CSA Home
Central Surgical Association

Back to 2021 Abstracts


PREDICTIVE MODELING OF IN-HOSPITAL MORTALITY FOLLOWING ELECTIVE SURGERY
H Janjua, M Rogers, A DeSantis, P Kuo
University of South Florida

Background
In-hospital mortality following elective surgery remains a challenge. Past research has focused on subcategories such as high-risk procedures, high-volume procedures, hospital and surgeon volumes, or patients with serious treatable complications. We hypothesize an accurate global elective surgical mortality model can be created.

Methods
Linked Florida AHCA and CMS Hospital Compare databases (2016-2019) were queried for in-hospital mortality following elective surgeries using ICD-10 procedure codes. A logistic regression model was employed using combination of backwards elimination and forward selection to eliminate insignificant variables with 47 variables (comorbidities, Post-op (PO) complications, other patient factors, and hospital factors (ownership and type, quality indicators etc.). This was followed by a gradient boosting machine (GBM) model comprised of 35 variables which described patients at risk for in-hospital mortality. 13 variables out of 35 contributed to 98% of model accuracy. Inverse probability of treatment weighting (IPTW) propensity matching of deceased and alive patients (1:2) using these 13 variables and principle procedures was performed followed by univariate analysis, and a multivariate logistic regression of the significant factors only.

Results
There were 511,897 admissions, 2266 in-hospital deaths (prevalence= 0.4%), 729 procedures codes (63 accounted for 50% of deaths), and 162 hospitals. The top 5 variables from GBM model (AUC 0.94, 95% CI 0.93-0.95) were PO respiratory failure, PO sepsis, length of stay 1-5 days, PO acute renal failure and PO electrolytes/fluid disorder. These factors contributed to 94% of overall model accuracy (59, 28, 4, 2.5 and 2% respectively). In the final multivariate logistic regression (AUC 0.63, 95% CI 0.61-0.66) with the propensity matched groups, age>70 (OR 2.87 95% CI: 1.36-6.67) and Hospital Compare 5 star rating (OR 0.47 95% CI: 0.29-0.71) were significant. Mortality among 5-star hospitals was 0.18% vs 0.57% for 1-star (p70 and Hospital Compare 5 stars rating predict in-hospital mortality in these high-risk patients following elective surgery.

Conclusion
Patient age cannot be altered. However, processes inherent to achieving a CMS Hospital Compare 5 star rating are associated with significantly decreased in-hospital mortality in high risk patients after elective surgery.


Back to 2021 Abstracts