1% at 1

and 3 years, respectively. A further example of the application of the model is provided in Fig. 2. The c-statistics of the MESIAH model for the derivation cohort was 0.77 (95% confidence interval [CI] = 0.74-0.80), whose interpretation is as follows. If two patients from the cohort are randomly selected, 77% of the time the score is able to identify correctly which one of the pair will survive longer. For the internal validation, patients in the derivation cohort were randomly divided into four groups and the coefficients were recalculated after removing a quarter of patients in the derivation cohort. The coefficients remained largely unchanged between iterations, with the average c-statistic of 0.77 (c-statistics of the 4 iterations: 0.73, 0.77, 0.76, and 0.81). As the MELD score was derived in patients with endstage liver disease, we tested the performance of the model in subgroups Vincristine of patients

with and without cirrhosis. The c-statistic was 0.77 (95% CI: 0.74-0.81) in patients with cirrhosis and 0.78 (0.70-0.87) in those without, indicating that the model works well in the noncirrhosis population as well. Table 3 summarizes the characteristics of the subjects in the validation cohort (n = 904). In contrast to the derivation cohort, HBV was the most common (75%) in the selleck chemicals llc validation cohort. Accordingly, fewer patients (73%) had evidence of cirrhosis and the MELD scores were lower in the validation cohort than in the derivation cohort. However, they tended to have more advanced tumors, with only 28% of patients meeting the Milan criteria.

TACE was the most common choice of initial treatment (n = 518, 57%), followed by resection (n = 121, MYO10 13%), systemic chemotherapy (n = 81, 9%), and ablation (n = 17, 2%). In 144 (16%), comfort care only was provided. Liver transplantation was not available for patients in the validation dataset. After a median follow-up of 15 months, 508 (56%) patients died. The MESIAH score had a high degree of discrimination in the validation cohort with a concordance statistic of 0.82 (95% CI: 0.80-0.83), which was even higher than that in the derivation cohort (median = 0.77, Table 4). The calibration of the model prediction was also satisfactory, as illustrated in Fig. 3 in which patients in the validation cohort were divided into three groups and their expected survival was found to match closely with observed survival, although the large sample size and number of events resulted in significant P-values for the comparison (P for overall observed versus expected <0.01; P for Tier 1 <0.01, Tier 2 = 0.50, and Tier 3 <0.01). The model performed equally well regardless of the underlying etiology (c-statistic for HBV patients = 0.81 [95% CI: 0.79-0.83], for HCV = 0.82 [95% CI: 0.76-0.88], and for non HBV/HCV = 0.82 [95% CI: 0.77-0.87]). In Table 4 the performance of the MESIAH score is compared with that of BCLC, CLIP, and JIS.