Is ANCA–Associated Vasculitis Ready for Personalized Medicine?

See Clinical Research articles on page 4199 and on Article 103776

Kidney involvement in antineutrophil cytoplasmic autoantibody–associated vasculitis (AAV) is common and increases the risk of kidney failure, death, infections, and other treatment related side effects.¹ More effective and safer treatments remain a priority. Two papers in KI Reports highlight the importance of renal involvement and its potential role in defining the best use of available treatments.^2,3 In essence, these data suggest that some treatments may be more or less efficacious depending on the severity or type of kidney involvement in patients with AAV—a form of personalized medicine—that could allow tailoring therapy to reduce kidney failure and avoid therapies (and their side effects) when they are unlikely to be efficacious.

Vignac et al.² studied 394 patients with AAV, of whom 105 were receiving dialysis, noting an approximately 10-fold increased risk of the composite outcome of death from any cause, or renal replacement therapy at 1 year in patients who required dialysis initially, after adjusting for several other characteristics including renal histologic lesions. About half the patients initially receiving dialysis recovered, and plasma exchange (PLEX) was attributed with a large reduction in the odds of the composite (odds ratio: 0.23, 95% confidence interval: 0.05–0.80), whereas there was no obvious effect in the overall cohort (odds ratio: 0.86, 95% confidence interval: 0.43–1.67).

Uzzo et al.³ studied 304 patients with AAV and examined outcomes by subgroups defined by renal histologic class. They noted that simple classification of the renal biopsy as focal, crescentic, sclerotic, or mixed was associated with kidney failure as noted in many previous studies. In addition, they noted that patients treated with rituximab monotherapy were less likely to recover than patients treated with cyclophosphamide (with or without additional rituximab; odds ratio: 0.235 compared with cyclophosphamide monotherapy, and odds ratio: 0.204 compared with the combination of rituximab and cyclophosphamide) only in those with crescentic class disease with similar results for the outcome of kidney failure (a lower power test).

Do these studies get us closer to individualizing therapy for patients with AAV and renal involvement? Should we use PLEX only in those who require dialysis initially? Should we use cyclophosphamide preferentially in patients with crescentic class biopsies?

Both papers suggest that the effects of a treatment differ based on patient characteristics; PLEX only works in patients who require dialysis, and rituximab monotherapy is less effective only in patients with crescentic class disease. In epidemiology and statistics, this is known as effect modification, a statistical interaction, or subgroup effects. Although some unimportant difference in treatment effects may realistically exist between different subgroups, qualitative interactions, where a treatment works in patients with a characteristic but not in patients without that characteristic, are uncommon when adequately tested in most areas of medicine.⁴ This does not prevent trialists and epidemiologists from finding many possible (spurious) qualitative interactions through chance or bias. Meta-epidemiologic analyses show that reliable and meaningful (qualitative) treatment-subgroup interactions are rare, with most reflecting only small quantitative interactions when studied in large meta-analyses. In AAV, long term studies identify strong prognostic factors; however, these do not predict differences in the effect of a treatment, underscoring that prognostic markers are far more common than effect modifiers. An initiative called ICEMAN (Instrument for assessing the Credibility of Effect Modification Analyses), developed a list of questions to help determine if the interactions observed in randomized trials and systematic reviews were credible.⁵

If we ask ICEMAN’s core questions of Vignac et al.’s² and Uzzo et al.’s³ studies (Table 1), we see that the subgroup effects they describe have low credibility. Importantly, neither uses the appropriate methodology to test for a subgroup effect. Tests of statistical interaction look for whether a treatment effect in one group of patients differs from the treatment effect in the other group of patients beyond the play of chance. Instead, both papers test whether there is a treatment effect in the subgroup of interest, which involves multiple testing in small groups of patients. By their very nature, small subgroups must exhibit (unrealistically) large effects to appear statistically significant, which further erodes their credibility (i.e., is it realistic that cyclophosphamide is 5 times more effective than rituximab in patients with crescentic class but similar in patients with all other classes, or that PLEX improved survival by 10 times in patients who required dialysis but had no effect in all other patients?)

