Secondary Logo

Journal Logo

Original Article

Detection of causal relationships between factors influencing adverse side-effects from anaesthesia and convalescence following surgery: a path analytical approach

Reuter, M.*; Hueppe, M.; Klotz, K. F.; Beckhoff, M.; Hennig, J.*; Netter, P.*; Schmucker, P.

Author Information
European Journal of Anaesthesiology: June 2004 - Volume 21 - Issue 6 - p 434-442

Abstract

Surgery is a stressful situation for patients [1,2] and anaesthesia is a major stress for the organism in addition to the surgery [3]. The psychological as well as the physical perioperative stress can influence convalescence of patients [4]. Therefore, it is of great importance to learn as much as possible about the factors that might reduce stress in the context of surgery. The anaesthesiologist is aware of this and tries to contribute to stress reduction by the choice of the appropriate anaesthetics and by an interview with the patient prior to surgery. The issue of which anaesthetics are best is always a critical matter of debate for anaesthesiologists and pharmacologists whereas the importance of the exchange of information between the patient and the anaesthesiologist [5,6] is often neglected. This is surprising because the interview prior to surgery has become a standard procedure in perioperative treatment and is a factor that can be easily modified by the anaesthesiologist.

The present study tries to elucidate the factors influencing adverse side-effects from anaesthesia and recovery from surgery, such as age and gender of the patients, duration of surgery and the patients' satisfaction with the information obtained from the anaesthesiologist. This will be achieved by means of structural equation modelling. Path analysis is a statistical method of the structural equation modelling family which permits detection of the multivariate causal relationships between observed variables; the term 'causal relationship' does not necessarily imply that the structural equation modelling approach could reveal causal relations in non-experimental data although causality is established by defining unidirectional paths in a model [7]. Yet, if this directionality is based on logic, theory, or on research design, then a causal interpretation of directional paths may be legitimate [8]. Moreover, we intended to analyse whether structural relations between variables vary across anaesthesiological regimens. Until now causal relationships between variables have hardly been reported in the context of anaesthesiology [9,10] and path analysis has never been applied for comparing different anaesthetics although there are over 1200 papers cited in MEDLINE which use the method of structural equation modelling. It is recommended to use structural equation modelling also in anaesthesiology because the method allows to detect and to quantify direct and indirect effects of variables relevant for the influence of anaesthesia on adverse side-effects of anaesthetics and on convalescence by simultaneously testing for structural differences across regimens of anaesthesia. The path analytical approach is demonstrated by establishing a structural equation model and by ensuing multiple group analysis testing for differences between anaesthesia obtained by propofol and by isoflurane + N2O. The target of interest is the role of the anaesthesiologist's interview on possible side-effects of the anaesthetics and on recovery from surgery. The question is whether structural equation modelling is a useful methodological approach for this multivariate problem.

Methods

Before the study was started study approval by the Institutional Ethical Review Board was acquired and written informed consent was obtained from all patients. The sample consisted of 710 surgical patients of whom 204 received propofol, 267 received isoflurane + N2O and 239 received mixed regimens for maintenance of anaesthesia. Anaesthesia regimens were applied according to clinical criteria. All patients received an elective operation mainly concerning the abdominal regions (35.5%), extremities (37%), throat (12.3%) or thorax (6.8%); 387 were males and 323 were females. Ages varied between 18 and 90 yr (mean, 54.32 yr; SD, 16.71 yr).

For the assessment of pharmacological side-effects, satisfaction with the anaesthesiologist's interview and satisfaction with convalescence, patients filled in the modified 'Anaesthesiological Questionnaire for Patients' (ANP II) [11] within the first week after surgery. The ANP II is an inventory, which measures emotional and physical states in the context of anaesthesia. It is a self-rating questionnaire that is based on the assumption that the most sensitive source of information relevant for convalescence is the patient. Based on theoretical considerations the following items were selected for model specification: difficulties in recovery from anaesthesia (DIFFWAKE), postoperative nausea and vomiting (PONV), postoperative pain (PAIN), feeling of physical discomfort (DISCOM), satisfaction with convalescence and present state (SATCON), satisfaction with the anaesthesiologist's interview (SATINTER), age (AGE), gender (SEX, with 1 = male and 2 = female) and duration (DURAT) of surgery. All items of the ANP II were scored on a four-point Likert-scale ('not at all', 0; 'a little bit', 1; 'quite', 2; 'strong', 3). The items DIFFWAKE, PONV, PAIN and DISCOM were dichotomized to correct for skewness. Values ≥1 were recoded into category 1 and zero values remained unchanged. For all items of the ANP II that were ordinal variables, the metric variables age and duration of surgery were also transformed into ordinal variables with three and four categories respectively. The categories of AGE (1 = 0-49 yr, 2 = 50-64 yr and 3 = 65-90 yr) and of DURAT were formed according to the criterion of equal distribution across groups. The higher the value of the category the older was the patient and the longer was the duration of surgery. Ratings of the interview (SATINTER) were scored on the original four-point scale and SATCON on a six-point scale.

