Models predicting day of surgery cancellation showed moderate discrimination with AUROC 0.75-0.92 in development and 0.70-0.74 in validation datasets.
Can statistical and machine learning models accurately predict day of surgery cancellation in elective procedures?
Current models to predict day of surgery cancellation demonstrate moderate discrimination ability but suffer from significant heterogeneity and limited calibration reporting.
Tasa de eventos absoluta: 0% vs 0%
BACKGROUND Day of surgery cancellation (DOSC) for elective surgery occurs in 18% of elective surgeries worldwide with resultant impacts on patients and healthcare systems. Accurate prediction of such cancellations could yield significant benefit. OBJECTIVE This systematic review synthesises evidence to inform future efforts at gold-standard statistical modelling. DESIGN Systematic review and qualitative synthesis using the ‘Synthesis Without Meta-Analysis’ (SWiM) framework. DATA SOURCES MEDLINE, Embase, Scopus and Web of Science, 2013–2024 ELIGIBILITY CRITERIA Studies were considered eligible if they presented the development, validation or update of a model to predict DOSC. Risk of bias for included studies was assessed using the prediction model risk of bias assessment tool (PROBAST). Data was collected on included variables, method of prediction, whether prediction was made at the level of the patient or the system, and training and assessment processes. RESULTS Literature searching identified 7154 unique studies, 6 of which were included in the final synthesis. These studies encompassed total of 759 337 elective surgical procedures, of which 47 609 were cancellations, across a variety of adult and paediatric surgery. Methods of prediction included logistic regression models, machine learning, and spatial regression models. These used demographic, socioeconomic, comorbidity, surgical, appointment and other factors as predictors. The best discrimination achieved by models in each study, quantified by area under the receiver operator curve, ranged from 0.75 to 0.92 in development and 0.70 to 0.74 in validation datasets. One study demonstrated better calibration at the census tract level of spatial regression models when incorporating local survey data compared with individual-prediction-aggregation models. CONCLUSIONS Models to predict DOSC identified in this review demonstrated moderate discrimination ability but there was significant heterogeneity between studies and reports of calibration were limited. This review serves as a valuable synthesis of current models to predict DOSC and should serve as a core reference for emerging studies in this field.
Sardesai et al. (Wed,) reported a other. Models predicting day of surgery cancellation showed moderate discrimination with AUROC 0.75-0.92 in development and 0.70-0.74 in validation datasets.