The Harvard community has made this article openly available. Please share how this access benefits you. Your story matters

Transcription

1 A Contemporary, Population-Based Analysis of the Incidence, Cost, Outcomes, and Preoperative Risk Prediction of Postoperative Delirium Following Major Urologic Cancer Surgeries The Harvard community has made this article openly available. Please share how this access benefits you. Your story matters Citation Ha, Albert Sangji A Contemporary, Population-Based Analysis of the Incidence, Cost, Outcomes, and Preoperative Risk Prediction of Postoperative Delirium Following Major Urologic Cancer Surgeries. Doctoral dissertation, Harvard Medical School. Citable link Terms of Use This article was downloaded from Harvard University s DASH repository, and is made available under the terms and conditions applicable to Other Posted Material, as set forth at nrs.harvard.edu/urn-3:hul.instrepos:dash.current.terms-ofuse#laa

2 Abstract Introduction Postoperative delirium is associated with poor outcomes and increased healthcare costs across numerous surgical and medical disciplines. Although characterized in other surgical fields, the population-based incidence, outcomes, and cost of delirium have not been assessed in major urologic cancer surgeries. We sought to evaluate the incidence, outcomes, and cost of postoperative delirium after major urologic cancer surgeries, specifically after radical prostatectomy (RP), radical nephrectomy (RN), partial nephrectomy (PN), and radical cystectomy (RC) in the USA. We have also developed a preoperative risk prediction model specific to major urologic cancer surgeries to identify patients at high risk for postoperative delirium. Methods Using the Premier Hospital Database, we retrospectively identified patients who underwent radical prostatectomy (RP), radical nephrectomy (RN), partial nephrectomy (PN), and radical cystectomy (RC) from 2003 to Postoperative delirium was identified using ICD-9 codes, as well as postoperative use of antipsychotics, sitters, and restraints. We constructed regression models to assess for mortality, discharge disposition, length of stay (LOS), and direct hospital costs. A preoperative risk stratification scoring system was also developed using known risk factors of delirium. The entire cohort was randomly divided into training (70%) and validation (30%) cohorts. Preoperative patient, hospital, and surgical characteristics associated with delirium were analyzed using multivariate regression, and a risk prediction score was developed 2

3 using the training cohort. Its performance was quantified using Receiver Operating Characteristic (ROC) analysis in both cohorts. Results We identified 165,387 patients representing a weighted total of 1,097,355 patients. 30,063 (2.7%) experienced postoperative delirium. The greatest incidence of delirium occurred after RC, with 6,268 cases (11%). Delirious patients had greater adjusted odds of in-hospital mortality (OR 3.65; p <0.001), 90-day mortality (OR 1.47; p = 0.013), discharge with home health services (OR 2.25; p <0.001), discharge to skilled nursing facilities (OR 4.64; p <0.001), and 0.9-day increase in median LOS (p <0.001). Delirious patients also experienced a $2,697 increase in direct admission costs (p <0.001), with the greatest costs in RC patients ($30,859 vs. $26,607; p<0.001). The largest driver of costs was in room and board across all surgeries (p<0.001). Our training and validation cohorts consisted of a weighted total of 767,408 and 329,926 patients, respectively. Our final model revealed many factors that increase risk for delirium, all which were used to create a preoperative risk score. The additive score was predictive of delirium in both the training (OR: 1.35, 95% CI, , p<0.001) and validation cohorts (OR: 1.34, 95% CI , p<0.001). The score also demonstrated good discrimination in predicting delirium in the training (AUC: 0.74, 95% CI, ) and validation (AUC: 0.75, 95% CI, ) cohorts. Conclusion Patients with postoperative delirium experienced worse outcomes, prolonged LOS, and increased admission costs following major urologic cancer surgeries. In particular, the largest incidence and costs occurred in delirious patients after RC. Moreover, the results of the pre- 3

4 operative risk prediction tool for delirium following major urologic cancer surgeries are promising given their consistency with published delirium risk factors and ease of use. Further testing will shed light on its clinical utility. Keywords postoperative delirium, risk prediction, cystectomy, prostatectomy, nephrectomy, nephrectomy, healthcare costs, outcomes 4

5 TABLE OF CONTENTS Section Page Abstract Glossary and Abbreviations Introduction Methods Results Discussion Conclusion, Summary, and Future Directions Tables and Figures References Appendix

6 Glossary of Abbreviations GU Genitourinary RP Radical Prostatectomy RN Radical Nephrectomy PN Partial Nephrectomy RC Radical Cystectomy ICU Intensive Care Unit ICD-9 International Classification of Disease, Ninth Revision CCI Charlson Comorbidity Index LOS Length of Stay ROC Receiver Operating Characteristic AUC Area Under the Curve BMI Body Mass Index CI Confidence Interval OR Odds Ratio IQR Interquartile Range IRR Incidence Rate Ratio USD United States Dollars 6

7 Introduction Delirium is defined as an acute decline in cognitive function and is associated with cognitive deficits in memory, perception, and altered thoughts or fluctuations in consciousness (1,2). It is a common complication among elderly hospitalized patients and has been associated with increased adverse outcomes across medical and surgical disciplines. For example, past studies have identified an increase in-hospital and one-year mortality in patients with delirium. In addition, delirium has been associated with postoperative complications and poor postoperative recovery resulting in increased length of stay and discharge to rehabilitation (3). Recent evidence also suggests that the cognitive impact of delirium may include long lasting mental health conditions such as cognitive impairment (4) and post-traumatic stress disorder (5). As a result, it is estimated that the management of delirium contributes $38 billion to $152 billion per year to healthcare spending in the US (6). Although delirium has been extensively studied in other surgical fields, research on the incidence and outcomes of postoperative delirium in urologic cancer surgeries remains sparse. Several studies have estimated the incidence of postoperative delirium to be between 6.0% and 9.0% following radical prostatectomy (7), and between 1.8% to 29% following radical cystectomy (8,9). However, such studies have been limited by small sample sizes and restricted to single institutions. Similar research into patient and hospital outcomes of postoperative delirium in urologic cancer surgeries have been limited by small sample sizes (8). To date, population-based approaches to analyze postoperative delirium in urologic cancer surgeries have not been performed. 7

