Applicability of risk scores for postoperative nausea and vomiting in adults to paediatric patients

L. H. J. Eberhart1,*, A. M. Morin1, D. Guber2, F. J. Kretz2, A. Schäuffelen3, H. Treiber3, H. Wulf1 and G. Geldner1

1 Department of Anaesthesiology and Intensive Care Medicine, Philipps University, Marburg, 2 Olga-Hospital, Children's Hospital Stuttgart and 3 Ambulatory Surgical Centre Söflingen, Ulm, Germany

* Corresponding author: Department of Anesthesiology and Critical Care Medicine, Philipps University Marburg, Baldingerstrasse, 35043 Marburg, Germany. E-mail: eberhart{at}mailer.uni-marburg.de

Accepted for publication March 25, 2004.


    Abstract
 Top
 Abstract
 Introduction
 Material and methods
 Results
 Discussion
 References
 
Background. Scores to predict the occurrence of postoperative vomiting (PV) or nausea and vomiting (PONV) are well established in adult patients. The aim of this survey was to evaluate the applicability of risk scores developed and tested in adult patients in 983 paediatric patients (0–12 yr) undergoing various surgical procedures.

Method. The predictive properties of five models were compared with respect to discriminating power (measured by the area under a receiver operating characteristic curve) and calibration (comparison of the predicted and the actual incidences of the disease by weighed linear regression analysis).

Results. The cumulative incidence of PV was 33.2% within 24 h. The discriminating power was low and insufficient in all models tested (0.56–0.65). Furthermore, the predicted incidences of the scores correlated only vaguely with the actual incidences observed.

Conclusion. Specialized scores for children are required. These might use the history of PV, strabismus surgery, duration of anaesthesia ≥45 min, age ≥5 yr and administration of postoperative opioids as independent risk factors.

Keywords: anaesthesia, paediatric ; model, mathematical ; vomiting and nausea, incidence ; vomiting and nausea, patient factors ; vomiting and nausea, surgical factors ; vomiting, patient factors


    Introduction
 Top
 Abstract
 Introduction
 Material and methods
 Results
 Discussion
 References
 
Life-threatening complications associated with anaesthesia have become very rare. This safety record has encouraged anaesthetists to focus attention on minor morbidity. Of these symptoms, postoperative nausea and vomiting (PONV) is the ‘big little problem’ after general anaesthesia.1 PONV may decrease satisfaction, increase the use of resources, including medical and nursing care, i.v. fluids, drugs and other supplies.25 Furthermore, in the ambulatory setting PONV is a major cause of unanticipated admission.6

The incidence of this distressing problem can be reduced by using an i.v. anaesthetic technique instead of inhalation agents and by administering antiemetics prophylactically. However, routine efforts to prevent PONV are not indicated because of the potential for adverse effects and increased costs, and the lack of evidence that patient satisfaction is affected.7

For these reasons, tools to predict an increased risk of developing nausea and vomiting are certainly useful in clinical practice. Several scores have been developed for adults in the past, and external validation has demonstrated acceptable predictive properties.8 9 However, such an evaluation has not been performed in paediatric patients. Therefore, in this prospective observational survey, five published risk scores1014 were compared with respect to their ability to predict postoperative vomiting (PV) in children.


    Material and methods
 Top
 Abstract
 Introduction
 Material and methods
 Results
 Discussion
 References
 
Data from 1150 children (0–12 yr old) included in this prospective survey were collected during a period of 22 months at a University hospital, a community children's hospital and an outpatient surgical centre. The local ethics committee approved the study and the parents gave informed consent and were asked about the susceptibility of the child with regard to motion sickness when driving in a car and the occurrence of PV when previous surgery had been performed. Older children answered these questions themselves and were asked if they smoked regularly. Children were fasted 2–4 h before surgery from clear fluids and at least 6 h from milk and solids. All received oral premedication with midazolam. The anaesthesia technique was not standardized but carried out according to the local standards. After surgery, the first oral intake was allowed depending on the length and site of operation. In most cases, this was within the first 4 h after surgery. All children received a non-opioid analgesic (paracetamol, metamizole and/or diclofenac) intraoperatively or immediately after operation. Intravenous piritramid or oral codeine was given on demand at the discretion of the anaesthetist or the nursing staff to treat postoperative pain. These children were classified as receiving postoperative opioid administration.

