Evaluating screening approaches for hepatocellular carcinoma in a cohort of HCV related cirrhosis patients from the Veteran’s Affairs Health Care System

Background Hepatocellular carcinoma (HCC) has limited treatment options in patients with advanced stage disease and early detection of HCC through surveillance programs is a key component towards reducing mortality. The current practice guidelines recommend that high-risk cirrhosis patients are screened every six months with ultrasonography but these are done in local hospitals with variable quality leading to disagreement about the benefit of HCC surveillance. The well-established diagnostic biomarker α-Fetoprotein (AFP) is used widely in screening but the reported performance varies widely across studies. We evaluate two biomarker screening approaches, a six-month risk prediction model and a parametric empirical Bayes (PEB) algorithm, in terms of their ability to improve the likelihood of early detection of HCC compared to current AFP alone when applied prospectively in a future study. Methods We used electronic medical records from the Department of Veterans Affairs Hepatitis C Clinical Case Registry to construct our analysis cohort, which consists of serial AFP tests in 11,222 cirrhosis control patients and 902 HCC cases prior to their HCC diagnosis. The six-month risk prediction model incorporates routinely measured laboratory tests, age, the rate of change in AFP over the past year with the current AFP. The PEB algorithm incorporates prior AFP screening values to identify patients with a significant elevated level of AFP at their current screen. We split the analysis cohort into independent training and validation datasets. All model fitting and parameter estimation was performed using the training data and the algorithm performance was assessed by applying each approach to patients in the validation dataset. Results When the screening-level false positive rate was set at 10%, the patient-level true positive rate using current AFP alone was 53.88% while the patient-level true positive rate for the six-month risk prediction model was 58.09% (4.21% increase) and PEB approach was 63.64% (9.76% increase). Both screening approaches identify a greater proportion of HCC cases earlier than using AFP alone. Conclusions The two approaches show greater potential to improve early detection of HCC compared to using the current AFP only and are worthy of further study. Electronic supplementary material The online version of this article (10.1186/s12874-017-0458-6) contains supplementary material, which is available to authorized users.


Appendix A Estimators of measures used to evaluate screening algorithms
For completeness, we include the estimators of the patient-level true positive rate (TPR), screening-level false positive rate (FPR) and positive predictive value that we use in our analysis.

Alternative parametric empirical Bayes (PEB) approaches
We explored multiple extensions of the PEB algorithm in the VA cohort to determine if incorporating additional patient information, improves the screening performance. The first modification of the PEB algorithm uses the linear predictor of a six-month risk prediction model as the biomarker (Y ij ). The risk prediction model is a simplification of the risk model in the laboratory-based algorithm and includes log 2 (AFP), log 2 (ALT), PLT, age at AFP test and two-way interactions between log 2 (AFP) and log 2 (ALT) and log 2 (AFP) and PLT.
The model is fit in the testing data using generalized estimating equations with a working correlation matrix that assumes independence and a sandwich variance estimator. This approach is referred to as the "PEB with Gastro 2014" screening algorithm in the results that follow.
The second modification to the PEB algorithm incorporates longitudinal log 2 (ALT) and PLT into the PEB algorithm through the hierarchical model assumed for Y ij = log 2 (AF P ij ) 4 in control patients through the mean structure as follows: The parametersθ, β 1 , β 2 , σ 2 and τ 2 can be estimated by fitting a linear mixed model with random intercept in the testing data. This approach is referred to as the "PEB with Adjusted AFP" in the results that follow.
The third and fourth modifications of the PEB algorithm allow both the mean and the variance components of the hierarchical model to depend on covariates. For k = 1, . . . K, we assume the following hierarchical model within each subgroup: The parameters can be estimated by fitting a linear mixed model with random intercept within each subgroup in the testing data. In the third modification of the PEB algorithm, the subgroups are defined based on demographic covariates age and race as follows: White Black Other/Unknown The four age categories are based on approximate quartiles. This approach is referred to as the "PEB with Age and Race" in the results that follow. In the fourth modification of the PEB algorithm, the subgroups are defined based on ALT and PLT levels as follows: The four ALT and PLT categories are based on approximate quartiles. This approach is referred to as the "PEB with ALT and PLT" in the results that follow.
In Table C, the screening-level FPR is fixed at 10% and in Table D, the screening-level FPR is fixed at 5%. There is some indication that the "PEB with Adjusted AFP" does increase the patient-level true positive rate when only screenings within six months (A1/A2) or one year (B1/B2) are considered true positive screens by a small amount (mostly 1-2%) compared to the original PEB algorithm. However for the other definitions, there is no indication of any difference between the modifications to the PEB algorithm and the original PEB algorithm. Table C: Comparison of the patient-level true positive rate (T P R(·, τ 1 , τ 2 )) when the threshold for each screening algorithm is chosen such that the screening-level false positive rate is 10%, i.e F P R(·, τ 1 ) = 0.1. In each definition, the choice of the parameters τ 1 and τ 2 varies.