A predictive model for identifying patients at risk of delayed transfer of care: a retrospective, cross-sectional study of routinely collected data.

Background: Delays to the transfer of care from hospital to other settings represent a significant human and financial cost. This delay occurs when a patient is clinically ready to leave the inpatient setting but is unable to because other necessary care, support or accommodation is unavailable. The aim of this study was to interrogate administrative and clinical data routinely collected when a patient is admitted to hospital following attendance at the emergency department (ED), to identify factors related to delayed transfer of care (DTOC) when the patient is discharged. We then used these factors to develop a predictive model for identifying patients at risk for delayed discharge of care.

Objective: To identify risk factors related to the delayed transfer of care and develop a prediction model using routinely collected data.

Methods: This is a single centre, retrospective, cross-sectional study of patients admitted to an English National Health Service university hospital following attendance at the ED between January 2018 and December 2020. Clinical information (e.g. national early warning score (NEWS)), as well as administrative data that had significant associations with admissions that resulted in delayed transfers of care, were used to develop a predictive model using a mixed-effects logistic model. Detailed model diagnostics and statistical significance, including receiver operating characteristic analysis, were performed.

Results: Three-year (2018-20) data were used; a total of 92 444 admissions (70%) were used for model development and 39 877 (30%) admissions for model validation. Age, gender, ethnicity, NEWS, Glasgow admission prediction score, Index of Multiple Deprivation decile, arrival by ambulance and admission within the last year were found to have a statistically significant association with delayed transfers of care. The proposed eight-variable predictive model showed good discrimination with 79% sensitivity (95% confidence intervals (CIs): 79%, 81%), 69% specificity (95% CI: 68%, 69%) and 70% (95% CIs: 69%, 70%) overall accuracy of identifying patients who experienced a DTOC.

Conclusion: Several demographic, socio-economic and clinical factors were found to be significantly associated with whether a patient experiences a DTOC or not following an admission via the ED. An eight-variable model has been proposed, which is capable of identifying patients who experience delayed transfers of care with 70% accuracy. The eight-variable predictive tool calculates the probability of a patient experiencing a delayed transfer accurately at the time of admission.