Balancing the benefits and harms from atrial fibrillation screening: insights from a novel patient-level computational model

Abstract Background Atrial fibrillation (AF) is a highly dynamic, progressive and heterogeneous disease, associated with higher risks of stroke and mortality. Screening can identify more AF patients and lead to early oral anticoagulation (OAC). However, clinical benefits and harms of screening-related early AF detection and OAC in the general population remain uncertain. Purpose We aim to simulate the dynamic occurrence of AF and AF-associated clinical outcomes using a patient-level computational model, and evaluate stroke and major bleeding risks of different screening strategies followed by OAC in the general population through virtual randomized clinical trials (V-RCTs). Methods The framework included two parallel Markov-like models; one with 7 clinical states (sinus rhythm, symptomatic/asymptomatic AF each with/without stroke and death) and one with 2 states (non-bleeding and major bleeding). AF-related atrial remodeling was incorporated and transition probabilities in two models were influenced by age, sex, and/or atrial remodeling (Figure 1A). Model calibration/validation was performed by reproducing a series of epidemiological and clinical data. Simulated OAC was initiated after clinical AF diagnosis (symptomatic AF episode of ≥3 hours) in control and screening groups or any screening-detected AF episode after 75 years old (only in screening groups). Kaplan-Meier (K-M) curves of stroke and major bleeding risks were modeled by Weibull distribution. Time-to-benefit/time-to-harm of OAC was estimated by Weibull models and Monte-Carlo simulations (4000 replications) in V-RCTs. Results The model can simulate all individual AF episodes and clinical outcomes during the lifetime follow-up in 10000 virtual patients with 30-minute resolution. It is consistent with a series of epidemiological data, including stroke incidence in the general population (Figure 1B) and in patients with device-detected AF (Figure 1C), as well as major bleeding incidence in the OAC-free general population (Figure 1D) and in patients with clinical AF and OAC (Figure 1E). Weibull models fitted the individual-patient-data K-M curves and both showed that screening groups had slightly lower stroke incidence and higher major bleeding risk during the long-term follow-up compared with the control group with only clinical AF diagnosis. Our analysis showed that 9.23, 14.61 and 11.04 years were needed to reduce 1 stroke per 100 screened participants, while only 6.59, 12.65 and 10.8 years were needed to observe 1 major bleeding for continuous rhythm monitoring, yearly 14-day Holter or daily 3 times 1-minute ECG (Figure 2A-B). Conclusions We present a novel model that reproduces epidemiological data of stroke and major bleeding in the general population and patients with AF. V-RCTs simulated by the model suggest that screening-triggered OAC in the general population will produce limited and delayed benefit from stroke reduction but causes early-onset harms from major bleeding.Figure 1 Figure 2