Abstract
Objective The Plasma Exchange and Glucocorticoids in Severe Antineutrophil Cytoplasmic Antibody–Associated Vasculitis (PEXIVAS) trial evaluated plasma exchange (PLEX) and glucocorticoid dose in antineutrophil cytoplasmic antibody–associated vasculitis (AAV). This study reanalyzed the PEXIVAS trial using Bayesian methods to focus on patients with diffuse alveolar hemorrhage (DAH).
Methods In the PEXIVAS trial, adults with AAV and kidney injury and/or DAH were randomized to receive PLEX or no PLEX and reduced- or standard-dose glucocorticoids. This study evaluated 1-year survival and severe infection among participants with DAH, no DAH, and severe DAH (oxygen saturation ≤ 85% on room air or mechanical ventilation) across multiple priors. The secondary outcome was severe infection (leading to hospitalization, intravenous antibiotics, or death). This study calculated mean hazard ratios (HRs) for survival and odds ratios (ORs) for infection, 95% credible intervals (CrIs), and probabilities of survival benefit (HR < 1) and infection (OR > 1).
Results Among 704 participants (191 with DAH and 61 with severe DAH), using noninformative priors, the probability of improved survival was 93% for participants with DAH receiving PLEX (HR 0.52; 95% CrI 0.21-1.26) and 67% for those without DAH (HR 0.86; 95% CrI 0.43-1.71). Participants receiving reduced-dose glucocorticoids had a 98% probability of improved survival without DAH (HR 0.46; 95% CrI 0.22-0.95) but a 12% probability with severe DAH (HR 2.02; 95% CrI 0.64-6.35). Plasma exchange increased the odds of severe infection (OR 1.25; 95% CrI 0.92-1.70; 92% probability of OR > 1), whereas reduced-dose glucocorticoids decreased the odds of infection (OR 0.85; 95% CrI 0.63-1.15; 85% probability of OR < 1). Results were consistent across priors.
Conclusion Patients with AAV and DAH may benefit from PLEX. Reduced-dose glucocorticoids improve survival for individuals without DAH but may be harmful in patients with severe DAH.
- antibodies
- antineutrophil cytoplasmic
- Bayes theorem
- glucocorticoids
- plasma exchange
- rapidly progressive glomerulonephritis with pulmonary hemorrhage
Diffuse alveolar hemorrhage (DAH) is a life-threatening manifestation of antineutrophil cytoplasmic antibody (ANCA)-associated vasculitis (AAV). The Plasma Exchange and Glucocorticoids in Severe Antineutrophil Cytoplasmic Antibody–Associated Vasculitis (PEXIVAS) trial was the largest randomized controlled trial (RCT) in individuals with AAV.1 Using a 2 × 2 factorial design, individuals were assigned to receive plasma exchange (PLEX) or no PLEX and a reduced-dose or standard-dose glucocorticoid (GC) regimen (in addition to either rituximab or cyclophosphamide). The primary outcome was a composite of end-stage kidney disease (ESKD) or death. The study concluded that PLEX did not reduce incidence of the primary outcome but reduced-dose GCs were noninferior to the standard regimen.
Recently, the results from the PEXIVAS trial were analyzed with a focus on DAH.2 Although the study did not identify statistically significant differences between those with and without DAH, there was a trend toward benefit with standard-dose GC dosing in those with severe DAH. The trial was not powered to detect differences in treatment in this subgroup, limiting the conclusions that could be drawn from this analysis.