8 In regards to risk prediction rules for delirium, several tools have previously been described in the perioperative setting. Marcantonio et al. was first to construct a clinical risk prediction rule to identify postoperative delirium in non-cardiac surgeries; however, urologic surgeries had been under-represented in the sample (10). Recently, Rudolph et al. developed a generalized risk stratification model for medical and surgical delirium at a tertiary Veterans Administration hospital; however, the study also under-represented urologic surgeries in the sample (11). Other delirium risk prediction tools for delirium have also been constructed exclusively for specific surgical specialties such as cardiac, vascular and orthopedic surgeries (12-15). While such models seem promising, most rely on resource-intense cognitive tests that may limit their clinical utility, especially in a busy clinical practice (10,14). Currently, no convenient preoperative risk prediction model specific to urologic cancer surgeries has been released. The purpose of this investigation is to perform a contemporary population-based analysis to estimate the incidence, cost, and outcomes associated with postoperative delirium after major urologic cancer surgeries, as well as construct a preoperative risk prediction rule to preemptively identify patients at high-risk for this condition. Specifically, we will first characterize the incidence of delirium after radical prostatectomy (RP), radical nephrectomy (RN), partial nephrectomy (PN), and radical cystectomy (RC) in the entire cohort. We will then delineate the association between postoperative delirium and negative outcomes that affect patients, as well as demonstrate the economic impact of this complication. In addition, to facilitate the construction of the preoperative risk prediction model, we will first randomly divide the cohort into both training and validation cohorts. We will then construct a preoperative risk prediction model for 8

9 postoperative delirium in the training cohort and finally assess the performance of this risk model in the validation cohort. 9

10 Methods Data Source The Premier Hospital Database (Premier Inc., Charlotte, NC, USA) is an all payer hospital discharge database that contains more than 45 million inpatient discharges from over 700 hospitals across the USA, amounting to approximately 20% of total annual discharges. Longitudinal analysis was possible because of unique patient identifiers. This study received exemption from our hospital Institutional Review Board. Overall Study Cohort and Covariates Using International Classification of Disease, ninth revision (ICD-9) diagnosis and procedure codes, we identified all adult patients who underwent RP, RN, PN, and RC from 2003 to 2013 (Table 8). Patient, hospital, and surgical characteristics associated with outcomes and costs were captured. Patient characteristics included age (<60, 60-69, 70-79, and 80 years), gender, race (White, Black, Hispanic, and other/unknown), marital status (unmarried and married), insurance status (Medicare, Medicaid, private, and other), and obesity. Hospital characteristics included teaching status, hospital size (<400, , and >600 beds), location (urban and rural) and geographical region (Northeast, Midwest, West, and South). Surgical characteristics included type of surgery (RP, RN, PN, and RC), operative time (0-4, 4-6, 6-8, and 8+ hours), and surgical approach (open vs. laparoscopic and robotic). 10

11 Identification of Postoperative Delirium Retrospectively identifying postoperative delirium using ICD-9 codes has been previously reported to be specific but not sensitive (16). To augment our capture rate, we utilized a composite of surrogate variables that identify unique elements of delirium management identified in a patient s hospital bill. These include ICD-9 diagnosis codes (Table 9), as well as postoperative use of sitters, restraints, and antipsychotics (2,17-21). To reduce misclassification bias, we excluded the use of restraints on the day of surgery or in the Intensive Care Unit (ICU). Antipsychotic medications such as haloperidol, risperidone, olanzapine, and quetiapine have previously been commonly used in the treatment of delirium (21), and their postoperative use was subsequently identified to capture management of postoperative delirium. Duration of delirium was calculated as the number of consecutive post-operative days in which at least one of the antipsychotics, sitters, or restraints was used. Outcomes and Cost Outcomes include direct hospital admission costs, in-hospital mortality, 90-day mortality, hospital length of stay (LOS), and discharge disposition such as home with healthcare, skilled nursing facility, and other (e.g. short-term general hospital, or hospice). In-hospital mortality was defined as Clavien 5 (postoperative mortality) of the Clavien classification system, which is recommended by the International Consultation on Urological Disease European Association of Urology (22). The Premier Hospital Database provides direct hospitalization costs that are distinct from hospital charges and do not need to be converted from hospital charges (23). Direct hospital costs were further subdivided into categories (e.g. room and board, radiology, or operative room costs) 11

12 to compare resource utilization among different types of urologic cancer surgeries. To facilitate comparison, all costs were adjusted to 2013 US dollars using the medical component of the Consumer Price Index. Construction of Training and Validation Cohorts and Identification of Preoperative Risk Factors The aggregate cohort was randomly divided into training (70%) and validation (30%) cohorts. Preoperative patient, hospital, and surgical characteristics were analyzed in both cohorts. Patient characteristics included age (<60, 60-69, 70-79, and 80 years), gender, race (White, Black, Hispanic, and other/unknown), marital status (unmarried and married), insurance status (Medicare, Medicaid, private, and other), obesity, and comorbid conditions. Hospital characteristics included teaching status, hospital size (<400, , or >600 beds), location (urban and rural) and geographical region (Northeast, Midwest, West, or South). Surgical characteristics included surgery type (RP, RN, PN, or RC), operative time (0-4, 4-6, 6-8, and 8+ hours), and surgical approach (open vs. laparoscopic/robotic). Predisposing risk factors of delirium such as history of sleep disturbances, hearing impairment, vision impairment, dementia, depression and/or anxiety, alcohol abuse, and preoperative immobility have been well-established in the literature (3,24,25). Preoperative patient, hospital, and surgical characteristics were analyzed in both training and validation cohorts (Table 2). Statistical Analysis Patient, hospital, and surgical characteristics for patients were summarized using descriptive statistics. Categorical variables were compared using chi-squared tests while 12

