Introduction: Cardiac amyloidosis accounts for 13% of all patients with heart failure with preserved ejection fraction (HFpEF). The National Amyloidosis Centre (NAC) scoring system (Stage 1-3), utilising eGFR and NT-proBNP correlates well with overall prognosis. Hypothesis: We wished to explore the feasibility of diagnosing disease severity using 12-lead ECG alone, using our database of >6,000 ECGs from 292 ATTR patients. Methods: Triplicate 12-lead ECG parameters (PR interval, QRS width, QT-interval, QRS morphology, axis, rhythm and low amplitude presence or absence), baseline NT-proBNP and baseline eGFR were extracted. Patients with pacing were excluded. For numeric baseline parameters, the mean triplicate ECG value was used. NAC stage was then calculated from eGFR and NT-proBNP (Stage 1: NT-proBNP <3000 AND eGFR >45, Stage 3: NT-proBNP >3000 and eGFR <45, Stage 2: meets neither stage 1 or 3 criteria). Data was assembled (Excel) and then machine learning models (multilayer perceptron, decision tree (DT), naïve bayes (NB), support vector machine and radius neighbours) were trained in a 80:20 split, with 5-fold stratified cross-validation (sklearn, Python 3.0) performed on the training set for tuning of hyperparameters, and the final model output performance measured on the separate test set. Performance was measured by F1 score. Results: 295 patients were originally included, and 37 were excluded due to pacing, leaving 258 patients in the final dataset. On cross-validation the NB classifier performed best (F1 average 0.6) and the worst performing model was DT (F1 average 0.52). On the final test set the NB model performed as follows: Stage 1 NAC = F1 score 0.7, Stage 2 NAC = F1 score 0.45, Stage 3 NAC = F1 score 0.0. Conclusions: Despite the small dataset size, a machine learning approach could detect stage 1 disease with reasonable accuracy but struggled to differentiate between stage 2 and 3 disease. This model requires further external validation. We have demonstrated it is feasible to use a 12-lead ECG alone to diagnose stage 1 disease severity in ATTR-CM but not more advanced disease using machine learning.