By continuing to browse this site you agree to us using cookies as described in About Cookies
Wiley Online Library is migrating to a new platform powered by Atypon, the leading provider of scholarly publishing platforms. The new Wiley Online Library will be migrated over the weekend of March 17 & 18. You should not experience any issues or loss of access during this time. For more information, please visit our migration page: http://www.wileyactual.com/WOLMigration/
Advanced hepatocellular carcinoma (HCC) patients who are not candidates for surgery or locoregional therapy are the focus of clinical trials of systemic therapy, as their overall prognosis remains poor. However, the current prognostic systems cannot reliably select appropriate candidates for systemic therapy trials based on the probability of 3-month survival. In this study, the authors constructed a new prognostic score system, the Advanced Liver Cancer Prognostic System (ALCPS), which can objectively predict the probability of 3-month survival.
Between 1990 and 2005, 1470 patients with advanced HCC who were not amendable to surgery or locoregional therapy were included in the analysis. The prognostic score system was developed from the multivariate Cox model through a point system and validated in an independent set. Okuda staging and Cancer of the Liver Italian Program (CLIP) score were also applied to the validation set to compare their predictive accuracy.
The ALCPS was based on 11 prognostic factors with different weights: ascites, abdominal pain, weight loss, Child-Pugh grade, alkaline phosphatase, total bilirubin, alpha-fetal protein, urea, portal vein thrombosis, tumor size, and presence of lung metastases. It stratified patients in both training and validation sets to different prognostic groups with significant difference in 3-month overall survival (P < .0001). By using the patients in the validation set with known 3-month survival status, the ALCPS showed significantly better predictive power (area under the curve [AUC], 0.77) than Okuda score (AUC, 0.66; P < .001) and CLIP score (AUC, 0.71; P = .002).
Hepatocellular carcinoma (HCC) is the fifth ranking cancer in the world, with greater than 80% of cases occurring in Asia.1 The most common causes of HCC are hepatitis B and C viral infections. Chronic hepatitis B viral infection is prevalent in Asian countries and accounts for most cases of HCC. In contrast, chronic hepatitis C viral infection is more common in Western countries. In recent years, there is an increasing incidence of HCC in Western countries, primarily because of an increase in the prevalence of hepatitis C viral infection.
Current effective treatments for HCC include liver resection, transplantation, and various local ablative and transarterial therapies. Surgical resection and liver transplantation are the main curative treatments. Unfortunately, only around 20% of patients, mostly diagnosed by regular screening, may benefit from these surgical therapies. Most other patients either present late with advanced tumor or have severe underlying cirrhosis, precluding any surgical or even locoregional therapies. Previously, these patients can only be palliated by chemotherapy or best supportive treatment alone. However, HCC is a relatively chemotherapy-resistant disease. Recently, single-agent sorafenib has been shown to improve the overall survival of patients with advanced HCC.2 Moreover, other molecular-targeted agents have also demonstrated promising results in a phase 2 setting3, 4 Therefore, there is now new enthusiasm among oncologists in developing novel systemic therapy for advanced HCC patients.
Several staging or prognostic systems have been developed to guide the prognosis and treatment of patients with HCC. These include the TNM,5 Okuda,6 Cancer of the Liver Italian Program (CLIP),7 Chinese University Prognostic Index (CUPI),8 and Barcelona Clinic Liver Cancer (BCLC) staging systems.9 All these staging systems were developed based on both surgical and nonsurgical patients, and the systems generally aim to classify patients with HCC broadly into early, intermediate, and advanced stages. However, advanced HCC patients are heterogeneous in characteristics, with substantial difference in overall survival. The available systems are not able to subdivide this heterogeneous population into different prognostic groups. Moreover, there is no worldwide consensus on which is the best system in staging and predicting the prognosis of patients with advanced HCC.
