Factors Predictive of Adverse Postoperative Events Following Tonsillectomy. A thesis submitted to the. Graduate School

Transcription

1

2 Factors Predictive of Adverse Postoperative Events Following Tonsillectomy A thesis submitted to the Graduate School of the University of Cincinnati in partial fulfillment of the requirements for the degree of Master of Science in Clinical & Translational Research In the Department of Environmental Health Division of Epidemiology & Biostatistics of the College of Medicine November, 2013 by Rajeev Subramanyam MBBS, Bangalore University, September 2000 MD, Postgraduate Institute of Medical Education and Research, June 2006 Committee Chair: Erin Haynes, DrPH

3 ABSTRACT Background: Tonsillectomy is the second most common surgery done in children in the United States. In spite of more than 500,000 surgeries done every year there is no evidence for postoperative disposition of patients following tonsillectomy. Our data shows that up to 10% of patients end up having unanticipated emergency room (ER) visits after tonsillectomy. The primary aim of the study is to predict the risk factors for the occurrence of unanticipated ER visits at 3-days and 3-weeks (after 3-days) following tonsillectomy. Methods: In this single-center prospective observational cohort study, all consenting patients < 17 years of age presenting for tonsillectomy with or without adenoidectomy were included. After obtaining the approval of the Institutional Review Board, patients undergoing tonsillectomy were recruited from the weekly operating room schedule, using a simple random sampling technique. Patients belonging to American Society of Anesthesiology physical status 4 and 5, undergoing any additional procedures with the adenotonsillectomy, and those who refused to consent were excluded. The preoperative, intraoperative, postoperative data on patient related, health system related, and anesthesia related factors were used as the predictive variables. The primary outcomes of interest were the prediction of risk factors for the occurrence of unanticipated ER visits at3-days and 3-weeks after tonsillectomy. Secondary outcomes were complications in post anesthesia care unit. The outcomes were assessed by a postoperative telephone survey at 24 hours and at 3-weeks post discharge and also by the electronic record. Reasons for the occurrence of adverse events were noted. Families were compensated $10 for their participation in the survey. Using REDCap database bolstered data accuracy. Univariate analysis was done and variable with a p value of <0.24 was included in the first model. Variables with a p value of ii

4 <0.05 with the first model was included in the final model. Goodness of fit was tested and c statistic obtained. Statistical analysis was done using SAS software. Results: Data on 469 patients < 17 year old who underwent tonsillectomy were available for analysis. The mean age was 6.5 years and body mass index was Pediatric sleep questionnaire was completed in all patients and sleep apnea was present in 56.5%. Multivariate logistic regression identified preoperative diastolic blood pressure (Odds Ratio (OR) 0.925; p- value=0.0097), intraoperative dexamethasone (OR 1.246, p-value=0.0078), preoperative neurologic comorbidity (OR 5.101, p-value=0.0561), presence of failure to thrive (OR 1.090, p- value=0.0421) as risk factors for unanticipated ER visits at3-days and preoperative systolic blood pressure (OR 0.939; p-value=0.005), intraoperative dexamethasone (OR 1.207, p-value=0.0058), use of home medications (OR 3.503; p-value=0.0148), presence of upper respiratory infections (OR 4.173; p-value=0.0098) at 3-weeks. Both the model had adequate goodness of fit with the variables. Conclusions: Our study identified preoperative blood pressure, intraoperative dexamethasone, preexisting conditions like upper respiratory infection, neurologic comorbidity, failure to thrive, and use of home medications as risk factors that predict unanticipated ER visits following tonsillectomy. These important and easily measurable variables provide valuable information for optimal management of patients undergoing tonsillectomy. iii

5 iv

6 ACKNOWLEDGEMENT I would like to thank the members of thesis committee Dr. Haynes, Dr. Varughese, and Dr. Hossain for their guidance and support for performing this research and submitting the thesis. In particular, Dr. Varughese s mentoring is invaluable for this research project as well as my future goal towards an independent clinical research career. I also thank Dr. Hossain, Department of Biostatistics and Epidemiology, Cincinnati Children s Hospital for the statistical analysis. I thank Jacqueline Knapke, Program Director with all her help and suggestions with the thesis and all through the Master s Program. I thank Dr. C. Dean Kurth, Chair Department of Anesthesia at Cincinnati Children s Hospital Medical Center for supporting my Master s Program. I especially thank the Clinical Research Coordinators Jessica Prim, Tara Buesking, and Kayla Stallworth for collecting and entering the data into REDCap database. I thank REDCap and ENT team for supporting us. I thank the patients and families who consented for the study without which this study could not be performed. I thank my wife Shikha Thapar and my daughter Ahana Iyer who spared their family time with me to complete this invaluable project. v

7 TABLE OF CONTENTS INTRODUCTION 1 Background 1 Specific aims 3 Significance 4 METHODS 5 Study design 5 Study sample 5 Study methods 5 Data management 8 Statistical analysis 9 RESULTS 10 Univariate analysis 11 Logistic regression 12 Secondary outcomes 13 DISCUSSION 14 CONCLUSIONS 18 TABLES 19 BIBLIOGRAPHY 36 APPENDICES 39 vi

