Risk scores to predict the occurrence of post-operative nausea and vomiting (PONV) are used increasingly to guide prophylactic antiemetic drug administration. Recent consensus guidelines seem to advocate the use of such models by stating that ‘the use of prophylactic antiemetics should be based on valid assessment of the patient's risk for post-operative vomiting or PONV’.1 In other words, antiemetic prophylaxis should be used only when the patient's individual risk is sufficiently high.
Although this recommendation seems logical at first sight, its clinical implementation is much less clear. First, there is no common consensus regarding the ideal threshold at which the risk for PONV is high enough [some anaesthesiologists are dissatisfied when the departmental incidence of PONV is higher than 10%, whereas others do not feel the need for additional therapy when it exceeds 40% or more (personal communication)]. Second, the optimal method of performing an individual risk assessment for the likelihood of PONV remains unclear.
One option for risk assessment and thus implementation of this recommendation is to apply one of the risk scores for PONV that have been developed in recent years.2–12 Obviously, the plethora of risk scores to predict PONV in adult patients shows that none of the available tools satisfies all needs. Furthermore, theoretical considerations have demonstrated clearly that the predictive properties of PONV scores are limited and it is unlikely that future scores will improve the situation.13
Additionally, there is fundamental criticism regarding the application of any risk model in this setting. This review summarises arguments against the use of PONV risk scores in clinical practice.
Lack of a systematic external evaluation of available post-operative nausea and vomiting scores
It is uniformly recommended that any risk model must undergo external validation before it can be recommended in clinical practice because risk factors may have been identified incorrectly and, even if correct, may not actually be useful as prognostic tools.14 Unfortunately, no systematic evaluation of the published scores has been performed to date. Thus, it is necessary to rely on the few available validation studies.15–19 One of these18 evaluated the agreement between different scores using the Bland–Altman method and found poor reliability of the PONV scores. In addition, caution must be applied because the datasets may be used repeatedly and, thus, results may not be independent.17,19,20
Criticism of the general application of PONV scores is based not only on the systematic lack of external validation of the scores but also on other methodological issues.
The eligibility for widespread use of these tools is based on three quality parameters: discrimination, calibration and applicability.
A predictive tool needs to allocate patients into a ‘disease group’ or a ‘non-disease’ group. Statistically, the discrimination power of a score is calculated from the likelihood that an individual with a higher PONV score will suffer from PONV and another individual with a lower score will not. Discrimination is calculated from the C-statistic and often presented as the area under a receiver operating characteristic (ROC) curve. Values of 0.8 and greater are usually regarded as good predictors assuming that a value of 0.5 represents a random guess. No PONV score has ever yielded an acceptable discriminating power of 0.8 even when highly sophisticated analysis techniques such as artificial neural networks have been employed.7,11 In most studies, the value of the C-statistic (indicating the discriminating power of a model) was between 0.6 and 0.7.16,17,19
The second quality parameter of a prediction tool is calibration, which is the ability of a score to predict the incidence of an event prospectively in a larger group of patients. Ideally, the predicted incidence of PONV matches the actual incidence, and thus a correlation between the two numbers has a slope of 1.0 with no offset.
All external validating studies that have been performed in relation to PONV scoring systems resulted in lower calibration properties of the tested scores in comparison to those described in the original papers. For instance, one validation study in more than 1500 patients17 documented acceptable calibration in the Koivuranta score,4 but highly erroneous prediction by the first available risk score published by Palazzo and Evans.2 A relative underestimation of risk in patients with a low-risk profile and a major overestimation in high-risk patients have been documented for the Apfel score, resulting in a flattened slope of only 0.39 (instead of 1.0).16 One reason for this disappointing result may be the low incidence of PONV (10%) predicted by the Apfel score in patients without any risk factors. In fact, these patients had an incidence of PONV of 18–22%.4,16Figure 1 shows this calibration curve.
The third quality indicator for a prediction tool is the ease of application in routine clinical practice. Although ‘ease of use’ is a subjective term, there is little doubt that a scoring system should have a limited number of predictors and should be relatively simple. This means that the calculation of a predicted risk must not require an electronic aid and every user should be able to calculate the score by mental arithmetic. The use of a risk score is simplified if the weight of each risk factor is equal or at least is an integer value. The requirement of applicability is fulfilled by only two PONV scores. Koivuranta et al.4 published the first score in 1997. Two years later, their data were compared with patient data from the University of Würzburg to create a modified version, the Apfel score.3 The latter tool has now gained widespread distribution based on good marketing (www.ponv.org) and by the fact that the predicted incidences of PONV fortuitously are close to round numbers (see Table 1).
