Skip to main content
. 2024 Apr 18;11(5):2490–2498. doi: 10.1002/ehf2.14796

Table 3.

Model performance parameters

AUC Sensitivity Specificity Precision score DeLong*
Centre 1 (training cohort) all ejection fraction
Random forest 0.99 (0.98–0.99) 93.3% 97.3% 93.8% N/A
Logistic regression 0.93 (0.91–0.95) 81.2% 95.0% 82.2% <0.001
ADHERE 0.68 (0.65–0.71) 75.9% 49.4% 15.0% <0.001
GWTG‐HF 0.69 (0.66–0.72) 64.5% 63.0% 14.6% <0.001
Centre 1 (training cohort) heart failure with reduced ejection fraction (ejection fraction ≤40%)
Random forest 0.99 (0.98–1.00) 94.1% 96.8% 95.1% N/A
Logistic regression 0.90 (0.86–0.94) 83.9% 86.9% 71.1% 3.11E‐03
ADHERE 0.68 (0.63–0.72) 59.3% 66.8% 18.1% <0.001
GWTG‐HF 0.64 (0.59–0.69) 52.5% 68.5% 15.8% <0.001
Centre 1 (training cohort) heart failure with mildly reduced ejection fraction (ejection fraction 41–49%)
Random forest 1.00 (1.00–1.00) 100.0% 99.6% 99.5% N/A
Logistic regression 0.73 (0.59–0.87) 79.0% 67.6% 31.6% 2.33E‐02
ADHERE 0.67 (0.54–0.80) 79.0% 49.6% 18.7% 1.59E‐03
GWTG‐HF 0.66 (0.53–0.80) 68.4% 62.3% 15.6% 2.59E‐03
Centre 1 (training cohort) heart failure with preserved ejection fraction (ejection fraction ≥50%)
Random forest 0.98 (0.97–1.00) 94.5% 93.9% 92.5% N/A
Logistic regression 0.91 (0.88–0.95) 77.9% 94.7% 77.1% 5.93E‐03
ADHERE 0.68 (0.64–0.72) 60.0% 66.7% 13.2% <0.001
GWTG‐HF 0.57 (0.53–0.62) 44.1% 71.1% 9.0% <0.001
Centre 2 (validation cohort) all ejection fraction
Random forest 0.96 (0.95–0.97) 87.3% 90.6% 79.9% N/A
Logistic regression 0.87 (0.85–0.89) 75.9% 86.9% 63.4% <0.001
ADHERE 0.70 (0.67–0.72) 69.0% 61.7% 11.4% <0.001
GWTG‐HF 0.69 (0.67–0.72) 66.8% 62.9% 11.1% <0.001
Centre 2 (validation cohort) heart failure with reduced ejection fraction (ejection fraction ≤40%)
Random forest 0.96 (0.94–0.98) 88.2% 91.0% 81.6% N/A
Logistic regression 0.85 (0.81–0.89) 64.6% 90.1% 54.8% <0.001
ADHERE 0.69 (0.65–0.74) 70.8% 60.9% 15.8% <0.001
GWTG‐HF 0.55 (0.50–0.60) 41.0% 68.5% 8.3% <0.001
Centre 2 (validation cohort) heart failure with mildly reduced ejection fraction (ejection fraction 41–49%)
Random forest 0.95 (0.88–1.00) 90.0% 95.3% 83.1% N/A
Logistic regression 0.80 (0.69–0.90) 75.0% 77.9% 20.2% 1.72E‐01
ADHERE 0.70 (0.58–0.81) 70.0% 72.6% 9.8% 2.17E‐02
GWTG‐HF 0.71 (0.60–0.83) 75.0% 67.3% 10.7% 3.53E‐02
Centre 2 (validation cohort) heart failure with preserved ejection fraction (ejection fraction ≥50%)
Random forest 0.95 (0.93–0.97) 87.4% 89.5% 76.9% N/A
Logistic regression 0.84 (0.81–0.88) 73.0% 84.5% 54.0% <0.001
ADHERE 0.70 (0.66–0.73) 66.5% 62.8% 9.9% <0.001
GWTG‐HF 0.59 (0.55–0.63) 35.2% 79.7% 7.7% <0.001
*

P value, compared with random forest.

Abbreviations: ADHERE, Acute Decompensated Heart Failure National Registry; AUC, area under the curve; GWTG‐HF, Get With the Guidelines Heart Failure.

HHS Vulnerability Disclosure