Nowadays, most systemic drug trials in advanced HCC require a life expectancy of more than 3 months as a basic inclusion criterion to fairly assess its activity. However, the currently available staging or prognostic systems provide little information that assists clinicians to reliably predict the survival of advanced HCC patients and select appropriate candidates for systemic drug trials. With the rapid development of new agents for HCC that will be evaluated in clinical trials, it is imperative to develop a prognostic score system that can more objectively predict the prognosis of patients with advanced HCC, so that appropriate patients can be recruited into such trials. In this study, we aimed to construct a prognostic score system that can predict the probability of 3-month survival in patients with advanced HCC not amendable to any surgical or local-ablative therapies, based on a large cohort of patients from a single tertiary referral center in Hong Kong.
MATERIALS AND METHODS
Between 1990 and 2005, there were 3805 adult patients (aged ≥18 years) diagnosed with HCC and presented to the hepatobiliary and liver transplant unit at Queen Mary Hospital, Hong Kong. The diagnosis of HCC was confirmed by either histology, cytology, elevated alpha-fetoprotein (≥400 ng/mL), or by typical radiological appearance. At our center, patients were staged mostly by computed tomography or ultrasound scan. All patients with adequate liver function and radiologically resectable tumor were initially evaluated for partial hepatectomy. Patients of aged 65 years or younger with early HCC and cirrhosis would be considered for liver transplantation if they were within the Milan10 or expanded University of California at San Francisco criteria.11 If the patients were not surgical candidates, transarterial chemoembolization or local ablative procedures, such as percutaneous ethanol injection and radiofrequency ablation, were offered, depending on tumor size, number, and position. The remaining 1470 patients who were not candidates for surgery, transplantation, or any locoregional therapies were included in the present analysis. These patients were followed with regular clinical examination, blood tests, and scanning until death or last follow-up in the same hospital. All data were entered prospectively into a database for analysis. Twenty variables that included patients' clinical, biochemistry, and tumor characteristics were selected for analysis. They were all previously reported as significant prognostic factors in other studies.6–8 The conventional threshold value for each continuous variable was applied and shown in Table 1. In this analysis, patients were classified to be asymptomatic upon presentation when their HCC was diagnosed incidentally by blood tests or ultrasound scanning and they did not have any symptoms attributable to HCC. Moreover, weight loss was defined as more than 15 lb loss in weight within 3 months before presentation.
Table 1. Demographic Data, Biochemistry, and Tumor Characteristics on Presentation
ALP indicates alkaline phosphatase; AFP, alpha-fetal protein; CI, confidence interval; SE, standard error.
Pearson chi-square test for categorical factors; Wilcoxon test for continuous factors and no further test was performed on the categorized variable; log-rank test to compare the survival curves.
Age, y Median (range)
Age >65 y
Asymptomatic upon presentation
Total bilirubin, μmol/L
>33 to ≤50
Prothrombin time, s
Portal vein thrombosis
Lymph node metastases
Median follow-up time, mo
Missing survival status at 3 mo
Median survival, mo (95% CI)
3-mo survival rate (SE)
All eligible patients were randomly allocated into a training set or a validation set with an approximately 3:1 ratio. The prognostic score system was constructed by using the training set and subsequently confirmed in the validation set. The survival time was defined as the time from the date of first registration in our unit to the date of death or last follow-up date. All the continuous variables were categorized using conventional threshold values on their original scales. Most variables were dichotomized into 2 groups, but a few were divided into 3 groups. All subsequent analyses were conducted using Statistical Analysis System Version 9.1.3 (SAS Institute, Cary, NC), and all P values resulted from the use of 2-sided statistical tests.
To check the similarity between the training and the validation sets, the clinical, biochemistry, and tumor characteristics of patients upon presentation in the 2 sets were compared using the chi-square test for categorical variables, the Wilcoxon rank sum test for continuous variables, and the log-rank test for overall survival.
