A Quest for Optimization of Postoperative Triage After Major Surgery

Introduction

The constant growth of health care expenditure, projected to upsurge from 17.9% of gross domestic product in 2016 to 19.7% of gross domestic product by 2026, poses the need for innovative strategies to reduce costs while improving quality of care and patient satisfaction. ¹ Intensive care units (ICUs) represent a significant expenditure portion, ² as the costs of a surgical ICU can range from $2000 to >$10,000 per day.^3,4 In addition, patients recovering in an ICU are exposed to an increased risk of overly aggressive treatments, iatrogenic injuries or nosocomial infections, pain and discomfort, deconditioning, cognitive impairment, and psychological problems such as post-traumatic stress disorder and depression. ⁵ Therefore, optimizing admission to the ICU is both economically and clinically important.

At present, due to the lack of well-established guidelines, postoperative triage decision-making is highly variable among different hospitals, leading to greatly different admission rates to ICUs after major surgery, ranging from 2.7% to 88%. The process is based on clinical judgment of patient factors (e.g., age, comorbidities, type of surgery, and intraoperative events), systemic factors (e.g., bed availability, operating room schedule, local clinical culture, and medicolegal climate), and physician factors (e.g., personality, specialty, training, type of practice, prior experience, and input from consultants)^6,7 (Fig. 1).

OPEN IN VIEWER

FIG. 1.

Conceptual model.

Current available guidelines and recommendations for ICU triage from the Society of Critical Care Medicine, from the Council of the World Federation of Societies of Intensive and Critical Care Medicine,^5,6 and from different national societies including the European Society of Intensive Care Medicine, ⁸ the American College of Critical Care Medicine, ⁶ the Italian Intensivist Group, ⁹ are used for the triage process in general but are not specific for the surgical nor for the postoperative triage of the hemodynamically stable patient. ¹⁰ Several models have also been proposed to aid in the prediction of postoperative outcomes: the Veterans Affairs Surgical Quality Improvement Program, National Surgical Quality Improvement Program, Charlson comorbidity index (CCI), revised cardiac risk index (RCRI), and surgical Apgar score (SAS) are just some among the many.

Since the risk of postoperative complications is directly related to both preoperative conditions, magnitude of surgery, and intraoperative events, our study aimed at investigating the role of commonly used surgical risk scores in identifying patients who would benefit from postoperative admission to the ICU. Of the new risk scores, we decided to include the SAS, since we previously validated it in a similar population. ¹¹

Materials and Methods

Currently, no objective criteria for appropriate ICU admission have been codified in the literature, thus we developed a list of “index events” (Table 1) that would usually grant admission to the ICU independently of the hospital system. We analyzed correlation between occurrence of at least one of those “index events,” commonly used clinical risk scores (e.g., CCI, RCRI, and SAS), objective patient data (age, blood losses, heart rate, etc.), and semiqualitative patient data (e.g., presence or absence of a certain condition). More than 200 quantitative and semiquantitative variables have been collected for each patient and a decision tree modeling (DTM) technique was used to investigate any correlation between clinical variables and the need for ICU.

Results

Using DTM techniques on preliminary data on 368 patients, we found that although SAS and the other surgical risk scores were factors highly correlating with the need of ICU, they were not the top predictors identified by the DTM. In testing the model, patients were triaged to an appropriate level of care with an accuracy of 86% compared with the 37% of our current practice. In addition, we validated the DTM against explanatory models, such as multiple logistic regression and least absolute shrinkage and selection operator, and found it to be superior in accuracy by 2.6% and 4.5%, respectively.

Discussion

DTM is a form of predictive modeling that results in a classification system that can be used to predict future observations using a set of decision rules.^12,13 The resulting tree diagrams are easy to interpret and offer a highly streamlined process for complex decision-making. Predictive modeling has several advantages over explanatory modeling (e.g., bivariate analysis and multiple logistic regression). For example, it can deal with numerous variables types, account for missing data using surrogate variables, and, most importantly, provide variable rank, which measures the importance of each predictive factor. For this reason, predictive modeling does not require an a priori hypothesis as to which variables are most important. Instead, it is able to consider every single variable of interest and then sort them by importance based on the model. Conversely, explanatory modeling requires variable selection and can only assess a restricted set of variables at once as it is limited by the large p/small n problem, where the number of predictors exceeds the number of subjects. Consequently, the selection of variables to be included in explanatory modeling would introduce a concerning source of bias. Furthermore, relevant variables can be missed if they are not selected for inclusion in the analysis. For all these reasons, we found DTM to be the ideal technique to develop an objective clinical model for postoperative triage.

There are many additional considerations to be addressed for postoperative triage optimization after major surgery. For instance, the accuracy of clinical judgment (the accepted standard of care) in determining the appropriate level of postoperative care is unknown.

As a second part of our study, we planned a prospective pilot observational study to investigate whether objective clinical parameters may outperform subjective clinical judgment in triaging patients to the most appropriate level of care. The main challenge with this type of study, however, is to identify which patients would benefit from an ICU admission. For instance, an ICU patient may receive a timelier resuscitation compared with a patient in the regular ward. Whether that timely intervention would result in the avoidance of a complication remains impossible to determine, despite prospective data collection. To partially overcome this limitation, for such pilot study we decided to focus solely on those patients for whom ICU admission was clearly clinically indicated. The downside of this approach is the loss of sensitivity in identifying who benefits from ICU admission, along with a corresponding increase in specificity.

This study has the potential to be very significant for several reasons. First of all, the study provides measurable parameters that define an objectively appropriate ICU admission. It assesses the accuracy of both measurable data and subjective clinical judgment for postoperative triage. This type of study also provides the background data for future research on the efficacy of implementing objective decision-making tools to improve postoperative triage of patients undergoing major surgery. Ultimately, this research provides substantial contribution to enhance patient care, cut hospital costs, and improve measurement of performance quality while clarifying the clinical decision-making process of postoperative patients.