Who Is at Risk for Postdischarge Nausea and Vomiting after Ambulatory Surgery?

• Postoperative nausea and vomiting (PONV) can be predicted with a simplified risk score; however, there is no simple model to predict post-discharge nausea and vomiting (PDNV)

• Depending on the number of the following factors, i.e. , female gender, age <50 yr, a history of nausea or vomiting, and opioid administration or nausea in the postanesthesia care unit, the patient's risk for PDNV can be predicted as 10%, 20%, 30%, 50%, 60%, or 80%

AMONG the millions of patients undergoing surgery with general anesthesia each year, many suffer from postoperative nausea, vomiting, or both (PONV).1,–,3Severe nausea can be so draining and debilitating that patients have rated it as more serious than postoperative pain.4Vomiting increases the risk of pulmonary aspiration of gastric contents and suture dehiscence, and may even lead to esophageal rupture, subcutaneous emphysema, and bilateral pneumothoraces.5,–,7In addition, PONV can delay patient discharge from the postanesthesia care unit (PACU), and it is a leading cause of unexpected hospital admission after ambulatory surgery.8As a result, PONV has a considerable economic impact on the U.S. healthcare system.9Current consensus guidelines recommend that the use of prophylactic antiemetics be tailored to the patient's risk of developing PONV.10,11 

The patient's risk of developing PONV can be estimated using a predictive model like our simplified PONV risk score (also known as the Apfel score), which was developed in European inpatients undergoing balanced inhalational anesthesia. According to this score, the risk factors (female gender, history of motion sickness and/or PONV, nonsmoking, and use of postoperative opioids) are each assigned a value of 1, and the incidence of PONV associated with 1, 2, 3, and 4 risk factors is 10%, 21%, 39%, 61%, and 79%, respectively.12 

Ambulatory surgery procedures are generally less invasive and less extensive than inpatient procedures. Consequently, they entail less exposure to emetogenic inhalational anesthetics and perioperatively administered opioids,13which may lead to a lower incidence of nausea and/or vomiting in the PACU. However, nausea and vomiting may also occur after the ambulatory patient has left the hospital. This postdischarge nausea and/or vomiting (PDNV) may be particularly hazardous for ambulatory surgery patients because they no longer have immediate access to fast-onset intravenous antiemetic rescue medication, and they may be unable to tolerate oral medication. In fact, a U.S. study of 154 patients undergoing ambulatory surgery under general anesthesia reported that 35% of patients were significantly distressed by PDNV.14 

With more than 60% of surgeries in the United States now performed on an ambulatory basis, accounting for millions of procedures with general anesthesia annually, the true incidence and risk factors of PDNV warrant closer scrutiny.15We sought to measure the incidence of PONV among ambulatory surgery patients during two distinct periods: in the PACU and outpatient facility before discharge, an outcome we defined as PONV, and after discharge, an outcome we defined as PDNV. It should be noted that both Phases I and II are included in our definition of the PACU period. In addition, we sought to characterize risk factors and/or protective factors, and develop, simplify, and validate a risk score for PDNV that would help clinicians tailor prophylactic regimens to at-risk patients before they are discharged from the hospital.

With the approval of the local institutional review boards of 12 ambulatory surgery centers in the United States (Medical College of Georgia, Augusta, GA; Brigham & Women's Hospital, Boston, MA; University of Virginia, Charlottesville, VA; St. Luke's Episcopal Hospital, Houston, TX; The University of Texas MD Anderson Center, Houston, TX; University of Kansas Medical Center, Kansas City, KS; University of Kentucky, Lexington, KY; Loma Linda University, Loma Linda, CA; University of Louisville, Louisville, KY; University of California – San Francisco (UCSF) Ambulatory Surgery Center, San Francisco, CA; UCSF Medical Center at Mt. Zion, San Francisco, CA; Mayo Clinic Scottsdale, Scottsdale, AZ), 2,493 adult patients scheduled for an elective outpatient surgical procedure during general anesthesia requiring tracheal intubation or a laryngeal mask airway gave their written informed consent to participate in this prospective cohort study from July 16, 2007, to August 28, 2008. Of those patients, 180 were excluded before surgery because they no longer met the eligibility criteria (e.g. , because of cancellation of surgery or conversion from planned general anesthesia to sedation); 75 patients were admitted to the center overnight (e.g. , because they had obstructive sleep apnea or they lived too far away to go home late in the evening); 16 had medical complications (e.g. , they remained intubated and ventilated overnight); and 52 patients could not be reached for postoperative follow-up. This analysis is based on data from the 2,170 patients who completed the study.

