Table 4 Performance of predictive models for incident AF risk during follow-up period in overall general population (age, sex, and BMI-adjusted models).

From: Long-term PM2.5 exposure and the clinical application of machine learning for predicting incident atrial fibrillation

Predictive models*

c-index (95% CI)

NRI (95% CI)

IDI (95% CI)

Traditional regression analysis

0.604 (0.598–0.611)

Ref

Ref

Machine learning models

Support vector machine

0.699 (0.688–0.710)

0.280 (0.220–0.340)

0.002 (0.001–0.003)

Decision tree

0.786 (0.771–0.800)

0.806 (0.747–0.866)

0.010 (0.009–0.011)

Random forest

0.787 (0.772–0.801)

0.764 (0.701–0.827)

0.006 (0.005–0.007)

Naïve Bayes

0.790 (0.776–0.805)

0.792 (0.732–0.853)

0.009 (0.008–0.010)

Deep neural network

0.779 (0.768–0.790)

0.218 (0.182–0.253)

0.002 (0.001–0.003)

Extreme gradient boosting

0.794 (0.780–0.807)

0.536 (0.484–0.589)

0.005 (0.004–0.006)

  1. AF atrial fibrillation, BMI body mass index, CI confidence interval, IDI integrated discrimination improvement index, NRI category-free net reclassification improvement index.
  2. *Age, sex, and BMI were used for constructing these predictive models (age, sex, and BMI were adjusted for traditional regression analysis, and these variables were used as input variables for training the listed machine learning models).
Back to article page