Machine-learning models integrating preoperative clinical factors and circulating tumor DNA features predict lymph node metastasis in esophageal carcinoma

Table 3 Comparison of two machine-learning models with computed tomography in evaluating lymph node metastasis in patients with esophageal cancer

Statistic	Training set		Validation set
	Value	95%CI	Value	95%CI
Computed tomography predicted
Sensitivity	42.6%	29.25%-56.8%	40.9%	20.7%-63.7%
Specificity	77.8%	67.8%-85.8%	85.0%	70.2%-94.3%
PPV	53.5%	41.2%-65.4%	60.0%	38.1%-78.6%
NPV	69.3%	63.6%-74.4%	72.3%	64.3%-79.1%
Accuracy	64.6%	56.2%-72.4%	69.4%	56.4%-80.4%
Model-1 predicted
Sensitivity	85.45%	73.34%-93.50%	66.67%	44.68%-84.37%
Specificity	88.76%	80.31%-94.48%	92.11%	78.62%-98.34%
PPV	82.46%	72.18%-89.49%	84.21%	63.45%-94.25%
NPV	90.80%	83.82%-94.95%	81.40%	71.15%-88.59%
Accuracy	87.50%	80.97%-92.42%	82.26%	70.47%-90.80%
Model-2 predicted
Sensitivity	89.09%	77.75%-95.89%	75.00%	53.29%-90.23%
Specificity	85.39%	76.32%-91.99%	89.47%	75.20%-97.06%
PPV	79.03%	69.34%-86.27%	81.82%	63.38%-92.12%
NPV	92.68%	85.56%-96.44%	85.00%	73.75%-91.95%
Accuracy	86.81%	80.16%-91.87%	83.87%	72.33%-91.98%

NPV: Negative predictive value; PPV: Positive predictive value.