8 List of Tables Table 1: Demographic Data Table 2: Data on comorbidity Table 3: Data on sleep apnea Table 4: Preoperative risk factors and hemodynamics Table 5: Data on premedication Table 6: Data on intraoperative anesthesia variables Table 7: Data on other intraoperative medications Table 8: Data on intraoperative adverse events Table 9: Data on post anesthesia care unit related variables Tables 10: Univariate analysis of the 3 day and 3 week preoperative categorical variables Table 11: Univariate analysis of the 3 day and 3 week categorical intraoperative and postoperative variables Table 12: Univariate analysis of the 3 day and 3 week continuous variables Table 13: Logistic Regression Model Results for unanticipated emergency room visits within 3- days after tonsillectomy with p value less than or equal to 0.24 on univariate analysis. Table 14: Logistic Regression Model Results for unanticipated emergency room visits within3- days after tonsillectomy with p value less than or equal to 0.05 in the first multivariate logistic model Table 15: Logistic Regression Model Results for unanticipated emergency room visits within 3- weeks after tonsillectomy with p value less than or equal to 0.24 on univariate analysis Table 16: Logistic Regression Model Results for unanticipated emergency room visits within 3- weeks after tonsillectomy derived from the previous model Table 17: Secondary Outcome data on post anesthesia care unit complications vii

9 INTRODUCTION Background Tonsillectomy is a surgical procedure performed wherein the tonsil with its capsule is completely removed by dissecting the peritonsillar space between the tonsil capsule and the muscular layer. Adenoidectomy is the surgical removal of adenoids and is commonly performed in combination with tonsillectomy. Tonsillectomy is most commonly performed for the presence of sleep disordered breathing (obstructive sleep apnea) and recurrent infections both of which compromise quality of life. Tonsillectomy in patients is associated with significant health benefits, including prevention of the sequel of obstructive breathing such as behavioral problems, neurocognition and improves quality of life scores. 1 On the other hand, tonsillectomy can be associated with both immediate and late postoperative adverse events such as respiratory and/or airway complications (11%), delay in discharge ( %), readmission at 28 days of 3.9 to 12%, and return to the operating room due to hemorrhage (1%). 2,3 In rare instances, it is also associated with catastrophic complications such as permanent brain damage or death. 4 The severity of sleep apnea is important as it can predict the risk of postoperative adverse outcomes. 5 Currently, high-risk patients for the presence of sleep apnea are frequently identified using polysomnography testing. Although considered by many to be the gold standard for evaluating the severity of sleep apnea, polysomnography is an imperfect tool as it is expensive, not-practical for all patients, can be associated with a long wait times and involve significant family and health system resources. 6 On the other hand, a simple 22-point based validated sleep questionnaire is available, that can be used 1

10 to identify sleep disordered breathing. The questionnaire has a high sensitivity (81%) and specificity (87%) when 8 or more responses to the 22-point questionnaire are positive. 7 The questionnaire evaluates subjectively the child s current and past sleep problems based on parental response. The performance of the questionnaire does not vary with participant age and has good internal consistency and reliable test retesting. 8 Due to lack of evidence-based guidance for selection of appropriate patients for outpatient tonsillectomy, significant variation for postoperative disposition exists. 9 At Cincinnati Children s Hospital Medical Center, criteria for hospitalization after tonsillectomy include age younger than 3 years, presence of severe obstructive sleep apnea and/or associated co morbidities such as Down s syndrome, prematurity and morbid obesity. However, often, patients designated to go home after the procedure, experience either, prolonged stays in the post anesthesia care unit (PACU), are admitted to the hospital, or are re-admitted to the hospital after discharge. Other studies have also shown adverse events such as desaturation, need for airway intervention, pain, vomiting, hemorrhage, poor oral intake, fever and dehydration resulting in delayed PACU discharge/ readmissions. 10,11 Identification of risk factors for prolonged recovery a priori may make outpatient surgery less cost-effective than overnight admission. 3 To the best of our knowledge, the criteria for postoperative disposition of patients following tonsillectomy are not well described. Data from 2006 showed that less than 3% of tonsillectomies are done in the inpatient setting. 2 In 2011, a panel for the American Society of Otolaryngology published clinical practice guidelines for tonsillectomy patients. Their intent was to provide clinicians with guidance for identifying patients who are the best candidates for surgery, optimizing perioperative management, and reducing 2

11 unnecessary variation. The panel, however, acknowledged there was a knowledge gap in the identification of and postoperative disposition of high-risk patients. 12 Specific Aims The overall objective of this project is the development of a predictive model for postoperative unanticipated emergency room (ER) visits in patients undergoing tonsillectomy. This study will help to understand, identify, and quantify factors associated with adverse immediate and late postoperative outcomes. The central hypothesis is that patient and health system related factors could be used to develop a structured, evidence-based predictive model for adverse events in the pediatric tonsillectomy population. This predictive model is the first step to developing a clinical rule to improve decision-making. This hypothesis was developed based on our preliminary data, indicating a gap in the current practice. 13 The rationale of this project is that developing a model to predict postoperative adverse events will lead to accurate selection of at risk patients for outpatient vs. inpatient procedures resulting in improved clinical outcomes and enhanced patient safety. It also reduces family psychosocial distress associated with unanticipated hospital visits or re-admission, and the burden on the health care system. The primary aim of this project was to identify and quantify the patient (demographic and clinical) and health care system factors associated with unanticipated ER visits in the first3-days and 3-weeks after tonsillectomy in patients < 17 years of age. The hypothesis is that demographic factors, clinical factors, and health system related factor was predictive of unanticipated ER visits. Demographic factors include age, race, time-of-year, body mass index. Clinical factors include American Society of 3

