Nomogram for the Prediction of Delayed Colorectal Post-Polypectomy Bleeding

Abstract

Background: Delayed colorectal post-polypectomy bleeding (PPB) is a fairly common complication after polypectomy. The present study aimed to build a novel nomogram-based model of delayed PPB.

Methods: A cohort of 2494 patients who had undergone colonoscopic polypectomy between January 2016 and April 2020 were consecutively enrolled. The patient demographics, polyp characteristics, laboratory factors, and pathological parameters were collected. The least absolute shrinkage and selection operator (LASSO) regression was applied for selecting potential variables. Multivariate logistic regression was used to develop the nomogram. A bootstrapping method was employed for internal validation. The performance of the nomogram was evaluated on the basis of its calibration, discrimination, and clinical usefulness.

Results: Of 2494 patients undergoing colonoscopic polypectomy, 40 (1.6%) developed delayed PPB. The LASSO regression identified 6 variables (age, gender, polyp location, polyp morphology, antithrombotic medication use, and modality of polypectomy), and a predictive model was subsequently established. The area under the curve (AUC) of the predictive model and the internal validation were 0.838 (95% CI: 0.775-0.900) and 0.824 (95% CI: 0.759-0.889), respectively. The predictive model provided acceptable calibration, and a decision curve analysis (DCA) showed its clinical utility.

Conclusion: This predictive model may enable clinicians to predict the risk of delayed PPB and optimize preoperative decision-making, for effective treatment.

Conflict of interest statement

Conflict of Interest: The authors involved declared no conflict of interest.

Figures

Figure 2.

ROC curve of the established model and in the internal validation. AUC (A) shows the discrimination in the model, and AUC (B) of the internal validation. ROC, receiver operating characteristic; AUC, area under the curve.

Figure 3.

Nomogram for prediction of delayed PPB risk and its predictive performance. First, find the points for each predictor (variable) of a patient on the uppermost rule; then, add all points to calculate the “total points;” finally find the corresponding predicted probability of delayed PPB on the lowest rule. Codes annotation: gender, 0 = female, 1 = male; antithrombotic medication use, 0 = none, 1 = aspirin, 2 = clopidogrel, 3 = warfarin; location, 0 = left hemicolon, 1 = right hemicolon; morphology, 0 = sessile, 1 = pedunculated; modality, 0 = snare/forceps, 1 = EMR. PPB, post-polypectomy bleeding; EMR, endoscopic mucosal resection.

Figure 4.

Calibration curve of the model. The calibration of the model in line with the agreement between predicted and observed outcomes of delayed PPB. The Y-axis represents the actual delayed PPB rate. The X-axis represents the predicted risk of delayed PPB. The shadowed line represents a perfect prediction by an ideal model. The dotted line represents the performance of the model, with a closer fit to the shadow line representing a better prediction. PPB, post-polypectomy bleeding.

Figure 5.

Decision curve analysis for the predictive model. The net benefit was produced against the high-risk threshold. The red solid line represents the predictive model. The decision curve indicates that when the threshold probability is less than 40%, the application of this predictive model would add net benefit compared with either the treat-all or the treat-none strategies.

Figure 1.

Predictor selection using the LASSO regression analysis. (A) Tuning parameter (lambda) selection in the LASSO regression used 10-fold cross-validation. Binomial deviance was plotted versus log (lambda). The dotted vertical lines were plotted at the optimal values according to the 1-SE criteria; (B) LASSO regression coefficient profiles of variables. A coefficient profile plot was created against the log (lambda) sequence. A total of 6 non-zero coefficients were filtered and used to construct predictive model. SE, standard error.