To the Editor:

With great interest, we read the recent article by Kose et al1 assessing the prevalence and factors affecting postoperative delirium (POD) in the neurosurgical intensive care unit. Because POD has been significantly associated with adverse outcomes and high costs after neurosurgery,2 this study has potentially practical implications. Other than the limitations described by the authors in discussion, however, there are several issues in this article on which we wish to invite the authors to comment.

First, all study objects are patients undergoing neurosurgery. The sedated patients and those with antipsychotic drugs, alcohol or drug addiction, kidney or liver failure, or neuropsychiatric diseases were excluded, but preoperative neurocognitive assessments of included patients were not performed. In fact, neurocognitive disorders are common among neurosurgical patients. Tucha et al3 report that approximately 91% of patients with brain tumor demonstrate impairment in at least 1 area of cognition and approximately 71% have impairment in more than 3 areas of cognition. Other than systemic water electrolyte disorders, moreover, many patients undergoing intracranial surgery frequently have the existing disorders of cerebral physiological processes that can decrease preoperative cognitive function, such as elevated intracranial pressure, a diffuse or regional disorder of cerebral circulation, abnormal homeostasis of cerebral cells, and others.2 Given that this study is performed in patients undergoing neurosurgery and preoperative cognitive impairment is a well-established risk factor for the development of POD,2 we argue that preoperative neurocognitive assessments of included patients should be the important contents of the design.

Second, in this study, only POD occurring within the 3 days after surgery was assessed by the Intensive Care Delirium Screening Checklist. However, newest "Recommendations for Nomenclature of Cognitive Change Associated With Anesthesia and Surgery-2018" need that POD occurring within the hospital up to 1 week after surgery or until discharge should be assessed.4 Thus, we are concerned that use of a 3-day assessment time in this study would have underestimated the incidence of POD in the neurosurgical intensive care unit.

Third, it must be emphasized that the development of POD after neurosurgery is the result of complex interactions among many perioperative predisposing and precipitating factors.2 This study provided the basic features and surgery type of patients but did not provide the details of anesthetic, intraoperative, and postoperative managements. Thus, it is difficult to determine the extents to which anesthetic and perioperative managements might have influenced the occurrence of POD. It has been shown that intravenous fentanyl administration during a neurosurgical procedure can result in a dose-dependent increased risk of POD.5 Furthermore, operation time, intraoperative hypotension and blood transfusion, postoperative pain control with opioids, inadequate postoperative pain management, and presence of immobilizing events have been significantly associated with the development of POD in patients undergoing intracranial and spinal operations.6,7 Thus, we believe that findings of this study would be more conclusive, if the possible influences of anesthetic and perioperative managements on the development of POD had been included in the design.

Finally, when multivariate logistic regression analysis was used to determine the factors affecting the prevalence of POD, the details of modeling were not provided. We noted that all the variables with statistical significance (P < .05) between patients with and without POD in the initial analyses, which were shown in Table 1 of Kose et al's article, were taken into the multivariate model. This method of multivariate modeling is questionable. According to modeling principles of multivariate logistic regression analysis, the variables with statistical significance in the initial analyses should first be entered into the univariate model to examine multicollinearity among them. Then, the variables with large P values (P < .2) in the univariate analysis are included into the multivariate model using POD as the dependent outcome variable to determine the independent risk factors of POD and obtain their P values, adjusted odds ratios, and 95% confidence intervals. On the basis of the adjusted odds ratios and 95% confidence intervals of variables in the multivariate model, the readers can determine the independent contributions of significant factors.8 Because a univariate analysis, the first step of multivariate logistic regression modeling, was not performed in this study, we argue that the results of multivariate logistic regression analysis are subject to bias because of multicollinearity.

References