Table 1.

Core questions from the ICEMAN applied to studies suggesting subgroup effects in ANCA–associated vasculitis

ICEMAN Core Question	Vignac et al.	Uzzo et al.
Was the direction of effect modification specified a priori?	No – but consistent with previous suggestion PLEX most effective in patients receiving dialysis	No – but consistent with previous hypotheses that rituximab has a slower onset of action
Was the effect supported by prior evidence?	No – meta-analyses show similar relative risk reduction across kidney function	No – no direct evidence for or against
Does a test for interaction suggest that chance is unlikely?	No – subgroups tested separately without test of interaction	No – subgroups tested separately without test of interaction
Did the authors test only a small number of effect modifiers?	Unclear – dialysis emphasized but multiple baseline variables in model	No – multiple subgroups with multivariable models created
If the effect modifier was continuous, were arbitrary thresholds avoided?	Yes – dialysis is a natural categorization of kidney function	N/A – histologic class is naturally categorical

ICEMAN, Instrument for assessing the Credibility of Effect Modification Analyses; N/A, not applicable; PLEX, plasma exchange.

In addition, there is little evidence external to the present studies to support either subgroup effect. PLEX has a wealth of randomized trial data at lower risk of bias and with larger sample sizes that found no evidence of a subgroup effect across a broad range of kidney function (i.e., the relative effects of PLEX are similar, whether someone is on dialysis or not at diagnosis).⁶ Vignac et al.² also found an effect on a composite of death and kidney failure that was not observed in the randomized trials, where the risk of kidney failure is reduced but there is no important effect noted on mortality.⁶ Vignac et al.’s² study does however highlight that though patients who initially require dialysis are at high risk of death and permanent dialysis, half recover; and safe, effective treatment is still a priority for them.

Similarly, Uzzo et al.³ suggest that rituximab alone may be less effective in patients with crescentic disease. There is only 1 randomized trial comparing rituximab alone to cyclophosphamide, and this was done in a patient population with modest kidney involvement.⁷ Therefore, there is no direct evidence supporting or refuting this finding. Although some experts suggest that cyclophosphamide exerts effects more quickly than rituximab and may therefore be superior in rapid, fulminant AAV, this is largely theory without empiric evidence to support or refute it and at odds with data that show much more rapid and complete peripheral B-cell completion with rituximab than with cyclophosphamide.⁸ More important are the potential biases that lead to some patients being treated with rituximab alone compared with cyclophosphamide based or combination regimens, which cannot be easily measured through patient characteristics. Although the findings are intriguing, they should not drive decision-making without further high-quality evidence.

While neither of these observational studies are sufficient to drive clinical practice guidelines for the treatment of AAV, they provide important steps forward. Considering the likelihood that AAV is not a single disease, we can almost certainly improve treatment efficacy and reduce harm if we disentangle differences in the mechanisms of diseases among individuals. Both studies emphasize this goal and the extensive patient data available to achieve it. We can now make more informed decisions about who is most likely to benefit from PLEX (i.e., those at significant risk of kidney failure), as highlighted in current guidelines. In addition, future trials may help us better understand which immunosuppression regimens may be more or less effective for certain individuals.^9,10 As the number of potential therapies to treat AAV increases, their mechanisms of actions diversify, and our knowledge of disease processes and evaluation improves, studies like these will become increasingly important to advance both clinical trials and care.

Disclosure

MW is supported by a career award from the Population Health Research Institute; and has received consults for, or serves on steering committees, safety monitoring boards, or adjudication committees for trials by Otsuka, Glaxo-Smith Kline, Bayer, AstraZeneca/ Alexion, Novartis, GlomCon Foundation, Hansa Pharmaceuticals, Roche, and Novo Nordisk. All monies are paid to a third party nonprofit foundation. CAP declared no competing interests.