After informed consent was obtained, relevant preoperative, intraoperative, and postoperative data were collected on standardized forms by trained study personnel. As participants in a cohort study, all patients received standard of care (including use of prophylactic antiemetics perioperatively and postdischarge) according to the patients' local clinical care teams. The patients' experience of nausea and/or vomiting was assessed 30, 60, and 120 min after surgery by trained study personnel. At discharge from the hospital, all patients were instructed to enter their experiences of nausea and/or vomiting into a standardized diary. Information entered into the diary was obtained during a telephone interview on the afternoon or evening of the first and second days following surgery, i.e. , at least 24 and 48 h after emergence from anesthesia.

Nausea was measured using the clinically standard, 11-point verbal rating scale, for which 0 represents “no nausea” and 10 represents “worst nausea imaginable.” Vomiting was quantified as the number of emetic episodes occurring at least 1 min apart during a given time interval. Severe nausea was defined as nausea of 7 or greater on the verbal rating scale, and severe vomiting was defined as three or more emetic episodes during a given time interval. Retching was recorded separately but included with vomiting in the analysis because it involves the same reflex, the only difference being that no gastric content is expelled.

The primary outcome was the proportion of patients with nausea and/or vomiting and/or retching after discharge from the hospital until 48 h after emergence from anesthesia (PDNV). Secondary outcomes were the: (A) proportion of patients with vomiting and/or retching after discharge from the hospital; (B) proportion of patients with nausea after discharge from the hospital; (C) proportion of patients with postoperative nausea and/or vomiting and/or retching in the PACU (PONV); (D) severity of nausea before and after discharge from the hospital; and (E) severity of vomiting before and after discharge from the hospital.

We expected about 20% of patients to develop PDNV, and the odds ratios (ORs) of risk factors to be in the range of about 2.0.12,16To have 80% power at a two-sided P  of 0.05 would require 1,486 patients.17However, after adjustment for a multiple correlation coefficient of R²= 0.25 with other covariates, a total of 1,982 patients would be required.17We planned to enroll approximately 2,000 patients and, in order to account for some patients not completing the study, we planned to obtain consent from about 20% more patients, for a total of approximately 2,400 patients.

After all data were verified and the database was locked, the following steps were taken:

1. Raw incidences and unadjusted ORs according to outcomes and time intervals were calculated.

2. We used data from all but the highest enrolling non-UCSF center as our development cohort. Logistic regression analyses were applied to identify independent predictors for nausea and vomiting for both time intervals, i.e. , in the PACU and postdischarge. We investigated a wide range of independent predictors, primarily those that have been shown to be associated with PONV as well as others that have been suspected to influence nausea and/or vomiting: patient-specific variables (female gender, young age, nonsmoking, history of PONV, history of motion sickness, concomitant medications, preexisting diseases); intraoperative variables (use of volatile anesthetics, type and dose of narcotics, supplemental electrolyte infusion, antibiotics, duration of anesthesia, type of surgery, antiemetics including use of glucocorticoids); and postoperative variables (postoperative type and dose of narcotics, incidence of nausea and/or vomiting in the PACU, crystalloid infusion, time and type of first drink and first food intake, time and length of ride home, home activities, and postoperative pain). All narcotic doses were converted to morphine equivalent units.18 

3. In stepwise forward logistic regression analysis, coefficients of statistically significant independent predictors were calculated and used in the development of a predictive model for PONV in the PACU and for PDNV in ambulatory patients. In our analysis, we used an inclusion threshold of less than 0.05 and an exclusion threshold of more than 0.1, using a forward conditional test for removal. Subsequently, the area under the receiver operating characteristic curve (ROC-AUC) was determined.

4. Next, a simplified risk score was developed by assigning one point to each of the identified independent predictors. We tested the hypothesis that there would be less than a 0.05 absolute difference between the ROC-AUC of the simplified score and that of the coefficient-based prediction model developed in the previous step. It is important to note that the 0.05 absolute difference was chosen as a threshold of clinical relevance, not statistical significance.

5. To validate the simplified score, we determined the ROC-AUC with the data of the highest-enrolling non-UCSF center, and defined a valid score as having an ROC-AUC within 0.05 of that for the development dataset. Furthermore, we plotted corresponding incidences for any nausea, moderate, severe nausea, and for any vomiting and severe vomiting.

