However, the goal of this study is to develop classifiers capable of making accurate and comprehensible predictions rather than simply identifying significant predictive factors, as in most previous research [ 7- 9].
Using selected factors that significantly predicted OS in Kaplan Meier analyses, multivariate analyses of independent and significant predictive factors for OS for all patients and for those alive at 1, 2, and 3 years of follow-up were performed using a multivariate Cox proportional hazards model (Table 5).
Pathological node metastasis, perineural invasion, and extracapsular spread were the most significant predictive factors for 5-year overall survival.
We extracted the following information from included studies: study design, sample size, inclusion and exclusion criteria, subject demographics, prevalence rate of aspiration pneumonia, type of outcome, type of statistical analysis and other significant predictive factors for each outcome.
Significant predictive factors for post-CABG stroke included; (i) carotid bruit (OR 3.6, 95% CI 2.8 4.6), (ii) prior stroke/TIA (OR 3.6, 95% CI 2.7 4.9) and (iii) severe carotid stenosis/occlusion (OR 4.3, 95% CI 3.2 5.7).
In the multivariate analysis (Cox proportional hazards models) MR-proADM levels and MR-proADM clearance at the 5th day following admission, were statistically significant predictive factors for mortality in the ICU and at 90 days (Figure 3). Figure 3 multivariate analysis mortality 90 days.
Univariate analysis of the endpoint: cardiovascular mortality (not shown in Table 2) resulted in the same significant predictive factors, but with higher HRs.
Multivariate analysis of cardiovascular mortality resulted in the same significant predictive factors with higher HRs (SST) as well: albuminuria 1.83 (95% CI 1.17 2.89); history of cardiovascular disease 2.59 (95% CI 1.56 4.28).
Among demographic variables, no significant age difference existed, while it was found that 6 months after the earthquake, symptoms of anxiety, depression and PTSD were significantly more common among students in grades 4 and 5 than those in grade 3, Initial exposure to death, bereavement and extreme fear were significant predictive factors for the occurrence of anxiety, depression and PTSD.
Re-identifying significant predictive factors would help in decreasing this incidence.
As no significant predictive factors were found, we refrained from multivariate testing.