Another way to evaluate the risks and benefits of an intervention is through Bayesian analysis.3 Whereas the traditional frequentist paradigm estimates an unknown, fixed, true value of a population parameter of interest using sample data (eg, a single hazard ratio [HR] for the entire population), the Bayesian paradigm treats each parameter as a random variable that follows an unknown probability distribution (the posterior distribution) and estimates it by combining the observed data (the likelihood) with prior knowledge (the prior distribution).4
The Bayesian approach provides several benefits, including the ability to interpret study results using probabilistic summaries, limiting difficulties in interpreting treatment effects when results do not meet frequentist statistical significance or when the study is underpowered5,6; avoiding incorrect conclusions of statistical significance when multiple hypotheses or subgroups are tested7; and allowing incorporation of a wide range of prior information into the analysis.8 Moreover, the Bayesian posterior distribution directly addresses effect size and can estimate the probability of clinically relevant effects without increasing the study error rate. This approach is also similar to how physicians address clinical uncertainty.9,10 The Bayesian approach to such situations has been successfully used in the rheumatology literature.11-18
This study reanalyzed the results of the PEXIVAS trial for GC dose and PLEX, across those with and without DAH and by DAH severity using Bayesian methods. A wide range of priors was incorporated to evaluate treatment effects across a range of plausible clinical beliefs regarding efficacy. Using these methods, the potential effects of these interventions were contextualized in individuals with this uncommon but important manifestation of a rare disease.
METHODS
Trial procedures. Individuals aged > 15 years with new or relapsing granulomatosis with polyangiitis or microscopic polyangiitis were enrolled if they had kidney and/or pulmonary involvement, defined as reduced glomerular filtration rate (< 50 mL/min/1.73 m2) and/or DAH, respectively. DAH was defined by a combination of compatible radiographic findings that could not be otherwise explained and at least 1 of the following: increasing bloody returns on bronchoalveolar lavage, hemoptysis, unexplained anemia, or increased diffusing capacity of carbon monoxide. DAH was classified as severe if it was associated with oxygen saturation ≤ 85% on room air or requiring mechanical ventilation. In a 2 × 2 factorial design, participants were randomized in a 1:1:1:1 ratio to receive either PLEX, defined as 7 treatments within 14 days of randomization, or no PLEX, and a standard- or reduced-dose GC taper. GC treatment consisted of 1- to 3-gram methylprednisolone pulse for 1 to 3 days, followed by approximately 1 mg/kg of prednisone or prednisolone per day for all participants. Thereafter, participants in the reduced-dose GC taper group had an approximately 50% reduction in total GC dose, which tapered over a 52-week course (see Walsh et al1 for details). In addition, participants received an induction agent—oral cyclophosphamide, intravenous cyclophosphamide, or rituximab—per investigator preference.1
Groups were stratified by age (< 60 or ≥ 60 years), kidney function (creatinine < 5.6 or ≥ 5.6 mg/dL), ANCA subtype (proteinase 3 or myeloperoxidase), induction agent (oral cyclophosphamide, intravenous cyclophosphamide, or rituximab), and severity of pulmonary hemorrhage (no hemorrhage, nonsevere hemorrhage, or severe hemorrhage). The trial was designed to test superiority of PLEX compared with no PLEX and the noninferiority of reduced-dose GC regimen compared with a standard-dose regimen. The primary outcome in the original study was a composite of all-cause mortality and ESKD.1,2 The primary outcome in this analysis was 1-year mortality rather than the composite outcome used in the main trial, as the aim was to evaluate the efficacy of PLEX and a reduced-dose GC taper in individuals with DAH and AAV, who may or may not have coexisting kidney disease as a result of AAV. This outcome was chosen to mirror the frequentist analysis of PEXIVAS outcomes among those with DAH, which also used 1-year mortality as the primary outcome.2 Secondary outcomes included death at 3 months and severe infections within 1 year, defined as infection leading to hospitalization, intravenous antibiotic therapy, or death.
