Predicting outcomes of gastric endoscopic submucosal dissection using a Bayesian approach: a step for individualized risk assessment

Abstract

Background and study aims: Efficacy and adverse events probabilities influence decisions regarding the best options to manage patients with gastric superficial lesions. We aimed at developing a Bayesian model to individualize the prediction of outcomes after gastric endoscopic submucosal dissection (ESD).

Patients and methods: Data from 245 gastric ESD were collected, including patient and lesion factors. The two endpoints were curative resection and post-procedural bleeding (PPB). Logistic regression and Bayesian networks were built for each outcome; their predictive value was evaluated in-sample and validated through leave-one-out and cross-validation. Clinical decision support was enhanced by the definition of risk matrices, direct use of Bayesian inference software and by a developed online platform.

Results: ESD was curative in 85.3 % and PPB occurred in 7.7 % of patients. In univariate analysis, male sex, ASA status, carcinoma histology, polypoid or depressed morphology, and lesion size ≥ 20 mm were associated with non-curative resection, while ASA status, antithrombotics and lesion size ≥ 20 mm were associated with PPB. Naïve Bayesian models presented AUROCs of ~80 % in the derivation cohort and ≥ 74 % in cross-validation for both outcomes. Risk matrices were computed, showing that lesions with cancer at biopsies, ≥ 20 mm, proximal or in the middle third, and polypoid are more prone to non-curative resection. PPB risk was < 5 % in lesions < 20 mm in the absence of antithrombotics.

Conclusions: The derived Bayesian model presented good discriminative power in the prediction of ESD outcomes and can be used to predict individualized probabilities, improving patient information and supporting clinical and management decisions.

Fig. 1

Example of an online platform that can readily usable in clinical practice ( http://servicosforms.gim.med.up.pt/form_test/esdbayes.html ). This example shows the posterior probability of curative resection (97 %) in a ASA II patient, with a non-polypoid non-depressed < 20 mm lesion located in the lower third of the stomach, with high-grade dysplasia on pre-resection biopsies, as well as the posterior probability of PPB (2 % without antithrombotics). The predicted probability should be interpreted along with those predicted from risk matrixes, taking into account credibility intervals.

Fig. 2

Example of Bayesian inference software that can be used in clinical decision support and information. This example shows the posterior probability of curative resection (83 %) in a patient with a non-polypoid non-depressed lesion greater than 20 mm located in the middle third of the stomach, with high-grade dysplasia on pre-resection biopsies.

Fig. 3

AUROC curves of the Bayesian and logistic regression models. AUROC curves (derivation cohort; leave-one-out and cross-validation) of Naïve Bayesian models and logistic regression for prediction of curative resection and post-procedural bleeding (PPB).

Fig. 4

Risk ( posterior probabilities ) matrix for curative resection based on morphology, localization, size and pre-resection histology, using a Bayesian model.

Fig. 5

Risk ( posterior probabilities) matrix for post procedural bleeding based on morphology, localization, size, pre-resection histology and antithrombotic therapy, using a Bayesian model.