| ||||||||||||||
|
|
|||||||||||||
Department of Anesthesiology and Intensive Care Medicine, Justus-Liebig-University Giessen, Germany
Address correspondence and reprint requests to Axel Junger, Dr med, Abt. Anaesthesiologie und Operative Intensivmedizin, Justus-Liebig-Universität, Rudolf-Buchheim-Str. 7, D-35392 Giessen, Germany. Address e-mail to Axel.Junger{at}chiru.med.uni-giessen.de
| Abstract |
|---|
|
|
|---|
Implications: The aim of this study was to investigate predictors for postoperative nausea and vomiting by using online anesthesia records. With the help of computerized data evaluation, 7 of 34 variables could be detected as risk factors. By implementing an automatic score into the record keeping system, an individual risk calculation could be made possible.
| Introduction |
|---|
|
|
|---|
We focused on this subject because PONV is a well known problem. Extensive results from prospective studies, therefore, are available for comparison. Palazzo and Evans (1) and Apfel et al. (2) used logistic regression analysis for single selected factors in calculating the individual occurrence risk of PONV and postoperative vomiting (PV) after peripheral orthopedic and otorhinolaryngological surgery. Sinclair et al. (3) studied outpatients to identify predictors for PONV.
Contrary to these studies, it was our intention to include a rather wide patient spectrum, scanned for patient-related factors as well as surgical- and anesthesia-related factors. Furthermore, we wanted to investigate whether data collected with an AIMS are suitable for comparable clinical investigations.
| Methods |
|---|
|
|
|---|
The anesthetists and anesthesia nursing teams responsible for perioperative patient care documented both the premedication visitation and the entire perioperative procedure in real time. Data from vital and respiratory variables were automatically entered. At the end of anesthesia care, the protocol file is closed and stored in a database (Oracle 7®, Oracle Corporation, Redwood Shores, CA) after several plausibility and integrity checks to prevent errors in documentation. The database is designed to conform to the principles of the relational data model and consists of 163 tables: 90 tables with 450 attributes for time-dependent data and 73 tables with 230 attributes for configuration and data description.
Evaluations and statistics were completed with the help of the Voyant® software (Brossco Systems, Espoo, Finland), a standard structured query language (SQL) tool. This program enables non-SQL specialists to generate queries graphically by using a user-friendly interface by linking different tables and setting criteria, thus facilitating statistical analyses.
The following antiemetics were used for treating nausea, retching, and vomiting: metoclopramide, droperidol, and dimenhydrinate. Prophylactic preoperative and intraoperative antiemetic therapy was administered according to the anesthesiologists judgement. Five-HT3 antagonists were not used for either antiemetic prophylaxis or treatment.
On admittance and during the stay in the PACU, patients were repeatedly asked by nurses if they had nausea. An antiemetic was administered each time a patient complained of nausea and each time retching or vomiting was observed. Hence, the antiemetic medication recorded with the AIMS allows an estimation of the PONV incidence during this period in the PACU. Patients were discharged from the PACU when their physical status was equivalent to an Aldrete score > 8 (6).
Nausea and vomiting in the PACU defines the event to be predicted. The following variables were investigated for predictive power. Patient-related variables included:
Surgical variables included:
301 of the German Social Legislation (http://www. dimdi.de/germ/klassi/ops301/fr-ops301.htm); Anesthetic variables included:
Postoperative variables included:
To statistically evaluate the data, they were exported from the database into the statistics program SPSS® (SPSS Software GmbH, Munich, Germany).
The dichotomous variable PONV in the PACU (yes/no) was used as the target criterion. Data sets were randomly split into an evaluation (n = 11,046) and a validation set (n = 16,580). The consistency of distribution between the data sets was determined by using the t-test for independent samples for metrically scaled variables as well as the
2 test for categorical variables.
The evaluation set was examined by using stepwise logistic regression analysis. Logistic regression is used to filter predictor variables that have a significant influence on "nausea and vomiting" within a multivariate model. Furthermore, logistic regression is used to estimate the coefficients (ß) of these variables. Based on these results, the probability (score) of the event "nausea and vomiting" may be estimated by a logistic function. A forward stepwise algorithm (inclusion criteria: log likelihood test ratio based on maximum likelihood function) was used for developing a logistic regression model. At each step, independent variables not yet included in the equation are tested for possible inclusion. The variable with the strongest significant contribution (P < 0.05) to improving the model is included. Variables already included in the logistic regression equation are tested for exclusion, based on the probability of a log likelihood test ratio. The analysis ended when no further variables for inclusion or exclusion were available. Variables in the logistic regression equation were defined as 1 if present and 0 if absent; constant values were calculated as absolute values.
