To Pulse or Not to Pulse in ANCA Associated Vasculitis? Save

Randomized trials typically get top billing at ACR Convergence, but the past three years have seen “target trial emulations” land coveted plenary sessions and oral abstracts. These are not actually trials; instead, they are a new-ish approach to generating causal inferences from observational datasets. One particularly interesting oral presentation from the meeting used this idea to investigate a critical question in ANCA associated vasculitis: should we be giving pulse dose steroids? 

What the Heck is a Target Trial Emulation? 

The target trial is a framework for how to approach answering causal questions with observational datasets. A few years back the word “causal” applied to any observational dataset would have been triggering to me, but after reading this essay and noodling on the results of the RCT DUPLICATE initiative, I changed my mind. The main reason is that I agree with the essay – using scientific euphemisms like “association” and “link” allow people to play fast and loose with their methods while still implying causality. Instead of shying away from causality, observational research that suggests a cause should embrace it. 

The target trial emulation is one approach to doing this. You start by designing a theoretical trial to answer your clinical question. Who would you enroll and who would you exclude? What is the treatment/comparator and when does follow up begin? When would you start counting outcomes and what would they be? From there, you bring the latest and greatest of epidemiology to bear, trying to solve the problems that observational data present. 

Target Trialing Pulsing People 

Instead of just comparing people who got pulses to those who did not, Abstract 0929 “emulated” a trial of pulse vs no-pulse. First, they created a lead in period from when induction began to 14 days later, establishing the “time zero” that randomization gives you. They next used some fancy stats (stabilized inverse probability of treatment weights) to adjust for baseline characteristics. As in a trial, they assessed our standard outcome measure (BVAS) and investigated safety. 

What did they find? The adjusted odds of remission were no different for people who received a pulse as compared to those who did not (aHR 0.96, 95% CI 0.44–2.08). Those are pretty wide confidence intervals, but it landed near “not helpful at all” and a large benefit seems unlikely. Notably, pulses did seem to cause an increase in serious infection, which affected 31% of those who received a pulse and 14% of those who did not (aHR 1.88, 95% CI 1.04–3.40). I have long wondered if pulse steroids help; this study suggests they do not and come with a substantial cost.  

Missing the Mark? 

Had this been a trial, it would have been immediately practice changing. Instead, this will temper my enthusiasm for pulse dose steroids, but I will still be using them in severe cases. Patients who received pulse steroids in this study were much sicker, with higher rates of diffuse alveolar hemorrhage or rapidly progressive glomerulonephritis (78% vs 28% for non-pulse patients). It seems obvious that some confounding by indication is happening here, which would erroneously deflated the perceived benefit of pulse steroids (if a sicker population does “just as well” as a healthier population… isn’t that a win for the pulse?) and harm the perceived safety of steroids (of course sicker patients had more severe infections!). 

Theoretically, fancy stats and thoughtful design mitigate these issues. The part of me that has always been skeptical of observational data finds it hard to believe they are sufficient. “Residual” confounding seems plausible (if not likely) and the authors have not presented evidence against it. Positive and negative controls as well as some sensitivity analyses to test the robustness of these findings would be a nice addition. For now I will still be pulsing my patients with severe findings, but I may dial it back a bit for healthier folks.