We thank Casal Moura et al.1 for their comment on our study.2 We found that plasma exchange (PLEX) was not associated with a better primary outcome in the whole study population, but we identified a subset of patients who could benefit from PLEX. A prediction model on the basis of the average treatment effect of PLEX was developed to estimate in which patients the use of PLEX could lower the probability of death or kidney replacement (RRT) by 12 months. However, these findings must be validated before being utilized in clinical decision making. Very recently, a systematic review and meta-analysis of randomized, controlled trials did not show any effect of PLEX on all-cause mortality (relative risk, 0.90; 95% confidence interval, 0.64 to 1.27), but PLEX reduced the risk of ESKD at 12 months (relative risk, 0.62; 95% confidence interval, 0.39 to 0.98). On the basis of these data, clinical practice guidelines were proposed, making a weak recommendation against PLEX in patients with low or low-moderate risk of developing ESKD and a weak recommendation in favor of PLEX in patients with moderate-high or high risk of developing ESKD.3,4 Overall, this guideline supports the need to identify subsets of patients who could benefit from PLEX. In their comment, Casal Moura et al.1 discussed the interpretation of the results from our study.
First, they stressed the absolute risk difference for death or RRT at 12 months that was higher in the PLEX-recommended group, but this corresponds to the same relative risk reduction between groups. Their point that the relative risk is similar for PLEX-recommended and PLEX-nonrecommended groups is well taken. Nevertheless, this does not mean that the effect of PLEX was similar in both groups. In situations where prognostic factors exist, it is not possible to have a constant non-null treatment effect both on the absolute risk difference and on the relative risk difference scales. There will always be heterogeneity of treatment effect on at least one of those scales. This forms the basis of the risk modeling approach of the Predictive Approaches to Treatment Effect Heterogeneity (PATH) statement.5 In particular, authors5 of the PATH statement noted that “risk models can be useful in identifying ‘clinically important heterogeneity in treatment effects,’ which is evaluated on the absolute risk difference scale.”
Furthermore, it is the absolute risk reduction in which the patient is ultimately interested.6 Being asymmetric, the relative risk scale is also misleading; although the relative risks for death or RRT at 12 months are estimated at 0.65 and 0.69 in the PLEX-nonrecommended and PLEX-recommended groups, respectively (very similar to what the meta-analysis showed), the relative risks of being alive and free of RRT at 12 months (which are simply the opposite view of the exact same outcomes) are 1.11 and 1.33, respectively.
Another important issue is that the average treatment difference in both PLEX-recommended and PLEX-nonrecommended groups does not show the heterogeneity of treatment effect within these groups. We had arbitrarily but prior to analysis defined the PLEX-nonrecommended group as individuals with a predicted benefit on the absolute risk reduction scale <5% to require at least a certain amount of benefit to recommend PLEX. The average observed benefit presented in supplemental table 4 in our study2 is close to 5% (4.8%), but we would like to underline that this was obtained after bootstrap bias correction. Even so, there are individuals with worse predicted outcomes with PLEX than without PLEX in this group, and the individualized relative risk for them is not about 0.65 because it is >1. Again, all group-level results understate between-patients variability, which can be taken into account for individualized decisions.
The other point of Casal Moura et al.1 relates to the prediction of PLEX benefit. To estimate PLEX benefit conditional on patients’ characteristics in an observational study, we used an augmented weighted approach, which combines a propensity score and a prognostic model and focuses on estimating the individualized treatment effect, treating the purely prognostic covariates as a nuisance.7 We are aware that such a treatment effect modeling approach may overestimate heterogeneity in treatment effects, and we, therefore, used regularization by the lasso, as recommended.8 We agree with the authors that variables were both prognostic and associated with the effect of PLEX. This is not surprising because we used the same variables as potential confounders, potential prognostic factors, and potential treatment effect moderators. Effects could have been different, but it turned out that factors associated with poor prognosis were also associated with the highest benefit. Because we used an absolute risk reduction scale to define treatment benefit, this was amplified. Again, the absolute risk reduction scale was chosen because we do think it is more clinically relevant.
Disclosures
All authors have nothing to disclose.
Funding
None.
Footnotes
See related letter to the editor, “Predicting Kidney Response to Plasma Exchange in ANCA-Associated Vasculitis: Need for Plausible Models” on pages 1223–1224, and original article “Kidney Histopathology Can Predict Kidney Function in ANCA-Associated Vasculitides with Acute Kidney Injury Treated with Plasma Exchanges,” in Vol. 33, Iss. 3, 628–637.