6. In addition, to explore whether the use of a new simplified PDNV risk score could significantly improve the clinician's ability to predict PDNV compared with established PONV risk scores, we determined the ROC-AUC of the highest-enrolling non-UCSF center for both the simplified PDNV score and the simplified PONV score. We defined an absolute difference of 0.05 as a clinically relevant improvement in prediction.

Analyses were conducted using SPSS version 19 (SPSS Inc., Chicago, IL) and STATA Intercooled version 10 (StataCorp LP, College Station, TX).

The average age of outpatients studied was 49.5 yr; 64.7% were women, 84.8% were nonsmokers, and 29.3% had a history of PONV (table 1). The four largest surgical groups were general surgery (20.3%), gynecological surgery (11.0%), knee arthroscopy (10.7%), and breast surgery (10.3%), together accounting for more than 50% of all surgeries conducted. Although none of the centers used propofol-based total intravenous anesthesia, some patients received propofol doses of 100–400 mg (n = 752) and more than 400 mg (n = 130) as an additional infusion or as boluses. Among all patients, 66.4% received sevoflurane and 77.4% received a prophylactic serotonin antagonist; all but two patients who received a prophylactic serotonin antagonist received ondansetron. Furthermore, 749 patients (34.5%) and 262 (12.1%) received two and three intraoperative antiemetics, respectively.

1. In the PACU, 19.9% of patients had nausea, 3.9% had vomiting, and 20.7% had nausea and/or vomiting (fig. 1, table 2). After discharge, 36.6% had nausea, 11.9% had vomiting, 37.1% had nausea and/or vomiting, 13.3% had severe nausea, and 5.0% had severe vomiting (fig. 1).

2. In the development dataset (n = 1,913), stepwise forward logistic regression analysis showed that female gender, age less than 50 yr, history of PONV, duration of surgery more than 1 h, 125 or more mcg fentanyl, ondansetron, arthroscopy, laparoscopy, and opioids administered in the PACU were statistically significant independent predictors for nausea and/or vomiting in the PACU (table 3, fig. 2). After discharge, i.e. , for PDNV, statistically significant independent predictors were female gender, age less than 50 yr, history of PONV, opioids in the PACU, and nausea in the PACU, but not nonsmoking status or ondansetron (table 4). Risk factors (e.g. , history of motion sickness or migraine, American Society of Anesthesiologists physical status, drinking status, adjuvant peripheral nerve block) that were not statistically significant are not listed in tables 3and 4.

3. During the stepwise forward logistic regression analysis, five factors were shown to predict the patient's risk of PDNV. According to this analysis, the patient's risk for PDNV could be estimated by p =1/(1+e  ^−z), in which P  is the probability that PDNV will occur and z = 0.43 (if female) + 0.77 (if age less than 50 yrs) + 0.41 (if history of PONV) + 0.66 (if opioids were needed in the PACU) + 1.14 (if nauseated in the PACU) − 2.42. The prediction model based on these coefficients had an ROC-AUC (95% CI) of 0.737 (0.715 to 0.759; fig. 3).

4. A simplified risk score in which each factor counted as one point led to an ROC-AUC of 0.706 (0.681 to 0.730; fig. 3). The absolute difference between 0.737 and 0.706 was less than the predefined clinically relevant absolute difference of 0.05. Furthermore, the ROC-AUC of the simplified PONV score – which was previously developed for inpatients in Europe – was only 0.630 (0.603 to 0.656). According to this simplified PDNV score, when zero, one, two, three, four, or five of these five risk factors was present, the associated PDNV incidences were 10.9%, 18.3%, 30.5%, 48.7%, 58.5%, or 79.7%, respectively. In addition, when patients were grouped according to their predicted risk into six groups based on the five predictors, the calibration plot of the predicted and actual incidences of PDNV resulted in a calibration line having a slope of 0.942 and an intercept of 0.006 (fig. 4A).

5. In the validation cohort (n = 257), when patients were grouped according to their predicted risk into six groups, the calibration plot of the predicted and actual incidences of PDNV resulted in a calibration line having a slope of 1.075 and an intercept of negative 0.044 (fig. 4B). The ROC-AUC of the simplified risk score was 0.721 (0.657 to 0.785; fig. 5) in the validation cohort. Figure 6displays the incidences for any nausea, moderate, severe nausea, and for any vomiting and severe vomiting. Note that the ROC-AUC of the smaller validation dataset was higher than the ROC-AUC of the development dataset, and that the incidences differ somewhat from the PDNV incidences of the development dataset listed in point 4, which reflects typical random variation.