12 Anesthesiologists (ASA) classification, preoperative diagnosis, presence of upper respiratory infection (URI) and co morbidities. Health system related factors include premedication, anesthesia induction/extubation technique, intraoperative anesthesia adverse events, use of opioids, postoperative disposition. The other outcomes were the immediate postoperative adverse events in PACU like desaturation, somnolence, laryngospasm/airway obstruction, airway instrumentation, and drug administration. Significance The contribution of this work is expected to be the basis for the development of a clinical predictive model for unanticipated ER visits in tonsillectomy patients. This work is significant because the identification and quantification of risk for patients undergoing tonsillectomy will allow for targeted interventions and can lead to improved outcomes, patient safety and family satisfaction by reducing ER visits, while reducing the burden on the health care system. 4

13 METHODS Study Design In this single-center prospective observational cohort study, all consenting patients presenting for tonsillectomy with or without adenoidectomy were included. Study Sample After obtaining the approval of the Institutional Review Board, patients undergoing tonsillectomy at Cincinnati Children s Main and Liberty (ambulatory) campuses were approached. The patients were recruited from the weekly operating room schedule, using a simple random sampling technique. Informed consent was obtained from families for their participation in the study on the day of surgery. Inclusion criteria were age < 17 years, American Society of Anesthesiology (ASA) physical status 1, 2 and 3. Patients belonging to ASA physical status 4 and 5, undergoing any additional procedures with the adenotonsillectomy, and those who refused to consent were excluded. Study Methods This study involved the observation of patients during the course of routine practice at our institution. Preoperative phase: As is our routine practice, healthy patients for outpatient and outpatient-admit procedures are evaluated on the day of surgery by a pediatric nurse practitioner and anesthesiologist. Complex patients are evaluated in the Pre-Anesthesia Consultation Clinic by an anesthesiologist prior to the day of surgery. Our patients and their family usually arrive 90 minutes ahead of the scheduled surgery start. A research coordinator approached the patient/family to explain the study purpose and procedures, and obtained informed consent. Where possible, families received recruitment materials 5

14 (consent form or generic recruitment flyer) during the preoperative ENT clinic visit. During the preoperative phase demographic, clinical, and system related data were collected. Demographic and clinical data included age, race, weight and body mass index, type of procedure (tonsillectomy, adenotonsillectomy), time-of-year (seasonal), presence of URIs, ASA classification (1, 2 or 3) preoperative diagnosis (adenotonsillar hypertrophy, obstructive sleep apnea, recurrent infections or a combination), associated co morbidities (prematurity, down s syndrome, craniofacial anomalies, asthma, seizure disorders, cardiac disease, sickle cell disease, failure to thrive, mucopolysaccharidosis etc.,), use of sedative premedication, and planned postoperative disposition (outpatient, 23 hour admission, Intensive Care Unit admission). Decisions on premedication administration were made according to an algorithm developed at our institution and took into account age, level of child preoperative anxiety, history of previous anesthesia, attendance at the preoperative tour and the time spent in the preoperative area. 14 If indicated, midazolam (0.5mg/kg orally) was administered 15 to 30 minutes prior to surgery. Data was obtained primarily from the anesthesia preoperative evaluation. A simple 22-point validated sleep questionnaire was administered (Appendix I) to a parent by the research coordinator and the responses recorded. Responses to the questionnaire will be graded as yes =1, no =0 and don t know =missing. Intraoperative Phase: The anesthetic technique for tonsillectomy at our institution consists of general anesthesia with endotracheal intubation. Patients are induced, either with intravenous or inhalational agents, depending on age of the child, patient preference, and discretion of the attending anesthesiologist. The use of an antiemetic (ondansetron 0.1 mg/kg) and steroid (dexamethasone 0.5 mg/kg, up to a maximum of 16 mg) is a 6

15 standard procedure. Opioid type (morphine, fentanyl or hydromorphone) and dose/kg is at the discretion of the anesthesiologist. Patients were extubated either deep or awake in the operating room, or transported intubated to the PACU and extubated when fully awake. The induction technique (inhalational vs. intravenous), medications used, fluids used (type and dose in ml/kg), type and dose of opioids and type of extubation (awake vs. deep) were documented. Adverse anesthesia events were defined as desaturation (SpO 2 < 90%) airway obstruction, difficult mask ventilation, difficult intubation, bronchospasm, and delayed emergence from anesthesia (use of naloxone) and were noted. Although the tonsillectomy surgical technique is standardized at our institution, the level of experience of the surgeon (resident, fellow or attending), level of experience of anesthesia staff, and the duration of surgery were noted. Postoperative Phase: Patients were discharged from PACU once standard criteria have been met (Appendix II). The average time spent in the PACU for tonsillectomy patients is about 90 minutes 15. All data on postoperative events, medications administered, and duration of PACU stay were recorded by a PACU nurse and extracted from the patients electronic medical record. The primary outcome of interest was the occurrence of unanticipated ER visits at3-days and 3-weeks after tonsillectomy. These measurements were obtained from families and electronic health record at 24 hours and at 3-weeks. In addition sleep concerns and snoring/airway obstruction were measured as subjective Yes/No answers from families. Secondary outcomes were measured in PACU. This included a binary measure of increased work of breathing (defined as presence of abnormal airway sounds, chest retraction or nasal flaring; Appendix III), desaturation (defined as SpO 2 80% at any 7

