DEVELOPMENT AND VALIDATION OF A RISK CALCULATOR FOR MORTALITY AMONG GENERAL SURGERY PATIENTS AT THE TIME OF INTER-HOSPITAL TRANSFER
Corey K. Gentle*, Sayf Al-deen Said, Kelly Nimylowycz, Mir Shanaz Hossain, Miguel D. Regueiro, Toms Augustin
Cleveland Clinic, Cleveland, OH
Background:
Inter-hospital transfers mark a critical decision point in the patient care continuum. Despite evidence that patients transferred between healthcare systems are more complex and experience greater morbidity and mortality, there are no available tools to correctly triage surgical patients based on their disease acuity. We hypothesized that readily available, low complexity patient parameters at the time of transfer could predict mortality after transfer.
Methods:
All patients transferred into general and colorectal surgery services at a quaternary care hospital between January-2016 and August-2022 were included. Demographics, laboratory values, vital signs, intensive care unit (ICU) admission, and vasopressor use were extracted from the medical record. Variables were chosen for easy availability, decreased variability, and feasible collection by trained non-physician transfer center personnel with nursing input from the transferring center. The primary outcome was admission-related mortality, defined as death during the admission or within 30 days post-discharge. Univariate differences were tested between the outcome groups. Logistic regression, penalized regression, gradient boosting regression, and deep neural network predictive models were trained on the training dataset and their performance was compared on the validation dataset.
Results:
A total of 4,664 adult transfers were included. Admission-related mortalities were 280 (6.0%): 142 (50.7%) occurred during the admission and 138 (49.3%) occurred in the 30-days after discharge. On univariate analysis, differences in all components of complete blood count, basic metabolic panel, and vital signs were statistically different between the two outcome groups. In addition, compared to survivors, patients who suffered mortality were more likely to be transferred from or into an ICU [153 (54.6%) vs. 758 (17.3%), P<0.001] and more likely to require vasopressor support [87 (31.1%) vs. 202 (4.6%), P<0.001]. While all variables were tested, only 12 commonly collected variables were included in the final model based on the penalized regression method. The model coefficients are numerated in Table 1. When validated, the final model achieved an area under the receiver operating characteristic curve, sensitivity, specificity, and accuracy of 0.846, 0.79, 0.72, and 0.72, respectively. After bias-correction, the Hosmer-Lemeshow C-statistic for the model was 8.22, P=0.412 indicating strong prediction and calibration.
Conclusion:
In an inter-hospital transfer setting, it is possible to predict the risk of mortality for general surgical patients based on readily available clinical parameters. Utilizing such a risk score could assist accepting hospitals in triaging patients to prioritize transfer acuity, improve resource allocation, and standardize care.
Table 1. Clinical patient variables predictive of mortality after inter-hospital transfer with corresponding coefficients at minimum Lambda and odds ratios.
Back to 2023 Abstracts