13 continuous variables were assessed via Students t-test. Binary and multivariable logistic regression was used to analyze 90-day mortality and discharge disposition, respectively. Hospital direct admission costs were analyzed using a multivariable generalized linear model conforming to a gamma distribution to account for skewed distribution of costs. Hospital LOS was modeled using negative binomial regression. Preoperative delirium risk factors from relevant patient and hospital characteristics; comorbidities; and surgical characteristics were analyzed using univariate and multivariate logistic regression for the training cohort. The results of the multivariable model from the training cohort were used to construct a risk prediction score based on the non-transformed beta coefficients. The risk score as a continuous variable was further stratified into categorical risk groups (low, intermediate, high, and very high risk) roughly based on the corresponding mean and standard deviation. Performance of the risk score was quantified and validated using Receiver Operating Characteristic (ROC) analysis in both the training and validation cohorts. All models were survey-weighted, clustered by hospitals to achieve nationally representative estimates, and adjusted for all relevant patient, hospital, and surgical characteristics. Statistical analysis was performed on Stata version 14.1 (College Station, TX). All tests were two-sided, and a p-value <0.05 was considered statistically significant. 13

14 Results Overall Characteristics of Hospitals and Patients Baseline patient, hospital, and surgical characteristics of patients with and without postoperative delirium are outlined in Table 1. The final cohort was made up of 165,387 patients, representing a weighted population of 1,097,335 patients in the United States from 2003 to Following stratification by type of surgery, a weighted total of 630,353 patients undergoing RP, 305,503 patients for RN, 104,217 patients for PN, and 241,263 for RC were identified. Statistically significant, preoperative characteristics were identified in patients with postoperative delirium. The mean age of patients experiencing delirium was 68.3 years vs years in those without delirium (p <0.001). A greater percentage of delirious patients were male (66.5% vs. 60.7%; p <0.001), white (75.4% vs. 71.2%; p = 0.001), and unmarried (46% vs. 33%; p <0.001). Such patients tended to have a greater number of comorbidities, reflected by a higher proportion of patients with Charlson Comorbidity Index (CCI) score greater than or equal to 4 (36% vs. 11%; p <0.001). These patients were also more likely to utilize either Medicare (69.7% vs. 39.5%) or Medicaid (5.6% vs. 2.4%; p <0.001). Patients with delirium experienced longer operative times of either 6 to 8 hours (12.1% vs. 6.5%) or greater than 8 hours (13.7% vs. 73%; p <0.001), and underwent a greater percentage of open surgical approaches vs. laparoscopic or robotic approaches (64.2% vs. 51.1%; p <0.001). In regards to pre-disposing risk factors, patients with postoperative delirium had increased sleep difficulties (p<0.001), hearing loss (p<0.001), vision loss (p = 0.003), dementia (p<0.001), depression/anxiety (p<0.001), alcohol abuse (p<0.001), and baseline immobility (p<0.001). Higher rates of comorbidities that include congestive heart failure, chronic pulmonary disease, cerebrovascular disease, diabetes, liver 14

15 disease, metastatic solid tumors, peripheral vascular disease, and renal disease (p <0.001) were also found. No statistically significant differences were identified among hospital size, location, region, or teaching hospital status in patients with and without delirium. Incidence, Onset, and Duration of Delirium A weighted total of 30,063 patients (2.7%) experienced postoperative delirium (Table 3). When stratified by type of surgery, 5,986/630,353 (1.0%; 95% CI, 0.8%-1.1%) of RP patients experienced postoperative delirium while 14,431/305,503 (4.7%; 95% CI, 4.4%-5.1%) and 3,377/103,217 (3.2%; 95% CI, 2.8%-3.8%) patients experienced this after RN and PN, respectively. The highest incidence of delirium was identified to be 6,268/57,262 (11.0%; 95% CI, 10.0%-11.9%) in the in the RC cohort. The median duration of delirium was similar across all types of surgeries at 2 days, albeit with different interquartile ranges of 1 to 3 days (RP), 1 to 5 days (RN), 1 to 4 days (PN), and 1 to 5 days (RC). Cost and Outcomes Patients with delirium also experienced a $2,697 (95% CI $2,250-$3,3144; p <0.001) increase in direct hospital admission costs compared with those without delirium. When subdivided by types of hospital costs, patients with delirium incurred a higher total cost of inpatient lab, pharmacy, radiology, room and board, operating room, supply, and other expenses across all urologic cancer surgeries (p <0.001) (Figure 1). In particular, the greatest difference in direct hospital admission costs is seen in delirious patients after RC ($30,859 vs. $26,607; p<0.001). 15

16 Patients with postoperative delirium had a greater risk of adverse outcomes and increased hospital costs (Table 4). After adjusting for relevant patient, surgical, and perioperative characteristics, patients with delirium were more likely to experience both in-hospital mortality (OR, 3.71; 95% CI, ; p <0.001) and 90-day mortality (OR, 1.48; 95% CI, ; p = 0.012) than those without delirium. Moreover, delirious patients were more likely to be discharged to higher level care facilities, which include disposition to home with additional health care (OR, 2.39; 95% CI, ; p <0.001) or disposition to skilled nursing facilities (OR, 4.98; 95% CI, ; p <0.001). Length of stay also varied significantly based on the presence of delirium. On adjusted analysis, patients with delirium experienced a 0.90-day increase in hospital stay (95% CI, ; p <0.001). Training and Validation Cohorts Our aggregate cohort was randomly subdivided into training and validation cohorts of 115,770 and 49,617 patients from 490 hospitals, respectively. These values represented a weighted population of 767,408 patients in the training cohort and 329,926 patients in the validation cohort. Following stratification by surgery type, the training cohort consisted of 440,684 patients undergoing RP; 214,099 patients for RN; 72,354 patients for PN, and 40,271 for RC. The validation cohort consisted of 189,669 patients undergoing RP; 91,403 for RN; 31,863 for PN, and 16,991 for RC. Baseline characteristics, risk factors, and preoperative co-morbidities of patients of training and validation cohorts are outlined in Table 2. Overall, the training and validation cohorts were similar in their preoperative patient, hospital, and surgical characteristics. Despite random assignment, the training cohort consisted of fewer people with obesity (p = 0.02), a 16