After surgery, vomiting or retching was assessed in the postanaesthetic care unit (PACU) by specially instructed nurses or anaesthetists. The children and/or their parents were interviewed 24 h after surgery. Additionally, all medical recordings were screened and the nursing staff were asked not to miss an emetic episode. The parents of patients having day-care surgery were interviewed by telephone using a structured interview on the first postoperative day. The main end-point of the survey was the cumulative incidence of PV within 24 h after surgery.

Several risk models to predict PV or PONV have been published (see references 9 and 15 for an extensive overview). Of these models, a selection was made covering five models that were published as a full paper and allow calculation of an exact risk of PV or PONV during the first 24 h after surgery.1014 Table 1 provides an overview of these identified models. PV was chosen as the main end-point of the survey because nausea is a subjective phenomenon, and the smaller child often may not be able to describe it.16


View this table:
[in this window]
[in a new window]
 
Table 1 Overview of the five risk models evaluated in this survey. In the rightmost column the areas under the ROC curve for children are listed. The higher the value the better the discriminating power of the risk model to forecast whether a child will vomit

 
The five scores were judged according to the practicability, e.g. ease of use in daily practice, and the predictive properties (discriminating power and calibration).

Discriminating power
For each patient, two probabilities of PV11 12 and four probabilities of PONV10 1214 were calculated. Using these predicted probabilities and the actual incidences of PV, a receiver operating characteristic (ROC) curve was drawn. The ROC curve can be constructed correlating true- and false-positive rates (sensitivity and 1–specificity, respectively) for a series of cut-off points for a test. Here the cut-off point is a predicted risk. In the case of a child exceeding the cut-off point, it is classified as suffering from PV. The area under the ROC curve (AUC) represents the probability that a random pair of test results will be ranked correctly as to their disease state.17 Theoretically, a 45° bisector is a score that predicts no better than a random guess. Thus, the area under this ‘random score’ would be 0.5. A score predicts significantly better than randomly when the lower limit of the 95% confidence interval for the AUC exceeds 0.5. All statistical calculations were performed using MedCalc (MedCalc Software, Mariakerke, Belgium).

Calibration curves
The calibration curves were constructed by correlating the predicted incidence of PV (or PONV) with the actual incidences of PV. To allow comparison of the different scores, the predicted risk of PONV or PV was clustered into four or five groups. The simplified scores10 12 provided these risk groups without further calculation. For the other three models,11 13 14 the predicted risks were categorized into five groups: 0–20%, 20–40%, 40–60%, 60–80% and 80–100%. This approach resulted in unequal numbers of children in each of the five groups, since most were predicted to have a low to moderate risk. Therefore, weighted linear regression analysis was used to compare the predicted and real incidences (Fig. 1).



View larger version (26K):
[in this window]
[in a new window]
 
Fig 1 (AF) Calibration curves of the six prediction models drawn using weighted linear regression analysis. The relative number of patients in a single risk group (predicted incidence of PV 0–20%, 20–40% ...) is characterized by the size of the plots.

 
To investigate the differences in prediction between the different scores, we performed an additional logistic regression analysis. For this purpose, all risk factors used in any of the available models and the types of surgery were included in a backward logistic regression procedure using the maximum likelihood method. Continuous data (age, duration of anaesthesia and surgery) were dichotomized using a cut-off value that provided the optimal discrimination in a univariate analysis. The significant risk factors remaining in the regression model are presented with their odds ratio (OR) and the 95% confidence interval. The goodness of the fit of the regression model was judged using Nagelkerkes's R2. The latter calculations were performed using SPSS 11.0 for Windows.