Published online ahead of print. Publication date available at www.jasn.org.
Contributor Information
Collaborators: The French Vasculitis Study Group, Vincent Cottin, Stanislas Faguer, Loïc Guillevin, Noémie Jourde-Chiche, Alexandre Karras, Luc Mouthon, Antoine Néel, Xavier Puéchal, Grégory Pugnet, Maxime Samson, Camille Taillé, and Benjamin Terrier
Author Contributions
F. Grolleau, D. Nezam, R. Porcher, and B. Terrier conceptualized the study; F. Grolleau, D. Nezam, R. Porcher, and B. Terrier were responsible for data curation; F. Grolleau, D. Nezam, R. Porcher, and B. Terrier were responsible for investigation; F. Grolleau, D. Nezam, R. Porcher, and B. Terrier were responsible for formal analysis; F. Grolleau, D. Nezam, R. Porcher, and B. Terrier were responsible for methodology; F. Grolleau, D. Nezam, R. Porcher, and B. Terrier were responsible for project administration; F. Grolleau, D. Nezam, R. Porcher, and B. Terrier were responsible for resources; F. Grolleau, D. Nezam, R. Porcher, and B. Terrier were responsible for software; F. Grolleau, D. Nezam, R. Porcher, and B. Terrier were responsible for validation; F. Grolleau, D. Nezam, R. Porcher, and B. Terrier were responsible for visualization; F. Grolleau, D. Nezam, R. Porcher, and B. Terrier were responsible for funding acquisition; F. Grolleau, D. Nezam, R. Porcher, and B. Terrier provided supervision; F. Grolleau, D. Nezam, R. Porcher, and B. Terrier wrote the original draft; and F. Grolleau, D. Nezam, R. Porcher, and B. Terrier reviewed and edited the manuscript.
References
- 1.Casal Moura M, Soler MJ, Sethi S, Fervenza FC, Specks U: Kidney histopathology in ANCA-associated vasculitides treated with plasma exchange. J Am Soc Nephrol 33: 1223–1224, 2022 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Nezam D, Porcher R, Grolleau F, Morel P, Titeca-Beauport D, Faguer S, et al. ; French Vasculitis Study Group : Kidney histopathology can predict kidney function in ANCA-associated vasculitides with acute kidney injury treated with plasma exchanges. J Am Soc Nephrol 33: 628–637, 2022 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Zeng L, Walsh M, Guyatt GH, Siemieniuk RAC, Collister D, Booth M, et al. : Plasma exchange and glucocorticoid dosing for patients with ANCA-associated vasculitis: A clinical practice guideline. BMJ 376: e064597, 2022 [DOI] [PubMed] [Google Scholar]
- 4.Walsh M, Collister D, Zeng L, Merkel PA, Pusey CD, Guyatt G, et al. ; Plasma exchange and glucocorticoid dosing for patients with ANCA-associated vasculitis BMJ Rapid Recommendations Group : The effects of plasma exchange in patients with ANCA-associated vasculitis: An updated systematic review and meta-analysis. BMJ 376: e064604, 2022 [DOI] [PubMed] [Google Scholar]
- 5.Kent DM, Paulus JK, van Klaveren D, D’Agostino R, Goodman S, Hayward R, et al. : The Predictive Approaches to Treatment Effect Heterogeneity (PATH) statement. Ann Intern Med 172: 35–45, 2020 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Alhazzani W, Walter SD, Jaeschke R, Cook DJ, Guyatt G: Does treatment lower risk? Understanding the results. In: Users’ Guides to the Medical Literature: A Manual for Evidence-Based Clinical Practice, 3rd Ed., edited by Guyatt G, Rennie D, Meade MO, Cook DJ, New York, McGraw-Hill Education, 2015 [Google Scholar]
- 7.Chen S, Tian L, Cai T, Yu M: A general statistical framework for subgroup identification and comparative treatment scoring. Biometrics 73: 1199–1209, 2017 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.van Klaveren D, Balan TA, Steyerberg EW, Kent DM: Models with interactions overestimated heterogeneity of treatment effects and were prone to treatment mistargeting. J Clin Epidemiol 114: 72–83, 2019 [DOI] [PMC free article] [PubMed] [Google Scholar]