17 higher rate of cerebrovascular disease (p = 0.05), and a lower rate of peripheral vascular disease (p = 0.03). Identifying Preoperative Risk Factors of Delirium Multivariate logistic regression model was developed using preoperative patient-specific and surgical characteristics suggested to be associated with postoperative delirium (Table 5) (26,27). This adjusted model demonstrated in the training cohort that male, white, and unmarried patients had significantly greater adjusted odds of developing postoperative delirium in all genitourinary (GU) surgeries. Compared to patients younger than 60 years old, patients aged 70 to 79 years (OR: 3.03; 95% CI: ; p <0.001) and patients aged greater or equal to 80 years (OR: 1.90; 95% CI: , p<0.001) had a significantly higher risk of postoperative delirium. A history of alcohol dependence (OR: 4.43; 95% CI: ; p <0.001), dementia (OR: 3.38; 95% CI: ; p <0.001), and immobility (OR: 4.05; 95% CI: ; p <0.001) had the strongest association with postoperative delirium. Many additional medical comorbidities were also significantly associated with delirium; however, significant associations with postoperative delirium were not found in patients with obesity, peripheral vascular disease, and liver dysfunction. Many of the risk factors found to be significant in our multivariable model remained significant after stratification by type of surgery. Delirium Risk Prediction Model Our preoperative risk prediction model constructed from the training cohort data is shown in Table 6. The mean score in patients without delirium was 3.9 (SD 2.3) compared to 6.7 (SD 3.6) in patients with delirium (p <0.001). Numerical scores of the preoperative delirium risk 17

18 prediction model were further categorized into low (risk score 0-2), intermediate (risk score 3-5), high (risk score 6-8), and very high (risk score 9) risk groups (Figure 3). The majority of patients undergoing RP were found to be in the low to intermediate risk group (88.3%), while those undergoing RC had the highest percentage of patients at the high and very high risk category (58.5%). Figure 3B illustrates the percentage of patients that have developed postoperative delirium by surgery type and risk category. In each of the four surgeries, increasing delirium risk was associated with increased delirium. In particular, RC had the highest incidence of delirium across all surgeries, with 4.30% at low-risk, 5.25% at intermediate risk, 11.61% at high risk, and 20.46% at very high-risk. Evaluation of the Clinical Prediction Score Each added point on the risk score was associated with an increased predicted probability of developing postoperative delirium across all GU surgeries (OR: 1.35; 95% CI: ; p <0.001, Figure 2A in blue) in the training cohort. Additional analysis on the validation cohort yielded similar results (OR: 1.34; 95% CI: ; p <0.001, Figure 3A in red). The risk score was also predictive of delirium by surgery type in both the training and validation cohorts (Figure 2B). The risk prediction model demonstrated good calibration in predicting postoperative delirium across all GU surgeries in both the training (ROC area: 0.74; 95% CI: ) and validation cohorts (ROC area: 0.75; 95% CI: , Figure 2B). Its predictive power was maintained with corresponding ROC values for both training and validation cohorts when stratified by surgery type. 18

19 Discussion This study performed the first contemporary, population-based analysis of postoperative delirium in major urologic cancer surgeries. The overall incidence of postoperative delirium was 2.7%. Stratification by type of surgery revealed a similar duration of delirium but different estimates of incidence, with the highest incidence of 11% in RC patients. Patients who developed postoperative delirium following urologic cancer surgeries had worse outcomes in mortality, disposition to higher level care facilities, length of hospital stay, and hospital costs. In particular, the high direct hospital admission costs were the result of increasing laboratory, room and board, and hospital staffing expenses required to care for patients with delirium. Moreover, our preoperative risk prediction score of postoperative delirium has revealed a wide range of clinically relevant risk factors that contribute to the risk of postoperative delirium (Table 6). Each 1-point increase in the risk score corresponds to a 1.34 (validation) to 1.35 (training) increase in odds of developing delirium. Further stratification of the score in risk categories reveals a stepwise increase in delirium incidence from the low to very high risk groups (Figure 3B). In particular, patients following RC are found to be in the high to very high risk categories, which correlate with higher rates of postoperative delirium. Although research on postoperative delirium has been extensive in other surgical specialties, prior investigations in delirium following urologic cancers surgeries have been less robust and have been differentiated by methods used to measure this condition. For example, in the context of bladder cancer and radical cystectomy, Shabsigh et al. estimated the incidence of postoperative delirium to be 1.8% via a retrospective, chart-based approach (28). This estimate suggests that such postoperative complications are coded as rare events in a patient s hospital 19

20 course. In contrast, in a single institution, prospective pilot study, Large et al. demonstrated that the incidence of postoperative delirium, when measured with standardized criteria such as the Confusion Assessment Method (29), was much higher at 29% (23). Our population-based analysis identified the incidence of delirium to be 11% after RC, a finding that is not only between the two previous estimates, but also one that affirms the significant burden of delirium following RC. The incidence of delirium after urologic cancer surgeries follows a similar pattern in other procedures, with lower reported incidences when using chart-based, retrospective assessments (particular nephrectomy at 0.25% (3) and radical prostatectomy at 6% to 9%(28)). Because our approach combines information from ICD-9 coding, acute medication use, postoperative restraints, and use of sitters to increase the sensitivity of our capture rate, our estimates falls in-between retrospective and prospective estimates. Understanding the financial impact of delirium remains essential to effectively utilize resources and improve clinical management. In the surgical setting, the financial impact of delirium has only been demonstrated after elective orthopedic surgery. Franco et al. demonstrated an increase in the technical, or supply-related costs associated with delirium but did not examine the granular detail outlining specific costs increases by procedure type (30). Our results, in contrast, provide a population-based analysis that highlights the effects of postoperative delirium on the utilization of hospital resources across major urologic cancer surgeries in the US. In particular, our results illustrate the $4,252 increase in direct hospital admission costs to care for patients with delirium following RC. With the growing number of elderly patients with high perioperative risk undergoing this procedure for muscle-invasive bladder cancer, these results affirm the need to devise novel strategies to effectively prevent and manage this complication. 20