16 time in the PACU), laryngospasm/airway obstruction (defined as requiring continuous positive airway pressure with 20 cmh 2 O and/or succinylcholine and/or propofol), airway instrumentation (defined as use of nasal trumpet and/or endotracheal intubation), drug administration (defined as use of bronchodilators, racemic epinephrine, diuretics, naloxone), and prolonged PACU stay. The prolonged PACU stay was due to either inadequate pain control (defined as 8 year old Numerical Rating Scale (NRS) > 3, < 8 year old or developmental delay Face, Leg, Activity, Crying, Consolability Scale (FLACC) > 3; comfortable pain level for developmentally delayed patients will be discussed with caregivers), nausea and vomiting requiring treatment, oxygen requirement to maintain a saturation of > 94%, and somnolence assessed with Ramsay Sedation Scale (Appendix IV). A phone call template was used to contact families (Appendix V) Families were compensated $10 for their participation in the survey. Data Management Using REDCap database bolstered data accuracy. 16 REDCap (Research Electronic Data Capture) is a secure, web-based application designed to support data capture for research studies, providing: 1) an intuitive interface for validated data entry; 2) audit trails for tracking data manipulation and export procedures; 3) automated export procedures for seamless data downloads to common statistical packages; and 4) procedures for importing data from external sources. Initial analyses will be undertaken to inspect data for errors, inconsistencies, and incomplete information. 8

17 Statistical analysis Statistical analysis was done with SAS software (Version 9.3). Univariate statistics were performed with Mann-Whitney U test for continuous variables and Chi-square or Fischer exact test for categorical variables. Analysis was done as a two sided test and a p value of < 0.05 was considered statistically significant. We developed multivariate models for unanticipated ER visits at3-days and at 3- weeks after tonsillectomy. The preoperative, intraoperative, and postoperative factors were used as predictor variables. Variables that were significant at a p-value of < 0.24 from the univariate logistic regression analysis were included in the multivariable logistic regression. Multivariable logistic regression model with backward elimination procedure was performed for unanticipated ER visits both at3-days and 3-weeks. Variables were retained in the final multivariable model if they were significant at the 0.05 levels in the first multivariable model. Adjusted odds ratios (ORs) with 95% confidence interval (CI) were calculated for significant predictors. We checked the fit of models with the goodness-of-fit test and developed prediction models from the fitted models on unanticipated ER visits using ROC curves. 9

18 RESULTS Data on 469 patients < 17 year old who underwent tonsillectomy were available for analysis. The mean age of patients was 6.5 years and the body mass index was kg/m 2. Most patients had adenotonsillar hypertrophy. About 5% had tonsillectomy alone. Outpatient disposition (same day discharge) was planned in 67.6% of patients and most belonged to ASA class 2. Other demographic information is provided in table 1. Comorbidity was present in about 42% of patients. Most of them had one comorbidity. The individual system based comorbidities are summarized in table 2. Sleep study was available in only 6.8% of patients. The pediatric sleep questionnaire was completed in all patients and sleep apnea was present in 56.5%. The data on the 22 points in the pediatric sleep questionnaire is summarized in table 3. Preoperative URI was present in 12%, failure to thrive in about 2.4%, and prematurity in about 10.5%. The mean age at birth in our patient population was about 35 weeks. Hemodynamic data are summarized in table 4. Premedication was administered in 8% of patients. The data on the drugs used and their dosages are provided in table 5. Anesthesia was induced with inhalational anesthetic agent in about 91%. The mean dosages of intraoperative medications are summarized in table 6. Most patients were extubated in PACU. Attending surgeons performed most of the surgeries and the others were done under their supervision. CRNA s performed most of the anesthetic care and attending anesthesiologist supervised all the cases. The mean duration of surgery was 13 minutes. Additional intraoperative medications were used in about 5% per the attending anesthesiologist based on the patient conditions are summarized in table 7. Intraoperative adverse events related to anesthesia occurred in 10

19 about 8.5% of patients of which airway obstruction was the most common. The other complications are summarized in table 8. PACU related variables are summarized in table 9. Analgesics were administered in about 58% of patients and the most common medication used was fentanyl. Most patients had one analgesic intervention in PACU. Nausea and vomiting occurred in about 5% and 6% of patients respectively. Antiemetic in PACU was administered to about 1.7% of patients. A total of 47 patients (10%) had unanticipated emergency visits from all causes in the first 3-weeks. After excluding those visits due to hemorrhage there were 11 patients (2.3%) who had unanticipated emergency visits in the first3-days and 23 patients (4.9%) in the first 3-weeks (after day 3), and these patients were the events of primary interest in our study. Univariate analysis for the primary outcome variables The univariate analyses for the categorical variables are summarized in table 10 and 11. The univariate analyses for the continuous variables are summarized in table 12. The outcomes are unanticipated ER visits at 3-days and at 3-weeks. Logistic regression for 3-day unanticipated emergency visits The factors significant in the univariate analysis were entered into the first logistic regression model with a p-value of < 0.24 (Table 13). The goodness-of-fit of the fitted model was assessed by the value of Akaike s information criterion (AIC). The first model for unanticipated ER visits at 3-days AIC was The c-statistic as a measure of area under the ROC curve for the binary responses for the first model was 86.5%. The percent of concordant is 86.5% and discordant is 13.5%. Hosmer and Lemeshow test showed 11