    Results
 Top
 Abstract
 Introduction
 Material and methods
 Results
 Discussion
 References
 
Patient characteristics and surgical data are shown in Tables 2 and 3. Of the 1150 patients included in the survey, 128 children receiving total i.v. anaesthesia with propofol or prophylactic antiemetics, including corticosteroids, had their data withdrawn. A further 39 patients were lost to follow-up or were excluded from analysis due to incomplete data. Of the 983 children included in the final analysis, 326 suffered from PV within the first 24 h postoperatively (33.2%; 95% confidence interval 30.2–36.2%). Table 1 shows the values of the area under the ROC curve for the models that were evaluated. Regardless of the predicted end-point (PV or PONV), all five scores predict significantly better than could be expected from chance, and even in the worst case (assuming the lower limit of the 95% confidence interval) the AUC was above 0.5.


View this table:
[in this window]
[in a new window]
 
Table 2 Patient characteristics. Normally distributed data are shown as arithmetic means and standard deviation, otherwise as median and 25th and 75th percentiles (in brackets). Dichotomous data are expressed as absolute and relative frequencies

 

View this table:
[in this window]
[in a new window]
 
Table 3 Surgery performed in the 983 children used to validate the risk models for PONV and PV respectively. Data are absolute and relative frequencies (in brackets)

 
Comparing the discriminating power of the different scores, the best results were achieved with the models published by Koivuranta and colleagues12 and Sinclair and colleagues.14 Both had significantly higher AUC values than the scores from Palazzo and colleagues13 (P<0.001) and the two scores published by Apfel and colleagues10 11 (P<0.003). Other comparisons across the different scores did not show statistical differences.

The calibration curves (Fig. 1AF) compare the predicted with the actual incidences of PV/PONV. This type of analysis is useful when a group of patients rather than a single child is concerned. The curve is expressed by the slope and the offset, where y is the actual incidence of PV observed in the described paediatric population and x is the predicted incidence of PV or PONV calculated according to the risk score (Fig. 1).

Koivuranta's score to predict vomiting (Fig. 1E) shows the best calibration properties of all evaluated models. However, even with this risk score, the equation of the calibration curve yactual risk of PV=1.18xpredicted risk of PV+8 underestimates the actual risk by up to a quarter (e.g. the actual incidence of PV may be 80% when the predicted risk is 61%).

The explorative logistic regression analysis of potential relevant risk factors for the occurrence of PV in our paediatric population revealed that five variables were independent predictors of PV in children. These five risk factors are listed in Table 4 with their odds ratios and 95% confidence interval.


View this table:
[in this window]
[in a new window]
 
Table 4 Risk factors obtained in the present population of children by logistic regression analysis. The five risk factors account for approximately 31% of the variance within the data (Nagelkerkes's R2=0.314). As a result of univariate analysis, the incidences of PV are listed when the risk factor is present or not

 
Ophthalmic procedures, duration of anaesthesia >45 min, previous PONV or a positive history of motion sickness, and postoperative use of opioids were risk factors that were also used in risk scores for adults, and are thus responsible for the discriminating properties of the adult models in children. Female gender was removed at an early stage of the stepwise procedure as not being statistically significant. Young age (<6 yr) was a protective factor in children in contrast to adult patients.11 14


    Discussion
 Top
 Abstract
 Introduction
 Material and methods
 Results
 Discussion
 References
 
Prediction of PV or PONV is now widely accepted to be useful in clinical practice. Identifying patients with a high baseline risk allows goal-directed use of antiemetic measures that may not be indicated for routine practice. For instance, a recent consensus conference outlined that patients at low risk are unlikely to benefit from antiemetic prophylaxis and would be put at unnecessary risk from potential side-effects of antiemetics. Thus, prophylaxis should be reserved for patients at moderate to high risk of PONV.18