21 The population undergoing urologic cancer surgeries has become increasingly vulnerable to postoperative complications such as delirium, in part due to issues such as increased number of comorbidities, increased frailty, and decreased physiologic reserve (31). In the setting of bladder cancer, radical cystectomy remains one of the most complex procedures in urologic cancer surgery. Despite improvements in perioperative care and surgical techniques, radical cystectomy remains burdened with high rates of morbidity (28% to 57%) and mortality (2% to 5%) (32). Preoperative identification of the risk of these patients including the risk of postoperative delirium would enable targeted delirium surveillance, implementation of preventative strategies, and a thorough disclosure of risk versus benefit. Marcantonio et al. was the first to design a postoperative delirium risk prediction rule for non-cardiac surgeries, (10) and subsequently, other risk prediction tools have been designed for use in other types of medical and surgical fields (11,14,15). Similar to prior models and analyses of preoperative risk factors for delirium, our model highlights predisposing risk factors such as cognitive impairment, older age, male gender, functional impairment, sensory impairment, depression, and increased comorbidities were highly associated with delirium in urologic cancer surgeries. In particular, dementia, alcohol abuse, and immobility were the most significant risk factors for delirium (26,33). Finally, while most prior risk prediction models are oftentimes labor intensive and require a battery of clinical instruments, our risk prediction model incorporates readily available patient information without the use of involved assessments to enhance perioperative planning for patients at high risk of delirium. At present, the etiology of postoperative delirium remains incompletely understood. Recent evidence, however, suggests that delirium can be triggered by the release of inflammatory mediators resulting in neuroinflammatory changes (34). The relationships between 21

22 surgical approach and postsurgical inflammatory cytokine pathways with delirium may provide additional insights. For example, Fracalanza et al. has demonstrated increased levels of C- Reactive Protein (CRP) and Interleukin 6 (IL-6) following open radical prostatectomy vs. robotic-assisted laparoscopic prostatectomy, suggesting that an open surgical approach may contribute to worsened post-surgical inflammation (35). Other studies have corroborated these findings of higher postoperative IL-6 levels in patients undergoing open colon surgery and cholecystectomy compared to laparoscopic approaches (36,37). Higher preoperative and postoperative levels of inflammatory markers such as CRP and IL-6 have been associated with postoperative delirium (8). Therefore, the increased inflammatory mediators released after an open surgical approach may be the underlying mechanisms driving the association with delirium found in this study. However, the decision to proceed with an open vs. minimally invasive approach often hinges on patient-specific and tumor-specific characteristics, and it is reasonable to suggest that patients who are not candidates for minimally invasive surgery have additional, unmeasured confounders that increase the risk of delirium. Therefore, further investigation is necessary to elucidate the mechanisms behind postoperative delirium and surgical approach in urologic cancer surgeries. In the era of healthcare quality, the management of postoperative delirium has become an increasing priority. In the surgical setting, early geriatric medicine consultation has been shown to reduce the incidence and severity of postoperative delirium in hip fracture patients (38), a finding that increasingly emphasizes the multidisciplinary role of delirium management. Other programs such as the Hospital Elder Life Program, which utilizes multidisciplinary teams to implement strategies such as reorientation, sensory support, hydration protocols, and sleep optimization, have shown considerable benefit and cost savings of roughly $9000 per year (39). 22

23 Moreover, several different pharmacological approaches have been considered to target specific features of delirium. For example, strategies such as the use of melatonin have been proposed to regulate the circadian rhythm of postsurgical patients (40,41). Additional pharmacologic strategies have also focused on reducing medications that increase risk of delirium such as opioids, anticholinergics, and sedatives (7). Finally, emerging data on the perioperative management of high risk patients suggest that reducing anesthetic depth and initiating prophylactic low-dose haloperidol may reduce postoperative delirium (30). Overall, the predictive risk stratification model that we developed in the current study may serve to better identify those patients at risk and to apply targeted interventions to prevent postoperative delirium among patients undergoing major urologic cancer surgery. Several important limitations must be taken into consideration. Classically, postoperative delirium has been diagnosed by applying diagnostic criteria after a comprehensive mental status interview (28,32). Moreover, retrospective identification of delirium using features of an administrative database including ICD-9 diagnostic codes has been reported to be specific but not sensitive due to the lack of clinical granularity required for a sufficient diagnosis (16,42). As a result, prior studies that have utilized this approach have suffered from limited interpretability. In contrast, our population-based approach used a combination of ICD-9 codes and other elements that represent postoperative delirium management and treatment to improve our capture rate. Nonetheless, it is possible that this combination still under-represents the true rates of postoperative delirium. Moreover, certain risk factors such as immobility, hearing loss, and vision loss have traditionally been quantified using interviewer-based clinical instruments(11) and may be subject to unmeasured confounding. 23

24 Other limitations of our study include sampling of US hospitals, a consequence that may not accurately represent the incidence, outcomes, and cost of postoperative delirium in other countries. Moreover, to minimize sampling bias, clustered statistical analysis while accounting for survey weights were used. Misclassification bias due to coding errors may also exist due to the secondary analysis performed on administrative data. Finally, detailed oncological and clinical information such as tumor staging, grade, and pathological characteristics, chemotherapy use, American Society of Anesthesiologists score, and use of postoperative care pathways and outpatient management strategies could not be reliably captured with the available data. 24

25 Conclusions, Summary, and Future Directions In conclusion, this contemporary evaluation of postoperative delirium in patients undergoing radical prostatectomy, radical nephrectomy, partial nephrectomy, and radical cystectomy in the US found that patients with postoperative delirium experience increased mortality, prolonged LOS, and higher direct hospital costs. In particular, the greatest incidence and adverse outcomes of postoperative delirium occur in patients undergoing radical cystectomy. Our results also illustrate the first population-based, preoperative risk prediction model for postoperative delirium in patients undergoing major urologic cancer surgeries. The results of our risk prediction model are promising given its overall consistency with published delirium risk factors and ease of use to identify patients at differing levels of risk for postoperative delirium. Our score could be used to help identify patients at a high risk of developing postoperative delirium in order to initiate targeting interventions and prevention strategies. However, further testing of this model will ultimately shed insight on its overall clinical utility. Most importantly, continued research in developing preventative strategies for delirium is essential for increasing the quality of contemporary urologic surgical care. 25

