We first conducted a retrospective secondary analysis of data from the aforementioned clinical trial (6) entailing nerve blocks/RA for joint replacement (primary hip or primary knee arthroplasty), but not involving ITM [as did our more recent report (2)]. This was an institutional review board (IRB)-approved amendment, exempt from patient research consent above and beyond original study consent, as well as original clinical surgery and anesthesia consent. Institutional approval was obtained (approval/exemption 1617210, VA Pittsburgh Healthcare System, Pittsburgh, PA, USA), allowing for tracking and external reporting of the described, newly-queried outcome data. For n = 115 Veterans (6), both electronic medical records (EMR) and database-archived study data were queried. For PONV prophylaxis, these patients received standardized perphenazine 8 mg per os (PO) (before case), ondansetron 4 mg IV (end of case), and possibly discretionary aprepitant 40 mg PO (before case). Of the described n = 115 Veterans, 48 (42%) experienced at least one episode of PONV (of any severity) during the joint replacement hospitalization study period, and these cases were compared against those who were PONV-free on postoperative days (POD) #0–2. The novel factors explored were based on our prospective study data case report forms or informed by recent Veteran-centric and active duty-specific publications from 2021 to 2022 (see Supplementary Tables 1a,b) (7–9). This analysis was conducted using bivariate analyses via cross-tabular chi-square tests (or Fisher’s exact tests, where appropriate; Supplementary Table 2) and then (due to the small sample size of available case data) multivariable parsimonious logistic regression (Table 1; SAS, SAS Institute Inc., Cary, NC, USA). Finally, despite the low number of African-American patients in the RA PONV (n = 12) retrospective analysis of prospective data, race was identified in the most parsimonious model as an associated predictor of PONV. Interaction terms of race with all other variables were explored during development of the most parsimonious model (data not shown), and no interaction terms were found to be associated with (or a signal of) PONV (with allowance of P < 0.2, for signals). Assuming low sample size as the most likely reason for these results, we then conducted exploratory bivariate analyses of race with all other measures from the most parsimonious model. The results of these analyses are presented in Supplementary Table 3.

We then conducted analyses of data from an ongoing observational QI study (2) regarding Veterans receiving GA (entailing propofol induction as the institutional standard of care, at ∼1–2 µg/kg over 30–60 s, depending on health status) with or without undergoing (before surgery) an ITM procedure (with the ITM procedure typically accompanied by a five-drug AEPPx, vs. case-matched historical controls not receiving ITM and not receiving a five-drug AEPPx). Institutional approval was obtained recently on 19 February 2025 (approval/exemption 1670098, VA Pittsburgh Healthcare System, Pittsburgh, PA, USA), allowing for tracking and external reporting of QI outcome data. In the previous publication (2), we had shown in this retrospective, case-matched QI series that a five-drug AEPPx was significantly protective against PONV (vs. non-ITM and typically four drugs or less for PONV prophylaxis). The five-drug AEPPx regimen included (predominantly) palonosetron, perphenazine, aprepitant, diphenhydramine, and dexamethasone (2). After VA Pittsburgh IRB approval/exemption, for n = 468 Veterans, we collated and analyzed EMR and subsequently used SPSS (Version 29, IBM, Armonk, NY, USA) to generate cross-tabular chi-square tests (data not shown) and multivariable logistic regression (data shown below), with a liberal P-value (P < 0.1) to also identify potential signals (P ≥ 0.05 and P < 0.1), to inform future risk factor research, and to compare PONV-free patients with patients that experienced PONV, first on POD#0–1 after surgery and (separately) on POD#2 (after no PONV on POD#0–1). The overview of results from both the RA and GA analyses (Table 2) is displayed adjacent to each other.

The inclusion and exclusion criteria for the RA and GA studies are previously described (2, 6).

Significant RFs for higher PONV risk, common to both RA and GA on POD#0–1, included (i) absence of diabetes mellitus, (ii) African-American/non-Caucasian race/ethnicity, (iii) peptic ulcer/gastric ulcer/any upper gastrointestinal conditions, and (iv) alcohol use (Table 2 and Supplementary Table 2).

Significant RFs relevant and specific to RA on POD#0–2 (Table 2) included (i) oral morphine equivalents consumed postoperatively, as an incremental (step) function, (ii) any recent nausea or vomiting (within ∼2 weeks of the joint replacement nerve block study date), (iii) baseline opioid use, (iv) history of any vomiting that was not related to PONV, and (v) female gender.

We identified significant RFs relevant to GA on both POD#0–1 and POD#2 (separately, specifically when there was no PONV POD#0–1; Tables 2–4). These RFs, both POD#0–1 and POD#2, included (i) history of any headache diagnosis and (ii) vertigo or history of any vertiginous diagnoses (such as Ménière’s disease, associated with more PONV). Significant RFs relevant to GA on POD#0–1 specifically (Tables 2, 3) included (i) younger age on date of surgery (associated with more PONV); (ii) a “bundled” factor entailing ITM, five-drug AEPPx, and no IV opioids on day 0 (associated with less PONV); (iii) history of past nausea (on EMR query) not necessarily occurring in the postoperative setting (associated with more PONV); and (iv) postoperative peak pain score difference from baseline pain score preoperatively (higher pain scores above baseline associated with more PONV). Signals/RFs, all relevant to increasing PONV risks after GA on POD#2 specifically (Tables 2, 4) after encountering no PONV POD#0–1, entailed (i) Vietnam War Veteran, (ii) step function related to higher ITM doses, (iii) history of any vomiting (in EMR review) not related to PONV, (iv) anxiety, (v) past PONV, and (vi) headache diagnoses (per EMR).

