Table 3.
Model performance measures for mortality and MACE risks in the discovery phase
Predictive model | AUC | IDI (95% CI) | Continuous NRI (95% CI) |
---|---|---|---|
Prediction of death | |||
Metabolomic + clinicala | 83.7 | ||
Metabolomicb | 80.9 | 0.072 (− 0.067 to 0.238) | 0.013 (− 0.259 to 0.263) |
TMAO + clinicalc | 76.6 | 0.096 (0.031–0.235) | 0.230 (− 0.032 to 0.446) |
Clinicald | 77.0 | 0.096 (0.012–0.231) | 0.121 (− 0.127 to 0.369) |
Prediction of MACE | |||
Metabolomic + clinicale | 67.4 | ||
Metabolomicf | 66.0 | 0.066 (0.005–0.124) | 0.097 (−0.049 to 0.238) |
TMAO + clinicalg | 59.8 | 0.068 (0.029–0.118) | 0.144 (0.005–0.324) |
Clinicalh | 58.4 | 0.072 (0.034–0.128) | 0.106 (− 0.001–0.321) |
The metabolic variables screened from LASSO, TMAO, and the 17 traditional clinical factors including age, sex, AST, eGFR, DM, HyperT, CHOL, HDLC, PPI, ACEI, BB, CCB, current smoking, family history of CVD, SYNTAX, SBP, and GLUC were input into multivariate Cox proportional hazards regression analysis to fit model, using a forward and backward stepwise process based on AIC (Akaike information criterion). The model with the smallest AIC value was considered the best and variables with P < 0.1 were retained. IDI (integrated discrimination improvement) and continuous NRI (net reclassification improvement) were calculated by comparing the Metabolomic + clinical model with TMAO + clinical and Clinical model, and the Metabolomic model with Clinical model, 95% CIs were calculated by 1000 bootstrap resampling
aMetabolomic + clinical model = Dulcitol, β-Pseudouridine, 3,3ʹ,5-Triiodo-l-thyronine, Kynurenine, age, current smoking, GLUC, AST, SBP
bMetabolomic model = Dulcitol, Kynurenine, Cyclic AMP, 3,3ʹ,5-Triiodo-l-thyronine, β-Pseudouridine
cTMAO model = TMAO, age, AST, current smoking, SBP, GLUC
dClinical model = age, AST, HDLC, CCB, current smoking, SYNTAX, SBP, GLUC
eMetabolomic + clinical model = lysoPC 20:2, 5-methyluridine, kynurenine, L-tryptophan, AST, DM, PPI, SYNTAX
fMetabolomic model = lysoPC 20:2, 5-methyluridine, kynurenine, l-tryptophan, d-sorbitol, phenyllactate
gTMAO + clinical model = TMAO, AST, DM, PPI, CCB, SYNTAX
hClinical model = AST, DM, PPI, CCB, SYNTAX. AUC = area under the curve, other abbreviations are as in Table 1