Development and Validation of a Nomogram for Early Detection of Malignant Gallbladder Lesions

Abstract

Objectives: Preoperative decision-making for differentiating malignant from benign lesions in the gallbladder remains challenging. We aimed to create a diagnostic nomogram to identify gallbladder cancer (GBC), especially for incidental GBC (IGBC), before surgical resection.

Methods: A total of 587 consecutive patients with pathologically confirmed gallbladder lesions from a hospital were randomly assigned to a training cohort (70%) and an internal validation cohort (30%), with 287 patients from other centers as an external validation cohort. Radiological features were developed by the least absolute shrinkage and selection operator logistic regression model. Significant radiological features and independent clinical factors, identified by multivariate analyses, were used to construct a nomogram.

Results: A diagnostic nomogram was established by age, CA19.9, and 6 radiological features. The values of area under the curve in the internal and external validation cohorts were up to 0.91 and 0.89, respectively. The calibration curves for probability of GBC showed optimal agreement between nomogram prediction and actual observation. Compared with previous methods, it demonstrated superior sensitivity (91.5%) and accuracy (85.1%) in the diagnosis of GBC. The accuracy using the nomogram was significantly higher in GBC groups compared with that by radiologists in the training cohort (P < 0.001) and similarly in each cohort. Notably, most of the IGBC, which were misdiagnosed as benign lesions, were successfully identified using this nomogram.

Discussion: A novel nomogram provides a powerful tool for detecting the presence of cancer in gallbladder masses, with an increase in accuracy and sensitivity. It demonstrates an unprecedented potential for IGBC identification.

Figure 1.

Radiological features of computed tomography scan selection using the least absolute shrinkage and selection operator (LASSO) logistic regression model. (a) Identification of the optimal penalization coefficient lambda in the LASSO model with 10-fold cross-validation. (b) Optimal lambda resulted in 6 nonzero coefficients. AUC, area under the receiver operating characteristic curve.

Figure 2.

Nomogram for estimating the probabilities of gallbladder cancer. PVP, portal vein phase.

Figure 3.

Calibration curve demonstrating how predictions from the model to the actual observed probability: (a) training cohort; (b) internal validation cohort; and (c) external validation cohort.

Figure 4.

Receiver operating characteristic curve for malignancy detection sensitivity and specificity in the external validation cohort: (1) ROC curve of clinical factors alone; (2) ROC curve of the ΔCT value (Zhou et al. (23)); (3) ROC curve of the ΔCT value combined with clinical factors; (4) ROC curve of radiological features; and (5) ROC curve of radiological features combined with clinical factors (our novel diagnostic model). AUC, area under the receiver operating characteristic curve; CT, computed tomography; ROC, receiver operating characteristic.

Figure 5.

Radiologist and nomogram accuracy in the benign, GBC groups of the training, internal, and external cohorts: (a) in training cohort; (b) in internal validation cohort; and (c) in external validation cohort; total scores of the diagnostic nomogram with the cutoff value (82) in training and internal and external validation cohorts: (d) in training cohort; (e) in internal validation cohort; and (f) in external validation cohort (All P < 0.001). GBC, gallbladder cancer.