Analysis. The primary analysis of this study compared the overall survival HR for both PLEX and reduced-dose GCs across 5 patient subgroups: all patients, patients with no DAH, patients with DAH, patients with nonsevere DAH, and patients with severe DAH. Participants with DAH may also have had concomitant kidney manifestation of AAV. For the primary analysis, a noninformative prior was used. As in prior analyses,2 estimates were adjusted for age, ANCA subtype, serum creatinine, and induction agent. Given the purpose of this analysis, the likelihood distribution was derived assuming a log-normal distribution based on the published mean and SD.2 Conjugate normal-normal updating was then used to obtain the posterior distribution (method details are provided in the Supplementary Materials, available with the online version of this article). This approach produces an unbiased Bayesian estimate incorporating log-normal prior information and the likelihood distribution from the frequentist estimate. Similar methods were used for the secondary outcomes: adjusted odds ratio (OR) of severe infection within 1 year and the OR of mortality at 3 months.2
For each subgroup, the posterior mean point estimate and 95% credible interval (CrI) were obtained. In addition, the area under the curve (AUC) of the posterior probability distribution beyond a clinically relevant threshold was calculated, representing the probability that the HR or OR for a given intervention exceeded the specified threshold. For the primary analysis, a threshold < 1 was used to estimate the probability of improved survival with each intervention. The AUC was also evaluated for HR thresholds of < 0.9, < 0.8, and < 0.67. This evaluated the probability of an effect of that magnitude or more extreme, consistent with prior studies.19 Additional sensitivity analyses were performed by evaluating the posterior distribution in each of the 5 subgroups across different informed priors. Analyses followed the Reporting of Bayes Used in Clinical Studies (ROBUST) criteria.20 All analyses were performed using R, version 4.3.3 (R Foundation for Statistical Computing; reproducible R code is provided in the Supplementary Materials, available with the online version of this article).
Priors. For the primary analysis of both PLEX and reduced-dose GCs, a noninformative distribution with the same characteristics (mean HR 1, SD of logarithm of HR 10) was assumed. Several informative priors based on prior observational studies were also used (Table 1).21 Given the greater uncertainty associated with observational studies compared with RCTs, these priors were downweighted by 50% (ie, the variance from the study was doubled).
Prior distributions and rationale for use.
Three additional priors were used in sensitivity analyses to evaluate the effect of PLEX. Prior observational studies formed the optimistic22 and pessimistic priors.23 The prior distribution was generated from the natural logarithm of the mortality risk ratio in the observational study, using the CIs to generate a log-normal distribution. This study used an additional sensitivity prior assuming a null effect (HR 1) of PLEX with variability similar to that of the optimistic prior.
Informative priors to assess GC dose were obtained from an observational study that compared the GC taper used in the PEXIVAS trial with a standard GC taper and found a nonsignificant increase in mortality.24 The same procedure described above was used in this study to obtain the prior distribution, using the point estimate and CIs to generate a log-normal distribution. The same variability was used for the other informative priors, with different mean effects for a null prior (HR 1) and an optimistic prior, assuming an HR of equal magnitude in the opposite direction (see Table 1 for further details). This analysis was conducted after the PEXIVAS trial but before the Bayesian analysis and therefore provided a suitable prior for interpreting the PEXIVAS results. Priors for secondary outcomes followed similar reasoning as outlined above (Supplementary Tables S1-S2, available with the online version of this article).
RESULTS
A total of 704 individuals were enrolled across 16 countries between 2010 and 2016, of whom 191 had DAH and 61 had severe DAH, with approximately equal numbers in each treatment group.2 In this study, individuals were followed for 1 year after randomization. During the first year after randomization, 57 individuals (8.1%) died, including 23 (12%) with DAH and 16 (26.2%) with severe DAH. The likelihood distributions derived from the trial are shown in Supplementary Figures S1-S2 (available with the online version of this article).
Results of the primary analysis are depicted in Table 2. The overall posterior mean HR for the effect of PLEX using a noninformative prior was 0.75 (95% CrI 0.44-1.27). Individuals without DAH had a mean HR of 0.86 (95% CrI 0.43-1.71), whereas those with DAH had a mean HR of 0.52 (95% CrI 0.21-1.26). Among participants receiving PLEX in the overall cohort, there was an 86% probability of decreased mortality (HR < 1), which varied across subgroups, with a 67% probability among those without DAH and a 93% probability among those with DAH (Figure 1).
Mean HRs and 95% CrIs under each prior belief by subgroup.
Posterior distribution of HRs for the effect of PLEX on mortality assuming multiple possible priors including (A) noninformative prior, (B) optimistic prior, (C) null prior, and (D) pessimistic prior. DAH: diffuse alveolar hemorrhage; HR: hazard ratio; PLEX: plasma exchange.
