Acute kidney injury (AKI) is a common and serious complication after a surgery which is associated with morbidity and mortality. The majority of existing perioperative AKI risk score prediction models are limited in their generalizability and do not fully utilize the physiological intraoperative time-series data. Thus, there is a need for intelligent, accurate, and robust systems, able to leverage information from large-scale data to predict patient's risk of developing postoperative AKI. A retrospective single-center cohort of 2,911 adult patients who underwent surgery at the University of Florida Health has been used for this study. We used machine learning and statistical analysis techniques to develop perioperative models to predict the risk of AKI (risk during the first 3 days, 7 days, and until the discharge day) before and after the surgery. In particular, we examined the improvement in risk prediction by incorporating three intraoperative physiologic time series data, i.e., mean arterial blood pressure, minimum alveolar concentration, and heart rate. For an individual patient, the preoperative model produces a probabilistic AKI risk score, which will be enriched by integrating intraoperative statistical features through a machine learning stacking approach inside a random forest classifier. We compared the performance of our model based on the area under the receiver operating characteristics curve (AUROC), accuracy and net reclassification improvement (NRI). The predictive performance of the proposed model is better than the preoperative data only model. For AKI-7day outcome: The AUC was 0.86 (accuracy was 0.78) in the proposed model, while the preoperative AUC was 0.84 (accuracy 0.76). Furthermore, with the integration of intraoperative features, we were able to classify patients who were misclassified in the preoperative model.
Acute kidney injury (AKI), previously known as acute renal failure, is one of the most common postoperative complication of many inpatient procedures (1). AKI is associated with increased risk of morbidity, mortality and with high financial costs due to prolonged hospital stay (2,3). AKI has been growing significantly at a rate of 14% per year since 2001 and the in-hospital deaths due to AKI rose by 16% in the US between 2001 and 2011 (4). Early detection of perioperative AKI risk is clinically challenging and accurate prediction of AKI has garnered a significant attention recently.
A number biomarkers -NGAL, Cys-C, KIM-1, IL-18, L-FABP, etc. (5)have been proposed for early detection of AKI based on serum, plasma or urine. However, some biomarkers (e.g., NGAL) reflects the severity of disease than being specific to the kidney injury and some of them are not significantly better than the standard clinical evaluations in early stages (6). Also, applying biomarkers to low risk patients will increase the health care cost, hence they are rarely used in everyday practice. With the advancement of technology and availability of abundant electronic health records (EHR), a number of predictive models also have been developed to estimate postoperative AKI risk in different clinical settings (7). Most of the existing AKI risk score calculators are limited to the preoperative factors (8), applicable only to a specific surgery type (9,10). Some of the available online prognostic calculators (11) are designed only for ICU patients (under surgical or medical category) without taking any surgical features into account. However, several studies have investigated the association between intraoperative data (such as the duration of MAP (Figure 1) in (12) and combination of low haemoglobin and severe hypotension in (13)) and the risk of AKI. The AKI prediction model in (14) integrates only few intraoperative variables (i.e., procedure duration, fluid balance, plasma and platelet transfusion) and specific to vascular surgeries. Therefore, the majority of existing perioperative AKI risk models do not fully utilize the available rich physiologic intraoperative data found in EHR. Thus, there is a need for intelligent, accurate, and robust systems, able to leverage information from operative period to predict postoperative AKI risk. The aim of this study was to develop a machine learning algorithm that could integrate the intraoperative features and improve the classification performance of preoperative prediction models.
The method of this study has been designed to evaluate the effectiveness and efficiency of AKI prediction before and after a major surgery with the integration of intraoperative data. The study was designed and approved by the Institutional Review Board of the University of Florida and the University of Florida Privacy Office. The statistical analysis and machine learning were performed using Python, R, and SAS software.
Using the University of Florida Integrated Data Repository, we have previously assembled a single center cohort of perioperative patients by integrating multiple existing clinical and administrative databases at UF Health (8). The billing database for UF Health, established in 1990, provides detailed information on patient demographics, outcomes, comprehensive hospital charges, hospital characteristics, insurance status and physician identity. International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes for up to fifty diagnoses and procedures are listed for each admission. We included all patients admitted to the hospital for longer than 24 hours following any type of operative procedure between January 1, 2000 and November 30, 2010. This dataset was integrated with the laboratory, pharmacy and blood bank databases and intraoperative database (Centricity Perioperative Management and Anesthesia, General Electric Healthcare, Inc.) to create a comprehensive perioperative database for this study.
We identified patients with age greater or equal to 18 years admitted to the hospital for longer than 24 hours following any type of inpatient operative procedure between January 1, 2000 and November 30, 2010. We chose only the first procedure of patients with multiple surgeries for further analysis. We excluded patients with chronic kidney disease (CKD) stage five on admission as identified by the previously validated ICD-9-CM diagnostic and procedure codes and those with missing serum creatinine, resulting in a 2,911 patient population for the analysis. We obtained institutional review board approval through the UF Gainesville Health Science Center Institutional Review Board and UF Privacy Office (#5-2009).
Main outcome of interest is postoperative acute kidney injury defined using the recent KDIGO (Kidney Disease: Improving Global Outcomes) criteria. A non end-stage renal disease patient (non-ESRD) will be diagnosed by AKI if one of the following holds: (1) If patie
This content is AI-processed based on open access ArXiv data.