Based on the results, some existing and some novel PONV RFs seem to warrant further attention. To date, non-Caucasian race has not been associated with an increased risk of PONV [except for the recently declared higher risk in the Chinese population (10); patients of Asian descent were not present in our case series described]. In our analyses, non-Caucasian race/ethnicity status was statistically significant for more PONV associations in both RA and GA contexts, compelling future analyses in both Veteran and civilian populations to uncover potential pharmacoequity-related considerations.

Some traditional RFs for PONV remained significant for some of the analyses run, but no traditional Apfel factors (11) were found predictive across all analyses. Other traditional RFs, such as non-smoker, were not significant in our models, which may introduce the possibility of some of these Veteran-centric (or RA-related) factors being potentially more dominating than some traditional RFs (11), pending further study. Again, we are not aware of either the Veterans or the RA populations having had representative RF-based analyses regarding PONV. Further refinement of these factors in Veterans, including those which appear to be novel, may prove to be more statistically robust than traditional (11) factors, which then could compel re-examination of these novel RFs in the general population.

Recent studies have shown other predictors of PONV, not necessarily specific to POD#2, after not encountering PONV on POD#0–1. First would be semaglutide and glucagon-like peptide-1 (GLP-1) agonists promoting PONV (12), and another would be cannabis-related products (13, 14). Our group's work showed oral sustained-release dextromethorphan as a POD#2 PONV predictor after no PONV on POD#0–1 (15), particularly (consistent with its package insert) related to cumulative doses of opioid-sparing dextromethorphan over consecutive days. This suggests the potential meaningful value of proposed booster AEPPx throughout the hospital stay, as non-opioid analgesic efforts, such as the dextromethorphan and celecoxib, still appear to be associated with unwanted PONV at POD#2 (15). Furthermore, the same recent study (15) found that a “bundle” of ITM including magnesium, as well as diabetes mellitus, was associated with protectors against PONV on POD#2 (15) after no PONV on POD#0–1.

Intraoperative “usual opioids” (i.e., fentanyl and hydromorphone) and our efforts to reduce or eliminate their use intraoperatively over time are reported elsewhere (15). Strategies for ITM dosing are also reported elsewhere (16).

Larger-sample prospective analyses involving newly identified predictive variables, and potential signals, may be warranted in both Veteran/active duty and civilian populations undergoing RA and/or GA to allow for better generalizability. It is also important to recognize that RFs differed significantly from POD#0–1 to POD#2 when analyzing the GA population, and one limitation of the data included herein is the lack of such analysis in the RA population. However, the RA risk factor query (which took place in 2023) was, in part, designed to inform our future queries in the wider, larger-sample GA population (with these GA case EMR reviewed in 2024).

Veterans may be anecdotally considered a low-risk population for PONV. However, Veterans have not been specifically studied in the present or past PONV consensus guidelines, nor have patients who specifically received ITM. Therefore, forecasts of five-drug PONV prophylaxis success in ITM cases may differ in the non-Veteran (with or without ITM) population. To offset this limitation, we allowed for signals (with P ≥ 0.05 but ≤0.1) into our final regression equations, often representing non-traditional but potentially interacting risk factors, to query for potential epidemiologic influence (or “bedside habit” influence) beyond traditional PONV risk factors.

Other “mixed signals” from our data include seizure disorder (possibly PONV-protective, data not shown) and peptic/gastric ulcer history (possibly PONV-predictive) and should be queried during history-taking (Table 2). Another recommendation, related to our recent report (2), is that a low-cost, off-patent five-drug AEPPx could be administered before any/every anesthetic, to offset insufficient sensitivity/specificity of past and present PONV consensus guidelines not yet incorporating these emerging factors, further promoting equity and inclusion, via primary prevention (pending further refinement of data).

Next, the RA cohort had case report form data that were completed in real time (including graded severity by the patient), while the GA cohort’s PONV was only determined via post hoc EMR QI review (without patient-reported severity grading). This forced a simpler overall dichotomous capture (no/yes) for all PONV data presented, although it seems conceivable that RA-based nausea would likely seem less severe than GA-based nausea if coexisting opioid load was otherwise constant.

Finally, we acknowledge that we were unable to derive and present expected a priori sample size determinations and power analyses. Our observations are limited to a single-center US Veterans Affairs population, which may not generalize to broader, diverse patient populations. For the GA/QI observational study, we could only “case match” cases that were performed and assess similar historical control cases performed at the same institution in the described time periods as comparators. It is difficult to expand an observational sample size (with limited resource support) to cases that are neither present in one's institution nor present in the institution's medical record archive. Future work by other research teams expanding the cohort and including non-Veteran populations would likely enhance applicability. Using the described consecutive caseload, our goal was to provide preliminary data for external researchers to create their own cohorts (such as in non-Veteran populations) or prospective randomized study groups incorporating the described paradigm shifts, as soon as the significant findings and signals were noted in analysis.