Table 2 Performance summary for best models on 1, 3, 5-year(s) expenditure.

From: Prediction of future healthcare expenses of patients from chest radiographs using deep learning: a pilot study

Model

1 year

3 years

5 years

Training set size

16,399

9324

1328

Classification

ROC-AUC (95% CI)

Corresponding Model

0.806 (0.793–0.819) Model TX1

0.771 (0.750–0.794)

Model X

0.729 (0.667–0.729)

Model TX2

Classification

F1 (95% CI)

0.779 (0.766–0.791)

0.775 (0.756–0.794)

0.781 (0.736–0.826)

Regression Spearman ρ (95% CI)

Corresponding Model

0.561 (0.536–0.586)

Model TX2

0.524 (0.489–0.559)

Model TX2

0.424 (0.324–0.529)

Model X

Regression

Pearson R (95% CI)*

0.557 (0.532–0.581)

0.523 (0.487–0.558)

0.421 (0.314–0.530)

  1. CI confidence interval.
  2. *Pearson R is calculated using log10-transformed data.
Back to article page