Prediction of early death after atrial fibrillation diagnosis using a machine learning approach: A French nationwide cohort study

Prognostic value of glycaemic variability for mortality in critically ill atrial fibrillation patients and mortality prediction model using machine learning

BACKGROUND

The burden of atrial fibrillation (AF) in the intensive care unit (ICU) remains heavy. Glycaemic control is important in the AF management. Glycaemic variability (GV), an emerging marker of glycaemic control, is associated with unfavourable prognosis, and abnormal GV is prevalent in ICUs. However, the impact of GV on the prognosis of AF patients in the ICU remains uncertain. This study aimed to evaluate the relationship between GV and all-cause mortality after ICU admission at short-, medium-, and long-term intervals in AF patients.

METHODS

Data was obtained from the Medical Information Mart for Intensive Care IV 3.0 database, with admissions (2008-2019) as primary analysis cohort and admissions (2020-2022) as external validation cohort. Multivariate Cox proportional hazards models, and restricted cubic spline analyses were used to assess the associations between GV and mortality outcomes. Subsequently, GV and other clinical features were used to construct machine learning (ML) prediction models for 30-day all-cause mortality after ICU admission.

RESULTS

The primary analysis cohort included 8989 AF patients (age 76.5 [67.7-84.3] years; 57.8% male), while the external validation cohort included 837 AF patients (age 72.9 [65.3-80.2] years; 67.4% male). Multivariate Cox proportional hazards models revealed that higher GV quartiles were associated with higher risk of 30-day (Q3: HR 1.19, 95%CI 1.04-1.37; Q4: HR 1.33, 95%CI 1.16-1.52), 90-day (Q3: HR 1.25, 95%CI 1.11-1.40; Q4: HR 1.34, 95%CI 1.29-1.50), and 360-day (Q3: HR 1.21, 95%CI 1.09-1.33; Q4: HR 1.33, 95%CI 1.20-1.47) all-cause mortality, compared with lowest GV quartile. Moreover, our data suggests that GV needs to be contained within 20.0%. Among all ML models, light gradient boosting machine had the best performance (internal validation: AUC [0.780], G-mean [0.551], F1-score [0.533]; external validation: AUC [0.788], G-mean [0.578], F1-score [0.568]).

CONCLUSION

GV is a significant predictor of ICU short-term, mid-term, and long-term all-cause mortality in patients with AF (the potential risk stratification threshold is 20.0%). ML models incorporating GV demonstrated high efficiency in predicting short-term mortality and GV was ranked anterior in importance. These findings underscore the potential of GV as a valuable biomarker in guiding clinical decisions and improving patient outcomes in this high-risk population.

Machine learning approach for prediction of outcomes in anticoagulated patients with atrial fibrillation

BACKGROUND

The accuracy of available prediction tools for clinical outcomes in patients with atrial fibrillation (AF) remains modest. Machine Learning (ML) has been used to predict outcomes in the AF population, but not in a population entirely on anticoagulant therapy.

METHODS AND AIMS

Different supervised ML models were applied to predict all-cause death, cardiovascular (CV) death, major bleeding and stroke in anticoagulated patients with AF, processing data from the multicenter START-2 Register.

RESULTS

11078 AF patients (male n = 6029, 54.3%) were enrolled with a median follow-up period of 1.5 years [IQR 1.0-2.6]. Patients on Vitamin K Antagonists (VKA) were 5135 (46.4%) and 5943 (53.6%) were on Direct Oral Anticoagulants (DOAC). Using Multi-Gate Mixture of Experts, a cross-validated AUC of 0.779 ± 0.016 and 0.745 ± 0.022 were obtained, respectively, for the prediction of all-cause death and CV-death in the overall population. The best ML model outperformed CHA2DSVA2SC and HAS-BLED for all-cause death prediction (p < 0.001 for both). When compared to HAS-BLED, Gradient Boosting improved major bleeding prediction in DOACs patients (0.711 vs. 0.586, p < 0.001). A very low number of events during follow-up (52) resulted in a suboptimal ischemic stroke prediction (best AUC of 0.606 ± 0.117 in overall population). Body mass index, age, renal function, platelet count and hemoglobin levels resulted the most important variables for ML prediction.

CONCLUSIONS

In AF patients, ML models showed good discriminative ability to predict all-cause death, regardless of the type of anticoagulation strategy, and major bleeding on DOAC therapy, outperforming CHA2DS2VASC and the HAS-BLED scores for risk prediction in these populations.

Multimorbidity in patients with atrial fibrillation and community controls: A population-based study

Background

Multimorbidity is common in patients with atrial fibrillation (AF), yet comorbidity patterns are not well documented.

Methods

The prevalence of 18 chronic conditions (6 cardiometabolic, 7 other somatic, 5 mental health) was obtained in patients with new-onset AF from 2013-2017 from a 27-county region and controls matched 1:1 on age, sex, and county of residence. For AF patients and controls separately, clustering of conditions and co-occurrence beyond chance was estimated (using the asymmetric Somers' D statistic), overall and for ages <65, 65-74, 75-84, and ≥85 years.

Results

Among 16,509 patients with AF (median age 76 years, 57% men), few (4%) did not have any of the 18 chronic conditions, whereas nearly one-quarter of controls (23%) did not have any chronic conditions. Clustering of cardiometabolic conditions was common in both AF patients and controls, but clustering of other somatic conditions was more common in AF. Although the prevalence of most condition pairs was higher in AF patients, controls had a larger number of condition pairs occurring together beyond chance. In persons aged <65 years, AF patients more frequently exhibited concordance of condition pairs that included either pairs of somatic conditions or a combination of conditions from different condition groups. In persons aged 65-74 years, AF patients more frequently had pairs of other somatic conditions.

Conclusion

Patterns of co-existing conditions differed between patients with AF and controls, particularly in younger ages. A better understanding of the clinical consequences of multimorbidity in AF patients, including those diagnosed at younger ages, is needed.