Clinico-radiological nomogram for preoperatively predicting post-resection hepatic metastasis in patients with gastric adenocarcinoma

Abstract

Objective: To establish and validate a model comprising clinical and radiological features to pre-operatively predict post-resection hepatic metastasis (HM) in patients with gastric adenocarcinoma (GAC).

Methods: We retrospectively analyzed 461 patients (HM, 106 patients); and non-metastasis (NM, 355 patients) who were confirmed to have GAC post-surgery. The patients were randomly divided into the training (n = 307) and testing (n = 154) cohorts in a 2:1 ratio. The main clinical risk factors were filtered using the least absolute shrinkage and selection operator algorithm according to their diagnostic value. The selected factors were then used to establish a clinical-radiological model using stepwise logistic regression. The Akaike's information criterion and receiver operating characteristic (ROC) analyses were used to evaluate the prediction performance of the model.

Results: Logistic regression analysis showed that the peak enhancement phase, tumor location, alpha-fetoprotein, cancer antigen (CA)-125, CA724 levels, CT-based Tstage and arterial phase CT values were important independent predictors. Based on these predictors, the areas under the ROC curve of the training and testing cohorts were 0.864 and 0.832, respectively, for predicting post-operative HM.

Conclusion: This study built a synthetical nomogram using the pre-operative clinical and radiological features of patients to predict the likelihood of HM occurring after GAC surgery. It may help guide pre-operative clinical decision-making and benefit patients with GAC in the future.

Advances in knowledge: 1. The combination of clinical risk factors and CT imaging features provided useful information for predicting HM in GAC.2. A clinicoradiological nomogram is a tool for the pre-operative prediction of HM in patients with GAC.

Figure 1.

Flowchart of the study enrollment. GAC, gastric adenocarcinoma; HM, hepatic metastasis; NM, non-metastasis.

Figure 2.

A 53-year-old female with hepatic metastasis after surgery for gastric adenocarcinoma. Multiple hepatic metastases were found on enhanced CT 91 days post-operation for gastric antrum cancer. a, b, and c are pre-operative CT images of gastric cancer in the arterial, venous, and delayed phases, respectively; the drawn areas are ROI diagrams to measure CT values for gastric cancer. d, e, and f are CT images of the arterial, venous, and delayed phases, respectively; hepatic metastasis was first identified after the operation, clearly demonstrating multiple hepatic metastatic tumors. ROI, region of interest.

Figure 3.

(a) LASSO coefficient profiles of the 23 hepatic metastasis-related features. A coefficient profile plot was produced against the log (λ) sequence. (b) Feature selection using the LASSO binary logistic regression model. The differentiation performance of the radiologic features was explored on the ROC curve. Tuning parameter (λ) selection in the LASSO model employed fivefold cross-validation via the maximum AUC. A λ value of 0.016 was selected (1-SE criteria) according to fivefold cross-validation. AUC, area under the curve; LASSO, least absolute shrinkage and selection operator; ROC, receiver operating characteristic.

Figure 4.

The developed clinical and radiologic features nomogram for predicting the probability of HM. For the PEP, 1 stands for AP, 2 stands for the VP, and 3 stands for DP. For the location, 1 stands for upper, 2 stands for middle, and 3 stands for lower. For the AFP, CA125, and CA724 levels, 0 represents within the normal value while 1 represents above the normal value. For the cT stage, 1 is for T1 stage, 2 is for T2 stage, 3 is for T3 stage, and 4 is for T4 stage. To locate the patient’s PEP, draw a line straight up to the x-axis to establish the score associated with that site. Repeat the same for the other covariates (tumor location, pT stage, AFP level, CA125 level, CA724 level). By summing the scores of each point and locating on the total score scale, the estimated probability of HM could be determined. AFP, alpha-fetoprotein; AP, arterial phase; Arterial.CTV, arterial CT values; DP, delayed phase; HM, hepatic metastasis; PEP, peak enhancement phase; VP, venous phase.

Figure 5.

For the decision curve, the y-axis represents the net benefit. This was calculated by expected benefit (gaining true positives) and subtracting expected harm (deleting false positives). The highest curve at any given threshold probability is the optimal prediction to maximize net benefit.