6. The ROC-AUC of the simplified PONV score in the validation cohort was 0.674 (0.607 to 0.741; fig. 5). With a sample size of 257, the validation cohort was too small to detect whether there was a statistically significant difference between the ROC-AUCs of the simplified PDNV score and the simplified PONV score. However, there was an absolute difference of 0.047 between the ROC-AUCs of the two risk scores, which was very close to our 0.05 threshold for clinical relevance.

In this large, multicenter cohort study, 37.1% of patients had PDNV, 13.3% had severe nausea, 11.9% had vomiting, and 5% had severe vomiting. These incidences were significantly higher than expected, especially given the patient and surgery profile of ambulatory procedures, as well as the relatively low incidence of PONV in the PACU (fig. 1, table 2). Considering that about one third of the approximately 35 million ambulatory surgeries performed in the United States annually use general anesthesia,15these findings translate into approximately 4.3 million patients experiencing PDNV every year.

The high incidence shown in our study is almost identical to the 35% incidence reported in a study of 154 ambulatory surgery patients conducted in the United States more than a decade ago.14Although the incidence of PDNV in a Canadian study was only 9.1%,19several factors may have contributed to this relatively low incidence. For example, PONV is triggered primarily by inhalational anesthetics and opioids used for general anesthesia,13and it is well-known that using a peripheral regional nerve block instead of general anesthesia significantly reduces the likelihood of PONV.20Although all patients enrolled in our study underwent general anesthesia, about half of the patients in the Canadian study received a regional nerve block or monitored anesthesia care. Furthermore, in the Canadian study, many patients who did receive general anesthesia underwent very brief procedures like dilation and curettage, and therefore received only intravenous propofol for maintenance instead of inhalational anesthetics.

The incidence and severity of PDNV after ambulatory surgery with general anesthesia appears to have been greatly underestimated, most likely because PONV in the PACU is less frequent and rarely severe for outpatients compared to inpatients. In our study, only 3.6% of outpatients had severe nausea and 0.2% had severe vomiting in the PACU, compared with 13.3% who had severe nausea and 5.0% who had severe vomiting postdischarge (fig. 1). Indeed, despite the high incidence of PDNV, only 4.4% of patients in our study received antiemetic prophylaxis acting long enough to prevent PDNV before they were discharged from the ambulatory care center – regardless of their risk for PDNV.

The risk factors for PONV in the PACU identified in this cohort study are consistent with those previously reported for inpatients.12,16,21,22Patient-specific independent predictors were female gender, age less than 50 yr, and history of PONV. Anesthesia- and surgery-specific factors were higher doses of intraoperative and postoperative opioids, duration of surgery more than 1 h, and a laparoscopic surgical approach. Although cholecystectomies were associated with the highest incidence of PONV in the PACU (table 2), the multivariable analysis suggests that this was because of the predominantly laparoscopic approach, so that type of surgery no longer reached statistical significance (table 3, fig. 2).

Although the patient-specific risk factors of female gender, age less than 50 yr, and history of PONV were predictors for both PONV in the PACU and PDNV, nonsmoking status was not an independent predictor for PDNV. The lower incidence of PONV in the PACU in smokers is probably not because of an acute antiemetic effect of nicotine. In fact, the use of the nicotine patch has been shown to increase, not prevent, nausea23; its use has never been associated with a reduced incidence of PONV.24Instead, smokers may have adapted to nicotine-induced and γ-aminobutyric acid-mediated increases of intrasynaptic dopamine release, and are thus likely to have relatively lower dopamine levels immediately after surgery. Similar to the antiemetic effect of dopamine receptor antagonists like metoclopramide,25reduced dopaminergic stimulation may have protective effects that disappear after patients are discharged and resume smoking.

Another difference between risk factors for PONV in the PACU and PDNV is that surgical approach was not statistically significant for PDNV. Considering that this statistical significance observed for PONV in the PACU came primarily from the increased incidence after laparoscopy, it is possible that increased arterial carbon dioxide and HCO³levels associated with laparoscopy may equilibrate in the PACU, so that this emetogenic effect is no longer relevant after patients are discharged.

The main difference between risk factors for PONV and PDNV was that patients who experienced nausea in the PACU had a 3-fold increased risk for PDNV.