Assuming a noninformative prior distribution, participants receiving reduced-dose GCs had a posterior mean HR of 0.69 (95% CrI 0.41-1.17; Table 2). The presence of severe DAH increased the mean HR of those with reduced-dose GCs to 2.02 (95% CrI 0.64-6.35). Evaluation of the AUC of these posterior distributions showed a 92% probability of decreased mortality (HR < 1) with reduced-dose GCs. However, when stratified by DAH, participants without DAH had mean HR 0.46 (95% CrI 0.22-0.95) and participants with nonsevere DAH had mean HR 0.62 (95% CrI 0.12-3.20), corresponding to 98% and 72% probability of improved survival when receiving reduced-dose GCs, respectively. Individuals with severe DAH had mean HR 2.02 (95% CrI 0.64-6.35), corresponding to 12% probability of improved survival when receiving reduced-dose GCs (ie, an 88% probability of reduced survival when receiving reduced-dose GCs; Figure 2).
Posterior distribution of HRs for the effect of reduced-dose GCs compared with standard-dose glucocorticoids on overall survival using (A) noninformative prior, (B) optimistic prior, (C) null prior, and (D) pessimistic prior. DAH: diffuse alveolar hemorrhage; GC: glucocorticoid; HR: hazard ratio.
This study used informative priors as sensitivity analyses, as shown in Table 2. Using the optimistic prior, the overall effect of PLEX was a mean HR of 0.72 (95% CrI 0.46-1.13). This result was similar in all subgroups, with mean HRs ranging from 0.58 to 0.77. Evaluation of the AUC of the posterior distribution showed that the probability of improved survival in this population ranged from 83% to 94%. Greater variability was observed with the pessimistic prior; however, the probability of improved survival in the posterior distribution remained elevated at 83% in all participants and 89% in those with DAH (Figure 1).
Informative priors were also used to evaluate the effect of reduced-dose GCs. Using a pessimistic prior, the benefit of reduced-dose GCs was still observed in the combined group, with a mean HR of 0.73 (95% CrI 0.44-1.22), corresponding to an 89% probability of reduced mortality (Table 2, Figure 2). However, the results were heterogeneous: in participants without DAH, the mean HR was 0.53 (95% CrI 0.27-1.06), whereas those with DAH had a mean HR of 1.40 (95% CrI 0.63-3.12), largely driven by those with severe DAH, who had a mean HR of 2.03 (95% CrI 0.73-5.66). These correspond to posterior probabilities of reduced mortality of 96% in participants without DAH, 20% in those with DAH, and 9% in those with severe DAH. The effect of reduced-dose GCs was enhanced with the null and optimistic priors, as shown in Table 2. However, even when assuming an optimistic prior, participants with severe DAH had a mean HR of 1.51 (95% CrI 0.54-4.21), corresponding to a 22% probability of improved survival.
In addition to evaluating the probability of an HR < 1, the probability of attaining HR thresholds of < 0.9, 0.8, and 0.67 was also evaluated. Using a noninformative prior, the probability of an HR < 0.67 with PLEX was 71% for participants with DAH and 24% in those without DAH (Supplementary Table S3, available with the online version of this article). With reduced-dose GCs under a noninformative prior, the probability of HR < 0.67 was 84% in participants without DAH but only 3% in those with severe DAH (Supplementary Table S4).
The odds of severe infection were increased in participants receiving PLEX, with a mean OR of 1.25 (95% CrI 0.92-1.7), corresponding to a 92% probability of increased infection (Supplementary Figure S3, available with the online version of this article). The probability of an OR > 1 for infection was 65% in participants with DAH and 90% in those without DAH. Reduced-dose GCs decreased the risk of infection with a mean OR of 0.85 (95% CrI 0.63-1.15), corresponding to an 85% probability of an OR < 1 (Supplementary Figure S4).
