Data management and statistical analysis

Statistical analyses were performed using R version 4.2.0 software. Firstly, the multiple imputation method was used to fill in missing data, and variables with more than 20% missing data were deleted [11]. Shapiro-Wilk tests were performed to assess the distribution of variables. For continuous data, variables with normal distribution were reported as mean ± standard deviation, and variables with skewed distribution were reported as median and interquartile range. Count data were expressed as numbers and percentages. Student′s t tests and Mann-Whitney U test tests were used to compare the continuous data. The chi-squared tests were applied to compare the categorical variables between the two groups. Prognostic factors were assessed by univariate analyses. To adjust for potential confounders, variables related to 28-day death in univariate analysis (p < 0.1) were entered into analysis by backward stepwise Cox proportional hazards regression modeling. And Akaike′s information criterion was used to select the final model. Interaction tests were further performed between the prognostic variables. Variance inflation factor (VIF) and Pearson’ s correlation coefficients to assess collinearity for continuous data. VIF > 5 was considered to indicate collinearity [12], and we also included variables with a correlation coefficient < 0.5 [13]. The 28-day survival was estimated using the Kaplan-Meier and compared by log-rank test.

Patients enrolled with SIMI were randomly distributed into the training cohort and validation cohort at the ratio of 7:3. And a nomogram for predicting 28-day mortality of SIMI was constructed in training cohort according to Occam′s Law, in which fewer variables should be included to achieve the aim [14] Concordance index (C-index) and the area under the receiver operating characteristic curve (AUC) were performed to assess the effectiveness of the nomogram [15], the integrated discrimination improvement (IDI) and the net reclassification improvement (NRI) were carried out to evaluate the accuracy of nomogram [16, 17]. Calibration plots by bootstrap method were used to evaluate consistency between the predicting probability and the actual probability, and decision curve analysis (DCA) was used to assess the net benefit of nomogram at different threshold probabilities [18].