Contrary to previous thoughts,26ondansetron is equally efficacious against nausea as it is against vomiting, with an OR of 0.6827,28similar to the 0.70 OR in this multicenter study. However, intraoperatively administered ondansetron did not reduce the risk of PDNV (fig. 2), most likely because of its short plasma half-life of about 3 h. In fact, patients who received ondansetron had a somewhat higher risk for PDNV, which almost looks like a rebound effect (fig. 2). Almost half of all patients received intraoperative dexamethasone (table 1). Glucocorticoids did not appear to reduce PONV in the PACU but significantly reduced PDNV (fig. 2).

Several studies have reported that propofol has an antiemetic effect at subhypnotic doses.29,–,31However, Scuderi et al.  were unable to demonstrate such effects at similar concentrations,32and Hvarfner et al.  suggest that the antiemetic effect of propofol is detectable only with concurrent sedation,33in which case the effect is similar to that of lorazepam. Furthermore, given the short half-life of propofol, and the similarity between total intravenous anesthesia and inhalational anesthesia in the late postoperative period,13total intravenous anesthesia does not appear to be a substitute for an effective antiemetic for PDNV.

The five statistically significant independent risk factors for PDNV were female gender, age less than 50 yr, history of PONV, opioids administered in the PACU, and nausea in the PACU. We used the coefficients of these factors to create a prediction model for PDNV with an acceptable discriminating power of an ROC-AUC curve of 0.74. However, for practical purposes, simplifying the calculation to one point for every risk factor present permits a risk prediction with an ROC-AUC curve of 0.72, which we consider clinically similar to that of the more complex calculation. To make the prediction model even simpler to remember, the incidence of PDNV was approximately 10%, 20%, 30%, 50%, 60%, or 80% when zero, one, two, three, four, or five of the independent risk factors were present, respectively. It is interesting that – even though the risk factors for nausea and for vomiting apparently were not identical – those factors also give a rough estimate of patients' risk for vomiting or severe nausea after being discharged home, which is about one third of their risk for PDNV in general. Moreover, the risk for severe vomiting after discharge can be estimated to be about half the risk for severe nausea or vomiting in general. Thus, the risk for severe vomiting is about one sixth of the risk for PDNV in general.

Because the size of the validation cohort was only 257 patients, future studies conducted by independent investigators at other centers will be needed to confirm the value of our newly developed PDNV score. However, considering that this simplified score was based on data from 12 centers across the United States, and that the ROC-AUC of the validation cohort was quite similar to that of the development cohort, it is fair to assume that this score will have a reasonable degree of external validity in other centers. Furthermore, because the absolute difference between the ROC-AUCs of the simplified PONV and PDNV scores was about 0.05, we believe that the use of a new risk score specific to outpatients is warranted, even though the validation cohort was not actually sufficiently powered to detect a statistically significant difference between the two simplified scores. However, a difference would not be surprising given that our simplified PONV score was developed in European inpatients for PONV through 24 h, whereas our PDNV score was developed in U.S. outpatients for PDNV through 48 h. Because of differences in length of the procedure, exposure to anesthetics, patient mobility and access to cigarettes, some of the predictors – likely the triggers – are different for PONV and PDNV.

Based on the results from the validation cohort, the use of our simplified PDNV risk score is useful to identify at-risk patients who are likely to benefit from long-acting prophylactic antiemetics like dexamethasone, aprepitant, palonosetron, and transdermal scopolamine, either alone or in combination. However, the efficacy of these agents for PDNV needs to be confirmed in future studies.

PDNV is a common and sometimes severe adverse outcome for ambulatory patients. By identifying the five most important independent predictors – female gender, age less than 50 yr, history of PONV, opioids administered in the PACU, and nausea in the PACU – we developed a new risk score for estimating the individual patient's risk of PDNV. The incidence of PDNV is approximately 10%, 20%, 30%, 50%, 60%, or 80% when zero, one, two, three, four, or five of these predictors are present, respectively. Clinicians whose patients undergo general anesthesia for ambulatory surgery might find this information useful when making decisions about the need for prophylactic antiemetics before patients are discharged from the hospital.

The authors thank all of the investigators and research coordinators for their important contributions. Specifically, the authors thank Alice Stader, M.A. (Clinical Research Manager, Department of Psychiatry, University of California San Francisco, San Francisco, California), for her diligent center coordination; Edmond Eger, M.D. (Professor Emeritus, Department of Anesthesia & Perioperative Care, University of California San Francisco), for his valuable comments on the manuscript; and Susan Eastwood, ELS(D), (1943–2010), for her invaluable editorial assistance. The authors also thank Suzanne DeVandry, Ph.D. (Senior Clinical Scientists, Merck and Co., Inc., North Wales, Pennsylvania), for supporting this study.