The odds of death within 3 months of intervention were similar to those observed at 1 year. The mean OR of death at 3 months after PLEX in the overall sample was 0.85 (95% CrI 0.44-1.64), corresponding to a corresponding to 69% odds of improved survival (Supplementary Figure S5, available with the online version of this article). The mean OR of death at 3 months with reduced-dose GCs was 0.68 (95% CrI 0.35-1.32), corresponding to an 87% probability of improved survival at 3 months (Supplementary Figure S6).
DISCUSSION
In a Bayesian reanalysis of the PEXIVAS trial, the largest RCT of individuals with AAV, the findings vary from those of the original frequentist analysis. Results were more favorable toward PLEX, suggesting that individuals with AAV receiving PLEX had an 86% probability of reduced mortality compared with those not receiving PLEX, which ranged from 82% to 92% when combined with prior observational data, depending on the strength of the priors used. Although reduced-dose GCs seemed favorable in individuals with kidney injury, those with severe DAH had an 88% probability of increased mortality when receiving reduced-dose GCs. Uncertainty remains regarding whether individuals with nonsevere DAH would benefit from reduced-dose GCs. These findings suggest the need to review these treatment modalities and highlight the importance of individualizing GC treatments for patients with AAV.
The pathophysiology of AAV is not completely understood but relies on activated neutrophils (eg, by tumor necrosis factor or interleukin-1β). These neutrophils express cytoplasmic antigens, myeloperoxidase, and proteinase 3 on their surface, to which ANCA can bind. Simultaneously, the Fc region of ANCA can bind to the Fcγ receptor on neutrophils, promoting excessive neutrophil activation and the release of reactive oxygen species, lytic enzymes, and neutrophil extracellular trap formation. These mechanisms ultimately cause endothelial damage and end-organ dysfunction.25 ANCA levels generally increase with disease activity and decrease when remission is induced, supporting their role in disease pathogenesis.26,27 PLEX has been used in AAV for decades, as evidence suggests it reduces ANCA levels and other humoral mediators.25
Before the PEXIVAS trial, the MEPEX trial evaluated the role of PLEX compared with intravenous methylprednisolone in 137 patients with kidney disease due to AAV and a creatinine level > 500 μmol/L.28 The original MEPEX trial found that PLEX increased rates of kidney recovery; however, subsequent reanalysis of the data showed that this benefit was not sustained long term and raised concern for a higher rate of infection in patients receiving PLEX.29 Unlike the MEPEX trial, the PEXIVAS trial included broader eligibility criteria, including patients with DAH without kidney involvement or those with milder kidney disease. The broader patient population in PEXIVAS may explain differences in results across subgroups. Notably, both MEPEX and PEXIVAS demonstrated a trend toward increased risk of infections requiring hospitalization or intravenous antibiotic therapy with PLEX.1,29
Placing the results of this study in the context of these findings, it may be the case that PLEX is beneficial in individuals with DAH, especially those with severe DAH or severe kidney injury (as in the MEPEX trial), but it may not be necessary in individuals with milder disease. There is also biological plausibility for this observation: cyclophosphamide or rituximab with GCs may be sufficient to reduce ANCA levels and induce remission in less severe disease but may be insufficient in more severe disease. Although this reasoning has not been empirically demonstrated, it may account for the disparate findings in these RCTs and supports a personalized approach to treatment in AAV. Similar arguments for targeted treatment in patients living with AAV have been proposed previously.30
Although the reduced-dose GC regimen was noninferior to a standard GC regimen, there were clear differences across patients with DAH and without DAH. A recent observational study of individuals with AAV compared those receiving either the PEXIVAS reduced-dose GC taper or a standard-dose GC regimen.24 The investigators found that those treated with the reduced-dose GC regimen had a higher rate of a composite of death, ESKD, AAV progression, or relapse within 1 year compared with the standard-dose GC regimen. In a subgroup analysis of those receiving rituximab induction, individuals with creatinine > 300 μmol/L or receiving low-dose GCs had a higher rate of death or ESKD. In this reanalysis, individuals without DAH benefited from the reduced-dose GC regimen. However, there was an increased probability of mortality when the reduced-dose GC regimen was used in individuals with severe DAH. As with PLEX, this suggests the need for a personalized approach to treatment depending on disease presentation, and perhaps a more personalized immunosuppressive regimen.