20 good model predictability within the subgroups of the study population with a p- value=0.79. The factors with a p-value of less than or equal to 0.05 from the first model was entered into the final model (Table 14). The AIC of the second model was and showed a little improvement. The c-statistic of the final model was 80.2%. The percent of concordant was 80% and discordant was 19.7%. Hosmer and Lemeshow test showed a p- value=0.9375, indicating an improvement in risk classification. There were four significant variables that predicted the risk of unanticipated ER visits at 3-weeks: preoperative diastolic blood pressure (Odds Ratio (OR) 0.925; p-value=0.0097), intraoperative dexamethasone (OR 1.246, p-value=0.0078), preoperative neurologic comorbidity (OR 5.101, p-value=0.0561), and presence of failure to thrive (OR 1.090, p- value=0.0421). Logistic regression for 3-week unanticipated emergency visits The factors significant in the univariate analysis were entered into the first logistic regression model with a p value of < 0.24 (Table 15). The goodness-of-fit was assessed by the AIC. The first model for unanticipated ER visits at 3-weeks AIC was The c-statistic for the first model was 82.8%. The percent of concordant is 82.8% and discordant is 17.2%. Hosmer and Lemeshow test showed insignificant risk classification predictability since the p-value was The factors with a p value of less than 0.05 from the first model were entered into the final model. The AIC of the second model was The c statistic of the final model was 75.2%. The percent of concordant was 75.1% and discordant was 24.7%. Hosmer and Lemeshow test showed a better risk classification predictability than the first model with the p-value= Although the AIC values were very similar for these two models, it is because of the improved p-value 12

21 for the Hosmer and Lemeshow test we prefer to report the results for the second model. There were four significant variables that predicted the risk of unanticipated ER visits at 3-weeks: preoperative systolic blood pressure (OR 0.939; p-value=0.005), intraoperative dexamethasone (OR 1.207, p-value=0.0058), use of home medications (OR 3.503; p- value=0.0148), and presence of URIs (OR 4.173; p-value=0.0098) (Table 16). Secondary outcomes Secondary outcomes are PACU complications and are summarized in table 10. Significant desaturation occurred in < 1%, but oxygen had to be restarted in 75% of these patients. No patients required reintubation in PACU due to laryngospasm or airway obstruction. PACU discharge criteria were met in 77 minutes and the total PACU stay was 93.5 minutes. PACU discharge was prolonged in about 21% of patients and the most common reason was inadequate pain control. Most patients (66.6%) were discharged home. 13

22 DISCUSSION Tonsillectomy is the most commonly performed surgical procedure in the United States with over half a million procedures performed in patients. 17 Ambulatory tonsillectomy procedures, in particular, have almost doubled over the last decade. 2 Despite being a common procedure, tonsillectomy can be associated with significant morbidity and/or death, particularly in the setting of hypoxic events and bleeding. 4,18 The incidence of fatal respiratory events following tonsillectomy in patients is twice that of adults, and these events occur more frequently in younger patients and those with co-morbid conditions. 4,19,20 As a procedure, outpatient tonsillectomy is considered safe. However, evidencebased guidance to support the accurate selection of patients who can safely undergo this procedure on an outpatient basis is lacking. 21,22 Significant variation exists between institutions and geographic locations with regard to criteria for postoperative disposition. 9,21 Arbitrary criteria to identify high-risk patients are younger age, presence of co morbidities (e.g., Downs syndrome), and severity of sleep apnea evaluated using polysomnography. 5 However, the high cost of polysomnography precludes its use for all high-risk patients. On the other hand, a validated sleep questionnaire developed to identify sleep disordered breathing, could serve as a more practical and cost-effective tool to identify the severity of sleep apnea. 7 Patients with sleep apnea often get low dose of narcotics due to their sensitivity to these medications. Postoperative pain can be considerable problem in these patients and may predispose them to ER visits. As a result of the lack of evidenced based criteria to evaluate risk, patients designated to go home after the procedure have unexpected ER visits. 10,11 This 14

23 unanticipated morbidity leads to family psychosocial distress, and can place an additional burden on the health care system. In this setting, the urgent need for the development of a method to identify patients at risk for adverse postoperative outcomes in this population is underscored. A predictive model using information that is practical, easy to obtain and avoids expensive testing, can provide a useful and cost-effective tool for the management of tonsillectomy patients. There has been no predictive model for unanticipated ER visits after tonsillectomy. We have identified numerous risk factors for this adverse event. Preoperative adequate blood pressure was identified as a protective factor for the occurrence of ER visits. Blood pressure in patients is age and weight dependent. Patients with OSA have preoperative hypertension and there is a reduction in diastolic blood pressure load in 24 hours after adenotonsillectomy. 23 Preoperative hypotension may indicate fluid deficit and since pain following tonsillectomy can prevent adequate hydration, this may be a risk factor for ER visits after patients go home. 24 Intraoperative dexamethasone was found to be associated with ER visits with odds of 1.2 both at 3-days and 3-weeks. This variable was assessed as a continuous variable in the model. Dexamethasone is an important medication administered during tonsillectomy and helps with pain, swelling of the airway, and reduces vomiting. 25 Dexamethasone has not been associated with an increase in postoperative bleeding. 26 In our study a higher dose was associated with ER visits. More recently a wait and watch approach has been proposed for dexamethasone instead of prophylactic administration. 27 Preexisting conditions like URI, neurologic comorbidity, failure to thrive, and use of home medications as risk factors that predict unanticipated ER visits following 15

