional international cohorts.
  • The XGBoost models yielded high performance across cohorts, with AUC values exceeding 0.810.
  • FIND-AFDAS demonstrated excellent external validation performance with AUCs ranging from 0.770 to 0.979.
  • Achieved 100% sensitivity and 88.1% specificity with an optimized risk threshold during clinical trials.

Study Design

Type

Observational (n=72,207)

Multicenter

Yes

Structured PICO

Does the FIND-AFDAS meta-machine learning algorithm accurately predict incident atrial fibrillation in patients after stroke presentation?

P
Population
72,207 patients with stroke from multiple international cohorts used to derive and validate a meta-machine learning model for predicting incident atrial fibrillation.
E
Exposure
FIND-AFDAS meta-machine learning algorithm (stacking XGBoost models) using age, sex, ethnicity, and five comorbidities
O
Outcome
Incident atrial fibrillation after stroke presentation (measured by AUC for prediction performance)

Main Result

Effect estimate: AUC 0.981 (95% CI 0.927-0.995)

The FIND-AFDAS meta-machine learning algorithm accurately identifies individuals at high risk for atrial fibrillation after stroke, potentially guiding targeted extended monitoring.

Cite This Study

Nadarajah et al. (2025) conducted an observational in Stroke and Atrial Fibrillation (n=72,207). FIND-AFDAS meta-machine learning model was evaluated on Incident atrial fibrillation after stroke (AUC 0.981, 95% CI 0.927-0.995). The FIND-AFDAS meta-machine learning model accurately predicted incident atrial fibrillation after stroke (AUC 0.981; 95% CI 0.927-0.995 in the PER DIEM cohort), reducing the number needed to screen from 10 to 2.

synapsesocial.com/papers/698585548f7c464f230088cchttps://doi.org/10.1093/eurheartj/ehaf784.512
View Full Paper
Ask AI
Bookmark
Share