Intraoperative phlebotomies and bleeding in liver transplantation: a historical cohort study and causal analysis

Abstract

Background Liver transplantation is associated with major bleeding and red blood cell (RBC) transfusions. No well-designed causal analysis on interventions used to reduce transfusions, such as an intraoperative phlebotomy, has been conducted in this population. MethodsWe conducted a historical cohort study among liver transplantations performed from July 2008 to January 2021 in a Canadian centre. The exposure was intraoperative phlebotomy. The outcomes were blood loss, perioperative RBC transfusions (intraoperative and up to 48 hr after surgery), intraoperative RBC transfusions, and one-year survival. We estimated marginal multiplicative factors (MFs), risk differences (RDs), and hazard ratios by inverse probability of treatment weighting both among treated patients and the whole population. Estimates are reported with 95% confidence intervals (CIs).

Observational studies: a perpetual quest for a signal among the noise

“…Dans ce nume ´ro du Journal, Carrier et coll. 3 rapportent les re ´sultats d'une e ´tude observationnelle concluant a `un effet be ´ne ´fique potentiel de la phle ´botomie parmi la population entie`re de receveurs de greffes de foie en termes de re ´duction des saignements et des transfusions (facteur multiplicatif marginal de saignement, 0,85; intervalle de confiance [IC] a `95 %, 0,72 a `0,99; diffe ´rence de risque pe ´riope ´ratoire [DR] pour la transfusion, -15,2 % [IC 95 %, -26,1 a `-0,8]; et RD perope ´ratoire, -14,7 %; IC 95 %, -23,3 a `-2,8), tout en ne montrant aucun effet sur la population traite ´e (nous de ´crirons la distinction entre population entie `re et traite ´e plus loin). Le principal mode d'analyse a consiste ´en l'analyse de la probabilite ´inverse de ponde ´ration du traitement (ou IPTWinverse probability of treatment weight), une forme de me ´thode du score de propension (SP).…”

“…In this issue of the Journal, Carrier et al 3 report findings from an observational study on a possible beneficial effect of phlebotomy among the whole population of liver transplant recipients in terms of reduced bleeding and transfusion (bleeding marginal multiplicative factor, 0.85; 95% confidence interval [CI], 0.72 to 0.99; perioperative risk difference [RD] for transfusion, -15.2% [95% CI, -26.1 to -0.8]; and intraoperative RD, -14.7%; 95% CI, -23.3 to -2.8) while showing no effect on the treated population (more on the distinction between whole and treated later). The primary mode of analysis was inverse probability of treatment weight (IPTW) analysis-a form of propensity score (PS) method.…”

“…In the case of the study by Carrier et al, 3 the effect estimates are measures of association between the hypothetical use of phlebotomy and blood loss with respect to two target populations of patients (i.e., just the treated population ''ATT'' or the whole population ''ATE'' created for the purpose of the analysis) using IPTW. This method uses weighting to create a pseudosample in which the distribution of potential confounding factors is conditionally independent of exposure, allowing a more accurate, less biased estimation of the relationship between exposure to phlebotomy and outcome.…”

“…Naturally, the PS model cannot consider factors that are unknown or were not measured, which potentially could influence the treatment selection. Therefore, in the Carrier et al 3 study, bias due to unmeasured or unknown differences between patients who received phlebotomy treatment vs those who did not cannot be ruled out because, despite its strengths, the PS method is clearly not a substitute for randomization.…”

“…In their current analysis, the researchers not only have interrogated their local database for a fresh clinical question regarding the role of intraoperative phlebotomy, integrating a causal framework, but also have extended their period of observations to January 2021, consequently including 30% more patients compared with their last trackable report. 13 Accordingly, the clinical relevance of Carrier et al's study 3 depends not only on the statistical issues discussed, but also on whether the study sample is representative of patients to whom the results could be generalized. This is particularly important given it was a single centre study.…”