24 tonsillectomy. Recent URI is a very controversial area in pediatric anesthesia. 28 Scientific data have confirmed that patients who have recent URIs are at an increased risk for perioperative complications. 29,30 Our study results are novel in showing that this risk also increases ER visits up to 3-weeks by odds of 4. Anesthesiologists should weigh the risks and benefits of having surgery in patients with recent URI as patients have about 4-12 episodes of URI per year. Most agree that any child with severe symptoms should have their surgery postponed by at least 4 weeks. 28 The possibility of observing them for prolonged period or even overnight hospitalization has to be considered given the significant risks of ER visit. Failure to thrive is a sign that a child is not getting adequate nutrition for growth and development. This is documented by use of growth charts. These patients also tend to have severe underlying comorbidities that render them incapable of having adequate nutrition. 31 The prevalence of obstructive breathing is higher in patients in failure to thrive as compared to general pediatric population. 32 Inadequate nutrition with inability to hydrate after a tonsillectomy predisposes to patients to having increased risk of ER visits. Preexisting neurologic comorbidity predisposes patients to muscle weakness, immobility, and even failure to thrive. 31 These patients may also have swallowing dysfunction and this may predispose them to lung aspiration. They also can develop secondary complications that may be associated with inadequate hydration and nutrition. Swallowing problems and pain after tonsillectomy prevent adequate hydration and these patients are a high-risk group for ER visits due to either dehydration or pain. 16

25 In parallel to underlying preexisting problems, these patients have a risk of being on home medications. The underlying condition predisposes patients to dehydration, pain, or vomiting necessitating ER visits. The other main reason for ER visits after tonsillectomy is bleeding. Bleeding is a leading cause of deaths and anoxic brain injuries. 33 We have previously shown that bleeding is the leading cause of malpractice claims against surgeons. 25 Age above 35 years has been found to be a risk factor for postoperative bleeding. 34 We excluded cases with bleeding from our predictive model, as we believe that bleeding cannot be predicted with the current candidate predictors. Analyzing bleeding risk may require genetic testing. Hence excluding bleeding will ensure the validity of our model. There are a few limitations in this study. Although the low event rate indicates an excellent clinical care provided to patients, this would be a limitation on the precision of the model. With an event rate of 11 out of 469 patients and a minimum ratio of 1:5 for precision (1 predictor for 5 events), our model has a precision to identify only 2 predictor variables. A larger sample size is required and would be analyzed in future for rebuilding the model. Second, for the same reason of a low event rate, an internal validation of the model with an internal set of data could not be performed. An internal validation would be done in future as a part of larger data collection. Third, due to the observational nature of this study, variation in anesthetic management and surgical technique may exist and may be difficult to control. However, anesthesia events including extubation techniques as well as surgical skill as stratified by experience were recorded. Fourth, there were many possible predictor variables and hence development of model required variable selection to avoid redundancy and over specification of the model. Fifth, the 17

26 anesthesia related factors are correlated, and as the multi-colinearity increases so does the standard error. To avoid this the final model was built first with a univariate analysis, then with a multivariate based on p value of 0.24 from the univariate, and then finally the multivariate based on p value of 0.05 from the first multivariate model. The model was tested to be a good fit. Last, bleeding which is one of the other major cause of unanticipated ER visits could not be studied with the current predictors. CONCLUSIONS Our study identified preoperative blood pressure, intraoperative dexamethasone, preexisting conditions like URI, neurologic comorbidity, failure to thrive, and use of home medications as risk factors that predict unanticipated ER visits following tonsillectomy. These important and easily measurable variables provide valuable information for optimal management of patients younger than 17 years of age undergoing tonsillectomy. These risk factors should be explored during the perioperative care of patients undergoing tonsillectomy. Patients may benefit by using a lower dose of dexamethasone. Appropriate patient or family counseling about the risk associated with preexisting neurologic disorders and failure to thrive should be a part of perioperative care. 18

27 TABLES Table 1: Demographic Data Variable Results Age (years) 6.51 ± 3.49 Gender Male/Female (n) 214/254 BMI (kg/m 2 ) ± 4.43 Diagnosis (%) Adenotonsillar hypertrophy Severe OSA Recurrent infections Surgery type (%) Tonsillectomy and Adenoidectomy Tonsillectomy Race (%) Caucasians Non Caucasians Respondent (%) Mother Father Others Planned postoperative disposition (%) Outpatient 23-hour admission ASA (%) I II III Home medications (%) Yes No Data in Mean and SD unless specified. ASA = American Society of Anesthesiology; OSA = Obstructive Sleep Apnea. 19

28 Comorbidity (%) Present Absent Variables Table 2: Data on comorbidity Results Total number of comorbidities (n) 1/2/3/4/5/6/ /9.01/2.15/0.43/0.64/0.21/0.21 Respiratory comorbidity (%) Yes No Cardiovascular comorbidity (%) Present Absent Endocrine comorbidity (%) Present Absent Neurologic comorbidity (%) Present Absent Musculoskeletal comorbidity (%) Present Absent Metabolic comorbidity (%) Present Absent Genetic comorbidity (%) Present Absent Hematologic comorbidity (%) Present Absent