However, simplification can create problems because the relative importance of different risk factors is neglected. Of possibly greater importance, the clinical information that can be drawn from obtaining the entire history of a patient is ignored. An example may illustrate this issue. Applying the Apfel score to a male smoker who has undergone some surgical procedures under general anaesthesia with very severe and prolonged nausea and vomiting after each procedure assigns only one risk factor when no post-operative opioids are administered. A non-smoking woman (no post-operative opioids) who has undergone the same number of operations without any signs of nausea after one operation will be also scored with one risk factor. Despite the apparently identical risk for PONV, clinical experience tells us that the male patient is much more likely to suffer from PONV than the female patient. One possible explanation is that the men are extremely prone to nausea and vomiting for reasons of which we are not yet aware (for example genetic or epigenetic constitution/other biological factors), whereas the women seem to be immune to emetogenic triggers (for the same unknown endogenous reasons). Clinical data support this hypothetical model: male patients with a positive history of PONV are more likely to develop PONV than female patients who did not suffer from PONV after repeated general anaesthesia despite an identical risk for PONV according to the Apfel score (re-analysed data from Eberhart et al.17 and Kranke et al.21). Thus, simplified models may be too simple for such complex interrelationships.
Another concern regarding the use of simplified risk scores is the exclusion of proven risk factors. For example, the Apfel score substituted the risk factor ‘duration of surgery longer than 60 min with ‘administration of post-operative opioids’ and condensed two risk factors of the Koivuranta score (history of PONV and motion sickness) into one factor (history of PONV or motion sickness). However, duration of surgery has been found to have a major influence on PONV and is a risk factor in several other PONV scores,5,7,8,11 including an earlier non-simplified risk score from Apfel et al.10 This may be one reason for the generally weaker discriminating power of the simplified Apfel score compared to the Koivuranta model.7,9,16 Furthermore, neither of the simplified scores uses the site of surgery as a risk factor for prediction. Although site of surgery used as a single predictor has a limited discriminating power,20 other studies have identified specific types of surgery as independent risk factors for PONV.5,6,12,22 Surprisingly, even a validation study by Apfel et al.23 reported a statistically significant (P = 0.002) increase in the incidence of PONV in patients after thyroid surgery, which could not be explained by patients’ characteristics alone. Thus, it remains questionable whether any risk model with a limited number of relatively weak risk factors (no risk factors for PONV has an odds ratio >3)13 will ever provide enough predictive power for a substantial improvement in prediction of PONV. Finally, there seem to be differences in the risk factors associated with nausea alone, and vomiting.24
Another problem is that PONV scores cannot be applied uniformly in every group of patients. Recently, a paper from South Africa25 reported a highly significant difference between the PONV incidences in black and white patients surveyed at the same department (black 27% vs. white 45%) and concluded that black South African ethnicity is an independent protective factor against PONV. Another study26 concluded that PONV scores have no value in prediction of PONV symptoms occurring 24–72 h after the end of anaesthesia. Even if specific risk scores were available for each population, we do not know whether scores developed for inpatients can be applied to the increasing number of outpatients. Post-discharge nausea and vomiting is an issue which is gaining increasing attention.27 As some risk factors, such as a history of motion sickness, might be more important for ambulatory patient during their travel home, an evaluation of PONV risk scores in these patients is mandatory but has not yet been performed.
The prediction of PONV is further complicated by successful antiemetic prophylaxis. At first glance, this seems to be contradictory, but becomes clearer if we understand that patients who report uneventful recovery after (several) previous anaesthetics might be rated incorrectly as low risk because no emetic event has occurred in the past. It is difficult to know whether no PONV has occurred because the patient is immune to emetic triggers or because he or she has received multimodal antiemetic prophylaxis. Thus, the more score-based decisions that are made to provide antiemetic prophylaxis, the weaker becomes the importance of a history of previous PONV, reducing still further the already limited discriminating power of PONV scores.
