Portal de Periódicos da CAPES

Venovenous Bypass Associated With Acute Kidney Injury Prevention in Liver Transplantation: An Ode to the Retrospective Data Researcher

2017; Lippincott Williams & Wilkins; Volume: 125; Issue: 5 Linguagem: Inglês

10.1213/ane.0000000000002449

1526-7598

Teus H. Kappen, Stuart A. McCluskey,

Renal and Vascular Pathologies

Predictable and avoidable acute kidney injury (AKI) should never occur. That was one of the principal recommendations of the “Adding Insult to Injury” report by the National Confidential Enquiry into Patient Outcome and Death.1 Within the perioperative period, predictability may be one thing, but identifying appropriate preventative strategies for AKI has been proven difficult.2,3 Ideally, one would like to test each candidate intervention in a separate randomized clinical trial. Unfortunately, that would be too costly and ethically difficult to do because of the large number of candidate interventions and what is considered standard clinical practice. In this edition, Xia et al4 took a common approach to this problem by performing a retrospective study to determine if there was an association between venovenous bypass and AKI. Their results confirm that a randomized trial may be warranted to study the effects of venovenous bypass on AKI in liver transplant patients, differentiating those effects between patients with normal and compromised pretransplant renal function, a result that is clearly demonstrated by the analysis of their data, but also a result that is difficult to interpret with our current understanding of the etiology of perioperative kidney injury. As reviewers of this work, we think it is informative if we describe: (1) what the authors had to endure from us; (2) how the authors persisted; and (3) how their persistence paid off and improved the interpretation of the results. It may sound familiar to those who regularly submit manuscripts to peer-reviewed journals and get to read the reviewers’ comments: have the reviewers truly been scrutinizing my manuscript or are they just nit-picking? There are many different strategies for retrospective analyses. When authors use strategies that a reviewer is more familiar with, the reviewer is likely to look more favorably on the analysis. So, what exactly was our issue with the manuscript? The authors reported a beneficial effect of venovenous bypass on AKI in patients with compromised pretransplant renal function, even though they did not observe any effect of venovenous bypass on the overall incidence of AKI in the entire population. The authors did not explain why and how they performed this post hoc subgroup analysis. Moreover, a proper exploration of the balance in patient characteristics within the subgroups was missing, and the authors had not presented what the effects of venovenous bypass were within the complementary subgroup of patients with normal pretransplant renal function. That led us to the belief that the analysis was very much data driven and made us very skeptical of the results—to the point of being tempted to completely reject the manuscript. Even though data-driven analyses should typically be avoided, a data-driven analysis does not per definition yield an incorrect result. If the results were true, a complete rejection would have been a waste of a lot of good data and an interesting hypothesis on an important subject. Expecting that additional analyses would render the initial subgroup results invalid, we asked for a major overhaul of the analysis. The first revision confirmed our suspicion: the cutoff value to define compromised pretransplant renal function was based on an optimal effect size. We thus insisted on an analysis where pretransplant creatinine was used as a continuous variable in the AKI regression model—including an interaction term with venovenous bypass. The authors—seemingly without any grudge—abided by our request, resulting in the very insightful Figure 2 of their final manuscript. In this figure, one can clearly observe: (1) a negative relation between pretransplant creatinine and the incidence of AKI; and (2) a suggested protective effect from venovenous bypass on AKI with higher pretransplant creatinine values. This analysis confirmed their initial result. Even the new analysis was still no proof of an unbiased result and a causal therapeutic effect of venovenous bypass. The results of a retrospective study can be hypothesis-generating at most. More importantly, the complex relation among pretransplant creatinine, venovenous bypass, and AKI seemed counterintuitive, and we asked the authors to provide us with a plausible explanation—still aware of the hypothesis-generating nature of retrospective studies. Again the authors complied. They cited an article by Filomia et al5 in which baseline creatinine was negatively associated with the incidence of contrast-induced AKI, concluding that it is not the first time such an effect was observed. The authors stipulated that the AKI criteria may not be equally sensitive in detecting AKI in patients with different baseline renal function. It might be easier to have a “short, quick hit” on renal function during the immediate postinsult period when your renal function is normal rather than compromised, especially since the newer criteria for AKI include relative changes in serum creatinine as compared to the use of only absolute changes in the older definitions for renal injury. Further analysis of the relation among baseline creatinine, postinsult time in days, and postinsult serum creatinine may be warranted within different patient populations, but this is beyond the scope of the authors’ manuscript. The perseverance of the authors paid off. Not only did they get their article published, but it has led to a much better methodological argumentation and clinical insight into their results and conclusions. At least that is our opinion as their reviewers; hopefully the authors agree with us. When we first read their manuscript, we thought there were only 2 explanations for their result: either it is true or it is residual bias due to the retrospective nature of the data and the data analysis. We would have never thought that this paper might give us some insight into the complexities of defining AKI and how that may influence estimated treatment effects.6 As reviewers, we have been reminded of a fundamental lesson in clinical research—results should be irrefutable. The authors’ well-placed faith in their data was contrasted by our skepticism in their interpretation and conclusions, resulting in a balanced—and hopefully unbiased—interpretation and discussion of the results. Rather than a peer-reviewed process, it was a peer-monitored process, not unlike the preprint publishing on arXiv.org, where peer-scientists comment on manuscripts before they are submitted for publication. This peer-monitored process was only made possible because of an open-minded editorial process allowing unexpected results to be duly considered. Looking back at the process, we realized why we often feel to have a love–hate relationship with retrospective studies: not because it is low-quality science, but because it is truly difficult to do well and stay unbiased toward the interpretation of the results. However, where 1 person cannot avoid having a colored view on things, multiple colored views together may shine a neutral light on the situation. DISCLOSURES Name: Teus H. Kappen, MD, PhD. Contribution: This author helped write the manuscript. Name: Stuart A. McCluskey, MD, PhD, FRCPC. Contribution: This author helped write the manuscript. This manuscript was handled by: W. Scott Beattie, PhD, MD, FRCPC.