F. M. Carrano1,2, Y. Fang5, D. Wang6, S. E. Sherman4, D. V. Makarov3,7, S. Cohen2, E. Newman1,2, H. Pachter2, M. Melis1,2 1VA New York Harbor Healthcare System,Department Of Surgery,New York, NY, USA 2New York University School Of Medicine, NYU Langone Medical Center,Department Of Surgery,New York, NY, USA 3New York University School Of Medicine, NYU Langone Medical Center,Department Of Urology,New York, NY, USA 4New York University School Of Medicine,Department Of Population Health,New York, NY, USA 5New Jersey Institute of Technology,Department Of Mathematical Sciences,Newark, NJ, USA 6Northwell Health,Department Of Surgery,New York, NY, USA 7VA New York Harbor Healthcare System,Department Of Urology,New York, NY, USA
Introduction:
Currently, there is a lack of standardized evidence-based criteria to determine which patients qualify for admission to a Critical Care Unit (ICU) after major surgery. Under-triage to regular floor can result in not recognizing serious post-surgical complications, which could have been prevented and treated expeditiously in the appropriate setting, while over-triage could lead to unnecessary strains on vital healthcare resources, not mentioning the cost of such miscalculations.The goal of this study is to identify objective criteria and create algorithms that may enhance post-operative triage to the appropriate level of care following major surgery.
Methods:
We performed a retrospective analysis of patients undergoing ENT, General, Urological and Vascular major surgery between 2014 and 2015 at a major VA Medical Center. Necessary ICU admissions were identified on the basis of any of 15 objective clinical events commonly observed in the post-operative period (e.g. use of pressors, re-intubation, sustained hypotension, cardiac arrest, etc.). We used 83 clinical variables and risk scores (including Charlson Comorbidity Index, Surgical Apgar Score, Mortality Probability Model, etc.) to generate a Decision Tree Model (DTM) that would objectively establish criteria as to which patients are deemed appropriate candidates for admission to an ICU post surgery. Overall quality and accuracy of the model were measured by examining the test misclassification rate.
Results:
Our study included a total of 358 patients (96% male with mean age of 67 years). Of those, 142 met at least one of the 15 objective criteria for ICU admission. Reliance on DTM for post-operative triage would have resulted in under-triage and over-triage in 29 and 21 patients respectively, for a total mistriage rate of 13.97%. In comparison to mistriage rates based on clinical judgement alone, 63% in our own experience, the DTM has resulted in a significantly lower mistriage rate. Sensitivity and specificity of the DTM were, respectively, 79.5% and 90.2%. Positive predictive value and negative predictive value were respectively 84.3% and 87.0%. Variables with most relevance within the DTM included functional status, amount of intra-operative blood losses, intra-operative administration of blood products, presence of malignancy, as well as patient ethnicity.
Conclusion:
Use of clinical judgment alone for post-operative admission to ICU after major surgery remains highly inaccurate and is associated with inordinately excessive mistriage rates. Statistical models such as DTM has proven in our hands to outperform clinical judgment in accuracy of post-operative triage. In the near future, such models, powered by artificial intelligence platforms, might be implemented in automated algorithms to enhance post-operative decision making.