Development of the prognostic score system
A prognostic score system was developed using the training set to identify patients with a high chance of survival beyond 3 months. First, the prognostic significance of each potential factor was evaluated by inspecting the Kaplan-Meier curves of overall survival of patients in different categories of the factor and compared by the log-rank test. All prognostic factors with P value <.1 in univariate analysis were identified and included in a multivariate Cox regression analysis.12 All statistically significant prognostic factors were then selected by backward elimination procedure as described in the SAS Procedure PHREG.13 A variable was removed if its Wald test result was not significant (ie, P value >.05). Afterward, insignificant variables in the univariate analysis were reintroduced 1 by 1 to examine their statistical significance in the presence of other variables. The proportional hazards assumption of the final model was assessed by the method described in Lin and Ying.14
The prognostic score system was then developed from the final multivariate Cox model using the point system described by Sullivan et al.15 All factors in the model were categorized. A point for each of the risk factors was determined by the relative magnitudes of the regression coefficients of the final Cox model. The prognostic score of an individual patient was calculated by summing up the assigned points over all relevant risk factors of the patient. The 3-month survival rate for each score was then estimated by the Cox regression model, using the score to approximate the regression portion of the Cox model as described in Sullivan et al.15 The training set was then divided into 3 different groups based on their estimated probability of 3-month survival. The poor prognostic group was defined as patients with a 3-month survival rate ≤35%, the good prognostic group was defined as patients with a 3-month survival rate >65%, and the rest were categorized as the intermediate prognostic group. The scores that corresponded to 35% and 65% survival rates were identified and used as the threshold values in the score system.
Validation of the prognostic score system
The prognostic score system was tested in the independent validation set to evaluate its predictive power. Patients were assigned to different prognostic groups based on the calculated score. Subsequently, pair-wise log-rank tests were performed to compare each prognostic group of the validation set to each prognostic group of the training set. The Kaplan-Meier curves for these patients grouped by Okuda stage and by CLIP score were also plotted.
To assess the discriminatory ability of the prognostic score system, the receiver-operating characteristic (ROC) curves were constructed for survival status at 3-month follow-up. Because only a small portion of patients were censored within 3 months, and procedures to construct ROC including censored observation such as the method proposed by Heagerty et al16 were not yet available in SAS 9.1, the ROC curve was constructed using only patients with at least 3 months follow-up time. The sensitivity-probability of score <cut-point given survival beyond 3 months, and specificity-probability of score >cut-point given death within 3 months, can be evaluated for each possible cut-point. The area under the ROC curve (AUC) represented the probability that a randomly selected patient who survived beyond 3months has a lower diagnostic score than a randomly selected patient who died within 3 months, and is used to represent the diagnostic ability of the score system. It was the c-index in the logistic regression using the diagnostic score as the only prognostic factor for 3-month survival and was obtainable from the SAS Procedure LOGISTC. The ROC curves for using Okuda score (range, 1-5) or CLIP score (range, 0-5) to predict 3-month survival were also plotted. The AUCs of these ROC curves were compared using the methods described by Hanley and McNeil.17
There were 1470 advanced HCC patients who were not amendable to locoregional therapy included in this analysis. Overall, 1006 patients were younger than 65 years, whereas 464 patients were older than 65 years (median = 59 years; range, 20 years-94 years). The majority of patients were men (84.4%). All except 4 patients were ethnic Chinese. Seventy percent of patients were hepatitis B virus carriers. Most patients (87%) presented with symptoms. Around 49.2 % of patients had underlying Child A cirrhosis, 36.3% of patients had Child B cirrhosis, and only 14.5% of patients had Child C cirrhosis. Half of the patients had underlying portal vein thrombosis, and for patients with data on tumor size, >80% patients had a tumor >5 cm or had diffuse tumor on presentation. Regarding treatment modalities, 57% of patients received best supportive treatment only, whereas 29% of patients were put on tamoxifen alone, and only 7% of patients had received chemotherapy treatment. The remaining patients received either thalidomide (3%) or octreotide (Sandostatin) (2%).
