Review Papers Against Their Own Results

 Quick Explanation Copied

 Long Explanation

Evidence and pooled estimates

The meta-analysis pooled 8 studies (2012-2016) totaling 256 HCA and 209 FNH lesions. The diagnostic signal rests on one imaging feature: hepatobiliary-phase (HBP) low signal. Pooled proportions of low signal in HCA were 2% (95% CI 0.01-0.05) in the arterial phase, 39% (95% CI 0.25-0.55) in the portal venous phase, and 89% (95% CI 0.80-0.94) in the HBP, versus 0%, 1%, and 3% in FNH. In 5 studies contributing 156 HCA and 209 FNH lesions, low HBP signal yielded pooled sensitivity 0.917 (95% CI 0.86-0.96), specificity 0.952 (95% CI 0.91-0.98), LR+ 15.028 (95% CI 7.10-31.82), LR- 0.105 (95% CI 0.07-0.17), and SROC AUC 0.9802 Phase-wise pooled low-signal proportions and diagnostic metrics.

Secondary HCA features were quantified as fat 37% (95% CI 0.27-0.49), bleeding 11% (95% CI 0.03-0.19), and central scar 10% (95% CI 0.03-0.27), each with substantial heterogeneity (I2 64.0%, 74.8%, 79.5%) Pooled secondary HCA imaging features.

Critical appraisal

Strengths: the authors transparently report heterogeneity, apply QUADAS, run trim-and-fill and leave-one-out sensitivity analyses (no single study changed pooled estimates), and explicitly warn that HBP low signal was "overvalued, especially for some HCA pathological subtypes." The mechanistic explanation linking OATP1B1/3 uptake and MRP3 efflux to HBP signal is grounded in a prior imaging-pathology study of 43 lesions where transporter expression correlated with subtype-specific enhancement Gadoxetic acid transporter-HCA subtype study.

Red flags and weaknesses:

• Publication bias confirmed: Egger test P = 0.040 for the diagnostic-accuracy outcome, with trim-and-fill imputing 2 studies Publication bias and trim-and-fill analysis. The near-perfect 1.000 sensitivity/specificity in the Mohajer study (6 HCA, 34 FNH) suggests spectrum and small-study effects inflating pooled accuracy.
• Reference-standard contamination: the authors admit some HCA and FNH lesions were confirmed only by imaging, and acknowledge the paper reported no imaging-to-pathology intervals anywhere; this creates incorporation bias that directly inflates sensitivity/specificity.
• Subtype dependence: inflammatory HCA can mimic FNH on gadoxetic acid MRI, and subtype-specific MRI accuracy is only ~85% even in dedicated studies MRI HCA subtype classification accuracy. The pooled 89% HBP low-signal rate conceals this subtype mix (e.g., only ~6 b-catenin lesions in the largest study).
• Internal data inconsistencies: the abstract states 90 studies were screened while the results section says seventy; Table 1 patient counts do not sum with HCA/FNH lesion counts (e.g., Grazioli: 82 patients but 43+68 = 111 lesions; Grieser: 68 vs 115); and the trim-and-fill result is written as "191.02% (95% CI: 73.17%-98.73%)", an impossible confidence interval where the upper bound (98.73) is below the point estimate (191.02) and the table footnote elsewhere gives the correct OR interval (73.17-498.73). These are transcription/typo-level errors that do not overturn the analysis but indicate weak editorial control.
• Scope limits: no HCC, angiomyolipoma, hemangioma, or hypervascular metastasis discrimination was quantified, so the headline specificity applies only to the HCA-vs-FNH comparison, not to real-world hypervascular lesion differentials.

Bottom line and falsifiability

Reported observations support low HBP signal as a useful HCA discriminator, but the pooled AUC of 0.98 is best read as an upper bound inflated by publication bias, imaging-based lesion confirmation, and FNH-enriched control cohorts. Author interpretation ("combining HBP signal with routine MRI and risk factors improves accuracy") is reasonable and consistent with the data; BGPT inference is that subtype-stratified accuracy (particularly I-HCA vs b-HCA) would fall well below the pooled values. The conclusions would change if: a prospective cohort with blinded pathology confirmation of every lesion reproduced substantially lower sensitivity/specificity; subtype-level analyses showed I-HCA HBP low-signal rates near FNH levels; or larger multi-center studies (post-2016) demonstrated materially different pooled estimates.

Author Review: Yongfei GuoAuthor Review: Wenjuan LiAuthor Review: Zhinan XieAuthor Review: Yi ZhangAuthor Review: Yijie FangAuthor Review: Wenli CaiAuthor Review: Guobin Hong

 BGPT Paper Review

Study Novelty

First (as of 2017) quantitative synthesis of gadoxetic acid MRI features specifically for HCA vs FNH, including pooled proportions across phases and subtype-aware caveats; conceptually incremental for a diagnostic-accuracy meta-analysis on an already-described imaging sign.

Scientific Quality

Transparent limitations and sensitivity analyses raise quality, but confirmed publication bias, imaging-based reference confirmation in some studies, subtype-mix heterogeneity, and multiple internal inconsistencies (90 vs 70 screened studies; Table 1 counts not summing; impossible trim-and-fill CI of 98.73%-191.02%) indicate weak editorial rigor. Only 8 studies, one prospective; QUADAS tool rather than QUADAS-2.

Study Generality

Narrow clinical question (HCA vs FNH only) with a rare disease and modest evidence base; generalizes poorly to other hypervascular liver lesions and does not advance broader scientific understanding beyond radiology practice.

Study Usefulness

Quantifies an underused diagnostic sign with clinically actionable pooled estimates and appropriately cautions against over-reliance, supporting combined-interpretation diagnostic strategies; directly useful for radiology reporting despite inflation risks.

Study Reproducibility

Search strategy, keywords, inclusion criteria, software (R, STATA 12.0, Meta-DiSc 1.4), and per-study data are reported, enabling reanalysis; however, no PROSPERO registration, unclear per-study raw TP/FP/TN/FN for proportion analyses, and internal count inconsistencies impair exact replication.

Explanatory Depth

Mechanistic rationale via OATP1B1/3 uptake and MRP3 efflux differences across HCA subtypes, plus explanation of fat and inflammation effects on HBP signal, exceeds typical diagnostic meta-analysis depth, though transporter claims rely on a small (43-lesion) study.