Evidence Paper Review

 Quick Explanation Copied

 Long Explanation

Paper Overview

Deng et al. (2023, Frontiers in Medicine) conducted a retrospective cohort study of 436 CIN3 patients who underwent total hysterectomy within 3 months after LEEP at two Wuxi hospitals (2009–2021). The cohort comprised 178 residual and 258 non-residual cases. Using univariable and multivariable logistic regression, the authors identified post-LEEP follow-up HPV, post-LEEP follow-up TCT, and glandular involvement as independent risk factors for residual lesions. A nomogram incorporating these three factors achieved a reported C-index of 0.975 (95% CI: 0.962–0.988) with internal bootstrap validation (1,000 resamples) and demonstrated clinical utility on decision curve analysis at thresholds >0.18.

Key Findings Supported by Evidence

The multivariable analysis yielded strong associations: post-LEEP HPV-16/18 positivity had an adjusted OR of 114.692 (23.947–549.298), while post-LEEP TCT HSIL showed an adjusted OR of 335.274 (43.075–2609.637), and glandular involvement had an adjusted OR of 68.742 (12.142–389.181), all with p < 0.05. Internal LEEP margin status differed between groups (p = 0.001) but was not independently predictive after adjustment (multivariable p = 0.141), consistent with known collinearity between margin status and other predictors. The calibration curve and Hosmer-Lemeshow test supported model fit, and DCA suggested net benefit over treat-all or treat-none strategies.

Reported discrimination metrics from the paper (Figure 3B and Results 3.4). No confidence intervals were provided for the individual factor AUCs.

Critical Appraisal

Temporal and causality concerns: Post-LEEP follow-up HPV and TCT are obtained after the LEEP procedure. Incorporating post-treatment test results into a model to predict residual disease risks conflating prediction with post-hoc diagnosis. The model is more a diagnostic triage tool after LEEP than a predictive model before LEEP, yet the title and framing suggest the latter.

No external validation: The authors acknowledge lack of external validation and generalizability limitations, a well-recognized major deficiency in prediction-model studies. The C-index of 0.975 is suspiciously high and strongly suggests possible overfitting, despite bootstrap internal validation; such performance rarely persists on independent cohorts.

Narrow population: Only CIN3 patients undergoing hysterectomy within 3 months after LEEP were included — a population already selected for definitive surgical management. Applicability to the broader LEEP-treated CIN3 population managed conservatively is unknown.

Missing biomarkers: HPV-mRNA testing (e.g., E6/E7) was not measured but is acknowledged by the authors as a promising residual-disease biomarker for future study.

Statistical concerns: The extremely wide confidence intervals on some multivariable ORs (e.g., TCT HSIL upper CI >2600) suggest sparse-data bias and unstable coefficient estimation. Stepwise selection was used, which is known to produce biased parameter estimates. Multiple comparisons across candidate predictors were not formally corrected.

Confidence and Implications

The descriptive findings (residual rate ~41%) and identified associations align with prior literature on post-LEEP HPV persistence. The nomogram's utility as an external-validation-required decision-support tool remains unproven. Its clinical application should await multicenter prospective external validation before informing management decisions.

 BGPT Paper Review

Study Novelty

While nomogram-based prediction of residual CIN after LEEP is not entirely novel — several prior studies have examined post-LEEP HPV and margin status as predictors — the specific combination of post-LEEP TCT and glandular involvement into a single nomogram for CIN3 patients adds modest incremental novelty in a niche area.

Scientific Quality

Adequate retrospective methodology with reasonable sample size and standard statistical techniques. However, the extremely high C-index (0.975), absence of external validation, temporal concern with post-treatment predictors in a 'predictive' model, sparse-data bias evident in wide OR confidence intervals, and use of stepwise selection all temper enthusiasm.

Study Generality

Highly specific to a niche population (CIN3 patients undergoing hysterectomy within 3 months of LEEP at two Chinese hospitals). The narrow inclusion criteria and single-country, single-center-pair setting limit general scientific understanding beyond this specific clinical scenario.

Study Usefulness

Moderately useful for gynecologic oncologists managing CIN3 patients post-LEEP, providing a structured risk-stratification framework. However, without external validation, the tool should not yet inform clinical decisions, and the conflation of post-treatment predictors with pre-treatment risk is problematic.

Study Reproducibility

Methods are described at a reasonable level of detail (SPSS 20.0, R 3.6.1, stepwise selection, 1000 bootstrap resamples). However, no raw data, code, or protocol is publicly shared, limiting independent reproduction.

Explanatory Depth

The paper is largely descriptive/predictive without mechanistic insight into WHY these factors predict residual disease beyond brief discussion of electrosurgical coagulation and variable collinearity. No biological pathway or causal mechanism is explored.