From the 1470 patients, 1109 patients (75.4%) and 361 patients (24.6%) were randomly assigned to the training set and the validation set, respectively. Table 1 shows 20 selected potential prognostic factors comprised of clinical, biochemistry, and tumor characteristics of the patients in these 2 sets upon presentation. Overall, they shared similar characteristics, including survival or censoring pattern. All patients' data were complete, except for tumor size and 3-month survival status. Tumor size was not available in 9.8% and 11.4% of patients in the training and validation sets, respectively; 3-month survival status was not available in 65 (5.9%) patients in the training set and 15 (4.2%) patients in the validation set. All but 3 of these patients were lost to follow-up. The 3 patients were registered less than 3 months before the cutoff time for data analysis. One thousand and eleven (91.2%) patients died in the training set; whereas 333 (92.2%) patients died in the validation set at the time of analysis. The median overall survival time was 2.6 months (95% confidence interval [CI], 2.3-2.8) in the training set after a median follow-up time of 2.3 months (range, 0 months-114 months), and 2.3 months (95% CI, 2-2.6) in the validation set after a median follow-up time of 2.2 months (range, 0 months-80 months).
Construction of a New Prognostic Score System
Table 2 shows the association between potential prognostic factors and overall survival in the univariate analysis. All the potential prognostic factors tabulated in Table 1, except for sex, encephalopathy, and bone metastases, showed a significant association with overall survival. In our dataset, all significant factors had a P value <.01 from the log-rank test, except for age, where P = .011. All the significant factors were entered in the multivariate analysis, and the final Cox regression model was selected by backward selection procedure as well as by Collett's strategy.12 The final model consisted of 11 significant prognostic factors: ascites, abdominal pain, weight loss, Child-Pugh grade, alkaline phosphatase, total bilirubin, alpha-fetoprotein, urea level, tumor size, portal thrombosis, and lung metastases. Table 3 shows the coefficients of these prognostic factors in the final Cox regression model based on the 1000 patients in the training set with information on tumor size. For total bilirubin, although only the higher level (>50 μmol/L) showed a significant association with survival independently in this study, we kept the original 3 categories in the final model. The assumption of proportional hazards of all the factors in the final model was assessed and was not rejected.
Table 2. Univariate Analysis of Potential Prognostic Factors on Presentation
HR indicates hazard ratio; β, coefficient in the multivariate Cox model; ALP, alkaline phosphatase; AFP, alpha-fetal protein.
P values are from 2-sided Wald's chi-square statistic.
Total bilirubin, μmol/L
Portal vein thrombosis
The new prognostic score system, termed the Advanced Liver Cancer Prognostic System (ALCPS), was based on the final Cox model. Table 3 shows the assigned points for each category of the 11 prognostic factors. Each patient's prognostic score was the sum of 11 points according to the patient's clinical, biochemistry, and tumor characteristics. The theoretical range of the prognostic score is 0 to 39. A higher score implies a lower chance of survival. The corresponding 3-month survival rate for each score is shown in Table 4. In this prognostic score system, patients with a score ≤8 are assigned to the good prognostic group and have a high chance (>65%) of 3-month survival; patients with a score in the range 9 to 15 are assigned to the intermediate prognostic group and have a chance between 35% and 65% of 3-month survival; and patients with a score ≥16 are assigned to the poor prognostic group, with a <35% chance of 3-month survival.
