Integrity Review Report: Construction of an Early Warning Surveillance Standard System for Lung Cancer in China Based on the Delphi Method (DOI: 10.12114/j.issn.1007-9572.2026.0066)

Academic fraud report · Geng Detector

Summary

This integrity review examines a 2026 Chinese General Practice article by Wei Zhimin et al. that uses the Delphi method to construct a lung cancer early-warning indicator system. The report rates the paper as highly suspicious, anchored primarily by an internal numerical inconsistency in Section 2.5.1 describing the first expert consultation round. The text first states the initial pool contained 73 third-level indicators and 17 second-level indicators, then claims that only 5 third-level and 3 second-level indicators were deleted, yet reports a residual pool of 28 third-level and 7 second-level indicators. The stated deletions cannot account for the reported reductions (a shortfall of 40 third-level and 7 second-level items), constituting a severe logic break. Automated statistical anomalies (Benford's law deviations, last-digit distribution) and image-forensic flags (PRNU co-sourcing, suspected copy-move) are reviewed and judged benign: the scoring uses bounded 1–10 integer ratings from 14 experts, and the PDF contains only text and tables, so template-driven PRNU similarity is expected. Confidence in the methodological finding is high; the statistical/image flags should not be treated as independent fraud evidence.

Verdict

Highly suspicious (Orange tier overall, with one critical Red finding). The principal concern is a severe internal inconsistency in the Delphi round-1 to round-2 indicator accounting. Two automated forensic signals (Benford deviation; page-level PRNU / copy-move flags) are present but, on review, are attributable to data structure and PDF rendering rather than misconduct. No fabrication of data figures, falsified figures, or plagiarism is established.

Key findings

Methodological / logical inconsistency (critical). In Section 2.5.1, the paper states the initial system comprised 4 first-level, 17 second-level, and 73 third-level indicators. It then reports deletion of 1 first-level, 3 second-level, and 5 third-level items. The resulting round-2 pool is reported as 3 first-level, 7 second-level, and 28 third-level indicators. The arithmetic 73 − 5 = 68 (not 28) and 17 − 3 = 14 (not 7) does not reconcile with the text. The shortfall is 40 third-level and 7 second-level indicators without explanation.
Benford / last-digit deviation flagged but benign. The Delphi survey uses 1–10 integer ratings from 14 experts, producing heavily bounded and clustered discrete data (e.g., constant-ratio values such as 14.29% and 21.43%). Such data violate Benford's assumptions by construction, so the alert is a false positive.
Image-forensic PRNU / copy-move flags benign. The article is a pure text-and-tables methodological paper with no experimental imagery. PDF page rasterisation produces shared template artefacts (margins, headers, footers, background), which fully explains the page-level PRNU co-sourcing and local copy-move matches. No figure duplication of scientific content is involved.

Evidence highlights

Section 2.5.1, first-round results:

73 third-level indicators

5 third-level indicators

7 second-level

28 third-level indicators

Method statement: 1–10 integer scoring scale; 14 experts across two rounds.
Automated signals (for context, not as evidence of misconduct): Benford joint test deviation (stats/benford, logLR=2.71); 5 matched offset pairs suggestive of copy-move (img-006.jpg, logLR=2.48); PRNU co-sourcing between img-000.jpg vs img-001.jpg and img-000.jpg vs img-005.jpg (NCC=1.000, PCE=128.0).
DOI: 10.12114/j.issn.1007-9572.2026.0066 (Chinese General Practice, 2026).

Notes

Counter-explanation considered: the “5 third-level indicators” may be a categorical grouping rather than a literal count. Even under this reading, the text explicitly states “delete 5 items,” which cannot yield a reduction from 73 to 28; an editorial explanation is required.
Automated Bayesian synthesis (BF ≈ 7.83×10^30, posterior ≥ 99.9%, 95% CI [100%, 100%]) is dominated by the two benign forensic signals and should not be interpreted as a fraud probability without further calibration; it is presented here only for transparency.
Recommended actions: request the authors' complete reconciliation table from round 1 to round 2; raise a PubPeer comment on the 73 − 5 ≠ 28 contradiction; recommend editorial review for corrigendum or investigation.
Limitations: this review relies on the supplied PDF and AI-assisted pattern matching; final determinations require institutional investigation and the authors' raw response data.