Structural equation modelling

Basic concepts. The objective of structural equation modelling, which is a well established statistical method, is to test if a theoretically derived model fits the empirical data. This implies that the question is addressed if a model is able to explain the patterns of covariance observed among the study variables. Therefore, structural equation modelling is inherently a confirmatory technique. Relationships between variables are represented by paths. Bidirectional paths mean noncausal, or correlational relations and unidirectional paths represent causal relations (standardized path coefficients are interpreted in the same way as beta weights in regression analyses). The relationships between the variables in a given model can be written as a set of structural equations which are solved by statistical software packages. The model parameters are estimated by means of an iterative process with the purpose to minimize the differences between the covariance matrix implied by the parameter estimates and the observed covariance matrix. The similarity between these two covariance matrixes is finally described by a fit index (see below). The assessment of 'absolute' fit is concerned with the ability of the model to reproduce the actual covariance matrix.

Structural equation modelling also allows modulating relationships between observed and latent variables (factors). It is possible to calculate the influence of observed predictors on observed criteria, which are mediated by latent factors. However, the present application of structural equation modelling is a path model which does not contain latent (not observable) constructs. As in regression analysis, calculations are based on correlations/covariances. In contrast to traditional regression analysis, structural equation modelling is able to handle many dependent criteria at a time. Although canonical correlations can also quantify the relationship between a group of predictors and a group of criteria, only structural equation modelling can modulate relationships within the group of criteria to explain variance in a 'second order' criterion. Furthermore, multiple regression models are not able to analyse all kinds of path models. For example, the above-mentioned bidirectional paths (variable A influences variable B but variable B also influences variable A) are not allowed, latent variables cannot be included into a model and residuals are presumed to be uncorrelated - whereas in structural equation modelling all these features are implemented.

Model specification

In the present study the model derived from theoretical considerations and clinical practice was as follows: it was assumed that age, gender, duration of surgery and satisfaction with the anaesthesiologist's interview (SATINTER) have an influence on the adverse side-effects. However, the adverse effects influence physical discomfort, and physical discomfort in combination with postoperative pain will reduce satisfaction with convalescence. Furthermore, it is expected that the so-called exogenous variables SATINTER, AGE, SEX and DURAT have direct effects on the so-called endogenous variables 'feeling of nausea and vomiting' (PONV), 'difficulties in waking up' (DIFFWAKE), 'postoperative pain' (PAIN), 'physical discomfort' (DISCOM) and 'satisfaction with convalescence' (SATCON). In a first step this recursive model (Fig. 1) was tested in the total sample of 710 subjects and thereafter the initial model was trimmed (i.e. n.s. paths were deleted) to reject spurious associations.

Figure 1
Figure 1:
Path model derived from theoretical considerations. The paths (long arrows) indicate the postulated causal relationships. Each of the exogenous variables (open rectangles) is assumed to influence each of the endogenous variables (dark rectangles). Furthermore, some of the endogenous variables influence other endogenous variables. Small arrows behind the dark rectangles indicate the 'disturbances' (unexplained variance of the endogenous variables).

Matrices, method of estimation and fit indexes

Given categorical data, path analyses are based on asymptotic covariance matrices and polychoric correlation matrices. Instead of the commonly used maximum likelihood (ML) estimation, the weighted least squares (WLS) method was used which is recommended for parameter estimation in models based on categorical data [12,13]. According to the recommendations in the literature [7] the following model fit indices will be reported besides the χ2 statistic: the standardized root mean square residual (SRMR), the comparative fit index (CFI) and the root mean error of approximation (RMSEA). A significant χ2 value relative to the degrees of freedom indicates that the observed and estimated matrices are different. Therefore, an n.s. χ2-test indicates a good model fit. Nevertheless an n.s. goodness-of-fit χ2-statistic is unlikely with large samples [14]. The CFI compares the researcher's model with a null model (the observed variables are assumed to be uncorrelated) and should not be smaller than 0.90 [14]. The SRMR which represents a standardized summary of the average covariance residuals should not be greater than 0.08 and the RMSEA which tests if the error of approximation is tolerable should be smaller than 0.05 and not greater than 0.08 [15]. All structural equation modelling analyses were conducted by LISREL 8.51 [16].