The validation set was used to confirm the models accuracy by calculating the probability of each factor and setting up a receiver operating characteristics (ROC) curve. The ROC curve plots the true positive values (sensitivity) based on the individual score against the false positive values (1 specificity). The area under the curve indicates the accuracy of the calculated model between 0 and 1. This may be interpreted as the probability of correct patient classification in one of the two categories (antiemetic at PACU yes/no). An area of 0.5 indicates that the prediction accuracy equals a random selection. The validation set data was classified into 10 percentiles according to their calculated risks. The medians of this risk percentile were compared and correlated with the actual PONV incidence in the PACU by using linear regression analysis.
| Results |
|---|
|
|
|---|
The PONV incidence in each sample was 7.7% in the evaluation set (n = 854) and 7.8% in the validation set (n = 1300). Both data sets are comparable with regard to the potential indicators ( Tables 1 and 2).
|
|
|
The individual risk for PONV probability may be calculated by the following logistic equation:
|
|
Odds ratios were determined by using a logistic function for the validation set. The resulting probabilities were used for calculating the ROC curve. The area under the ROC curve, acting as the measure of accuracy, was 0.76.
By using a probability of 0.15 as a value having an opportune relationship between sensitivity and specificity as a threshold for factor inclusion, a sensitivity of 0.513 and a specificity of 0.816 were obtained. The score of correctly categorized patients based on this threshold is 0.79. Other values can be used and their respective variablessensitivity, specificity and the totally correctly categorized patients based on other limitsbe ascertained by using the ROC curve ( Fig. 1).
|
|
| Discussion |
|---|
|
|
|---|
We obtained results that are comparable to those of prospective studies. In our hospital, we found an overall PONV incidence of 7.9% in the PACU in the first 90 minutes after surgery. Only a fraction of PONV occurred during this short period, but we had to focus on the PONV in the PACU because of the available data recorded with the AIMS. Lerman (14) reported a PONV incidence of 10% in the PACU and 30% during the first 24 hours after surgery; Sinclair et al. (3) observed a PONV incidence of 4.6% in the PACU and 9.1% in the first 24 hours. In comparison to previously published risk scores (13,15), which only describe patient-related (sex, age, smoking) and surgical (duration and form of surgery) factors as the main predictors, we could prove that N2O and intraoperative opioids had a strong influence on the incidence of postoperative nausea, retching, and vomiting, whereas IVA with propofol has a protective effect. This has relevance for anesthesiologists because they can influence these factors.
We could confirm the strong influence of female sex on the PONV incidence with an odds ratio (OR) of 2.45. This factor is the strongest variable in the score developed by Apfel et al. (2) (OR = 3.61). In the risk score developed by Sinclair et al.(3), female sex also had a more than twofold increased risk for PONV (OR = 2.78) together with the form of surgery.
Smoking had a protective influence on PONV in the PACU with an OR = 0.53. In the scores developed by Apfel et al. (2) and Sinclair et al. (3), this variable has a similarly strong influence with an OR = 0.48 and OR = 0.66, respectively. This association is also included in the score developed by Koivuranta et al. (15).
Increased age of patients was found to have a positive effect on the risk of postoperative sickness in the PACU, which was also included in the other scores (2,3). This significant influence was also found in the studies by Bellville et al. (16) and Cohen et al. (17), whereas in the multivariate analysis of Palazzo and Evans (1) as well as Koivuranta et al. (15), this could not be confirmed. The different associations of patient age in our risk score and that of Apfel et al. (2) may be explained on one hand with differences in the selected patient collective and on the other hand in the metric variables themselves. In their study, Apfel et al. (2) included patients from the age of 35 to 60 years, whereas in our study, there was no age limit. In our collective, the age ranges from infants to patients older than 80 years.
Duration of surgery is also included in our risk score as a surgical and anesthesia risk factor, comparable to Apfel et al. (2) and Sinclair et al. (3). The different associations may also be attributed to a difference in the patient collective. In our score, duration of surgery is a metric variable, whereas Apfel et al. (2) divided patients into two groups with a 60-minute limit. Sinclair et al. (3) used a 30-minute time scale.
In contrast to other risk scores (13,15) in our score for PONV in the PACU, three anesthetic variables were included. Opioids increased the risk more than fourfold, whereas the protective influence of IVA (OR = 0.40) showed a stronger association than sex (OR = 2.45), which has in the scores by Apfel et al. (2) (OR = 3.61) and Sinclair et al. (3) (OR = 2.78) one of the strongest associations. In their score, however, consideration of anesthetic factors is only marginal, which has been criticized by Kortilla (18) in his editorial for risk scores. The administration of N2O increasing the risk more than twofold was identified as a third anesthetic factor showing a stronger association for the probability of PONV in the PACU, than individual factors included in other risk scores (13,15).
The negative influence of opioids on postoperative sickness is well known, whereas the influence of N2O is still a matter of controversy. The use of N2O in combination with desflurane showed an increased incidence of PONV (19). No increased vomiting, however, was found in combination with propofol (20,21). This observation may be explained with the antiemetic effect of propofol. In comparison with sevoflurane administration during strabismus surgery, propofol induces less nausea and vomiting (22). These observations are also confirmed by our investigations. IVA has a protective effect on the individual risk of PONV in the PACU.