Individual risk factors
Although determination of gender is obviously no clinical problem, all other important risk factors for PONV are much less clearly defined. In particular, there is still no clear definition of a (non) smoker, a history of motion sickness or a history of previous PONV.
Some patients admit to smoking infrequently (for example 1–2 cigarettes per day) or irregularly (for example at parties or only during weekends). There are pipe smokers who do not inhale and patients who report having stopped smoking before their operation (in order to reduce pulmonary risks). Others have tried to stop smoking but have relapsed into previous habits. The available literature on PONV scores has not yet indicated whether these examples should be judged as smokers or non-smokers.
The same holds true for the definition of motion sickness. For example, many patients state that they suffered from motion sickness when they were children but now have no problems (except for travelling in a ship on rough sea). Some patients cannot sit in the back of a car, and others start feeling sick only when they are reading during a journey.
History of post-operative nausea and vomiting
A history of previous PONV is also less easy to define than it seems at first glance. Some examples might highlight this problem. A patient reports that he was absolutely fine after surgery and thus he ingested a lot of oral fluids and had lunch just 2 h after his procedure. For some minutes afterwards, he felt nauseous, but then recovered and had no more symptoms of PONV. One might argue that this patient has developed clear signs of post-operative nausea. However, he has recovered perfectly from anaesthesia and the short period of nausea was obviously triggered by the early oral intake. A patient who has recovered uneventfully from surgery under spinal anaesthesia but suffered from PONV after the spinal has worn off and opioid analgesics were administered intravenously would be characterised better as suffering from ‘opioid-induced nausea and vomiting’ and not primarily PONV. What about a patient reporting severe PONV decades ago, but who has had several uneventful anaesthetics recently?
Post-operative opioid administration
One of the risk factors in the simplified Apfel score3 is the administration of post-operative opioids. Whether a patient will require post-operative opioids or not remains speculative before the end of the surgical procedure. Furthermore, there is good evidence that side-effects of opioids such as nausea and vomiting are dose-dependent. Thus, risk of PONV is influenced considerably by whether a few milligrams of morphine are administered in the immediate post-operative period and pain therapy is mainly based on NSAIDs thereafter, or high doses of an opioid are administered during the first few post-operative days [for example using a patient-controlled analgesia system].28 Maleck and Piper29 have highlighted these limitations of the Apfel score. In reply, Apfel and Roewer argued that even some incorrectly matched patients would decrease the discriminating power of the risk score by less than 2% (AUC 0.684–0.668) if 10% of the patients were allocated wrongly.29 At first sight, this loss of less than 2% of predictive power seems to be negligible. However, if it is appreciated that there is only an absolute improvement of 18% resulting from the use of a simplified PONV score, this decrease in predictive power represents a relative decrease of more than 10%.
Most information gained from predictive scores is already known
The majority of patients undergoing surgery have one or two risk factors for PONV.21 Using simplified PONV scores (Apfel or Koivuranta), about 20–40% of these patients are forecast to suffer from PONV. However, this is not new information. Numerous epidemiological trials on PONV have documented this baseline risk when using a conventional balanced anaesthesia technique without prophylactic measures.30 Only for a small number of patients with three or more risk factors (predicted incidence of PONV >55%; see Table 1) will a simplified score provide ‘new’ information, but it may be possible to detect these high-risk patients simply by reviewing the medical history.
Information gained from risk models does not necessarily modify clinical practice
In addition to the methodological deficiencies of PONV scores, other problems arise relating to their implementation. This was highlighted by a study by Kooij et al.31 They studied the rate of administering antiemetics to patients at increased risk of PONV according to local standards and found that only 39% of patients were treated prophylactically. After implementing an electronic pop-up window reminding the (nurse) anaesthetist that PONV prophylaxis was indicated, this increased to 79%. However, after turning off the decision support, the proportion of patients treated decreased abruptly and returned to the baseline value (41%).
Two recent papers also highlight the problems of transforming the information gained from the predictive models into clinical practice. Klotz and Philippi-Höhne32 found that antiemetic prophylaxis was administered according to local guidelines in only 79% of paediatric patients. In adults,33 the adherence to a standing operating procedure was even lower; only 45% of patients received the designated treatment and patients with a high predicted risk often received inadequate treatment.
