Subscribe to RSS
DOI: 10.1055/s-0038-1637225
DEVELOPMENT OF A NEW SCORE SYSTEM TO DETERMINE THE RISK OF DELAYED BLEEDING AFTER ENDOSCOPIC MUCOSAL RESECTION OF LARGE COLORECTAL LESIONS
Publication History
Publication Date:
27 March 2018 (online)
Aims:
There are two predictive models published that estimate the risk of delayed bleeding (DB) after endoscopic mucosal resection (EMR) of large colorectal lesions, the Spanish Endoscopic Resection Group predictive model (GSEED-RE) and the Australian Colonic Endoscopic Resection model (ACER), but none of them has been subjected to external validation.
The aim was to validate both and to develop a more accurate new model.
Methods:
Discrimination determined by estimating the area under the ROC curve with its 95% CI and calibration assessed with the Hosmer-Lemeshow goodness-of fit-test of GSEED-RE and ACER models, were evaluated on new data of EMR of non-pedunculated colorectal lesions ≥20 mm prospectively included in a national database (March 2015-November 2016). All models were updated (re-calibrated and re-estimated, Moons et al 2012). A new global model was built with the predictors of DB derived from the previously published scores and the new cohort.
Results:
DB occured in 45/1034 EMR (4.5%) of the new cohort. Proximal location (OR 2.84; 95% CI: 1.31 – 6.16), antiplatelet (OR 2.51; 95% CI: 0.99 – 6.34), anticoagulants (OR 4.54; 95% CI: 2.14 – 9.63), difficulty of EMR (OR 3.23; 95% CI: 1.41 – 7.40) and comorbidity (OR 2.11; 95% CI: 0.99 – 4.47) were independent predictors of DB. Performance of GSEED-RE and ACE models was poor but re-estimated and re-calibrated performance of them yielded acceptable results (AUC 0.64; 95% CI 0.54 – 0.74 and AUC 0.65; 95% CI 0.57 – 0.73, respectively, and adequate calibration). Size, proximal location, comorbidity and antiplatelet/anticoagulant treatment were used to build the new global model, providing a better performance (AUC0.71; 95% CI 0.65,0.77). The percentage of DB in the low (0 – 3 points), medium (4 – 6 points) and high (7 – 9 points) risk category was 2.2%, 4.9% and 17.1% respectively.
Conclusions:
A new simple model provides a good performance for DB prediction and it may be used to guide the management of patients after large EMRs and be helpful to design clinical trials.