Using the guidelines for cancer biomarker validation suggested by Pepe et al. (23), we propose to perform a Phase 2 study of DCP for the detection of early stage HCC. In this proposal, we plan to perform a larger case-control study to compare the sensitivity and specificity of DCP and AFP alone and in combination in differentiating patients with all stages of HCC and more importantly those with early HCC from patients with cirrhosis. We plan to enroll consecutive patients with HCC seen at 7 centers in the United States. Controls are frequency matched to cases (all center combined) using the following criteria: age (±10 years), gender (+10%) and etiology of liver disease (viral vs non-viral (+5%). Within each participating institution, there will be an equal number (+20%) of cases and controls.
The aims of this proposal are: (a) to determine the sensitivity and specificity of DCP for the detection of early HCC, (b) to compare the accuracy of DCP and AFP for the detection of early HCC, and (c) to determine whether demographic or etiology of underlying liver disease alter the expression of DCP or AFP. In order to achieve the aims of the study, we will enroll patients at 7 liver centers to perform a case control study of those with cirrhosis and those with early HCC. We will obtain the demographics, medical history, history of liver disease, social history (attention to lifetime smoking and alcohol), etiology of liver disease, family history, and clinical and laboratory data of patients with cirrhosis and early HCC. In addition, we will obtain serum, plasma and DNA for the evaluation of markers for HCC. All the data obtained will be maintained in a web-based database. We plan to enroll 450 HCC patients and 450 cirrhosis controls in order to target 190 early-stage HCC cases to achieve 90% power. If 260 late-stage HCC cases are recruited before the target 190 early-stage HCC cases, then additional enrollment of late-stage HCC cases will be stopped. This is an important validation study of DCP, which may lead to the development of a clinically needed marker for HCC.
The primary goal is to compare the sensitivity and specificity of DCP to AFP for early stage HCC. We will test the null hypothesis that the ratio, TPF(DCP)/TPF(AFP) is larger than one, and similarly the null hypothesis FPF(DCP)/FPF(AFP) is less than one (Aim 2). 95% confidence interval for these two ratios, as well as that for differences between the two markers in TPF and FPF will be calculated. If both null hypotheses are rejected, we will conclude that DCP is superior to AFP in both TPF and FPF. If one null hypothesis is rejected (most likely for TPF) and another is inconclusive, the conclusion is that DCP is superior to AFP in TPF but inconclusive in FPF. A subsequent study will be required with a larger number of non-HCC cirrhosis patients to estimate the magnitude and direction of the difference in FPF. If neither null hypothesis is rejected, we will conclude that DCP is not superior to AFP in TPF or FPF. In that case, it is still of interest to see if the combination of two will improve the performance of either test. A joint confidence region of (TPF, FPF) for DCP (Aim 1), as well as that for AFP will be constructed. This will provide the performance measure of each biomarker for early stage HCC. To examine whether the performance of DCP depends on demographic or etiology of underlying liver disease (Aim 3), we will use logistic regression model. Taking ethnicity as an example, the logistic regression will have disease status as the dependent variable and independent variables will include the value of DCP, ethnicity indicator, and DCP by ethnicity interaction. The effect of clinical variables, e.g., etiology (no Hepatitis, HCV, HBV), and their interaction with DCP in predicting disease status will be examined in similar way. Of particular interest is to compare Hepatitis C-related HCC versus Hepatitis C-related controls and Hepatitis C negative HCC versus Hepatitis C negative controls due to the fact that Hepatitis C is the most common cause of cirrhosis and HCC in the USA. To examine whether the combination of DCP and AFP will perform better than either test alone, we will use logistic regression with DCP or AFP alone, or a model with DCP, AFP, and their interaction term in the model, and compare their performance. If the combination does improve the classification, we will look at the linear combination of regression coefficients to find out the optimal combination as a classification rule. AFP-L3 was added to this study as an ancillary study. The data analysis of AFP-L3, as compared to AFP, is similar to that for DCP described above. We will also use a logistic regression entering DCP, AFP, and AFP-L3 into the model in different order to assess each marker’s independent additional contribution to ROC curve in addition to other marker(s).
The primary goal is to compare the sensitivity and specificity of DCP to AFP for early stage HCC. We will test the null hypothesis that the ratio, TPF(DCP)/TPF(AFP) is larger than one, and similarly