Prediction scores of subclinical atrial fibrillation in patients with cryptogenic stroke and transient ischemic attack

Background: In patients with cryptogenic stroke (CS) or transient ischemic attack (TIA), prolonged cardiac monitoring is recommended to improve detection of atrial fibrillation (AF). Prediction scores have been proposed to identify patients with a high likelihood of post-stroke AF detection and some of them have been used to guide the selection of patients for implantable loop recorders (ILR), but few studies have externally assessed their performances. Aims: Aim of this prospective cohort study was to assess the performance of nine AF prediction scores in a cohort of CS and TIA monitored with ILR. Methods: Patients were included after a diagnosis of CS or TIA and ILR implantation between July 2018 and December 2023. Nine AF prediction scores were evaluated: STAF, LADS, HAVOC, Brown-ESUS AF, AS5 F, C 2 HEST, CHASE-LESS, AF-ESUS, and Empoli ESUS-AF. For each score we calculated sensitivity, specificity, negative (NPV) and positive predictive value (PPV), overall accuracy, and area under the receiver operating characteristic curve (AUROC). AUROCs were compared with DeLong’s test. Results: Of 1032 admitted patients, 270 (26.2%) were defined cryptogenic, 194 of whom (71.9%) received an ILR (43.3% women; median age 74.0 years [IQR 65.8-82.0]; median NIHSS score on admission 3.0 [1.0-6.0]; 182 (93.8%) ischemic stroke and 12 (6.2%) TIA). Median time from index event to ILR implant was 10 days (7-37). During long-term monitoring (median follow-up 23.0 months [12.0-37.3]), AF was detected in 62 patients (32%), with a median time from index stroke to AF diagnosis of 4.0 months (1.0-11.3). Sensitivity of the scores ranged between 12.9% and 95.2%, specificity 12.9-67.7%, PPV 37.3-48.1%, NPV 68.6-90.6%, and overall accuracy 45.4-66.3%. The Brown ESUS-AF score reached the highest AUROC (0.697 in the whole cohort, 0.707 in the ischemic stroke subgroup). In patients with ischemic stroke, AUROC was higher for Brown ESUS-AF compared to HAVOC (p = 0.014), CH 2 EST (p = 0.002), and Empoli ESUS-AF (p = 0.015) and for LADS (AUROC = 0.690) compared to CH 2 EST (p = 0.039) and Empoli ESUS-AF (p = 0.015). Conclusions: AF prediction scores based on clinical and cardiovascular imaging parameters do not predict AF detection with adequate accuracy in patients with CS or TIA and ILR. Brown ESUS-AF and LADS scores demonstrated a better performance compared to other prediction scores.