26 Tables and Figures Table 1 Patient, Hospital, and Surgical Characteristics for patients with and without postoperative delirium following radical prostatectomy, radical nephrectomy, partial nephrectomy, and radical cystectomy in the USA from 2003 to Radical Prostatectomy N = 630,353 Radical Nephrectomy N = 305,503 Delirium Status No Yes P-value No Yes Partial Nephrectomy N = 104,217 P- value No Yes N = 3,377 N = 50,993 Radical Cystectomy N = 57,261 P- value No Yes P-value N = 6,268 Variables N = 624,367 N = 5,986 N = 291,072 N = 14,431 N = 100,839 Patient Characteristics Age, (%) < < < < < Gender, (%) Male Female - $ Race, (%) White Non-white BMI (kg/m 2 ), (%) Obesity ( 30) Not obese (<30) Marital Status, (%) Married < < Not married Insurance Status, (%) Medicare Medicaid < < < <0.001 Private Other

27 <0.001 <0.001 <0.001 <0.001 Charlson Comorbidity index, (%) Hospital Characteristics Hospital beds, (%) < > Teaching Hospital, (%) Vicinity to city center, (%) Rural Urban Region, (%) Midwest Northeast South West Surgical Characteristics Lymph Node Dissection, (%) Lap/Robotic, (%) Open surgery, (%) Lap/Robotic, (%) < Operative time (hours) < < < PRBC Transfusion 0-4 units < < < < units ICU Admission < < < <

28 All patients received a baseline Charlson Comorbidity Index score of 2 due to presence of tumor without metastasis $Gender N/A for Radical Prostatectomy patients PRBC = Packed Red Blood Cells ICU = Intensive Care Unit 28

29 Table 2 Patient, Hospital, and Surgical Characteristics of the Training and Validation Cohorts Variable N = 767,408 Training N = 329,926 Validation P-value Patient Characteristics Age in yr., no. (%): 0.14 < (37.3) (37.8) (40.2) (40.3) (18.2) (17.7) (4.3) (4.3) Gender, no. (%): 0.85 Male (83.4%) (83.4%) Female (16.6%) (16.6%) Race, no. (%): 0.39 White (71.2) (71.5) Non-white (28.8) (28.5) BMI (kg/m 2 ), no. (%): 0.02 Obese ( 30) (93.1) (93.5) Not obese (<30) (6.9) (6.5) Marital Status, no. (%): 0.89 Married (66.7) (66.6) Not married (33.3) (33.4) Insurance Status, no. (%): 0.10 Medicare (40.5) (39.8) Medicaid (2.5) 8300 (2.5) Private (51.8) (52.4) Other (5.2) (5.3) Charlson comorbidity index, no. (%): (65.6) (65.5) (22.1) (22.2) (12.3) (12.3) Hospital Characteristics Hospital beds, no. (%): 0.48 < (11.3) (11.5) (39.8) (39.4) (27.4) (27.8) > (21.5) (21.3) Teaching Hospital, no. (%): (29.3) (29.4) 0.80 Vicinity to city center, no. (%): Rural (4.1) (4.1) Urban (95.9) (95.9) Region, no. (%): 0.39 Midwest (21.8) (22.2) Northeast (17.2) (17.3) South (39) (38.6) West (22) (22) Delirium Risk Factors Sleep Difficulties, no. (%): (4.5) (4.6) 0.52 Hearing Loss, no. (%): 5520 (0.7) 2634 (0.8) 0.09 Vision loss, no. (%): 8297 (1.1) 3613 (1.1) 0.81

30 Dementia, no. (%): 1701 (0.2) 578 (0.2) 0.12 Depression/Anxiety, no. (%): (2.2) 6904 (2.1) 0.51 Alcohol abuse, no. (%): 5730 (0.8) 2388 (0.7) 0.66 Immobility, no. (%): 379 (0.1) 122 (0) 0.39 Comorbidities Congestive Heart Failure, no. (%): (2.8) 9305 (2.8) 0.85 Chronic Pulmonary Disease, no. (%): (12.3) (12.3) 0.99 Cerebrovascular disease, no. (%): (1.6) 4845 (1.5) 0.05 Diabetes, no. (%): (17) (17.1) 0.81 Liver disease, no. (%): 2683 (0.4) 1070 (0.3) 0.51 Metastatic solid tumor, no. (%): (5.5) (5.7) 0.18 Peripheral vascular disease, no. (%): (2.2) 7761 (2.4) 0.03 Renal Insufficiency, no. (%): (5.6) (5.6) 0.95 Surgical Characteristics Surgery, no. (%): 0.57 Prostatectomy (57.4) (57.5) Radical Nephrectomy (27.9) (27.7) Partial Nephrectomy (9.4) (9.7) Radical cystectomy (5.3) (5.2) Operative time (hours), no. (%): (58) (57.2) (28) (28.5) (6.6) (6.7) (7.4) (7.7) Lymph Node Dissection, no. (%): (32.1) (32.1) 0.85 Surgical Approach, no. (%): 0.36 Open Surgery (51.3) (51.6) Laparoscopic/Robotic (48.7) (48.4) All patients received a baseline Charlson Comorbidity Index score of 2 due to presence of tumor without metastasis 30

31 Table 3 Incidence, Duration, and Onset of Postoperative Delirium in Four Major Urologic Cancer Surgeries Overall N = 1,097,335 RP N = 630,353 RN N = 305,503 PN N = 104,217 RC N = 57,262 Delirium Status No Yes No Yes No Yes No Yes No Yes Incidence (%) 1,067,272 (97.3) 30,063 (2.7) 624,367 (99.1) 5,986 (1) 291,072 (95.3) 14,431 (4.7) 100,839 (96.8) 3,377 (3.2) 50,993 (89.1) 6,268 (11) Duration (days), Median (IQR) - 2 (1, 4) - 2 (1, 3) - 2 (1, 5) - 2 (1, 4) - 2 (1, 5) RP = Radical Prostatectomy, RN = Radical Nephrectomy, PN = Partial Nephrectomy, RC = Radical Cystectomy 31

