A novel risk score for in-hospital perioperative mortality of five major surgeries

BACKGROUND: Risk-scoring tools for perioperative mortality adjustment are essential for inter-hospital quality comparisons, which are still lacking in China. Existing scores had significant disadvantages when applied in managerial practice. OBJECTIVE: This study aimed to develop a simple risk score using highly accessible information that could appropriately adjust the perioperative mortality from major surgeries across tertiary Chinese public hospitals and provide a reference for other underdeveloped countries with the same need. METHODS: A study cohort from 19 hospitals was randomly split into a development set and an internal validation set in the ratio of 7:3. Another cohort from six hospitals was used as an external validation set. All data were obtained from the military-hospital public services database of the National Engineering Laboratory of Application Technology in Medical Big Data. Patients aged above 18 years undergoing one of the five categories of major surgical procedures between 1 January 2010 and 31 December 2020 were identified. The multivariate logistic regression analysis was used to predict the risk of mortality and derive the risk score. The area under the receiver operating characteristic curve (AUC) and the Hosmer-Lemeshow goodness-of-fit test were used to assess the discrimination and calibration of the model, respectively. RESULTS: The study set included 45 558 cases, divided into a development set containing 31 891 cases and an internal validation set with 13 667 cases. Another cohort with 14 956 cases was used as an external validation set. The final included predictor variables were age, Elixhauser Comorbidity Index, condition at admission, admission route and the procedure. The predicted risk score ranged from -21.5 to 37.0 points. The model discriminated well in the development set, internal validation set, and external validation set. The AUC for them were 0.753 (Standard Error(SE) 0.016, 95% Confidence Interval(CI): 0.721,0.784), 0.790 (SE 0.025, 95% CI: 0.742,0.839), and 0.766(SE 0.019, 95% CI: 0.728, 0.804), respectively. P values in the Hosmer-Lemeshow goodness-of-fit test were all above 0.05, indicating a good calibration. CONCLUSIONS: This risk-scoring model was proved to have satisfactory performance, allowing the rapid and effective risk adjustment of perioperative mortality when comparing the surgical quality in tertiary hospitals in China and other underdeveloped regions.