29 Table 3: Data on sleep apnea Variables Yes No Patient had polysomnography (%) Was Pediatric Sleep Questionnaire completed? (%) Snore more than half the time? Always snore? Snore loudly? Have 'heavy' or loud breathing? Have trouble breathing or struggle to breathe? Seen your child stop breathing during the night? Tend to breathe through the mouth during the day? Have a dry mouth on waking up in the morning? Occasionally wet the bed? Wake up feeling un-refreshed in the morning? Have a problem with sleepiness during the day? Has a teacher/supervisor commented that your child appears sleepy during the day? Have difficulty waking up in the morning? Wake up with headaches in the morning? Did your child stop growing at a normal rate at any time since birth? Is your child overweight? Does not seem to listen when spoken to directly Has difficulty organizing tasks and activities Is easily distracted by extraneous stimuli Fidgets with hands or feel or squirms in seat Is 'on the go' or often acts as if 'driven by a motor' Interrupts or intrudes on others (e.g. butts into conversations or games) Obstructive sleep apnea (%) 56.5 All data in percentages. 21

30 Table 4: Preoperative risk factors and hemodynamics Variables Results Presence of upper respiratory infections (%) Yes No Prematurity (%) Yes No Born at how many weeks ± 1.57 Failure to thrive (%) Yes No Child previous surgery (%) Yes No Heart rate (beats per minute) ± Oxygen saturation (%) ± Respiratory rate (beats per minute) ± 3.00 Systolic blood pressure (mm Hg) ± Diastolic blood pressure (mm Hg) ± 8.56 Data in Mean and SD unless otherwise specified. 22

31 Table 5: Data on premedication Variables Did the patient take a premedication? (%) Yes No Drug (name) (%) Albuterol Acetaminophen Midazolam Other Dose (of each drug) Albuterol Acetaminophen Midazolam Other Did the patient receive another pre-medication? (%) Yes No Drug (name) (n) Midazolam Other Dose (of each drug in mg) Midazolam Other The drug doses are in Mean and SD. Other data are percentages. Results ± ± ± ±

32 Table 6: Data on intraoperative anesthesia variables Variables Results Induction (%) Inhalational Intravenous Intraoperative fentanyl (mcg) ± Intraoperative morphine (mg) 2.70 ± 1.80 Intravenous acetaminophen (mg) (%) Administered Not administered Dose of intravenous acetaminophen ± Other intraoperative analgesic used (%) Yes No Hydromorphone (mg) 0.02 ± 0.14 Intraoperative Antiemetic (%) Yes No Ondansetron (mg) 2.64 ± 1.55 Dexamethasone (mg) 9.72 ± 3.56 Total dose of intraoperative Propofol (mg) ± Extubated in operating room (%) Yes No Extubated status (%) Awake Deep Experience of surgeon (%) Attending Fellow Resident Experience of Anesthesia staff (%) Attending 100% supervision CRNA Fellow 6.14 Resident 0 Other 1.82 Duration of surgery (minutes) ± 7.16 Data in Mean and SD unless otherwise specified. CRNA = Certified Registered Nurse Anesthesiologist. 24

33 Table 7: Data on other intraoperative medications Variables Other relevant medications (%) Yes No Medication (Name) (%) Albuterol Anectine Chloridine Cisatracurium Dilaudid Glycopyrolate Lidocaine Marcaine Precedex Rubinul Sodium chloride Succihylcholine Vercronium Robinul Succinylcholine Dose Albuterol Anectine Chloridine Cisatracurium Dilaudid Glycopyrolate Glycopyrrolate Lidocaine Marcaine Precedex Rubinul Sodium chloride Succinylcholine Vercronium Robinul Succinylcholine Results ± ± ± ± ± 7.07 The drug doses are in Mean and SD. Other data are percentages. The drug names may be brand names and are mentioned as derived from the database. 25

34 Table 8: Data on intraoperative adverse events Complication Yes No Intraoperative adverse events to Anesthesia Desaturation to < 90% Airway obstruction Difficult mask ventilation Difficult intubation Bronchospasm Delayed emergence from anesthesia

35 Table 9: Data on post anesthesia care unit related variables Variables Results PACU extubation (%) 6.22 Total duration of sleep in PACU ± Analgesic administered in PACU (%) Yes No Name of analgesic in PACU (%) Tylenol Fentanyl Morphine Acetaminophen Dose Tylenol Fentanyl Morphine Acetaminophen Route (%) Intravenous Oral ± ± ± ± Number of IV Analgesic Interventions in PACU (%) 0/1/2/3/4 3.32/65.68/26.20/2.58/1.85 Total Number of Analgesic Interventions in PACU (%) 1/2/3/4/ /32.10/2.95/1.85/0.37 Nausea 4.94 Vomiting 6.24 Antiemetic Administered in PACU (%) 1.72 Drug (Name) (%) Ondansetron Phenergan Zofran Dose Ondanestron Phenergan Zofran ± 2.06 Another Antiemetic Administered in PACU (%) 0 Benadryl 0 The drug doses are in Mean and SD. Other data are percentages. PACU = Post Anesthesia Care Unit. 27

