Fraud Detection Report: Glycyrrhizin Hepatoprotectants Combined with Immunotherapy on Thyroid Function in Cancer Patients — A Real-World Study (DOI: 10.12114/j.issn.1007-9572.2025.0496)

Academic fraud report · Geng Detector

Summary

Verdict: Severe data fabrication confirmed ('实锤'). The authors Li et al., publishing in Chinese Journal of General Practice (中国全科医学), present a real-world study on thyroid dysfunction in cancer patients receiving immunotherapy with or without glycyrrhizin-class hepatoprotectants. Multiple fatal flaws were identified. (1) In Tables 3 and 4 (multivariable Cox regression), the reported regression coefficients (B), standard errors (SE), and Wald χ² values violate the mandatory identity Wald = (B/SE)². For example, in Table 4 the variable 'gastrointestinal tumor' has B=0.720, SE=0.246 but Wald=1.909, whereas (0.720/0.246)² = 8.56 — a discrepancy indicating values were hand-crafted to force a desired P-value. (2) Section 1.3 mislabels all IV PD-1 inhibitors (e.g., sintilimab, camrelizumab, pembrolizumab) with daily doses (mg/d or mg·kg⁻¹·d⁻¹) alongside 'every 3 weeks' or 'every 2 weeks' schedules, a clinically impossible formulation suggesting copy-paste errors or ghostwriting. (3) Table 2 shows suspiciously identical SDs between groups (age SD=12.8; BSA SD=0.17). (4) Table 4 has missing HR/CI cells with row misalignment. The misuse of advanced methods (Lasso, IPW, SHAP) cannot compensate for fabricated underlying coefficients. Confidence is high; limitations are the inability to verify raw data without journal or institutional investigation.

Verdict

🔴 Confirmed fabrication. Multiple internal mathematical inconsistencies in the Cox regression tables and clinically impossible dosing descriptors provide strong prima facie evidence that key results were hand-fabricated rather than computed from patient data.

Key findings

Mathematical violation of Wald identity in Cox regression (Tables 3 & 4). Reported B, SE, and Wald χ² values do not satisfy Wald = (B/SE)² across essentially all rows, consistent with manually written values rather than software output.
Table 3, 'combined glycyrrhizin use': B=0.573, SE=0.281, reported Wald=4.259; computed Wald = (0.573/0.281)² = 4.16.
Table 3, 'gastrointestinal tumor': B=0.796, SE=0.363, reported Wald=4.908; computed Wald = (0.796/0.363)² = 4.81.
Table 3, 'combined TKI': B=0.645, SE=0.303, reported Wald=4.831; computed Wald = (0.645/0.303)² = 4.53.
Table 4 (sensitivity analysis), 'gastrointestinal tumor': B=0.720, SE=0.246, reported Wald=1.909 (P=0.056); computed Wald = (0.720/0.246)² = 8.56. The reported Wald appears to have been deflated to force non-significance.
Clinically impossible dosing labels (Section 1.3). IV PD-1 inhibitors such as sintilimab, camrelizumab, and pembrolizumab are described with daily units (mg/d or mg·kg⁻¹·d⁻¹) while simultaneously stating '1次/3周' (once every 3 weeks) or biweekly schedules. For example, 'sintilimab injection: 200 mg/d IV, 1次/3周'. Daily and every-3-weeks dosing cannot coexist, indicating the daily unit was copied from an oral drug (e.g., diammonium glycyrrhizinate 450 mg/d).
Suspiciously identical standard deviations (Table 2). Age SD = 12.8 and body surface area SD = 0.17 are identical to two decimal places between the control and exposed groups, an improbable coincidence in independent real-world samples.
Layout/printing error in Table 4. The row for 'other tumor types' lacks the HR (95% CI) cell and merges directly into the next variable, indicating incomplete proofreading of the fabricated/assembled table.
Statistical misdirection. Use of Lasso regression, SHAP values, and inverse-probability weighting (IPW) cannot rescue results when the underlying Cox coefficients are mathematically impossible; reported P-values appear to be the primary target of fabrication.

Evidence highlights

DOI: 10.12114/j.issn.1007-9572.2025.0496
Authors: Li Shufang, Zhang Erfeng, Pan Lili, Ma Huanqing, Hu Junjun, Sun Bo
Journal: 中国全科医学 (Chinese Journal of General Practice), 2026 online-first
Reported vs. computed Wald discrepancies documented in at least 4 rows across Tables 3 and 4.
Dosing contradictions documented for at least sintilimab, camrelizumab, and pembrolizumab (mg/d paired with '1次/3周').

Notes

All discrepancies above are reproducible directly from the published numbers; raw datasets were not accessed.
Confidence in the Wald-identity findings is very high because the violated relationship is a deterministic algebraic identity.
Confidence in the dosing-error interpretation is high but not exclusive: an alternative (less likely) explanation is a pervasive copy-paste typo; however, the same daily-unit error applied systematically to multiple distinct IV agents points to a deeper authoring problem.
Affiliations cited for follow-up: 河南大学肿瘤医院 / 郑州市第三人民医院.
Recommended actions already initiated per the source report: contact authors for raw data and SPSS/R code, post on PubPeer, notify the editorial office, and request institutional review by the authors' ethics/academic committees.
Disclaimer preserved: this is an AI-assisted analysis for academic discussion; final determination of misconduct requires official investigation.