Table 4: Cox regression analysis for all-cause mortality.

Basic model Basic model + cTnT × 100 Basic model + GLS ≥ −15%Full model Full model + interaction termReduced model 1 Reduced model 2 (final model) Final model + interaction term

Albumin0.170.004**0.230.047**0.240.048*0.29 0.09
CAD2.620.04*2.430.04*1.960. —
DM3.170.03*2.100.352.970.04*1.720.401.490.50  —
HTN3.390.02*2.940.05*3.090.02*3.090.03*2.640.093.000.04*2.700.062.31 0.12
cTnT × 1001.160.001*1.130.009*1.190.01*1.150.001*1.140.002*1.21 0.001*
GLS ≥ −15%3.570.02*3.090.03**2.790.049*6.45 0.01*
(cTnT × 100) : (GLS ≥ −15%)#0.910.220.90 0.14

Data presented are based on the Cox regression analysis; 95% CIs for each HR are presented in Supplementary Table S3. In multivariate analysis, we firstly constructed the model including background coronary arterial disease (CAD) and diabetes and hypertension together with serum albumin concentration as a basic model; then, we added cTnT × 100 and/or a GLS ≥ −15% into the basic model to study their effects on mortality. Next, a backward stepwise procedure was used to choose the final reduced model, with variables significant at being retained in the model. To obtain an adequate reduced model, before dropping a covariate from the model, we confirmed that its absence did not result in a substantial change in the overall predicting power of the model.
; #(cTnT × 100) : (GLS ≥−15%), interaction between cTnT × 100 and GLS ≥ −15%.
Abbreviations: CAD: coronary arterial disease; cTnT: cardiac troponin T, DM: diabetes; GLS: global left ventricular peak systolic longitudinal strain (a less negative GLS was defined by a GLS ≥−15%).