36 Tables 10: Univariate analysis of the 3 day and 3 week preoperative categorical variables Variables 3 days 3-weeks Respondent mother Surgery by month Gender Diagnosis Adenotonsillar hypertrophy Obstructive sleep apnea Recurrent infection Snoring Restless sleeper Others Caucasian vs. non white race Planned disposition ASA Comorbidities (Yes/No) Comorbidities (Number) Respiratory Cardiovascular Endocrine Neurological Musculoskeletal Metabolic Genetic Hematological Others Sleep study (PSG) URI Prematurity Prematurity (per weeks) Failure to thrive Premedication ASA = American Society of Anesthesiologists; PSG = Polysomnography; URI = Upper Respiratory Infection 28

37 Table 11: Univariate analysis of the 3 day and 3 week categorical intraoperative and postoperative variables Variables 3 days 3-weeks Induction Intravenous acetaminophen Intraoperative antiemetic Other intraoperative medication Extubated in operating room (Awake vs. Deep) Intraoperative adverse events Desaturation < 90% Airway obstruction Difficult mask ventilation Difficult intubation Experience of surgeon Experience of anesthesia Staff PACU extubation PACU analgesia Name of analgesic PACU analgesic any time 1.0 Number of analgesic intervention in PACU Total number of analgesic intervention in PACU PACU nausea PACU vomiting PACU antiemetic Increased work of breathing Significant desaturation < Laryngospasm/ airway obstruction Airway instrumentation Use of drugs PACU discharge prolonged Inadequate pain control PACU Nausea and vomiting Oxygen requirement PACU sedation PACU = Post Anesthesia Care Unit 29

38 Table 12: Univariate analysis of the 3 day and 3 week continuous variables Variables 3 days 3-weeks Age Weight Height Body mass index Heart rate Saturation <0.001 < Respiratory rate Systolic blood pressure Diastolic blood pressure Intraoperative ondansetron Intraoperative dexamethasone Intraoperative propofol Intraoperative fentanyl Intraoperative morphine Duration of surgery Duration of sleep in PACU <0.001 < PACU discharge < PACU stay in minutes Intraoperative acetaminophen dose PACU = post anesthesia care unit. 30

39 Table 13: Logistic Regression Model Results for unanticipated emergency room visits within 3-days after tonsillectomy with p value less than or equal to 0.24 on univariate analysis. Odds Ratio 95% Confidence Limits P value Age Heart rate Systolic blood pressure Diastolic blood pressure Intraoperative dexamethasone Neurologic comorbidity (Yes vs. No) Genetic comorbidity (Yes vs. No) Upper respiratory infection (Yes vs. No) Failure to thrive (Yes vs. No) PACU analgesia (Yes vs. No) PACU increased breathing (Yes vs. No) ASA II vs. I ASA III vs. I ASA = American Society of Anesthesiology; PACU = Post Anesthesia Care Unit 31

40 Table 14: Logistic Regression Model Results for unanticipated emergency room visits within3-days after tonsillectomy with p value less than or equal to 0.05 in the first multivariate logistic model Odds Ratio 95% confidence limits P value Diastolic blood pressure Intraoperative dexamethasone Neurologic comorbidity (Yes vs No) Failure to thrive (Yes vs. No)

41 Table 15: Logistic Regression Model Results for unanticipated emergency room visits within 3-weeks after tonsillectomy with p value less than or equal to 0.24 on univariate analysis Odds Ratio 95% Confidence Limits P values Age Heart rate Systolic blood pressure Diastolic blood pressure Intraoperative dexamethasone Recurrent infections (Yes vs. No) Snoring (Yes vs. No) ASA II vs. I ASA III vs. I Neurologic comorbidity (Yes vs. No) Total comorbidity Neurologic comorbidity (Yes vs. No) Home medications (yes vs. No) Sleep study (Yes vs. No) Upper respiratory infection (Yes vs. No) Prematurity (Yes vs. No) Failure to thrive (Yes vs. No) Premedication (Yes vs. No) Intraoperative acetaminophen (Yes vs. No) PACU nausea (Yes vs. No) ASA = American Society of Anesthesiology; PACU = Post Anesthesia Care Unit 33

42 Table 16: Logistic Regression Model Results for unanticipated emergency room visits within 3-weeks after tonsillectomy derived from the previous model Estimate 95% confidence limits P values Systolic blood pressure Intraoperative dexamethasone Home medications Yes vs. No Upper respiratory infections Yes vs. No 34

43 Table 17: Secondary Outcome data on post anesthesia care unit complications Variables Yes No Increased work of breathing (%) Significant desaturation (%) Lowest percentage oxygen saturation on room air 70 ± Oxygen restarted for desaturations (%) Duration of O 2 desaturation < 80% in PACU requiring O2 (minutes) 95 ± Laryngospasm/airway obstruction requiring CPAP (%) Airway instrumentation (%) Nasal trumpet Endotracheal intubation Use of bronchodilators, racemic epinephrine, diuretics, naloxone (%) Time to meet PACU discharge criteria (minutes) ± PACU discharge prolonged (%) Inadequate pain control (%) Nausea and vomiting requiring treatment (%) Oxygen requirement to maintain a saturation of >94% (%) Excessive Sedation (%) Respiratory depression (%) Non-medical reasons (%) Duration of PACU stay (minutes) ± Discharged to (%) Floor Home Intensive care unit Did the patient have any adverse events? Data in Mean and SD. CPAP = Continuous Positive Airway Pressure; O 2 = Oxygen; PACU = Post Anesthesia Care Unit