Results

The polychoric correlation matrix of the total sample is presented in Table 1, and Figure 2 illustrates the path model of the proposed anaesthesia model after model trimming.

Table 1
Table 1:
Polychoric correlation matrix for the total sample (n = 710).
Figure 2
Figure 2:
Path diagram for the total sample (standardized solution). All paths were significant except for the dotted paths. Curved paths indicate the covariance between variables. χ2 = 4.29, d.f. = 11, P = 0.96082, RMSEA = 0.000.

The model fit was perfect, CFI = 1.00, SRMR = 0.017, RMSEA < 0.001 and even the goodness of fit χ2 statistic was n.s. (χ2 = 4.29, d.f. = 11, P = 0.961), a result which is not very likely with large samples. As demonstrated in Figure 2, not all paths from the exogenous variables to the endogenous variables are meaningful. Non-significant paths were omitted from the model unless their deletion would make the fit of the model worse. Paths which were not excluded from the model despite non-significance (t-values ≤ 1.96 indicate an n.s. path) were the paths from SATINTER to DIFFWAKE (t = −1.37), from SATINTER to DISCOM (t = −1.73), from SEX to PAIN (t = 1.04), from DURAT to SATCON (t = −1.01) and from PONV to SATCON (t = 1.16). Fourteen percent of the variance of 'difficulties in waking up', 17% of the variance of 'nausea and vomiting', 13% of the variance of 'postoperative pain', 47% of the variance of 'feeling physical discomfort' and 19% of the variance of 'satisfaction with convalescence and present state' were explained by the trimmed model. The amount of explained variance with respect to a given variable can be inferred from Figure 2 by subtracting the amount of the unexplained variance reported behind the endogenous variables from 1 (e.g. the explained variance of DISCOM is calculated as follows: 1 − 0.53 = 0.47).

Results of effect breakdown (Table 2) show that some exogenous variables influence endogenous variables only indirectly, e.g. although they have no direct influence they have an impact on them merely by mediator effects. For example, SATINTER has a slight negative effect on PONV which results from an indirect effect mediated by DIFFWAKE. AGE has exclusively indirect effects on PONV (−0.10), DISCOM (−0.22) and SATCON (0.08). While the indirect effect of age on PONV (−0.02) is exclusively mediated via DIFFWAKE, the indirect effects of age on DISCOM and SATCON are mediated by all other adverse side-effects either by one mediator or by their interactions (more than one mediator). Similarly, SEX exerts only indirect effects on DISCOM (0.18) and SATCON (−0.04) which were caused by first, second or third order mediator effects. The exclusively indirect effects of DURAT on endogenous variables are only small (indirect effect on PONV: 0.03 and indirect effect on PAIN: 0.01). The results corroborate results reported in the literature (for a review see [17]) that older patients suffer less from DIFFWAKE and PONV than younger patients (negative path coefficients) and that females report more complaints with respect to these two adverse side-effects than males (positive path coefficients).

Table 2
Table 2:
Breakdown of standardized effects for the 'anaesthesia model'.

In order to quantify the influence of 'satisfaction with the anaesthesiologist's interview' on the endogenous variables the other three exogenous variables were excluded from the model. Results showed that the nested model (i.e. that equivalent to the term hierarchical model: two path models are hierarchical if one is a subset of the other) - with SATINTER as the only exogenous variable - explains 1% of the variance of DIFFWAKE, 11% of the variance of PONV, 6% of the variance of PAIN, 45% of the variance of DISCOM and 18% of the variance of SATCON. Furthermore, this nested model had a very good fit (CFI = 1.00, SRMR = 0.025, RMSEA < 0.001, χ2 = 3.72, d.f. = 4, P = 0.445) but its explanation of variance of some endogenous variables was in part markedly smaller (DIFFWAKE: −13%, PONV: −6%, PAIN: −7%, DISCOM: −2%, SATCON: −1%) in comparison to the complete model which included the other three exogenous variables AGE, SEX and DURAT.

Testing the 'anaesthesia model' in the propofol- and isoflurane + N2O-samples

The polychoric correlation matrices of the propofol and the isoflurane + N2O subsamples are presented together in Table 3.

