Cannabis Use Is Associated With a Small Increase in the Risk of Postoperative Nausea and Vomiting

A Retrospective Machine-learning Causal Analysis

Wendy Suhre; Vikas O'Reilly-Shah; Wil Van Cleve

Disclosures

BMC Anesthesiol. 2020;20(115) 

In This Article

Results

After applying inclusion and exclusion criteria, 27,388 unique anesthetics at UWMC were available for analysis (Table 1). When stratified by self-described cannabis use, a higher proportion of daily users were ASA 3 (58%) than non-users (42.7%). Considering risk factors for PONV: daily cannabis users were more often male and more likely to smoke tobacco, but also had higher rates of prior PONV/motion sickness and higher rates of opioid use in the PACU when compared to non-users. The unadjusted incidence of PONV was higher in daily users (21.9%) and current users (18.8%) when compared to non-users (17.3%).

A probit BART model was created to model the probability of any PONV or rescue administration in the recovery room. Graphical depiction of the results of this model are provided in Figure 1 and Figure 2. The pooled relative risk of PONV was higher in daily users when compared to non-users, with a relative risk of 1.20 (95% CI 1.00–1.45, posterior probability RR > 1 = 97.6%), and slightly higher in current users compared to non users, with a relative risk of 1.07 (95% CI 0.94–1.21, posterior probability RR > 1 = 84.7%). As can be observed in Figure 1, the increased probability of PONV associated with daily cannabis use appeared to be moderated with increasing Apfel (PONV risk) score. In terms of absolute changes in probability of PONV, daily users were predicted to have a mean increase in risk of 3.3% (95%CI 0.4–6.4%) compared to non-users, while current users were predicted to have a mean increase in risk of PONV of 1.2% (95% CI -0.7 - 3.1%).

Figure 1.

Sample average treatment effect (SATE) measured as relative risk of any postoperative nausea/vomiting modeling entire sample as non-users, current (non-daily), or daily cannabis users. Estimates stratified by Apfel score. 95% Bayesian posterior credible interval for SATE generated from 1000 MCMC estimates. Pooled estimate across all Apfel scores showed at right of each grouping

Figure 2.

Sample mean predicted probability of PONV of postoperative nausea/vomiting stratified by Apfel score and conditioned on pattern of cannabis use. 95% Bayesian posterior credible interval for mean probability generated from 1000 MCMC estimates. Pooled mean probability estimate across all Apfel scores showed at right of each grouping

We validated our BART model's results using two techniques: first, we compared its predictions to a Bayesian logistic regression model using propensity score adjustment (Table 2). The model's odds ratio for daily cannabis use was 1.16 (95% CI 0.99–1.35), and the sample average treatment effect (calculated as a relative risk) was 1.13 (95% CI 0.97–1.30). We then replicated our BART modeling strategy using independently generated data at HMC (Figure 3). We observed a nearly identical sATE at HMC, with an estimated mean relative risk of 1.19 (95% CI 1.00–1.40, posterior probability RR > 1 = 97.7%) for daily cannabis users.

Figure 3.

Sample average treatment effect (SATE) measured as relative risk of any postoperative nausea/vomiting modeling entire sample as non-users, current (non-daily), or daily cannabis users. Estimates stratified by Apfel score. 95% Bayesian posterior credible interval for SATE generated from 1000 MCMC estimates. Pooled estimate across all Apfel scores showed at right of each grouping

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....