In this retrospective study, we identified HCC- patients treated at the Department of Medicine II of Munich's University Hospital between January 1998 and March 2009. The research study was approved by the ethics committee of the University of Munich and the need for written informed consent was waived, because the data were analyzed retrospectively and anonymously. Histological or radiological (AASLD radiologic criteria 
) confirmation of diagnosis was mandatory for inclusion. Baseline was defined as time of primary diagnosis of HCC, and certain baseline examinations including laboratory and imaging studies were required for inclusion in the study. Patients were excluded when showing too fragmentary documentation of the data (>4 parameters missing) or whenever the survival status was unknown. In total, 550 consecutive patients with HCC were identified, of these 145 had to be excluded because of lacking data, leaving a study population of 405 patients.
Patients were identified from a data base collection in our institution, by using the International Classification of Diseases (ICD) code 150.0 for primary liver cancer. Clinical, tumor related and laboratory data needed to stage patients in all seven staging systems were retrieved from our electronic medical records. Additionally, a wide range of other parameters was compiled in order to further characterize our HCC-collective. The following data were collected: Age, sex, date of initial diagnosis, date of initial therapy, survival status, date of death, end of observation, liver cirrhosis, etiology, mode of therapy, Eastern Cooperative Oncology Group status (ECOG), Karnofsky-index, histology, ascites, hepatic encephalopathy (HE), portal vein thrombosis, portal hypertension, tumor extension, tumor burden (>/<50% of liver), number of tumor nodes, macroscopic vascular invasion, distant metastasis, lymph node involvement, BCLC tumor features (: singular <2 cm, : 3 nodules ≤3 cm or 1 nodule 2- ≤5 cm, : multilocular, : Portal invasion, N1, M1). Furthermore, the following laboratory parameters were retrieved in order to be able to calculate all tested staging systems: AFP, bilirubin, alkaline phosphatase, Quick and albumin.
In those cases without histology, the diagnosis of liver cirrhosis was made dependent on typical clinical signs of portal hypertension or on unequivocal radiological signs. Portal hypertension was diagnosed, if an elevated hepatic vein pressure above 10 mm/Hg, esophageal varices, splenomegaly or a platelet count below 100.000/µl were noted. Classification of ascites was performed according to the Child-Pugh score. Ascites detected by imaging but not visible on physical examination was termed mild, while the ascites was classified as “massive”, if clinically visible. Whenever exact classification of HE was missing in medical records, clinical signs of HE like tiredness, confusion and coma were used to retrospectively classify the respective HE grades I–IV 
Whenever medical records did not include exact documentation of Karnofsky performance (KPS) and
Eastern Cooperative Oncology Group performance status (ECOG), these classifications were retrospectively estimated on the basis of the available data on the general health status of the patient. For patients with exact documentation of either KPS or
ECOG, the missing score was deducted on the basis of the following estimation 
: ECOG 0
KPS 100%, ECOG 1
KPS 80%–90%, ECOG 2
KPS 60%–70%, ECOG 3
KPS 40%–50% and ECOG 4
All treatment decisions were based on an interdisciplinary tumor composed of hepatologists, (interventional-) radiologists, oncologists and surgeons. Although the advent of staging systems including treatment recommendations according to specific stages like BCLC has had an impact on these boards, treatment allocation to date remains an individual approach.
All baseline tumor parameters necessary to characterize the HCC-cohort and to calculate the staging systems were obtained by reviewing radiology and pathology reports, respectively. When in doubt concerning certain tumor measurements a radiologist (C.Z.) with 8 years experience in abdominal CT and MRI reevaluated the baseline images. Regional lymph node involvement was assumed when suspect lymph nodes (>1 cm in diameter) were detected on MRI and CT, respectively. Information on survival was retrieved from the clinical records, whenever possible. In all other cases the primary care physician was contacted via telephone or fax.
Out of 405, 365 patients showed sufficient data to perform stratification according to Child-Pugh-score, 395 patients according to TNM, 373 patients according to Okuda, 352 patients according to CLIP, 341 patients according to BCLC, 358 patients according to JIS, and 304 patients according to GETCH. 290 Patients could be classified by all staging systems. In order to keep the numbers of patients with incomplete data as small as possible this cohort was enlarged to 354 patients by substituting missing values for laboratory parameters by the median (Bilirubin 1, Quick 2, AFP 11, Albumin 16, and AP 42 values). Ranking of scores was done for both cohorts of 290 and 354 patients, respectively. There were no substantial differences found, thus only values for the 354 patients are reported.
For statistical analysis SAS-Software [SAS V9.2, SAS Institute Inc., Cary, NC] was used. p<0.05 indicated statistical significance, with a p<0.0001 the parameter was considered to be of high statistical significance.
For univariate analysis overall survival was estimated by using the Kaplan-Meier method from the date of primary diagnosis of HCC to the date of death or last follow-up. Survival curves were compared using the log-rank test. Additionally to the p-value medians of survival time and 95% confidence intervals for the different strata are given. Both, single parameters and the whole scores were analysed concerning their prognostic significance. For Kaplan-Meier-analysis of continuous variables, one or more cut-off values are necessary; therefore, laboratory values were divided into quartiles.
While the univariate analysis was performed for all the patients showing the individual parameter, multivariate analysis relates only to the cohort of n
354 patients who could be classified in all staging systems as described above. This number reflects those patients who could be classified in all staging systems. In order to keep the numbers of patients with incomplete data as small as possible, for calculating the scores and for multivariate analysis missing values for laboratory parameters were substituted by the median. In those parameters showing significance in univariate analysis using Cox proportional hazards regression model was conducted in order to examine their independent prognostic relevance. To avoid arbitrary cut-off values in this model laboratory values were taken as base two logarithms and used as continuous variables.
Ranking of staging systems was achieved by the Akaike information criterion (AIC) 
derived from the Cox model and concordance- index (c-index) 
. AIC is a measure of relative goodness-of-fit and thus provides a means for comparing models, a lower AIC value indicating a better model fit. Calculating the c-index requires no model assumptions, it represents the proportion of concordance in all possible pairs of patients meaning that the patient with the better prognostic score has the longer survival time. A score with a c-index of 0.5 is not better than chance, a c-index of 1 indicates perfect prediction. C-indices together with 95% confidence intervals were calculated using the SAS macro 
. In cases with disconcordant values of AIC and c-index, the AIC-value was favoured.