Table 3
Table 3:
Polychoric correlation matrices for the propofol (n = 204) and the isoflurane + N2O (n = 267) samples.

In the sample of the patients who received propofol for maintenance of anaesthesia (n = 204) the model generated in the total sample could be confirmed. The fit indexes were very good (CFI = 1.00, SRMR = 0.037, RMSEA < 0.001) and even the goodness of fit χ2 statistic was n.s. (χ2 = 5.95, d.f. = 11, P = 0.877). Nevertheless further analyses revealed that some of the paths are dispensable in the propofol sample. By model trimming the following paths were eliminated from the model: the paths from SATINTER to DIFFWAKE, from SATINTER to DISCOM, from DURAT to PAIN, from DURAT to SATCON and from DISCOM to SATCON. Furthermore, an additional path from SATINTER to PONV was added which further improved the goodness of fit χ2 statistic (CFI = 1.00, SRMR = 0.037, RMSEA < 0.001, χ2 = 6.21, d.f. = 15, P = 0.976). The trimmed model for the propofol-group is presented in Figure 3.

Figure 3
Figure 3:
Trimmed path diagram for propofol group (standardized solution). The bold dotted path indicates a significant association not observed in the total sample. χ2 = 7.62, d.f. = 15, P = 0.93795, RMSEA = 0.000.

Results for the propofol group show that satisfaction with the anaesthesiologist's interview does neither directly influence difficulties in waking up nor physical discomfort. Furthermore, gender does not directly influence postoperative pain, and the duration of surgery as well as the amount of physical discomfort have no direct effects on satisfaction with convalescence.

Also in the isoflurane + N2O group (n = 267) the anaesthesia model derived from the total sample could also be confirmed (CFI = 1.00, SRMR = 0.034, RMSEA < 0.001, χ2 = 7.77, d.f. = 11, P = 0.733). Nevertheless, inspection of the modification indices revealed that some paths were also dispensable and that in analogy to the propofol group a path from SATINTER to PONV could be added. Results show that the trimmed models for the propofol and the isoflurane + N2O group are identical. The modifications further improved the goodness of fit χ2 statistic (CFI = 1.00, SRMR = 0.030, RMSEA < 0.001, χ2 = 7.62, d.f. = 15, P = 0.938). The trimmed model for the isoflurane + N2O group is presented in Figure 4. The more complex model of the total sample seems to be a result of the third subsample of 239 patients who received other anaesthesia regimens.

Figure 4
Figure 4:
Trimmed path diagram for isoflurane + N2O group (standardized solution). χ2 = 6.21, d.f. = 15, P = 0.97603, RMSEA = 0.000.

Multiple group analysis

In order to test if the relationships between variables obtained in the propofol and the isoflurane + N2O sample were significantly different, a multiple group path analysis was conducted with equality constraints on the path coefficients across both groups. The multi-sample analysis again yielded perfect fit indexes (CFI = 1.00, SRMR < 0.001, RMSEA < 0.001, χ2 = 18.58, d.f. = 45, P = 0.999). The χ2 statistic of the model with its path coefficients constrained to equality was then contrasted against the χ2 of the unconstrained model (χ2 = 13.83, d.f. = 30, P = 0.995). The difference test was not significant (P(Δχ2) = 0.994); Δd.f. = 15) indicating that the free estimation of the constrained parameters in each sample would not significantly improve the model, i.e. no path coefficients differ significantly between groups.

Discussion

The present study tries to detect the relationship of factors influencing adverse side-effects of anaesthesia and their influence on patients' convalescence from surgery. A general model was set up and tested by means of structural equation modelling which allows to detect and quantify causal relationships and mediator effects. The target of interest was the impact of the anaesthesiologist's interview on complaints related to anaesthesia such as nausea and difficulties in waking up [5,6]. It is assumed that adverse side-effects from anaesthesia contribute to a process which renders surgery into a traumatic or at least stressful life event [1,2], which worsens convalescence [4]. The evidence of a stress reducing effect of the anaesthesiologist's interview prior to surgery would provide a powerful tool to improve convalescence and well-being of the patients. Furthermore, it was intended to test if the 'anaesthesia model' is likewise valid in anaesthesia maintenance regimens using propofol or the volatile anaesthetic isoflurane + N2O.

