Skip to main content

Table. 3 Classification statistics of the selected thresholds for the recalibration-in-the-large of the validation cohort and subsets

From: Validation and update of a multivariable prediction model for the identification and management of patients at risk for hepatocellular carcinoma

Probability threshold1 Proportions divided Sensitivity
(95% CI)2
Specificity
(95% CI)2
PPV
(95% CI)2
NPV
(95% CI)2
NB
(95% CI)2
All patients HCC patients
For HCC of any etiology, [15.5% (157/1012)]
 1.3% 427/685 0/157 1 (1–1) 0.499 (0.466–0.532) 0.268 (0.256–0.282) 1 (1–1) 0.150 (0.125–0.173)
 13.1% 823/189 13/144 0.917 (0.873–0.955) 0.947 (0.931–0.963) 0.763 (0.71–0.818) 0.984 (0.976–0.992) 0.136 (0.113–0.160)
 15.5% 829/183 15/142 0.905 (0.854–0.949) 0.952 (0.937–0.966) 0.777 (0.724–0.832) 0.982 (0.973–0.99) 0.133 (0.112–0.155)
 1/3 856/156 20/137 0.873 (0.815–0.924) 0.978 (0.966–0.987) 0.879 (0.828–0.927) 0.977 (0.967–0.986) 0.126 (0.104–0.145)
 50% 874/138 26/131 0.834 (0.771–0.892) 0.992 (0.985–0.998) 0.950 (0.909–0.985) 0.970 (0.960–0.980) 0.123 (0.102–0.143)
 2/3 888/124 37/120 0.764 (0.694–0.828) 0.995 (0.991–0.999) 0.968 (0.934–0.992) 0.958 (0.947–0.969) 0.111 (0.090–0.132)
 98.3% 933/79 78/79 0.503 (0.427–0.580) 1 (1–1) 1 (1–1) 0.916 (0.905–0.928) 0.078 (0.061–0.095)
For single HBV-related HCC, [15.6% (122/782)]
 1.3% 367/415 0/122 1 (1–1) 0.556 (0.520–0.596) 0.294 (0.278–0.314) 1 (1–1) 0.151 (0.129–0.178)
 13.1% 646/136 8/114 0.934 (0.885–0.975) 0.967 (0.952–0.979) 0.839 (0.782–0.892) 0.988 (0.979–0.995) 0.142 (0.120–0.168)
 15.5% 652/130 10/112 0.918 (0.869–0.959) 0.973 (0.959–0.985) 0.862 (0.807–0.916) 0.985 (0.976–0.992) 0.139 (0.114–0.163)
 1/3 664/118 13/109 0.893 (0.836–0.943) 0.986 (0.977–0.994) 0.924 (0.875–0.966) 0.980 (0.970–0.989) 0.134 (0.109–0.160)
 50% 675/107 19/103 0.844 (0.779–0.910) 0.994 (0.988–0.996) 0.963 (0.924–0.991) 0.972 (0.960–0.984) 0.127 (0.102–0.153)
 2/3 685/97 28/94 0.771 (0.697–0.844) 0.995 (0.989–1) 0.969 (0.931–1) 0.959 (0.947–0.972) 0.113 (0.087–0.141)
 98.3% 716/66 56/66 0.541 (0.451–0.631) 1 (1–1) 1 (1–1) 0.922 (0.908–0.936) 0.084 (0.064–0.105)
For HCC of other causes, [15.2% (35/230)]
 1.3% 60/170 0/35 1 (1–1) 0.308 (0.246–0.369) 0.206 (0.192–0.222) 1 (1–1) 0.144 (0.100–0.193)
 13.1% 177/53 5/30 0.857 (0.743–0.971) 0.882 (0.836–0.928) 0.569 (0.475–0.680) 0.972 (0.949–0.994) 0.115 (0.073–0.159)
 15.5% 177/53 5/30 0.857 (0.743–0.971) 0.882 (0.836–0.928) 0.569 (0.475–0.680) 0.972 (0.949–0.994) 0.112 (0.067–0.157)
 1/3 192/38 7/28 0.800 (0.657–0.914) 0.949 (0.918–0.974) 0.737 (0.620–0.862) 0.964 (0.939–0.984) 0.100 (0.057–0.148)
 50% 199/31 7/28 0.800 (0.657–0.914) 0.985 (0.964–1) 0.906 (0.800–1) 0.965 (0.941–0.985) 0.109 (0.061–0.157)
 2/3 203/27 9/26 0.743 (0.600–0.886) 0.995 (0.985–1) 0.964 (0.880–1) 0.956 (0.932–0.980) 0.104 (0.061–0.152)
 98.3% 217/13 22/13 0.371 (0.229–0.543) 1 (1–1) 1 (1–1) 0.899 (0.878–0.924) 0.057 (0.030–0.091)
  1. CI confidence interval, HCC hepatocellular cancer, NB net benefit, NPV negative prediction value, PPV positive prediction value
  2. 11.3%, the median prediction probability of non-HCC; 13.1%, the optimal cutoff value; 15.5%, the incidence rate of HCC in the validation cohort; 1/3, the low threshold in the original ASAP model; 50%, the midpoint of the sigmoid curve; 2/3, the high threshold in the original ASAP model; 98.3%, the median prediction probability of HCC
  3. 2Computed with 2000 stratified bootstrap replicates