Balancing statistical rigor and clinical applicability in transfusion research on transfusion-dependent thalassemia: a clarification

We appreciate the thoughtful comments from Naveed and Abid regarding our article “Determinants of intertransfusion interval in children with transfusion-dependent thalassemia: a retrospective single-center cohort study in China” (1). Their constructive feedback provides an excellent opportunity to clarify our methodological rationale and further discuss the clinical implications of our findings.

Regarding the dichotomization of intertransfusion intervals

The correspondents raised a valid statistical concern about converting continuous variables into binary outcomes. We acknowledge that dichotomization may reduce statistical power in certain contexts (2,3). However, our analytical approach was deliberately designed as a multi-layered strategy rather than relying solely on binary classification.

First, we employed correlation analysis treating intertransfusion intervals as a continuous variable (Tab. 3), which revealed significant correlations between intervals and age (r=−0.341), weight (r=−0.298), genotype (r=0.175), and red blood cells (RBC) products (r=−0.197). Second, we constructed a multivariable linear regression model (Tab. 5) with pretransfusion hemoglobin at subsequent transfusion as a continuous outcome, which preserved the full variability of the data. The binary logistic regression (Tab. 4) served as a complementary analysis to identify clinically meaningful determinants, as the median-based grouping (≤18 vs. >18 days) aligns with clinical practice where transfusion scheduling often follows discrete intervals rather than continuous spectrums. Furthermore, our dichotomization threshold (18 days) was not arbitrary but derived from the median of our cohort data, representing a clinically relevant distinction between shorter and longer transfusion intervals in pediatric transfusion-dependent thalassemia (TDT) management. This approach enhances the clinical applicability of our findings for practitioners who must make categorical decisions about transfusion timing.

Regarding RBC product selection and potential confounding

The correspondents astutely noted that washed RBCs (WRBC) are typically reserved for patients with specific immunological indications. We fully agree that this represents an important clinical consideration. In our manuscript, we explicitly stated that “leukoreduced red blood cells (LRBC) is recommended as first-line therapy due to its superior hemoglobin preservation, while WRBC is reserved for patients with specific immunological indications” (1).

In our cohort, the selection of WRBC was primarily based on documented history of allergic transfusion reactions, rather than disease severity. As detailed in Tab. 6, patients No. 4, 10, 11, and 18 had documented histories of allergic reactions, which guided the use of WRBC. Importantly, our binary logistic regression analysis (Tab. 4) adjusted for multiple confounders including age, weight, genotype, and transfusion adequacy. After this adjustment, RBC product type remained an independent determinant [odds ratio (OR) =0.378, 95% confidence interval (CI): 0.186–0.770, P=0.007], suggesting that the observed association is not solely attributable to indication bias.

We also provided detailed product characteristics in Tab. S1, demonstrating that WRBC units have significantly lower total hemoglobin content (20.76 vs. 24.56 g, P=0.02) and reduced volume (117.38 vs. 150 mL, P<0.001) compared to LRBC units. These inherent differences in product composition offer a plausible biological mechanism for the observed shorter intertransfusion intervals with WRBC, independent of patient selection factors.

Regarding the independence of observations

The correspondents raised an important point regarding the correlation structure of repeated measurements within patients. We appreciate this methodological perspective and would like to offer some clarification.

While mixed-effects models or generalized estimating equations (GEE) are indeed valuable approaches for handling clustered data, their application requires careful consideration of sample characteristics. With only 21 patients serving as clusters in our study, implementing hierarchical models poses practical challenges. Simulation studies have demonstrated that mixed-effects models may yield biased variance estimates when the number of clusters falls below 30, particularly for random effect parameters (underestimation of standard errors by approximately 15%) (4). Given these constraints, we opted for analytical approaches that would provide more stable and interpretable results for our exploratory investigation. It is worth noting that our regression models incorporated patient-level variables (genotype, age, weight) as covariates, which partially accounts for between-patient heterogeneity. Furthermore, the consistency of our findings with established clinical knowledge and the Thalassemia International Federation (TIF) guidelines, which similarly recognize genotype severity as an independent predictor of transfusion requirements (5), lends support to the validity of our observations.

We acknowledge that future studies with larger patient cohorts would benefit from employing mixed-effects models to more precisely estimate effect sizes while accounting for the nested data structure.

Conclusions

We sincerely thank Naveed and Abid for their valuable methodological insights. Their comments highlight important considerations for future research in this field. We concur that prospective, multi-center studies employing mixed-effects models and larger sample sizes would strengthen the evidence base for personalized transfusion protocols in pediatric TDT. Our study, while exploratory in nature, provides preliminary evidence that genotype severity warrant consideration in individualized transfusion management, findings that merit validation in more rigorous study designs.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.