Anzeige
Erschienen in:
26.08.2019 | 2019 SAGES Oral
Can we better predict readmission for dehydration following creation of a diverting loop ileostomy: development and validation of a prediction model and web-based risk calculator
Erschienen in: Surgical Endoscopy | Ausgabe 7/2020
Einloggen, um Zugang zu erhaltenAbstract
Background
Dehydration is the most common morbidity following creation of a diverting loop ileostomy (DLI). We aimed to develop and validate a prediction model and web-based risk calculator for readmission for dehydration following DLI creation.
Methods
After institutional review board approval, we retrospectively reviewed the American College of Surgeons-National Surgical Quality Improvement Program (ACS-NSQIP) database between 2012 and 2017. Adult patients (> 18 years) who underwent DLI with a resection for colorectal cancer, inflammatory bowel disease, or diverticulitis were identified. Patient demographics, operative and postoperative data were collected. The final prediction model, developed in 60% of the cohort (training set) and which modeled the 30-day cumulative incidence of readmission for dehydration, was selected using highest area under the receiver operating curve (AUC) criterion. Model calibration was assessed with the Hosmer–Lemeshow goodness-of-fit test. The model was then assessed in validation and test sets, using 20% of the cohort for each.
Results
Of 25,638 patients in the ACS-NSQIP database who met inclusion criteria, 15,222 patients were randomly selected for the training set. The incidence of readmission for dehydration in this cohort was 2.1%. The final model with the highest AUC retained 12 candidate variables: age, sex, smoking status, diabetes, hypertension, American Society of Anesthesiologists score, type of admission, underlying diagnosis, procedure performed, operative time, index admission length of stay, and major morbidity. The model demonstrated good discrimination (AUC 0.76, 95% CI 0.74–0.79) and the Hosmer–Lemeshow goodness-of-fit test confirmed good calibration (p = 0.50). Five-thousand and seventy-three patients were available for the validation and test sets, respectively, and the model remained strong in both the validation and test sets (AUCs of 0.73 and 0.73, respectively). The prediction model was then converted into a web-based risk calculator.
Conclusions
A prediction model and web-based risk calculator for readmission for dehydration after DLI creation was developed and validated, demonstrating good predictive capabilities.