Related Experiment Videos
A laboratory-based selected logistic model for early prediction of 28-day mortality in patients with septic shock
Liying Zhang1, Pu Chen2, Jie Li3
1The Blood Transfusion Department, the First People's Hospital of Yibin City, Yibin, China.
Background:
Early risk stratification in septic shock requires transparent models based on variables available soon after clinical diagnosis of septic shock. We reanalysed a retrospective cohort using the original screening strategy: candidate predictors with univariable P < 0.05 were entered into multivariable logistic regression, and variables with multivariable P < 0.05 were retained in the final model.
Methods:
The analysis included 250 patients with complete 28-day outcome and laboratory data. The outcome was 28-day mortality. Candidate predictors comprised demographics, comorbidities, complete blood count, chemistry, coagulation, inflammatory markers, arterial pH, and PaO2/FiO2. Laboratory values were treated as the first available values within 24 h after clinical diagnosis of septic shock. The final fixed model and the complete model-development pipeline were internally validated using 1,000 bootstrap resamples. Discrimination, calibration, Brier score, ROC-derived cutoff, decision curve analysis, LAR comparison, restricted cubic spline analysis, VIF, and LASSO sensitivity analysis were reported.
Results:
Among 250 patients, 73 died by day 28 (29.2%). Univariable screening identified 12 candidate predictors, and multivariable screening retained lactate, D-dimer, and PaO2/FiO2. The final equation was logit(P) = 0.412145 + 0.250015 × Lactate + 0.046852 × D-dimer - 0.016820 × PaO2/FiO2. The apparent AUC was 0.902 (95% CI 0.864-0.940), and the optimism-corrected AUC for the fixed final model was 0.897. When the complete model-development pipeline was repeated within 1,000 bootstrap resamples, the optimism-corrected AUC was 0.875. The final model had higher discrimination than LAR alone [0.902 (0.864-0.940) vs. 0.719 (0.645-0.793); DeLong P < 0.001].
Conclusions:
A selected logistic model using lactate, D-dimer, and PaO2/FiO2 provided internally validated early risk estimates for 28-day mortality. External validation is required before routine clinical use.