Results showed that the proposed anaesthesia model with the exogenous variables 'satisfaction with the anaesthesiologist's interview', age and gender of the patients and 'duration of surgery' was able to explain 14% of the variance of 'difficulties to wake up', 17% of the variance of 'feeling of nausea and vomiting', 13% of the variance of 'postoperative pain', 47% of the variance of 'feeling physical discomfort' and 19% of the variance of 'satisfaction with convalescence and present state'. By breaking down the total effects of exogenous on endogenous variables it turned out that some of the exogenous variables exert merely indirect effects on certain endogenous variables. Here the power of structural equation modelling in detecting mediator effects is striking. Indirect effects which would not be disclosed by a bivariate correlational approach could be revealed. For example, the age of the patient strongly influences feelings of physical discomfort (−0.22) but this effect is completely mediated by the adverse side-effects and postoperative pain. Besides the reported mediator effects, the model also corroborates findings reported in the literature, e.g. higher postoperative nausea in females and younger patients (for a review see [17]). Furthermore, the importance of the anaesthesiologist's preoperative interview was corroborated. Eleven percent of the variance of 'nausea and vomiting', 6% of the variance of 'postoperative pain', 45% of the variance of 'feeling physical discomfort' and '18% of the variance of 'satisfaction with convalescence and present state' could be explained by satisfaction with the preoperative interview and its mediator effects. Concerning the last two variables, this is not much less than the amount of variance explained by the complete model. Interestingly, the assumption that patients with more postoperative side-effects would also be more likely to be dissatisfied with the preoperative interview (giving their rating the first week after surgery) could not be corroborated. Results show that the satisfaction with the interview is nearly uncorrelated with the side-effects. Only in the two drug sub-samples satisfaction with the interview is slightly correlated with PONV (−0.09 and −0.12, respectively).

Therefore, anaesthesiologists may contribute to a large extent to an amelioration of patients' physical comfort and convalescence by making great efforts in their preoperative conversation with the patient. They should be aware that through a high quality interview, the side-effects from anaesthesia could be reduced which, besides the direct effects of the interview, in turn influence the recovery process. Nevertheless, it cannot be decided if a preoperative interview by itself, or its quality, is responsible for the positive effects since there was no control situation without an interview. It has to be pointed out that the character of the study is observational which means that not all variables which potentially have confounded the results could be controlled. Nevertheless, the data show that postoperative pain is not influenced by the preoperative interview (neither directly nor by mediators). Further studies may investigate the reason for this unexpected result. As discussed at the outset, structural equation modelling is a method that is best suited to test for causal relationships even if no longitudinal data are available. In the present study one of the most essential criteria for causality is warranted, the temporal sequence between exposure and outcome. But this only refers to the impact of the exogenous variables (satisfaction with the interview, age, gender and duration of surgery) on the endogenous variables (adverse side-effects and satisfaction with recovery). The causal relationships within the endogenous variables have to be tested by comparing the present model with alternative models implying alternative causal relationships. Testing all potential alternative models would exceed the boundaries of an empirical study. So it is common practice to test a certain model against a so-called null-model (assuming no causal relationships at all) which has been done in the results presented.

Multiple group analysis revealed that there are neither structural differences nor differences in the strength of relationships between the variables in the anaesthesia model obtained in the propofol and the isofluane + N2O group. For both subsamples a nearly identical reduced model could be established which fitted the data perfectly. A recent publication [18] also reported similar properties of these anaesthesia regimens in a univariate approach: the only difference found was less nausea and vomiting after propofol which, however, is associated with higher monetary costs than volatile anaesthetics. In another study [19], total intravenous anaesthesia with propofol was also associated with higher economic costs but provided a more rapid recovery from anaesthesia, and less frequent postoperative side-effects. Although there are differences in the absolute values of the measured variables, structural equation modelling shows that the structural relationships between anaesthetic variables are the same for propofol and isoflurane + N2O treated patients.

Our results are important and suggest that anaesthesiologists may be able to improve patients' postoperative physical comfort and convalescence by improving patient satisfaction with the preoperative interview. While our results are provocative, additional studies are needed to determine whether improvements in the preoperative interview translate into improved patient outcomes and which components of the preoperative interview are crucial for its positive effects. Furthermore, the biological mechanisms underlying the potential beneficial relationship between the anaesthetist's preoperative interview and convalescence have to be further investigated. It may be assumed that the stress reducing effect of the interview has a positive influence on the activity of the hypothalamic-pituitary-adrenal axis and the immune system, which are involved in most psychological and physiological stress effects.

Structural equation modelling turned out to be a powerful tool to detect interdependencies, causal relationships and mediator effects which could be of great use in anaesthesiological and medical research. Traditional regression analysis and other statistical methods are not able to demonstrate the mutual relationships between criteria accounting for the explanation of additional variance in second order criteria (convalescence in our example). Moreover, structural equation modelling is able to compare complex theoretically derived models between different groups testing each relationship for significance. Therefore, the structural equation modelling method should be further established in such disciplines.