Finally, a computer simulation on the efficiency of 10 different potential antiemetic strategies (routine prophylaxis in every patient vs. score-adopted algorithms vs. simple dichotomous approaches) could not identify one single optimal strategy.21 Antiemetic success defined as absolute risk reduction in the simulated populations was a function mainly of the number of antiemetic interventions. The strategy applied was of minor importance. None of the tested strategies was universally applicable, highlighting the need for individual institutional policies that should be based on the local incidence of PONV and demands of patients and surgeons. The study also showed that an adequate reduction in PONV can be achieved only by liberal administration of multimodal antiemetic prophylaxis.
In the light of the imprecise prediction of PONV, the difficulties in implementing risk scoring systems into clinical practice and the urgent concerns of our patients not to suffer from PONV,34,35 we should adopt a much more liberal use of multimodal antiemetic prophylaxis. Acquisition costs and potential side-effects play only a very minor role and we should focus on improved speed of recovery, post-operative well being and patient satisfaction. Thus, it is not surprising that all multimodal rehabilitation concepts (‘fast-track’ recovery) include multimodal prophylactic measures against PONV.36 Our plea for the importance of liberal multimodal prophylaxis against PONV does not exclude individual therapeutic decisions. Without doubt, every anaesthesiologist should be informed about scientifically proven risk factors for PONV, but decisions should not rely on simple scores. We advocate the use of routine multimodal antiemetic prophylaxis (i.e. at least two antiemetic measures) in every patient. Anaesthesiologists are then free to judge the risk for each patient taking into account all additional information from the patient that affects the likelihood of PONV, which cannot all be included in simplified scoring systems. According to this personal evaluation, antiemetic prophylaxis can be augmented in high-risk patients (or reduced in special cases). Furthermore, individual concerns of the patient should be respected and multimodal antiemetic prophylaxis should be provided whenever a patient is fearful of suffering from emetic symptoms post-operatively.
1 Gan TJ, Meyer TA, Apfel CC, et al
. Society for Ambulatory Anesthesia Guidelines for the management of postoperative nausea and vomiting. Anesth Analg 2007; 105:1615–1628.
2 Palazzo M, Evans R. Logistic regression analysis of fixed patient factors for postoperative sickness: a model for risk assessment. Br J Anaesth 1993; 70:135–140.
3 Apfel CC, Läärä E, Koivuranta M, et al
. A simplified risk score for predicting postoperative nausea and vomiting. Anesthesiology 1999; 91:693–700.
4 Koivuranta M, Läärä E, Snåre L, Alahuhta S. A survey of postoperative nausea and vomiting. Anaesthesia 1997; 52:443–449.
5 Sinclair DR, Chung F, Mezei G. Can postoperative nausea and vomiting be predicted? Anesthesiology 1999; 91:109–118.
6 Gan TJ. Postoperative nausea and vomiting: can it be eliminated? JAMA 2002; 287:1233–1236.
7 Träger M, Eberhart A, Geldner G, et al
. Prediction of postoperative nausea and vomiting using an artificial neural network. Anaesthesist 2003; 52:1132–1138.
8 Junger A, Hartmann B, Benson M, et al
. The use of an anesthesia information management system for prediction of antiemetic rescue treatment at the postanesthesia care unit. Anesth Analg 2001; 92:1203–1209.
9 Engel JM, Junger A, Hartmann B, et al
. Performance and customization of 4 prognostic models for onset of postoperative nausea and vomiting in ear, nose, and throat surgery. J Clin Anesth 2006; 18:256–263.
10 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.
11 Peng SY, Wu KC, Wang JJ, et al
. Predicting postoperative nausea and vomiting with the application of an artificial neural network. Br J Anaesth 2007; 98:60–65.
12 Tropé A, Ræder JC. Can postoperative nausea or vomiting be predicted? Tidsskr Nor Lægeforen 2000; 20:2423–2426.
13 Apfel CC, Kranke P, Greim CA, Roewer N. What can be expected from risk scores
for predicting postoperative nausea and vomiting? Br J Anaesth 2001; 86:822–827.
14 Ware JH. The limitation of risk factors as prognostic tools. N Engl J Med 2006; 355:2615–2617.
15 Toner CC, Broomhead CJ, Littlejohn IH, et al
. Prediction of postoperative nausea and vomiting using a logistic regression model. Br J Anaesth 1996; 76:347–351.