The scores of Apfel et al. (2), Koivuranta et al. (15), Sinclair et al. (3), and Palazzo and Evans (1) include the variables "previous postoperative emetic history" and "motion sickness" in their risk calculation. Because of the fact that these variables were not standardized items in the record-keeping software, it was not possible to reliably generate the two variables from the database. Therefore, these items were not investigated as potential indicators. However, to insure the future improvement of the AIMS, this will have to be changed.
The area under the ROC curve, acting as measure of accuracy, was 0.76. Compared with the score by Apfel et al. (2) with a value of 0.77 and the score by Sinclair et al. (3) with a value of 0.78, our equation possesses acceptable predictive power.
PONV probability in the PACU calculated with this model, being 0.076, is at a relatively low level. If, for example, a calculated probability of 0.15 were chosen as an inclusion criterion, more than 50% (sensitivity) of patients suffering from PV would be classified for antiemetic therapy, but less than one fifth (specificity) of all nonclassified patients would have to vomit. Based on this inclusion threshold, the rate of all patients classified correctly lies at 0.79. In comparison, the study by Eberhart et al. (23) showed a threshold of 30% for PONV occurrence during the first 24 hours, giving a specificity of 0.80 and a sensitivity of 0.40 at a random score of just 0.20. With the help of the ROC curve, a basis for the decision at hand can be given for detecting patients having an increased PONV risk.
Because the investigation period was limited to the time spent in the PACU, validation of our risk score within a 24-hour postsurgical period remains to be demonstrated in a prospective study. Motion sickness and previous history of PONV will also have to be investigated for a possible inclusion in our model.
In the future, as computerized record-keeping systems proliferate, this type of data collection from existing databases will become increasingly important (24,25). The major requirements to enable the routine use of these systems for scientific purposes are the implementation of record keeping software for data collection into the daily clinical work flow and the selection of a standardized, clearly structured database. In our opinion, only SQL-based standard database systems can be useful for this purpose, provided that the database structure is consistent with the principles of the relational data model and that a clearly defined documentation of the database structure is available. Because of the availability of commercial and easy-to-use SQL tools, the time required for becoming familiar with the subject is acceptably short. The use of standards offers the potential to combine data with other databases from other institutions (e.g., for external quality assurance projects).
The retrospective analysis based on these online data routinely collected do not give the same objective comparison as does a prospective collection of complete (e.g., "previous postoperative emetic history" and "motion sickness") and uniform study material data. But it can be used to formulate a hypothesis or as a prompt to a planned prospective study. Integrity checks and logical algorithms are suitable for improving the quality of retrospective data and must be taken into consideration in further developments of AIMSs.
As part of the further development of the presented AIMS, an equation for risk calculation could be implemented into the program and an individual risk could be calculated automatically after entry of all data during premedication as well as during surgery. In this way, at-risk patients could profit from an adapted anesthesia technique and prophylactic antiemetic therapy. This may lead to a cost reduction and increased patient satisfaction.
| References |
|---|
|
|
|---|
This article has been cited by other articles:
![]() |
S. Muravchick, J. E. Caldwell, R. H. Epstein, M. Galati, W. J. Levy, M. O'Reilly, J. S. Plagenhoef, M. Rehman, D. L. Reich, and M. M. Vigoda Anesthesia Information Management System Implementation: A Practical Guide Anesth. Analg., November 1, 2008; 107(5): 1598 - 1608. [Abstract] [Full Text] [PDF] |
||||
![]() |
K. Leslie, P. S. Myles, M. T. V. Chan, M. J. Paech, P. Peyton, A. Forbes, D. McKenzie, and the ENIGMA Trial Group Risk factors for severe postoperative nausea and vomiting in a randomized trial of nitrous oxide-based vs nitrous oxide-free anaesthesia Br. J. Anaesth., October 1, 2008; 101(4): 498 - 505. [Abstract] [Full Text] [PDF] |
||||
![]() |
M. M. Vigoda, L. I. Rodriguez, E. Wu, K. Perry, R. Duncan, D. J. Birnbach, and D. A. Lubarsky The Use of an Anesthesia Information System to Identify and Trend Gender Disparities in Outpatient Medical Management of Patients with Coronary Artery Disease Anesth. Analg., July 1, 2008; 107(1): 185 - 192. [Abstract] [Full Text] [PDF] |
||||
![]() |
T. J. Gan, T. A. Meyer, C. C. Apfel, F. Chung, P. J. Davis, A. S. Habib, V. D. Hooper, A. L. Kovac, P. Kranke, P. Myles, et al. Society for Ambulatory Anesthesia Guidelines for the Management of Postoperative Nausea and Vomiting Anesth. Analg., December 1, 2007; 105(6): 1615 - 1628. [Abstract] [Full Text] [PDF] |
||||
![]() |
T. J. Gan Risk factors for postoperative nausea and vomiting. Anesth. Analg., June 1, 2006; 102(6): 1884 - 1898. [Abstract] [Full Text] [PDF] |
||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|