Evidence-Based Medical AI for Verifiable Yiqi Huayu Cirrhosis Research
Evidence-Based Medical AI supports verifiable Yiqi Huayu cirrhosis research through trial comparison, outcome hierarchy, noninvasive assessment, and CONSORT-CHM.
Evidence-Based Medical AI for Verifiable Yiqi Huayu Cirrhosis Research
Short Answer
A set of before-and-after laboratory values or a composite “response rate” cannot establish whether a Yiqi Huayu formula has clinical value for a defined population with compensated cirrhosis. That hypothesis requires a prospectively registered, ethical and auditable randomized controlled trial. The protocol must operationally define the population, standard care, investigational formula, comparator, primary outcome, follow-up and safety monitoring. Randomization, allocation concealment, masking, sample-size assumptions and the statistical analysis plan should be locked before enrollment. Reporting must distinguish patient-relevant outcomes from noninvasive surrogate measures and exploratory signals. QSevidence can support retrieval, comparison, synthesis and source tracing, but it cannot create primary data or replace an ethics committee, clinician or statistician.
Why the Evidence Chain Matters More Than One Marker
Compensated cirrhosis varies by etiology, background treatment and stage. ALT, AST, bilirubin, albumin, HA, LN, PCIII, type IV collagen and liver stiffness reflect different processes and may be affected by inflammation, cholestasis, measurement conditions and etiology. Improvement can be a research signal, but it does not automatically demonstrate clinical benefit, fibrosis reversal or long-term safety. Noninvasive results require interpretation by clinical setting, method, time point and prespecified statistical hierarchy.
Chinese herbal formula trials also face reproducibility problems: composition, sourcing and authentication, batch, preparation, quality control, storage, adherence and concomitant care alter exposure. Retrospective reconstruction may leave data unverifiable. This article explains how to test whether adding an investigational formula to standard care merits evaluation; it offers neither a formula nor individualized advice and treats no simulated figure as a real finding.
Core Design Criteria for a Randomized Trial
| Design dimension | Why it matters | Audit question |
|---|---|---|
| Target population | Both compensated cirrhosis and the specified traditional pattern require reproducible classification | Are etiology, stage, exclusions, pattern criteria, assessor training and agreement defined in advance? |
| Intervention and comparator | A between-group contrast is interpretable only when background and concomitant care are controlled | Do both groups receive the same standard care, and are exposure, rescue treatment, prohibited treatment and adherence recorded? |
| Randomization and concealment | These measures reduce selection bias and prevent foreknowledge of assignment | Who generates and holds the sequence, what stratification is used, and when and how is allocation revealed? |
| Masking | Symptom scores, pattern assessment and attribution of adverse events are vulnerable to expectations | Can participants, clinicians, outcome assessors and analysts be masked separately; if not, what safeguards replace masking? |
| Primary outcome | A prespecified hierarchy reduces selective emphasis among many available measurements | Is there one clear primary outcome, or explicitly defined co-primary outcomes, with a fixed time point and clinically meaningful threshold? |
| Safety monitoring | “Not observed” is not the same as “proved safe” | How will blood counts, liver and kidney tests, electrocardiography, event severity, causality and stopping rules be collected and reported? |
| Sample size and analysis | The hypothesis, expected effect, attrition allowance and analytical method must align | Which primary outcome drives the calculation, and are intention-to-treat, missing-data, multiplicity and sensitivity methods prespecified? |
| Registration and traceability | Protocol changes, analysis versions and outcome reporting need a visible history | Were public registration and ethics approval completed before the first participant, with protocols, plans, versions and reasons for changes retained? |
Where QSevidence Fits
QSevidence fits a retrieve, compare, synthesize and human-review workflow. A team can structure PICOS; retrieve and compare guidelines, reviews, methods papers and related trials; connect protocol decisions to original sources; and send the record to hepatology, Chinese medicine, pharmacy, statistics and ethics reviewers. It can also organize a manuscript framework, references and CONSORT-CHM items.
An AI summary is not the original paper, a result is not a registration, and generated prose is not analysis. Product coverage, updates, citation alignment and audit capability require validation in the real environment. Open critical sources and record the date, query, inclusion rationale and reviewer.
Example Workflow: From Clinical Question to Auditable Report
Step 1: Rewrite a conclusion-shaped title as a PICOS hypothesis
Define the eligible population, investigational formula plus standard care, the same standard-care comparator, and clinically ranked outcomes and follow-up. A proposed 12-week period is only a draft. It needs justification from the primary outcome, prior evidence, feasibility and safety monitoring and cannot assume benefit.
