Clinical transfusion-outcomes research: A practical guide

Clinical transfusion-outcomes research faces unique methodological challenges compared with other areas of clinical research. These challenges arise because patients frequently receive multiple transfusions, each unit originates from a different donor, and the probability of receiving specific blood product characteristics is influenced by external, often uncontrollable, factors. These complexities complicate causal inference in observational studies of transfusion effectiveness and safety. This guide addresses key challenges in observational transfusion research, with a focus on time-varying exposure, time-varying confounding, and treatment-confounder feedback. Using the example of donor sex and pregnancy history in relation to recipient mortality, we illustrate the strengths and limitations of commonly used analytical approaches. We compare restriction-based analyses, time-varying Cox regression, and inverse probability weighted marginal structural models using a large observational dataset of male transfusion recipients. In the applied example, restriction and conventional time-varying approaches suggested an increased mortality risk associated with transfusion of red blood cells from ever-pregnant female donors compared with male-only donors (hazard ratio [HR] 1.22; 95% CI 1.05-1.42 and HR 1.21; 95% CI 1.04-1.41, respectively). In contrast, inverse probability of treatment and censoring weighted analyses, which account for treatment-confounder feedback, showed no evidence of an association (HR 1.01; 95% CI 0.85-1.20). These findings demonstrate how conventional methods can yield biased estimates when complex longitudinal structures are not adequately handled. We provide practical guidance on study design, target trial emulation, and the use of g-methods, including a reproducible tutorial and example dataset, to support valid causal inference in clinical transfusion research.

💡 Research Summary

This paper provides a methodological roadmap for clinical transfusion‑outcomes research, a field characterized by repeated, time‑varying exposures where each blood unit originates from a different donor. The authors begin by outlining the causal inference challenges unique to transfusion studies: multiple transfusions per patient, donor‑specific characteristics (such as sex and pregnancy history), and external factors (calendar time, blood group, geography) that influence exposure probabilities. These complexities violate the assumptions underlying many standard observational designs, particularly the notion of a single, fixed exposure and baseline‑only confounding adjustment.

To address these issues, the authors introduce the target trial emulation framework. By explicitly defining eligibility criteria, treatment strategies, random assignment, start of follow‑up, outcome, and causal contrast, researchers can structure observational analyses to mimic a randomized controlled trial. The paper emphasizes the three core assumptions of the potential outcomes framework—exchangeability (no unmeasured confounding), positivity (non‑zero probability of each treatment level for all covariate patterns), and consistency (well‑defined interventions).

A central methodological obstacle highlighted is treatment‑confounder feedback. In the illustrative example, red blood cells from ever‑pregnant female donors deliver a smaller hemoglobin increment, prompting additional transfusions. The cumulative number of transfusions (a time‑varying confounder) is both affected by prior exposure and predictive of mortality, placing it on the causal pathway. Traditional adjustment methods (stratification, matching, standard Cox regression) would condition on this variable, creating collider bias and distorting the estimated effect of donor sex.

The authors advocate for g‑methods, specifically inverse‑probability‑of‑treatment weighting (IPTW) combined with inverse‑probability‑of‑censoring weighting (IPCW) to construct a marginal structural model (MSM). IPTW creates a pseudo‑population in which treatment assignment is independent of measured covariates at each time point, while IPCW corrects for informative censoring when patients deviate from their initial exposure category. By applying these weights, the MSM estimates the causal effect of receiving blood from ever‑pregnant female donors versus male‑only donors on all‑cause mortality.

In the empirical analysis of male transfusion recipients, conventional restriction analysis and a time‑varying Cox model both suggested an increased mortality risk (hazard ratios ≈ 1.21–1.22, 95 % CI ≈ 1.04–1.42). In contrast, the weighted MSM yielded a hazard ratio of 1.01 (95 % CI 0.85–1.20), indicating no meaningful association after properly accounting for treatment‑confounder feedback. This discrepancy illustrates how failure to handle complex longitudinal structures can produce biased estimates.

The paper also critiques restriction‑based approaches, which limit the cohort to patients who never change exposure status. While intuitively appealing, this strategy discards patients who cross over, leading to selection bias because those who remain on a single exposure are systematically different. Similarly, time‑varying Cox models that include the number of transfusions as a covariate without appropriate weighting cannot resolve feedback bias.

To facilitate adoption of advanced methods, the authors provide a reproducible tutorial, complete R/Stata code, and a synthetic dataset. The tutorial walks readers through data preparation, calculation of stabilized weights, assessment of positivity, model fitting, and diagnostic checks. By making these resources publicly available, the authors aim to empower transfusion researchers and, more broadly, investigators dealing with repeated, time‑varying exposures to conduct valid causal inference.

Overall, the manuscript serves as both a conceptual guide and a practical toolkit, demonstrating that marginal structural models and related g‑methods are essential for unbiased estimation of treatment effects in clinical transfusion research where treatment‑confounder feedback is likely.