Table 4. Probability of Patients Surviving at Least 3 Months Estimated by the Advanced Liver Cancer Prognostic System Score
3-Month Survival Rate
On the basis of this prognostic score system, the 1000 patients in the training set were stratified according to their respective prognostic score, and the Kaplan-Meier plot of each prognostic group is illustrated in Figure 1A. In the good prognostic group, the 200 (20%) patients had a 3-month survival rate of 82.0% (95% CI, 76.5%-87.5%), and the median overall survival was 7.9 months (95% CI, 6.6-10.2). In the intermediate prognostic group, the 459 (45.9%) patients had a 3-month survival rate of 53.4% (95% CI, 48.3%-57.7%), and the median overall survival was 3.2 months (95% CI, 2.8-3.5). In contrast, in the poor prognostic group, the 341 (34.1%) patients had a 3-month survival rate of 18.9% (95% CI, 14.7%-23.3%), and the median overall survival was 1.4 months (95% CI, 1.2-1.6).
Validation of the Prognostic Score System
There were 361 patients in the validation set, but the analysis was only performed in 320 patients with information on tumor size. After applying the ALCPS in these 320 patients, 64 (20%) patients were assigned to the good prognostic group, 149 (46.6%) patients to the intermediate prognostic group, and 107 (33.4%) patients to the poor prognostic group. The Kaplan-Meier curves of patients in the 3 prognostic groups are plotted in Figure 1B. The median overall survival and 3-month survival rates in the validation set were very similar to those for the training set. They were estimated as 7.5 months (95% CI, 5.7-9.4), 3.2 months (95% CI, 2.4-3.9), and 1.2 months (95% CI, 0.9-1.5), respectively; and 82.4% (95% CI, 72.5%-91.5%), 51.1% (95% CI, 42.7%-59.3%), and 21.1% (95% CI, 13.2%-28.8%), respectively. The pair-wise comparison of each prognostic group in the validation set to each prognostic group in the training set indicated that patients assigned to the same prognostic level in different sets had very similar survival curves, but patients assigned to different prognostic levels from different sets had very distinguishable survival curves. The P values of the log-rank tests (unadjusted for multiple comparisons) were presented in Table 5. In contrast, the Kaplan-Meier survival curves of these 320 patients grouped by Okuda stage and by CLIP score were illustrated in Figure 1C and D, respectively. Moreover, Table 6 shows the median overall survival and 3-month survival of patients with different Okuda stages and CLIP scores.
Table 5. P Values of the Pair-Wise Log-Rank Tests Comparing Each Prognostic Group in the Validation Set to Each Prognostic Group in the Training Set
Table 6. Median Survival and 3-Month Survival Rate in the Validation Set According to Okuda and Cancer of the Liver Italian Program Score
CI indicates confidence interval; CLIP, Cancer of the Liver Italian Program.
Median survival, mo (95% CI)
3-mo survival rates, % (95% CI)
Median survival, mo (95% CI)
3-mo survival rates, % (95% CI)
Furthermore, 306 patients of the validation set had complete information, that is, information on all prognostic factors including tumor size and the survival status at 3 months. The discriminatory ability of the ALCPS was assessed using these patients by generating a ROC curve for 3-month survival status. Figure 2 demonstrates the ROC curve with the AUC = 0.77 (standard error = 0.026) and shows that the ALCPS discriminated reasonably well by the survival status of the patients at 3 months of follow-up. Of note, the ROC curves generated using Okuda score and using CLIP score to predict 3-month survival were completely under the curve for the ALCPS (Fig. 2), with a significantly smaller AUC of 0.66 (P < .001) for Okuda score and 0.71 (P = .002) for the CLIP score system.
Patients with advanced HCC who are not candidates for surgery or locoregional therapy have poor overall prognosis. They are heterogeneous in character, including patients with extrahepatic metastases, extensive portal vein thrombosis, significant liver function derangement, poor performance status, or significant comorbidities. Single-agent doxorubicin has been shown to produce a response rate of about 10% to 15% but no proven survival benefit.18 Other systemic agents, including tamoxifen, somatostatin analogue, and other chemotherapy agents and their combinations have been extensively investigated with disappointing results, showing low response rates and no survival benefit.19, 20 Recently, a study of single-agent sorafenib has shown some promising results in improving the overall survival of patients with advanced HCC.2 This exciting development has attracted much academic and industrial interest in pursuing effective systemic treatment therapy for HCC. In this era of targeted therapy, proper patient selection is vital to the success of the clinical trials of new agents, as patients have to survive long enough for the targeted therapy to exert its effect. Therefore, in most clinical trials, life expectancy of at least 3 months is 1 of the inclusion criteria. However, assessment of life expectancy is often based on subjective “guessing” by the investigators rather than any objective prediction.