All risk scores that were evaluated in the present trial allow calculation of an exact risk. Previous external evaluations of these models using data from adult patients suggested that the discriminating power of the scores should exceed 0.65–0.7 before it can be judged as useful for clinical practice.8 9 Furthermore, it has been pointed out that it is of great importance that a risk model could be used easily in clinical practice to guarantee widespread use and acceptance by clinicians. In this context, two criteria are of great importance. First, there should be as few factors as possible that need to be remembered when the score is applied to a patient. Secondly, the recorded risk factors must be easily translatable into the patient's individual risk without complicated calculations. Only the score of Koivuranta and colleagues12 and the simplified score of Apfel and colleagues10 meet these requirements. To calculate both risk models, five factors must be explored. According to the number of factors present, the predicted risk can be determined (Table 1). For instance, using Koivuranta's score in adults, the expected incidence of PONV is 17% if no risk factor applies, and 18, 42, 54, 74 and 87% if one to five factors are present respectively.12 The other scores11 13 14 use a logit equation that cannot be calculated without the aid of a pocket calculator and thus hamper their clinical usefulness. The use of the model of Sinclair and colleagues14 is further complicated by the large number of factors (12) and the use of logical Boolean arithmetic.

Comparing the five available scores for predicting the probability of PV in paediatric patients undergoing general anaesthesia revealed that all models predict the occurrence of PV better than a random guess. This feature of a prognostic tool is called discrimination. It is represented by the area under a ROC curve that can be constructed correlating true- and false-positive rates (sensitivity and 1–specificity, respectively). In all evaluated models this value exceeds 0.5 (even if the worst scenario is assumed, e.g. the lower limit of the 95% confidence interval is used). This 50% chance is the pre hoc probability that a random pair of two children (one child who will vomit after surgery and the other one will not) will be correctly classified as to their disease state. However, though statistically significant, a model that improves this pre hoc probability from 50% to values between 56%13 and 65%14 is far from being clinically relevant, even though surveying the five risk factors is not very time-consuming, and no invasive or expensive diagnostic procedures are needed. The two simplified scores have a discriminating power of 0.5810 and 0.6112 respectively and are thus not useful in children despite their ease of use.

In addition to discriminating power, calibration is the other feature of a prognostic score. Calibration is defined as the fit of the predicted incidences of a group of patients with the actual incidence of an event. As seen in Fig. 1, there is evidence that an increased risk, as predicted by the different scores, correlates with the actual incidence of PV. However, calculating a linear equation indicates that the calibration of all models was far from optimal. An optimal fit of this curve would result in a slope of 1.0 with no offset (e.g. resulting in the linear equation y=1.0x+0). The slope of most evaluated models was flat between 0.51 and 0.64,10 11 13 14 indicating that these scores overestimate the increase in PV when more risk factors are present. Only the PV model of Koivuranta and colleagues had a slope greater than 1.0,12 resulting in systematic underestimation of the actual rate of PV. All scores had a positive offset of 8–30%, indicating that a notable incidence of PV was present, even though these scores predicted a low risk.

This means that extrapolation of the risk that has to be expected for children with a certain mix of risk factors is far from being exact. At best, this information can be used to prepare guidelines when antiemetic prophylaxis is advisable. For instance, if there is agreement that children who have a risk of PV of 50% or even higher should receive antiemetic prophylaxis, this should be performed in children who present with four or more of the risk factors identified in the simplified model of Koivuranta and colleagues (Table 1). Using the latter example, this decision criteria would result in a sensitivity of only 5% but the specificity would be 98%. This means that only a few children would be falsely classified as vomiters, and thus the likelihood that a child receives an antiemetic without indication would be very low. On the other hand, many children will be falsely classified as non-vomiters but will develop PV.