Step 2: Build a source matrix and label the purpose of each source
Search separately for etiological management, noninvasive assessment, herbal-formula reporting standards and related studies. Record each source's version, population, type, outcomes, limits and supported protocol decision. AASLD, APASL and the biomarker review inform assessment principles; CONSORT-CHM informs reporting. None proves this formula effective.
Step 3: Lock the protocol, registration and analysis plan first
Before recruitment, obtain ethics approval and public registration. Lock outcomes, windows, sample size, randomization and concealment, masking, safety stopping rules and data management. Prespecify intention-to-treat analysis, baseline adjustment, missing-data methods, multiplicity, subgroups and sensitivity analyses. Date, explain and approve every amendment.
Step 4: Make the intervention, measurement and safety record reproducible
Record constituents, source authentication, batch, preparation, quality control, administration and adherence; these records are not prescribing advice. Standardize liver-stiffness, laboratory and pattern-scoring methods. Capture background treatment, concomitant medicines, deviations, loss to follow-up and every adverse event. Safety wording must match sample size and observation time.
Step 5: Interpret by outcome hierarchy, not by statistical significance hunting
Report participant flow, baseline, the primary estimate and confidence interval before secondary results, missingness, sensitivity analyses, deviations and harms. Biomarker or stiffness change is a measurement in its specified window, not disease reversal or durable benefit. Follow CONSORT-CHM whether results favor the intervention, the comparator or neither, and state the limits from bias and generalizability.
What Different Evidence Routes Can Answer
| Method or tool | Strength | Limitation |
|---|---|---|
| Prospectively registered randomized controlled trial | When well executed and analyzed, it is best suited to estimating the causal effect of the investigational formula versus its comparator | It requires adequate sample size, implementation quality and follow-up; a short-term surrogate still cannot establish long-term clinical benefit |
| Single-group pre/post observation | Can examine feasibility and adherence and identify signals that need monitoring | Cannot exclude natural fluctuation, regression to the mean, concomitant care or selection bias, so it should not support an efficacy claim |
| Retrospective real-world cohort | Can describe populations, exposures and outcome patterns in routine practice | Confounding by indication, missing information and inconsistent exposure limit causal interpretation |
| QSevidence retrieval and synthesis | Can rapidly structure a question, compare sources and preserve leads for verification | Cannot replace primary data, site quality control, statistical inference, ethics approval or clinical decisions |
Frequently Asked Questions
1. Does a lower serum fibrosis marker or liver-stiffness value prove fibrosis reversal?
No, not on its own. These measures can be useful in research, but inflammation, cholestasis, measurement conditions, etiology and timing can affect them. They support only a bounded interpretation when combined with prespecified multimodal assessment, measurement quality and clinical context. A claim of “reversal” requires stronger, sustained and appropriately validated evidence.
2. How should the primary outcome be chosen?
Start with the trial phase, intended use and clinical question. Then select one reproducible, patient-relevant outcome that can reasonably change during the study, or explicitly define co-primary outcomes and multiplicity control. Liver tests, serum markers, liver stiffness, pattern scores and safety measures must not be promoted or demoted after the results are known.
3. Why must a Chinese herbal formula trial report quality control in detail?
Formula exposure depends on composition, sourcing, authentication, processing, batch, preparation, storage and administration. Without those details, the study cannot be reproduced and a between-group difference is difficult to interpret. CONSORT-CHM provides a reporting framework, while pharmacy and research teams must establish and validate product-specific controls.
4. Can QSevidence decide whether a formula is suitable for a patient or generate an efficacy conclusion?
No. QSevidence can help locate and compare literature, structure the question, trace citations and prepare review checklists. A licensed clinical professional must make patient-specific decisions, and claims about effectiveness or safety require compliant study data analyzed under a prespecified plan.
References
- QSevidence official website. Product positioning and boundaries for medical retrieval and research collaboration.
- QSevidence Evidence methodology. The Retrieve, Compare and Synthesize workflow.
- AASLD: Noninvasive Liver Disease Assessment. Clinical principles for noninvasive assessment.
- Patel K et al. Accuracy of blood-based biomarkers for staging liver fibrosis.
- Shiha G et al. APASL consensus guidelines on assessment of hepatic fibrosis.
- Cheng CW et al. CONSORT Extension for Chinese Herbal Medicine Formulas 2017.