Table 3.
Utility of significantly altered (P < 0.05) metabolites in accurately predicting hepatocellular carcinoma (hepatocellular carcinoma cases vs patients with cirrhosis)
| Ref. | Platform | Comparison | Class prediction methodology | Classification accuracy or sensitivity/specificity | AFP sensitivity/specificity |
| Patterson et al[29] | UPLC/ESI-QTOF-MS | HCC (n = 20) vs cirrhosis (n = 7) | Random forest | 96.3 | - |
| Chen et al[30] | Integrated GC/QTOF-MS + UPLC/QTOF-MS | HCC (n = 82) vs healthy (n = 71) | OPLS-DA | 100.0 | - |
| Wu et al[31] | SELDI-TOF MS | HCC (n = 48) vs cirrhosis (n = 54) or healthy (n = 42) | GRO-α + thrombin light chain PS20 Protein immunoassay | 89.6/89.6 | 69/83 |
| Cao et al[32] | UPLC/QTOF-MS | HCC (n = 23) vs cirrhosis (n = 22) | PLS-DA | 67.0 | - |
| Gao et al[33] | NMR | HCC (n = 39) vs cirrhosis (n = 36) | PLS-DA | 45.7 | - |
| Wu et al[34] | GC/MS | HCC (n = 20) vs healthy (n = 20) | PCA with ROC curve analysis | AUC=88.3; AUCAFP = 92.5 when combined with AFP | - |
| Soga et al[35] | LC/MS-MS | HCC (n = 32) vs HCV-only (n = 35) or cirrhosis (n = 18) | Multiple logistic regression; ROC curve analysis | 88.1 | 0.760 |
| Wang et al[38] | UPLC-MS | HCC (59) vs cirrhosis (20) or NHC (20) | PLS-DA, ROC curve analysis | CSA 79.3/100 CSA + AFP20 96.4/100 UPLC-MS 100/100 | AFP20 74/38 AFP200 52/90 |
| Zhou et al[39] | UPLC-QTOF-MS | HCC (n = 69) vs cirrhosis (n = 28) | PLS-DA, ROC curve analysis | AEA 88.0 PEA 82.0 AEA + PEA 88.0 | - |
| Nahon et al[40] | NMR | Small HCC (n = 28) vs cirrhosis (n = 93); Large HCC (n = 33) vs cirrhosis (n = 93) | OPLS | Small HCC: 61.0/100.0 Large HCC: 100.0/100.0 | - |
| Yin et al[41] | RPLC/QTOF-MS; HILIC/QTOF-MS | HCC (n = 25) vs cirrhosis (n = 24) or healthy (n = 25) | OPLS | RPLC: 61.8 HILIC: 57.0 RPLC + HILIC = 63.6 | - |
| Li et al[42] | UPLC/QTOF-MS | HCC (n = 8) vs cirrhosis (n = 6) or healthy (n = 6) (murine samples) | OPLS-DA | 88.2 | - |
| Budhu et al[43] | Training set1: GC/MS + UPLC/MS-MS; Testing set2: Affymetrix GeneChip | Training set: Stem-like aggressive HpSC-HCC (n = 15) vs Mature hepatocyte less aggressive MH-HCC (n = 15); Testing set: HpSC-HCC and MH-HCC (n = 217) | Multivariate analysis | 172.0/83.0, AUC = 0.830 272.0/91.0, AUC = 0.860 | - |
| Fitian et al[45] | UPLC/MS-MS + GC/MS | HCC (n = 30) vs HCV-cirrhosis (n = 27) | Random forest | 72% 12-HETE 73.3/69.2 | AFP20 63.3/83.6 |
| ROC analysis | 15-HETE 83.3/59.3 | ||||
| Aspartate 100/51.9 | |||||
| Glycine 83.3/63.0 | |||||
| Serine 73.3/85.2 | |||||
| Phenylalanine 73.3/81.5 | |||||
| Homoserine 70.0/85.2 | |||||
| Sphingosine 58.3/86.7 | |||||
| Xanthine 63.3/88.9 | |||||
| 2-Hydroxybutyrate 76.7/77.8 | |||||
| Gao et al[46] | GC-TOF/MS | HCC (n = 39) vs HBV-cirrhosis (n = 52) | Random forest (validation set) | 96.8% in HCC vs HBV-cirrhosis 100% in HBV-cirrhosis vs HBV | - |
| 100% in HBV vs NHC | |||||
| ROC analysis (validation set) | 100/95.2 HBV vs NC | ||||
| 83.3/100 HBV-cirrhosis vs HBV | |||||
| 76.9/83.3 HCC vs HBV-cirrhosis | |||||
| Bayes discriminant function model (validation set) | 76.9% HCC 100% HBV-cirrhosis | ||||
| 94.1% HBV | |||||
| 100% NHC |
Classification accuracy describes the capacity of the metabolomic classification technique to accurately predict the group of each study subject. UPLC: Ultrahigh-performance liquid chromatography; AEA: Anandamide; OPLS: Orthogonal projection to latent structure; PCA: Principal component analysis; PEA: Palmitylethanolamide; PLS-DA: Partial least squares-discriminant analysis; MS: Mass spectrometry; TOF: Time-of-flight; GC: Gas chromatography; HCC: Hepatocellular carcinoma; HBV: Hepatitis B virus; GRO-α: Growth related oncogene-alpha; ROC: Receiver operator characteristic; NHC: Normal healthy controls; AFP20: AFP performance at the cutoff of 20 ng/mL; AFP200: AFP performance at the cutoff of 200 ng/mL.