32 Table 4 Mortality, discharge disposition, direct admission costs, and hospital length of stay outcomes associated with postoperative delirium in the USA from 2003 to All costs were inflated to 2013 US dollars Categorical outcomes 90-d Mortality, no. (%): Disposition, no. (%): Unadjusted Analysis Adjusted Analysis $ Overall No Delirium Delirium OR (95% CI) P-value OR (95% CI) P-value (0.5) (0.4) (4.7) ( ) <0.001 ( ) Home Health Care 97,763 (8.9) 91,059 (8.5) 6,704 (22.3) 4.66 ( ) < ( ) <0.001 Skilled Nursing Facility 29,865 (2.7) 23,249 (2.2) 6,616 (22) 18.0 ( ) < ( ) <0.001 In-Hospital Death 3,936 (0.4) 2,801 (0.3) 1,135 (3.8) 25.7 ( ) < ( ) <0.001 Other 4061 (0.4) 3403 (0.3) 659 (2.2) 12.3 ( ) < ( ) <0.001 Unadjusted Analysis Adjusted Analysis $ Continuous Outcomes Overall No Delirium Delirium β (95%CI) P-value β (95%CI) P-value Admission Cost (USD), Median (IQR) 10,876 (8,131, 14,977) 10,771 (8,079, 14,718) 18,710 (12,170, 31,534) 17,261 (15,602-18,919) < ,697 (2,250-3,144) <0.001 Length of Stay (days), Median (IQR) 3 (2, 4) 3 (2, 4) 7 (4, 11) 6.18 (5.7, 6.7) < ( ) <0.001 LOS = Length of Stay, IQR = Interquartile Range, USD = United States Dollars, OR = Odds Ratio, CI = Confidence Interval, β = Beta Coefficient $Adjusted for delirium, patient (age, race, gender, marital status, obesity, Charlson Comorbidity Index, Insurance status), hospital (size, teaching status, region, urban vs. rural), and perioperative characteristics (type of surgery, open vs. minimally invasive, OR time, number of packed RBC use greater than 4 units, ICU admission, soft tissue infection, sepsis, urinary tract infection, other postoperative infections, pneumonia, aspiration, ileus, peritonitis, perforation, GI hemorrhage, acute renal failure, DVT, arrhythmias, myocardial infarction, cardiac arrest, respiratory failure, and cerebrovascular incident) 32

33 Table 5 Preoperative Risk Factors of Delirium for the Training Cohort Covariates Multivariate $ All GU Surgeries OR (95% CI), Significance Multivariate RP OR (95% CI), Significance Multivariate PN OR (95% CI), Significance Multivariate RN OR (95% CI), Significance Multivariate RC OR (95% CI), Significance Age <60 Ref Ref Ref Ref Ref [ ] 0.94 [ ] 1.26 [ ] 1.20 [ ] 1.18 [ ] [ ] *** 1.24 [ ] 1.86 [ ] *** 2.17 [ ] *** 2.09 [ ] *** [ ] *** 2.01 [ ] 4.39 [ ] *** 3.45 [ ] *** 2.74 [ ] *** White Race 1.15 [ ] * 1.16 [ ] 1.13 [ ] 1.15 [ ] 1.17 [ ] Male 1.29 [ ] *** [ ] 1.22 [ ] ** 1.61 [ ] *** BMI (kg/m 2 ) [ ] 1.01 [ ] 1.12 [ ] 1.14 [ ] 1.16 [ ] Unmarried 1.38 [ ] *** 1.97 [ ] *** 1.30 [ ] 1.20 [ ] ** 1.29 [ ] ** Sleep Disturbance 1.21 [ ] * 1.32 [ ] 1.85 [ ] * 0.96 [ ] 1.20 [ ] Hearing Loss 1.21 [ ] 1.47 [ ] 2.07 [ ] 0.99 [ ] 1.04 [ ] Vision Loss 2.27 [ ] * 1.23 [ ] 2.35 [ ] 3.50 [ ] ** 0.36 [ ] Dementia 3.38 [ ] *** 4.33 [ ] * 3.91 [ ] ** 2.88 [ ] *** 4.25 [ ] *** Alcohol Abuse 4.43 [ ] *** 5.16 [ ] *** 3.68 [ ] *** 4.81 [ ] *** 3.09 [ ] *** Depression and/or 1.65 [ ] *** 1.98 [ ] ** 1.53 [ ] 1.71 [ ] *** 1.23 [ ] Anxiety Immobility 4.05 [ ] *** [ ] ** 1.25 [ ] CHF 2.06 [ ] *** 4.43 [ ] *** 1.32 [ ] 1.79 [ ] *** 2.80 [ ] *** CPD 1.50 [ ] *** 1.96 [ ] *** 1.50 [ ] * 1.52 [ ] *** 1.17 [ ] CVD 2.57 [ ] *** 2.96 [ ] *** 2.63 [ ] *** 2.59 [ ] *** 2.50 [ ] *** Diabetes 1.18 [ ] ** 1.36 [ ] * 1.12 [ ] 1.14 [ ] 1.13 [ ] Liver Dysfunction 1.15 [ ] 2.27 [ ] 1.39 [ ] 0.87 [ ] 2.33 [ ] Renal Insufficiency 1.40 [ ] *** 1.87 [ ] ** 1.28 [ ] 1.53 [ ] *** 1.15 [ ] Metastases 1.17 [ ] * 1.55 [ ] 1.11 [ ] 1.24 [ ] * 1.06 [ ] PVD 1.12 [ ] 1.65 [ ] 1.37 [ ] 0.99 [ ] 1.14 [ ] Open Surgery 1.33 [ ] *** 1.41 [ ] * 1.21 [ ] 1.40 [ ] *** 1.07 [ ] $ Also adjusted for type of surgery Removed due to the absence of cases * p <0.05 ** p <0.01 *** p <0.001 RP = Radical Prostatectomy; PN = Partial Nephrectomy; RN = Radical Nephrectomy; RC = Radical cystectomy; CHF = Congestive Heart Failure; CPD = Chronic Pulmonary Disease; CVD: Cerebrovascular Disease; PVD = Peripheral Vascular Disease 33