In the present study, we have established a new prognostic score system, the ALCPS, which is able to objectively estimate the probability of 3-month survival for advanced HCC patients who are not amendable to surgery or other local ablative treatments. The ALCPS was established based on a large single-center prospectively maintained database with very few missing data. Although the study period spanned more than 15 years, the management algorithm for patients with advanced HCC who were not amendable to surgery or other local ablative treatments was similar throughout the study period. This is the largest unselected series consisting of only advanced HCC patients in the literature thus far. The low median overall survival of this patient population as demonstrated in the current study reflects the finding that many HCC patients presented with late disease, which is a reality not only in Asia but also in many Eastern and Western countries. In contrast, the patients recruited in reported systemic drug trials such as the SHARP trial2 were those with relatively less advanced tumors, good liver function, and performance status preselected for entry into the trials, and hence their median survival was expectedly much better than the current cohort. This actually reinforces the importance of a scoring system to select patients with reasonable prognosis for entry into clinical trials of novel agents so that the benefit of the agents can be demonstrated. Notably, although the overall median survival was low in our cohort, a substantial subgroup of patients can still be selected with reasonable prognosis for participation in clinical trials using the ALCPS system.
With most solid tumors, the prognosis of patients is directly related to the staging of the disease at the time of presentation. However, in HCC, the situation is more complicated, as it consists of 2 diseases: cancer and underlying cirrhosis. The prognosis of patients is affected by the extent of cancer, underlying cirrhosis, and organ function of patients. In the present prognostic system, all the aforementioned factors have been taken into account. The variables consisted of patients' clinical, blood biochemistry, and tumor characteristics. All these variables are routinely available in clinical practice and easily computed. Our final model consisted of 11 prognostic factors: ascites, abdominal pain, weight loss, Child-Pugh grade, alkaline phosphatase, total bilirubin, alpha-fetal protein (AFP), urea, portal vein thrombosis, tumor size, and presence of lung metastases. Most of these factors were previously reported as significant prognostic factors in other studies,7–9 except urea and presence of lung metastases. Moreover, from the assigned points, we see that Child-Pugh C cirrhosis has the largest weight contributing to a poor prognostic outcome, followed by AFP level >400 ng/mL and diffuse type of HCC. These findings support the notion that both the underlying cirrhosis and tumor burden significantly affect the prognosis of advanced HCC patients. The advantage of this score system is that it gives a more comprehensive assessment of patients rather than relying on a single or few clinical, biochemical, or radiological parameters alone. This will minimize the chance of excluding patients based on a single variable such as portal vein thrombosis who may otherwise be good candidates for clinical trials. Although performance status (PS) is an important parameter in assessing the likelihood of the patients to benefit from and tolerate systemic therapy, it is not taken into consideration in the current study, as our database only started to include PS as 1 of the parameters a few years ago, and thus there are many missing data in PS. In fact, the ALCPS provides an estimation of the chance of the patient surviving long enough to benefit from a trial agent, whereas PS helps to select patients who are likely to tolerate the treatment. These 2 parameters can be complimentary to each other in assessing the eligibility of HCC patients in clinical trials.