There are several reasons for the disappointing predictive properties when the scores developed for adult patients were applied to children. Some of the risk factors used in the available risk scores to predict PV or PONV in adults are difficult to assess or do not usually apply in children. For instance, smoking has been identified as a protective factor in several investigations.11 12 14 Among our paediatric patients there was only one 12-yr-old boy who admitted smoking regularly. However, we did not ask about passive smoking. Younger age is a risk factor identified in adults;11 14 this applies in reverse to children. In agreement with previous findings,19 our data demonstrate that toddlers are less susceptible to emetic stimuli than schoolchildren.19 The odds ratio (OR) for developing PV is 1.6 in children aged ≥6 yr compared with younger children. Furthermore, even in older children sex does not play a major role in the occurrence of PV.20 In our stepwise logistic regression analysis, gender was removed from the model early as not being significant.

Certain types of surgery have often been considered to be associated with an increased risk of PV or PONV. However, of the risk models that were evaluated in the present analysis, only that of Sinclair and colleagues included some of these types of surgery.14 Of these, only ophthalmic procedures were identified as relevant in our paediatric patients. Since more than 90% of these procedures were strabismus surgery, a second analysis with this specific procedure revealed that, in our population, strabismus surgery but not eye surgery per se was associated with an increased risk (odds ratio 3.6). This is comparable with the result of Sinclair's model, with which ophthalmic procedures had an odds ratio of 5.9. Obviously, this agreement is one reason why the Sinclair score provided the best discrimination power of all tested risk scores.

Despite these limitations of the scores initially developed for adults, there are also several parallels between risk factors in children and adults. One example is a positive history of PV after previous surgery or motion sickness, which was observed in all but one model for adults. Obviously, the incidence of patients who had undergone previous surgery is lower than in an adult population. Thus, children are more likely to be classified as having a negative history of PV than adult patients. Furthermore, a history of motion sickness may be difficult to elicit in smaller children. Nevertheless, in the explorative logistic regression analysis, a history of previous PV or motion sickness was identified as an independent risk factor present also in children (odds ratio 2.6).

The administration of postoperative opioids and a longer duration of anaesthesia are two further examples of risk factors present in children and in adults. As in other investigations,10 11 we did not differentiate between different kinds of opioid analgesics. This was done for practical reasons but also under the assumption that there is no relevant difference in the emetogenic effects of different opioids.21 Administration of postoperative opioids was associated with an odds ratio of 1.8. The duration of anaesthesia was an even stronger risk factor for PV in children. Although duration of anaesthesia and surgery are highly correlated, previous risk models uniformly use duration of surgery to describe a higher risk in adults.11 12 14 In our patients, 45 min was the optimal cut-off that achieved the best prediction. The odds ratio for this risk factor was 2.8.

Because of the ease of calculation, only the models of Koivuranta and colleagues and Apfel and colleagues are of potential clinical relevance. However, none of the scores discriminated sufficiently in our study. Thus, specialized scores for children are required. These future risk models should consider the history of PV and of motion sickness, strabismus surgery, duration of anaesthesia, school age and the administration of postoperative opioids as potentially highly relevant risk factors. Furthermore, additional attention should be paid to other environmental and genetic factors, e.g. passive smoking and the susceptibility of close relatives to PONV.


    References
 Top
 Abstract
 Introduction
 Material and methods
 Results
 Discussion
 References
 
1 Kapur PA. The big ‘little problem’. Anesth Analg 1991; 73: 243–5[ISI][Medline]

2 Cieslak GD, Watcha MF, Phillips MB, Pennant JH. The dose–response relation and cost-effectiveness of granisetron for the prophylaxis of pediatric postoperative emesis. Anesthesiology 1996; 85: 1076–85[CrossRef][ISI][Medline]

3 Patel RI, Davis PJ, Orr RJ, et al. Single-dose ondansetron prevents postoperative vomiting in pediatric outpatients. Anesth Analg 1997; 85: 538–45[Abstract]

4 Splinter WM, Rhine EJ, Roberts DJ. Vomiting after strabismus surgery in children: ondansetron vs propofol. Can J Anaesth 1997; 44: 825–9[Abstract]

