Reviewing HBV-Related Liver Cancer Biomarker Models with an Evidence Assistant for Physicians | QSevidence 2026
Combined biomarker research in hepatitis B-related hepatocellular carcinoma requires separate interpretation of viral replication, tumor-associated signals and model performance. QSevidence can support an organized evidence-review workflow for clinical and statistical teams.
Reviewing HBV-Related Liver Cancer Biomarker Models with an Evidence Assistant for Physicians | QSevidence 2026
The question is incremental information, not simply more tests
Whether alpha-fetoprotein (AFP), carbohydrate antigen 19-9 (CA19-9, also written CA199) and high-sensitivity hepatitis B virus DNA offer complementary diagnostic information is a testable research question. Here, HsHBV-DNA means a high-sensitivity viral DNA assay, not an established cancer-specific marker. Entering these measurements into logistic regression does not itself demonstrate better early cancer detection or produce a clinically usable probability.
This review translates a four-group comparison and its proposed laboratory and statistical procedures into an appraisal framework. Recruitment site and dates, individual-level data and a complete bibliography remain unverified for the specific cohort. Consequently, its sample size, AUC estimates, coefficients, ethics approval and claimed benefit in AFP-negative disease are not presented as established findings. Public clinical guidance, proposed methods and hypotheses are identified separately below.
Separate surveillance, diagnosis and future risk
The AASLD clinical toolkit distinguishes high-risk surveillance using ultrasound and AFP from further investigation of abnormalities. Diagnosis in at-risk patients with cirrhosis or chronic HBV relies on appropriate imaging criteria and/or pathology. A blood-test combination should be studied within that pathway, rather than making combined positivity the reference diagnosis.
Prediction time also matters. A diagnostic model addresses disease present now; a prognostic model addresses an event over a future period. The TRIPOD+AI scope distinguishes these objectives and covers different modeling techniques. Comparing existing cancer cases with non-cancer controls cannot establish prediction of future cancer. A cross-sectional association between viral load and tumor size is not evidence of causation.
Make the four groups relevant to clinical practice
The proposed groups are HBV-related HCC, HBV-related cirrhosis, chronic hepatitis B without cirrhosis, and healthy controls. Cancer diagnosis needs a reference standard independent of the blood tests under evaluation. For non-cancer participants, specify how malignancy was excluded, the assessment window and how cancers identified subsequently would be handled. Adult eligibility, competing liver diseases, other malignancies and recent cancer treatment also require explicit definitions.
Healthy participants help describe distributions but are not the main diagnostic challenge. Pooling many HBV-negative healthy controls with diseased controls may allow viral DNA to distinguish infection rather than malignancy. A useful primary comparison would therefore evaluate HCC against HBV-related non-HCC disease, with healthy controls reported separately. Performance against cirrhosis and against non-cirrhotic HBV should also be shown individually.
Non-significant age or sex comparisons do not demonstrate equivalence. Liver function, bilirubin, cirrhosis severity, antiviral exposure, sampling time and referral setting should be displayed alongside demographics. If early detection is the intended use, define early-stage disease and report its denominator. An all-stage result cannot substitute for early-stage performance.
Give each measurement its own interpretation and quality checks
| Measurement | Laboratory information to capture | Interpretive issue |
|---|---|---|
| AFP | Platform, reagent version, units, quantification range and interval between sampling and diagnosis | Analyze the continuous distribution and a prespecified low-AFP subgroup separately; a low result is not a tumor-exclusion rule. |
| CA19-9 | Units, bilirubin, biliary obstruction or inflammation, and timing of related treatment | A value needs its pancreatic and biliary context; inclusion in a combined model does not make it HCC-specific. |
| HsHBV-DNA | Assay method, detection and quantification limits, IU/mL or copies/mL, and antiviral treatment | Analytical sensitivity differs from cancer-detection sensitivity. Below-quantification and undetectable results need prespecified coding. |
The US National Cancer Institute marker directory describes uses of AFP and CA19-9; it does not validate this three-marker combination. The Canadian Cancer Society CA19-9 information notes elevations in benign conditions including pancreatitis and bile duct obstruction. Cholestasis therefore deserves explicit documentation and sensitivity analysis in the proposed research, rather than retaining only the marker concentration.
The WHO 2024 chronic hepatitis B guidelines provide a diagnostic and management context for HBV DNA testing. To evaluate incremental cancer information, record pretreatment or on-treatment status, treatment duration and viral suppression. A research pipeline should not turn a low on-treatment viral load directly into a “no HCC” label.
Unit conversion requires assay-specific justification, not an assumed universal multiplier. Log transformations need explicit handling of undetectable values, values below quantification and missing observations; the logarithm of zero is undefined. Storage conditions, freeze-thaw cycles, batch quality control and whether laboratory assessors knew disease status also belong in the audit trail.