34 Table 6 Preoperative Risk Score Model for Delirium Preoperative Risk Prediction Score 1 point 2 points 4 points White race Depression and/or Anxiety Vision Loss Male gender Chronic Pulmonary Disease Cerebrovascular disease Unmarried Renal Insufficiency 5 points Open surgery 3 points Age 80 years Sleep Disorder Age years Dementia Diabetes Congestive Heart Failure 6 Points Metastases Alcohol Abuse Immobility 34

35 Table 7 Distribution of Delirium Risk Prediction Scores by Surgery Surgery Overall No delirium Delirium P-value All Genitourinary Surgery 4 (2.4) 3.9 (2.3) 6.7 (3.7) <0.001 Radical Prostatectomy 3.4 (1.7) 3.4 (1.7) 4.5 (2.6) <0.001 Partial Nephrectomy 4.1 (2.7) 4 (2.6) 6.1 (3.8) <0.001 Radical Nephrectomy 4.7 (2.9) 4.6 (2.8) 7.1 (3.6) <0.001 Radical Cystectomy 6.4 (3.1) 6.2 (3) 8.3 (3.4) <0.001 Scores displayed as Mean (SD) Table 8 ICD-9 codes for Urologic Cancer Surgeries Surgery Type ICD-9 Procedure ICD-9 Diagnosis Prostate Cancer 60.5, Kidney Cancer/Renal Mass 55.4 $ 223.0, 223.1, 189.0, 189.1, 189.2, 189.8, Kidney Cancer/Renal Mass 55.51, 55.52, , 198.0, 198.1, , 593.2, 593.9, , , , V10.52, V16.51 Bladder Cancer 57.71, , 188.1, 188.2, 188.3, 188.4, 188.5, 188.6, 188.7, 188.8, 188.9, $Denotes partial nephrectomy 35

36 Table 9 ICD-9 Codes for Delirium and Major Risk Factors Variable Brief Description ICD-9 Codes Outcome Delirium Delirium and/or acute changes in cognition , , 290.2, 290.3, , , 290.8, 290.9, , , , , , , , 292.9, 293.0, 293.1, , , , , , 293.9, 294.9, 298.1, 298.2, 298.8, 298.9, , , 308.0, 308.1, 308.2, 308.3, 308.4, 308.9, 310.1, 310.9, , 348.3, , , , 437.2, 572.2, 768.7, , , , , , , Preoperative Risk Factors Alcohol Abuse Sleep Disturbance Hearing Loss Vision Loss Dementia Depression Anxiety Immobility Alcohol abuse and alcohol-related medical complications Sleep-related medical problems such as sleep apnea, insomnia, and circadian rhythm disorders Hearing difficulties and/or loss Vision impairment in one or both eyes Dementia Depression and/or anxiety Preoperative immobility 291.0, 291.1, 291.2, 291.3, 291.4, 291.5, , , , 291.9, 303.0, , , , 303.9, , , , 305.0, , , , 357.5, 425.5, 535.3, , 571.0, 571.1, 571.2, 571.3, 790.3, 977.3, , 307.4, , , , , , , , , , , , , , , , 327.2, , , , , , , , 327.3, , , , , , , , , 327.4, , , , , , , , , , 327.8, 780.5, , , , , , , , , , 387.0, 387.1, 387.2, 387.8, 387.9, 388.0, , , 388.1, , , 388.2, 388.3, , , 388.4, , , , , 388.5, 389.0, , , , , , , , 389.1, , , , , , , , , 389.2, , , 389.7, 389.8, 389.9, 744.0, , , , , , , 367.1, 368.3, , , , , , , , , , , , , , , , , , , , , , , 369.3, 369.4, 369.6, , , , , , , , , , 369.7, , , , , , , 369.8, , 290.1, , , 290.4, , 291.2, , 294.2, , 331.0, , , 331.2, , , , 331.9, , , , , , , , 296.3, , , , , , , 300.0, , , , 309.1, , 93.51, 93.52, 93.53, 93.54, 93.55, 93.56, 93.57, 93.58,

37 Figure 1 Adjusted hospital admission costs by delirium status and type of surgery in the United States from 2003 to Direct hospital admission costs were adjusted for delirium status, patient characteristics (age, race, gender, marital status, obesity, Charlson Comorbidity Index, insurance status), hospital characteristics (size, teaching status, region, urban vs. rural), and perioperative characteristics (type of surgery, open vs. minimally invasive, OR time, number of packed RBC use greater than 4 units, ICU admission, soft tissue infection, sepsis, urinary tract infection, other postoperative infections, pneumonia, aspiration, ileus, peritonitis, perforation, GI hemorrhage, acute renal failure, DVT, arrhythmias, myocardial infarction, cardiac arrest, respiratory failure, and cerebrovascular incident). Hospital admission costs were sub-divided by categories that include supplies, room and board, radiology, pharmacy, operating room, lab, and other hospital costs. All costs were inflated to 2013 US dollars. * Significance difference in total costs (p<0.001). 37

38 38

39 Figure 2 Predicted Probability and Receiver Operating Characteristic (ROC) Curve Analysis by the Preoperative Delirium Risk Score. A illustrates predicted probability curve of the preoperative risk prediction model. The blue shading represents the training cohort, while the red shading represents the validation cohort. B illustrates the Area Under the Curve (AUC) Analysis of the risk prediction model of the training (blue) and validation (red) cohorts. Additional sub-group analysis of the predicted probability curve and the AUC analysis are shown. 39

40 40

41 Figure 3 Percent of patients of postoperative delirium by delirium risk prediction in the training cohort. Part A illustrates the percent of patients at risk of postoperative delirium in low, intermediate, high, and very high risk categories as classified by the delirium risk prediction score. Part B illustrates the percent of patients with postoperative delirium in each, respective risk category. 41

42 42