References

1. Shalev AY, Schreiber S, Galai T, Melmed RN. Post traumatic stress disorder following medical events. Br J Clin Psychol 1993; 32: 247-253.
2. Schreiber S, Galai-Gat T. Uncontrolled pain following physical injury as the core-trauma in the post-traumatic stress disorder. Pain 1993; 54: 107-110.
3. Udelsman R, Chrousos GP. Hormonal responses to surgical stress. In: Chrousos GP, Loriaux DL, GOLD PW, eds. Mechanisms of Physical and Emotional Stress. Advances in Experimental Medicine and Biology, Vol. 245. New York, USA: Plenum Press, 1988: 265-272.
4. Brodner G, Van Aken H, Hertle L, et al. Multimodal perioperative management - combining thoracic epidural analgesia, forced mobilization, and oral nutrition - reduces hormonal and metabolic stress and improves convalescence after major urologic surgery. Anest Analg 2001; 92: 1594-1600.
5. Bolton V, Brittain M. Patient information provision: its effect on patient anxiety and the role of health information services and libraries. Health Libr Rev 1994; 11: 117-132.
6. Shuldham C. A review of the impact of pre-operative education on recovery from surgery. Int J Nurs Stud 1999; 36: 171-177.
7. Hoyle RH, Panter A. Writing about structural equation models. In: Hoyle RH, ed. Structural Equation Modeling, Concepts, Issues and Applications. Thousand Oaks, USA: Sage, 1995: 158-176.
8. Hoyle RH. The structural equation modeling approach, basic concepts and fundamental issues. In: Hoyle RH, ed. Structural Equation Modeling, Concepts, Issues and Applications. Thousand Oaks, USA: Sage, 1995: 1-15.
9. Kain ZN, Sevarino F, Alexander GM, Pincus S, Mayes LC. Preoperative anxiety and postoperative pain in women undergoing hysterectomy. A repeated-measures design. J Psychosom Res 2000; 49: 417-422.
10. Rhoton MF, Barnes A, Flashburg M, Ronai A, Springman S. Influence of anaesthesiology residents' noncognitive skills on the occurrence of critical incidents and the residents' overall clinical performances. Acad Med 1991; 66: 359-361.
11. Hüppe M, Klotz KF, Heinzinger M, Prussmann M, Schmucker P. Rating the perioperative period by patients. First evaluation of a new questionnaire. Anaesthesist 2000; 49: 613-623.
12. Muthén B. A general structural equation model with dichotomous, ordered categorical and continuous latent variable indicators. Psychometrka 1984; 49: 115-132.
13. Bollen KA. Categorial observed variables. In: Bollen KA, ed. Structural Equations with Latent Variables. New York, USA: Wiley, 1989: 433-448.
14. Kline RB. Principles and Practice of Structural Equation Modeling. New York, USA: Guilford Press, 1998.
15. Jöreskog KG. Testing structural equations models. In: Bollen KA, Long SJ, eds. Testing Structural Equations Models. Newbury Park, USA: Sage, 1993: 295-316.
16. Jöreskog KG, Sörbom D. LISREL 8.51. Lincolnwood, USA: Scientific Software International, 2001.
17. Apfel CC, Roewer N. Risk factors for nausea and vomiting after general anaesthesia: fictions and facts. Anaesthesist 2000; 49: 629-642.
18. Visser K, Hassink EA, Bonsel GJ, Moen J, Kalkman CJ. Randomized controlled trial of total intravenous anaesthesia with propofol versus inhalation anaesthesia with isofluranenitrous oxid: postoperative nausea with vomiting and economic analysis. Anesthesiology 2001; 95: 616-626.
19. Ozkose Z, Ercan B, Unal Y, et al. Inhalation versus total intravenous anaesthesia for lumbar disc herniation: comparison of hemodynamic effects, recovery characteristics, and cost. J Neurosurg Anesthesiol 2001; 13: 296-302.
Keywords:

ANAESTHESIA; ANAESTHETICS, INHALATIONAL, isoflurane, nitrous oxide; ANAESTHETICS, INTRAVENOUS, propofol; DATA COLLECTION, interviews; PATIENT SATISFACTION; SURGICAL PROCEDURES, OPERATIVE, preoperative care

© 2004 European Academy of Anaesthesiology