TL;DR
This study developed and validated a clinical prognostic model to predict the risk of new-onset atrial fibrillation (AF) after cardiac surgery. The model, built using a large UK electronic health record dataset, identifies patients at high risk who could benefit from targeted monitoring and preventative strategies, aiming to improve outcomes and reduce adverse events like stroke.
What Problem Does This Paper Address?
Atrial fibrillation (AF) is a common complication following cardiac surgery, affecting 30-50% of patients. This post-operative AF (AFACS) is associated with longer hospital stays, increased risk of stroke, thromboembolism, and heart failure. While interventions exist to prevent AFACS and its consequences, their effectiveness can be limited by side effects. Crucially, there's a lack of validated tools to accurately stratify patients by their risk of developing AFACS. This makes it difficult to tailor preventive strategies and monitoring efforts, leading to potential under- or over-treatment. Existing models often fall short of established reporting guidelines (like TRIPOD), necessitating the development of a robust, validated prognostic tool.
Methods
The researchers employed a prospective cohort study design using linked electronic health records from the UK's Clinical Practice Research Datalink (CPRD). For model development, they utilized multivariable logistic regression with the Least Absolute Shrinkage and Selection Operator (LASSO) for variable selection. This technique is effective in handling a large number of potential predictors by shrinking the coefficients of less important variables towards zero, effectively performing feature selection and regularization simultaneously. Internal validation was performed using the bootstrap method. Model performance was assessed using the c-statistic for discrimination and calibration plots for assessing how well the predicted probabilities matched observed frequencies. They also conducted sensitivity analyses using random forests for variable selection.
Datasets
The primary dataset used was the UK Clinical Practice Research Datalink (CPRD), a large electronic health record database containing data from approximately 5.6 million individuals. For this study, they focused on a cohort of 33,464 patients who underwent cardiac surgery between 1998 and 2016. This dataset was linked with hospital episode statistics, national cancer registry data, and death registry data (Office for National Statistics-ONS). No new benchmarks were introduced; the study leverages existing, well-established UK health data.
Key Insights and Ideas
- Predictive Power of Existing Risk Scores: The study confirmed that established risk scores like CHA2DS2-VASc, while designed for stroke prediction in AF patients, also hold predictive value for AFACS in the post-cardiac surgery context.
- Protective Factors Identified: The model revealed that statin use and belonging to an ethnic group other than white or black were associated with a protective effect against AFACS. This suggests potential avenues for intervention or highlights existing disparities in care.
- Valve Surgery as a Strong Predictor: Valve surgery emerged as the strongest single predictor of AFACS, underscoring the significant hemodynamic and inflammatory changes associated with these procedures.
- Importance of Procedural Complexity: The number of combined procedures performed during the index cardiac surgery (e.g., valve replacement with coronary artery bypass grafting) also significantly influenced AFACS risk.
- Need for Targeted Interventions: The model's ability to identify high-risk individuals supports the rationale for personalized management strategies, including long-term monitoring, risk factor management, and thromboprophylaxis.
Key Results
The final prognostic model included 13 baseline predictors: age, year of index surgery, elevated CHA2DS2-VASc score, congestive heart failure, hypertension, acute coronary syndromes, mitral valve disease, ventricular tachycardia, valve surgery, receiving two or three combined procedures, statin use, and ethnicity other than white or black.
The model demonstrated moderate discrimination, with an optimism-corrected C-statistic of 0.68 for both the derivation and validation cohorts. Calibration was reported as good. Among the 33,464 patients who underwent cardiac surgery, 6,902 (20.6%) developed new-onset AFACS. The model was able to identify a substantial proportion of patients at high risk for AFACS, with three-quarters of patients with valvular heart disease being classified as high or highest risk.
Limitations and Future Work
The authors acknowledge several limitations:
- Limited Biomarker Data: Detailed information on specific biomarkers and clinical AF types (paroxysmal vs. persistent) was not available for all patients, which could have further refined the model.
- Potential for Missing Information: While the EHR data is extensive, it may not capture all relevant clinical details or intra-operative variables.
- Generalizability: While the UK cohort is large, external validation in diverse populations (e.g., US-American and German cohorts) is planned to confirm generalizability.
Future work will focus on:
- Developing the PARADISE-1 Score: Integrating these findings into a comprehensive pre-operative risk score.
- External Validation: Testing the model in other international cohorts.
- Exploring Intra-operative Variables: Incorporating intra-operative data for the PARADISE-2 score.
- Investigating AF Burden: Further research into the impact of AF burden and clinical type on subsequent outcomes.
- Clinical Trial Design: Using the model to inform future clinical trials for preventive strategies and interventions for high-risk AFACS patients.
Summary updated