Dongwoo Chae (1,2)+, So Yeon Kim (3)+, Dongwoo Han (3)*, and Kyungsoo Park (1,2)*
(1) Department of Pharmacology, Yonsei University College of Medicine, Seoul, Korea (2) Brain Korea 21 Plus Project for Medical Science, Yonsei University, Seoul, Korea (3) Department of Anesthesiology and Pain Medicine, Anesthesia and Pain Research Institute, Seoul, Korea +Both authors contributed equally to this work, *Co-corresponding authors
Introduction: Postoperative nausea and vomiting (PONV) is often ranked by patients as one of the most undesirable postoperative outcomes [1]. While PONV is typically self-limited, it can lead to unanticipated admissions [2] and postoperative complications. Risk factors of PONV have been studied extensively, and two of the most well-known risk scores in adults have been proposed by Apfel [3] and Koivuranta [4]. Female gender, history of motion sickness or PONV, non-smoking, and the use of postoperative opioids appear to be the most robust risk factors. While more elaborate models have been proposed, difficulty in their implementation seems to have limited their widespread use. Regarding postoperative opioids, there is evidence that there exists a strong dose-response relationship between postoperative opioid dose and PONV [5]. With increased use of postoperative patient controlled analgesia (PCA), accurate prediction of Fentanyl’s nausea promoting effect would contribute to better patient management.
Objectives:
- Develop a predictive model of postoperative nausea and vomiting (PONV) and identify its risk factors.
- Investigate the time-varying effects of the risk factors.
- Develop a web application that facilitates the use of the predictive model.
Methods: Data from 22,144 postoperative patients who underwent general anesthesia and treated with intravenous Fentanyl based PCA were retrospectively collected from electronic medical records of two hospitals located in Seoul, South Korea. The dataset was randomly split into training and test sets in a ratio of 8:2. A stepwise logistic regression was performed on the training dataset to select the most significant covariates. This process was repeated 100 times with different random seeds, whereupon the frequency of each covariate inclusion was calculated. Using only those covariates that were selected in more than 95% of the resampled datasets, a multivariate logistic regression model predicting the incidence of PONV within 48 postoperative hours was built. Since effects of different types of surgery are often confounded by peculiarities of the hospitals, surgeons, and the treating physician, we also built an alternative mixed effects model that treated surgery type as a random effect. Non-linearity of the continuous covariates was explored in this step. Finally, the time-varying effects of the selected covariates were examined by performing multivariate logistic regressions for PONV occurring within 0~6 h, 6~12 h, 12~18 h, 18~24 h, and 24~48 h postoperative time periods. A web application that interactively outputs the probability of PONV was developed based on the final model using RShiny.
Results: As with previous studies, female gender, history of motion sickness and PONV, and non-smoking were identified as significant risk factors. In addition, the following covariates were found statistically significant: DM, ASA class, cancer surgery, laparoscopic surgery, use of total IV anesthesia (TIVA), postoperative pain severity, and use of postoperative Tramadol. A strong dose-response relationship between Fentanyl mg/kghr and PONV was found. The nausea-promoting effect of Fentanyl seems to be alleviated when ketorolac is added to the PCA regimen. The parameter estimates acquired from a mixed effects model yielded similar results, and the standard deviation of the random variability incurred by different surgery types in a logit scale was estimated as 0.65. Time varying effects of the selected covariates were quantified and visualized using a rectangular grid with each axis representing the covariates and the time bins.
Conclusions: Our findings successfully identified factors associated with both increased and decreased risks of PONV. The magnitude of random variability due to different surgery types was identified using a mixed effects modelling technique, and the time varying effects of the selected covariates examined. An interactive web application developed using the predictive model is expected to improve clinician’s accessibility and facilitate its use in the clinic.
References:
[1] Macario A, Weinger M, Carney S, Kim A. Which clinical anesthesia outcomes are important to avoid? The perspective of patients. Anesth Analg 1999; 89:652.
[2] Fortier J, Chung F, Su J. Unanticipated admission after ambulatory surgery–a prospective study. Can J Anaesth 1998; 45:612.
[3] Apfel CC, Läärä E, Koivuranta M, et al. A simplified risk score for predicting postoperative nausea and vomiting: conclusions from cross-validations between two centers. Anesthesiology 1999; 91:693.
[4] Koivuranta M, Läärä E, Snåre L, Alahuhta S. A survey of postoperative nausea and vomiting. Anaesthesia 1997; 52:443.
[5] Roberts GW, Bekker TB, Carlsen HH, et al. Postoperative nausea and vomiting are strongly influenced by postoperative opioid use in a dose-related manner. Anesth Analg 2005; 101:1343.
Reference: PAGE 28 (2019) Abstr 8882 [www.page-meeting.org/?abstract=8882]
Poster: Clinical Applications