Clinicians can use the ALCPS to predict the likelihood of any individual with advanced HCC to survive more than 3 months. It clearly separates this patient group into 3 groups: good, intermediate, and poor, based on their probability of 3-month survival, with significantly distinct overall survival curves in the validation set over the entire follow-up period. Okuda6 and CLIP7 staging systems are widely used worldwide to estimate the prognosis of HCC and were chosen for comparison with ALCPS in the current study. CUPI8 was not used for comparison, as 1 of its variables is TNM staging, which is a staging system for patients with surgical resection of HCC and is based on data from pathological examination of the tumors. Therefore, TNM stage may not be applicable to our patient populations with advanced HCC, and it is impossible to obtain reliable TNM stage information from this group of patients. More importantly, CUPI uses the International Union Against Cancer 1977 TNM staging system and gives different scores to stage I/II, stage III, and stage IV patients. However, the TNM staging system was updated in 2002 with significant changes, and the classification between stage III and stage IV is different from the 1997 system. This further complicates the use of CUPI in the current era. With respect to Barcelona Clinic Liver Cancer9 staging, it is mainly designed for guiding treatment choices ranging from transplantation, resection, ablation, and chemoembolization to systemic therapies. It is applicable only when the whole range of patients are included, including early, intermediate, and advanced HCC. Therefore, it was not chosen to compare with ALCPS, which is comprised of the advanced HCC patient population only. Okuda staging is widely used because of its simplicity. However, it only includes tumor involvement, ascites, jaundice, and serum albumin level in the system. Moreover, it is difficult to give an accurate estimation of tumor involvement (ie, >50% or <50%). When Okuda staging was applied to the cohort of patients in our validation set, the survival curves of the 3 groups of patients in Okuda staging overlapped. The CLIP scoring system was developed in 2000 based on 435 patients from 16 different Italian centers and is supposed to have better prognostic power than Okuda staging. Its scoring system was based on Child-Pugh stage, tumor morphology, AFP level, and the presence of portal vein thrombosis. Patients were stratified into levels 0 to 6 based on their calculated scores. Similar to the shortcoming of Okuda staging, tumor morphology is difficult to estimate in the CLIP score system. When we applied the CLIP scoring system to the cohort of patients in the validation set, the survival curves of 6 distinct CLIP score groups overlapped and crossed over. More importantly, when we used the ROC curve to assess the discriminatory ability of the prognostic score system for 3-month survival status, the ALCPS has reasonably good AUC (0.77) and is significantly better than Okuda score (AUC, 0.66; P < .001) and CLIP score (AUC, 0.71; P = .002). All the above results confirm that the ALCPS is a more sensitive and specific predictor of 3-month survival than the Okuda and CLIP scoring systems for patients with advanced HCC. Although the ALCPS contains a few more variables than the existing systems,6–8 it represents a more comprehensive assessment of patients and disease extent with higher predictive accuracy. Moreover, all the parameters are readily available and easily computed using the current scoring system.
Although the exact cutoff point for patient selection into new drug trials for HCC remains arguable, we propose that probably only patients with at least a 50% chance of surviving 3 months should be considered for recruitment into systemic drug trials. This implies that only patients with score 0 to 12 will be eligible for trial. However, the requirement in terms of predicted survival may vary in different trial protocols, and our prognostic score system provides the investigators with an objective system to predict different levels of chance of 3-month survival.
In the present study, ALCPS is constructed based on a large dataset, predominantly in a hepatitis B prevalent Chinese population. However, whether this system can be applied in Western countries, which consist of largely Caucasian populations with the etiology of HCC based mainly on hepatitis C and alcohol, is still unknown. Moreover, our current dataset is based on a single institutional experience. Therefore, future validation of this prognostic system by further studies in a multi-institutional setting including other Chinese and non-Chinese centers is needed before the generalized use of ALCPS can be recommended.
In conclusion, this study has established a novel prognostic system, the ALCPS, which can objectively predict the chance of 3-month survival in patients with advanced HCC and hence can assist clinicians to select appropriate advanced HCC patients for various drug trials.
We thank Professors Raymond Liang and Richard Epstein for supporting this collaborative work and Dr. Daniel Fong for providing his statistical opinions in the present study.