Although these results provide additional context for potential uses of PLEX and a standard-dose GC regimen in appropriate patient populations, these interventions are associated with higher odds of infections requiring intravenous antimicrobials or hospitalization. Infections are not commonly associated with PLEX,31,32 but were previously identified in the MEPEX trial.29 These potential risks should be considered when making individual treatment decisions.
Bayesian analyses can provide additional context to interpret frequentist results.9 Whereas Bayesian analyses are gaining traction in areas outside of medical research, including economics, political science, and ecology,33 they remain a small proportion of clinical research. Such analyses are particularly useful in settings in which studies are difficult to conduct or may be underpowered, providing a complementary approach for interpreting the literature.11,19 Another benefit of a Bayesian approach is that future studies can use data from the present study as priors for analyses involving new data, thereby updating the understanding of AAV. In this study, a framework is provided that could be used by other investigators applying Bayesian methods in the analysis of their results, along with scripts to reproduce the results.
This study had several limitations. First, only 191 individuals with DAH were included in the cohort, and only 61 with severe DAH, which may limit the generalizability of the results. Although Bayesian analysis may better accommodate small sample sizes than frequentist analysis,5 it does not altogether alleviate these limitations. However, as the PEXIVAS trial is a large, international, well-conducted RCT, the present study likely represents the best available evidence of AAV treatment to date.
There were several baseline differences between those with DAH and without DAH, including differences in induction treatments and creatinine levels, which may account for some baseline differences, although these factors were adjusted for in the proportional hazards model. Selection of informative priors was limited to heterogeneous observational studies with varying follow-up periods, and these priors may therefore be imperfectly calibrated for analysis of outcomes in 3 and 12 months. To mitigate this, the best available empirical evidence was used and priors were downweighted. Interpretation of results across different priors is encouraged, with particular emphasis on the noninformative prior, which is not subject to this limitation.
Results varied according to the strength of the priors used, which may be difficult to interpret. This likely reflects the limited sample size9 and highlights the importance of interpreting results from different prior assumptions. Although individual participant–level data were not available for this analysis, given the goals of reanalyzing the study, raw data would not be expected to significantly affect the planned analyses. The analytic technique used, Bayesian normal-normal updating using a conjugate prior, is mathematically identical to an individual participant–level analysis with the same likelihood and prior distributions.
The strengths of this study include the use of a well-conducted trial with rigorously collected data to address an important clinical question, as the PEXIVAS trial is the largest study of its kind in individuals with AAV. A wide range of priors was used, and Bayesian analysis reporting guidelines were followed.20 Finally, reproducible analysis scripts were provided to enable further reanalyses and repurposing by other investigators.
In a Bayesian reanalysis of individuals with AAV, this study found that PLEX was associated with a high probability of improved 1-year survival, which was greater in those with DAH. Reduced-dose GCs were beneficial in individuals with kidney disease but may be harmful in those with severe DAH. These findings suggest that patients may benefit from a targeted treatment approach for AAV.
Footnotes
A.S. Gershon and K. Liu contributed equally to this work as co-senior authors.
CONTRIBUTIONS
OAA: conceptualization, formal analysis, methodology, visualization, writing – original draft; LF, SRJ: writing – review and editing; ASG: supervision; writing – review and editing; KL: conceptualization, methodology, supervision, writing – review and editing.
FUNDING
OAA is supported by the University of British Columbia Clinician-Investigator Program and the Canadian Institute of Health Research.
COMPETING INTERESTS
The authors declare no conflicts of interest relevant to this article.
ETHICS AND PATIENT CONSENT
Ethics approval and patient consent were not required as this work used previously published data.
- Accepted for publication April 2, 2026.
- Copyright © 2026 by the Journal of Rheumatology
This is an Open Access article, which permits use, distribution, and reproduction, without modification, provided the original article is correctly cited and is not used for commercial purposes.
REFERENCES
SUPPLEMENTARY DATA
Supplementary material accompanies the online version of this article.
