Predicting new-onset atrial fibrillation after transcatheter aortic valve implantation using the FIND-AF machine learning algorithm
C Simela, J H Tan, U Nadeem, T Habtezghi, K Kazi, J Wu, C P Gale, R NadarajahAbstract
Background
New-onset atrial fibrillation (NOAF) after transcatheter aortic valve implantation (TAVI) is associated with poor outcomes, including death and stroke (1, 2). It would therefore be useful to risk stratify patients before TAVI to guide post-TAVI surveillance and clinical care. We have recently shown the FIND-AF machine learning algorithm can accurately predict short-term AF from primary care electronic health records (EHR) but its utility in the setting of post-TAVI NOAF is yet to be evaluated (3).
Purpose
To explore whether it is feasible and beneficial to use the FIND-AF machine learning algorithm to predict NOAF and all-cause mortality in the one year following TAVI using pre-TAVI patient characteristics.
Methods
We retrospectively analysed the electronic health record (EHR) data at a single tertiary cardiology centre to identify patients ≥ 18 years old without prior AF who underwent TAVI between 3rd January 2023 and 27th December 2023. Risk scores were assigned to patients using the FIND-AF machine learning algorithm. Scores > 90th centile were deemed high-risk. The primary outcome was NOAF, which was defined as a new AF diagnosis coded on hospital or primary care electronic health records at one year following TAVI. The exploratory outcome was all-cause mortality within one year of TAVI. Mantel-Haenszel tests were used to compare outcomes between the low-risk and high-risk groups with the former being the reference group. The negative predictive value (NPV) and positive predictive value (PPV) of a high FIND-AF risk category for NOAF after TAVI was also calculated.
Results
There were 247 patients identified (223 [90.3%] low-risk and 24 [9.7%] high-risk). Compared to the low-risk group, the high-risk group were older with a higher prevalence of multiple valve disease and heart failure (Table 1). Baseline AS severity was similar between groups. The primary outcome occurred in ten (4.5%) low-risk patients and one (4.2%) high-risk patient (HR 0.932; 95% CI: 0.127 to 6.855; p = 0.707). Twelve (5.4%) low-risk and two (8.3%) high-risk patients experienced the exploratory outcome (HR 1.711; 95% CI: 0.289 to 10.120; p = 0.554) (Figure 1). NPV and PPV of FIND-AF risk categorisation for NOAF at one year post-TAVI was 95.5% and 4.2% respectively.
Conclusion
After TAVI, AF detection in routine care was not associated with a risk score that includes common predictors of AF. AF after TAVI may be under-detected through routine practice and/or originate from different biopathology.