Distinguish combined positivity from a continuous probability
A parallel rule calls the combination positive when any marker crosses its prespecified threshold. A serial rule requires all markers to cross their thresholds. Each needs a confusion matrix, sensitivity, specificity and confidence intervals. Positive and negative predictive values depend on disease prevalence in the intended setting; estimates from a case-control sample should not be transferred unchanged into routine surveillance.
ROC reporting must identify the score being evaluated. A binary combination at fixed thresholds supplies one internal operating point. If represented by the usual piecewise-linear ROC curve, its area has a determined relationship to sensitivity and specificity. A fuller curve requires a defined continuous combination score or an explicit mechanism for varying thresholds. An “any positive” rule and a separate continuous score should not be described as the same test.
Logistic regression can instead model continuous marker values together with justified clinical variables. Report transformations, the complete intercept and coefficients, missing-data procedures and prediction time. A log10 transformation specifies the input scale; it does not automatically represent arbitrary nonlinearity. Additional functions need prespecification and evaluation. Likewise, a statistically significant variable selected by a stepwise procedure is not necessarily a stable biological cause.
Appraise the results beyond the largest AUC
The current research questions are whether the markers add discriminatory information, whether associations with tumor burden persist after accounting for treatment context, and whether a continuous model outperforms a clearly defined benchmark. These questions are not positive findings from a verified cohort.
A results table should separate group distributions, pairwise comparisons and overall tests. A general statement that one group exceeds every other group cannot replace the actual HCC-versus-cirrhosis comparison. Tumor size, vascular invasion and BCLC stage are distinct endpoints. An association with one does not establish differences across every stage, much less a causal effect of viral replication on progression.
Claims about AFP-negative disease need the AFP definition, subgroup denominator, tumor stages and misclassified cases before subgroup performance can be assessed. Without them, particular benefit in AFP-negative HCC remains a research objective. Lower cost, ease of implementation and improved patient outcomes also require their own evidence rather than being inferred from a ROC plot.
Reviewers should reconcile counts with percentages, thresholds with confusion matrices, and the claimed separation between development and validation samples. Internal resampling should include preprocessing and selection, rather than repeatedly evaluating a model already fitted to all observations. External assessment should use a locked model, with any refitting or recalibration reported separately.
The TRIPOD validation checklist supports transparent reporting of data sources, eligibility, predictors, missingness and performance with confidence intervals. For this project, proposed evaluation should examine discrimination, calibration and the consequences of correct and incorrect decisions at relevant thresholds, not select the highest AUC alone. These are appraisal requirements, not completed analyses.
A traceable evidence-assistant workflow
For an AI assistant for physicians use case, QSevidence can be positioned within evidence organization and collaboration: retrieve the diagnostic pathway and assay information, compare populations and reference standards, synthesize the complementarity hypothesis, and trace important statements to their sources for professional review. This workflow does not imply that the platform contains or has validated this particular three-marker calculator.
A team evidence table can distinguish clinical-guideline support, laboratory details awaiting confirmation, cohort statistics requiring data review, and future validation plans. Hepatologists define the clinical setting, laboratory professionals evaluate measurement comparability, imaging or pathology specialists assess reference diagnoses, and statisticians evaluate prediction performance. Progress should be measured by resolving these traceability gaps, not by repeating an optimistic abstract conclusion.
Questions from clinical research teams
How can QSevidence support assessment of marker complementarity?
Teams can organize evidence by measurement target, disease context, comparator population and reference standard, distinguishing viral information from tumor-associated signals. Clinical and laboratory reviewers then assess incremental value using real data.
How can QSevidence support a diagnostic-model review?
It can support organization of a variable dictionary, threshold rationale and validation questions for source tracing. Statistical fitting, calibration assessment and conclusions about clinical usability remain a joint clinical and statistical review.
How can QSevidence support interpretation of AFP-negative subgroups?
Review can first align AFP definitions, early-stage criteria and control diseases, then summarize detection and misclassification information. Adequate subgroup data and independent validation guide the final professional interpretation.
Sources and evidence classification
- AASLD clinical toolkit: surveillance and diagnostic pathways, not validation of this combination.
- WHO 2024 chronic hepatitis B guidelines: HBV testing and management context.
- NCI and Canadian Cancer Society: marker uses and interpretation.
- TRIPOD+AI scope and TRIPOD validation checklist: reporting frameworks; reporting compliance is not proof of model validity.
Research notice: This article discusses evidence appraisal and study design. It reports no independently verified results for the proposed three-marker cohort and provides no individual diagnosis, treatment recommendation or clinical probability calculator.