Notice of Retraction: This article has been formally retracted by the journal Frontiers in Surgery. The following summary is provided for archival and informational purposes regarding the study's original claims, but the findings should not be considered reliable scientific evidence.
Abstract
The ILAG predictive model, which integrates ground-glass nodule (GGN) diameter, mean CT attenuation, and heterogeneous uniformity of density via high-resolution computed tomography (HRCT), was proposed to show great potential for the early estimation of lung adenocarcinoma (LUAD) invasiveness.
Introduction
Lung adenocarcinoma (LUAD) remains a significant cause of cancer mortality. The identification of ground-glass nodules (GGNs) on CT scans presents a clinical challenge: determining the degree of invasiveness to guide surgical strategy. While non-invasive or minimally invasive nodules may be managed with limited resection to preserve lung function, invasive nodules typically require lobectomy. Currently, invasiveness is confirmed via post-operative pathology. This study aimed to develop a pre-operative predictive model (ILAG) using HRCT features to identify invasive LUAD.
Methods
A retrospective analysis was conducted on 301 surgical LUAD patients at Shanghai Chest Hospital (May–July 2020). The cohort included 171 invasive and 130 non-invasive cases.
- Imaging: High-resolution CT scans (1.0 mm slice thickness) were analyzed for 15 GGN-related features by two independent radiologists.
- Statistical Analysis: Univariate logistic regression identified features correlated with invasiveness. Significant features were then processed through multivariate logistic regression to determine independent risk factors.
- Model Development: The "Invasion of Lung Adenocarcinoma by GGN features" (ILAG) model was constructed using the identified independent factors.
Results
The study identified three primary independent risk factors for LUAD invasion:
- Diameter: Larger nodules showed a higher likelihood of invasion (OR = 1.490).
- Mean CT Attenuation: Increased density correlated with invasiveness (OR = 1.007).
- Heterogeneous Uniformity of Density: This was the strongest predictor (OR = 3.009), indicating that non-uniform internal nodule structure significantly increases the risk of invasion.
The combined ILAG model demonstrated superior performance compared to individual features, achieving an Area Under the Curve (AUC) of 0.919 (95% CI, 0.889–0.949), which is considered excellent for diagnostic modeling.
Discussion and Limitations
While the ILAG model suggested a high predictive accuracy, several limitations were noted. The study relied on a single-center retrospective dataset, which may limit generalizability. Additionally, the assessment of morphological features involves subjective radiologist interpretation. Most critically, the subsequent retraction of this paper by Frontiers in Surgery indicates that the underlying data or methodology may have significant flaws, and these results should be interpreted with extreme caution. Future research in this area would require multi-center validation and potentially automated AI-driven feature extraction to ensure consistency and reliability.
Summary updated