5 Watcha MF, Bras PJ, Cieslak GD, Pennant JH. The dose–response relationship of ondansetron in preventing postoperative emesis in pediatric patients undergoing ambulatory surgery. Anesthesiology 1995; 82: 47–52[CrossRef][ISI][Medline]

6 Patel RI, Hannallah RS. Anesthetic complications following pediatric ambulatory surgery: a 3-year study. Anesthesiology 1988; 69: 1009–12[ISI][Medline]

7 Scuderi PE, James RL, Harris L, Mims GR. Antiemetic prophylaxis does not improve outcomes after outpatient surgery when compared to symptomatic treatment. Anesthesiology 1999; 90: 360–71[ISI][Medline]

8 Eberhart LHJ, Högel J, Seeling W, Staack AM, Geldner G, Georgieff M. Evaluation of three risk scores to predict postoperative nausea and vomiting. Acta Anaesthesiol Scand 2000; 44: 480–8[CrossRef][ISI][Medline]

9 Apfel CC, Kranke P, Eberhart LHJ, Roos A, Roewer N. Comparison of predictive models for postoperative nausea and vomiting. Br J Anaesth 2002; 88: 234–40[Abstract/Free Full Text]

10 Apfel CC, Läärä E, Koivuranta M, Greim CA, Roewer N. A simplified risk score for predicting postoperative nausea and vomiting. Anesthesiology 1999; 91: 693–700[ISI][Medline]

11 Apfel CC, Greim CA, Haubitz I et al. A risk score to predict the probability of postoperative vomiting in adults. Acta Anaesthesiol Scand 1998; 42: 495–501[ISI][Medline]

12 Koivuranta M, Läärä E, Snåre L, Alahuhta S. A survey of postoperative nausea and vomiting. Anaesthesia 1997; 52: 443–9[ISI][Medline]

13 Palazzo M, Evans R. Logistic regression analysis of fixed patient factors for postoperative sickness: a model for risk assessment. Br J Anaesth 1993; 70: 135–40[Abstract]

14 Sinclair DR, Chung F, Mezei G. Can postoperative nausea and vomiting be predicted? Anesthesiology 1999; 91: 109–18[ISI][Medline]

15 Maleck WH, Piper SN. Predictive models for postoperative nausea and vomiting. Br J Anaesth 2002; 88: 339–42[CrossRef]

16 Rose JB, Watcha MF. Postoperative nausea and vomiting in paediatric patients. Br J Anaesth 1999; 83: 104–17[Abstract/Free Full Text]

17 Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982; 143: 29–36[Abstract]

18 Gan TJ, Meyer T, Apfel CC, et al. Consensus guidelines for managing postoperative nausea and vomiting. Anesth Analg 2003; 97: 62–71[Abstract/Free Full Text]

19 Cohen MM, Cameron CB, Duncan PG. Pediatric anesthesia morbidity and mortality in the perioperative period. Anesth Analg 1990; 70: 160–7[Abstract]

20 Rowley MP, Brown TCK. Postoperative vomiting in children. Anaesth Intensive Care 1982; 10: 309–13[ISI][Medline]

21 Breitfeld C, Peters J, Vockel T, Lorenz C, Eikermann M. Emetic effects of morphine and piritramide. Br J Anaesth 2003; 91: 218–23[Abstract/Free Full Text]





This Article
Abstract
Full Text (PDF)
All Versions of this Article:
93/3/386    most recent
aeh221v1
E-Letters: Submit a response to the article
Alert me when this article is cited
Alert me when E-letters are posted
Alert me if a correction is posted
Services
Email this article to a friend
Similar articles in this journal
Similar articles in ISI Web of Science
Similar articles in PubMed
Alert me to new issues of the journal
Add to My Personal Archive
Download to citation manager
Disclaimer
Request Permissions
Google Scholar
Articles by Eberhart, L. H. J.
Articles by Geldner, G.
PubMed
PubMed Citation
Articles by Eberhart, L. H. J.
Articles by Geldner, G.