16 van den Bosch JE, Kalkman CJ, Vergouwe Y, et al
. Assessing the applicability of scoring systems for predicting postoperative nausea and vomiting. Anaesthesia 2005; 60:323–331.
17 Eberhart LHJ, Högel J, Seeling W, et al
. Evaluation of three risk scores
to predict postoperative nausea and vomiting. Acta Anaesthesiol Scand 2000; 44:480–488.
18 Thomas R, Jones NA, Strike P. The value of risks scores for predicting postoperative nausea and vomiting when used to compare patient group in a randomised controlled trial. Anaesthesia 2002; 57:1119–1128.
19 Apfel CC, Kranke P, Eberhart LHJ, et al
. Comparison of predictive models for postoperative nausea and vomiting. Br J Anaesth 2002; 88:234–240.
20 Apfel CC, Kranke P, Eberhart LHJ. Comparison of surgical site and patient's history with a simplified risk score for the prediction of postoperative nausea and vomiting. Anaesthesia 2004; 59:1078–1082.
21 Kranke P, Eberhart LH, Gan TJ, et al
. Algorithms for the prevention of postoperative nausea and vomiting: an efficacy and efficiency simulation. Eur J Anaesthesiol
22 Eberhart LHJ, Geldner G, Kranke P, et al
. Development and validation of a risk score to predict the probability of postoperative vomiting in pediatric patients. Anesth Analg 2004; 99:1630–1637.
23 Apfel CC, Greim CA, Haubitz I, et al
. The discriminating power of a risk score for postoperative vomiting in adults undergoing various types of surgery. Acta Anaesthesiol Scand 1998; 42:502–509.
24 Stadler M, Bardiau F, Seidel L, et al
. Difference in risk factors for postoperative nausea and vomiting. Anesthesiology 2003; 98:46–52.
25 Rodseth RN, Gopalan PD, Cassimjee HM, Goga S. Reduced incidence of postoperative nausea and vomiting in black South Africans and its utility for a modified risk scoring system. Anesth Analg 2010; 110:1591–1594.
26 White PF, Sacan O, Nuangchamnong N, et al
. The relationship between patient risk factors and early versus late postoperative emetic symptoms. Anesth Analg 2008; 107:459–463.
27 Le TP, Gan TJ. Update on the management of postoperative nausea and vomiting and postdischarge nausea and vomiting in ambulatory surgery. Anesthesiol Clin 2010; 28:225–249.
28 Marret E, Kurdi O, Zufferey P, Bonnet F. Effects of nonsteroidal antiinflammatory drugs on patient-controlled analgesia morphine side effects: a meta-analysis of randomized controlled trials. Anesthesiology 2005; 102:1249–1260.
29 Maleck WH, Piper SN. Predictive models for postoperative nausea and vomiting. Br J Anaesth 2002; 88:339–340. Authors' reply 340–342.
30 Watcha MF, White PF. Postoperative nausea and vomiting: do they matter? Eur J Anaesthesiol 1995; 12(Suppl 10):18–23.
31 Kooij FO, Klok T, Hollmann MW, Kal JE. Automated reminders increase adherence to guidelines for administration of prophylaxis for postoperative nausea and vomiting. Eur J Anaesthesiol 2010; 27:187–191.
32 Klotz C, Philippi-Höhne C. Prophylaxis of postoperative nausea and vomiting in pediatric anesthesia: recommendations and implementation in clinical routine. Anaesthesist 2010; 59:477–478.
33 Franck M, Radtke FM, Baumeyer A, et al
. Adherence to treatment guidelines for postoperative nausea and vomiting. How well does knowledge transfer result in improved clinical care? Anaesthesist 2010; 59:524–528.
34 Gan TJ, Sloan F, Dear Gde L, et al
. How much are patients willing to pay to avoid postoperative nausea and vomiting? Anesth Analg 2001; 92:393–400.
35 Eberhart LHJ, Morin AM, Wulf H, Geldner G. Patient preferences for immediate postoperative recovery. Br J Anaesth 2002; 89:760–761.
36 Kranke P, Redel A, Schuster F, et al
. Pharmacological interventions and concepts of fast-track perioperative medical care for enhanced recovery programs. Exp Opin Pharmacotherap 2008; 9:1541–1564.