Introduction

Allogeneic hematopoietic stem cell transplantation (allo-HSCT) offers a curative potential for many hematological malignancies, yet its success is frequently limited by complications such as graft-versus-host disease (GVHD) and disease relapse [1,2]. Improving long-term outcomes requires the identification of robust, pre-transplant prognostic factors to enhance patient selection and tailor therapeutic strategies [3]. While the clinical utility of such factors is clear, the landscape of risk prediction is continually evolving [4,5].

Established prognostic variables include recipient characteristics like age and sex [6,7], pre-transplant laboratory values [8], and derived inflammatory markers such as the neutrophil-to-lymphocyte ratio (NLR) [9]. While the individual effects of such factors are well-documented [10], a deeper understanding of their interplay remains a critical area of investigation. Specifically, severe pre-transplant recipient neutropenia reflects not merely a transient laboratory abnormality, but a heavily pre-treated, depleted bone marrow microenvironment and profound host immunosuppression (the “soil”) [10]. Conversely, while routine donor parameters like baseline hemoglobin levels are strictly checked for donor safety, they are rarely investigated as predictors of recipient outcomes [11]. Recent literature has increasingly emphasized that non-HLA donor characteristics, including metabolic state and specific immune subsets mobilized during apheresis, play a profound role in both engraftment kinetics and the modulation of post-transplant complications [12–14]. From a biological standpoint, higher donor hemoglobin within the physiological range may act as a surrogate marker for robust hematopoietic reserve and an inherently vigorous graft (the “seed”). Therefore, it is plausible that the prognostic impact of a recipient’s intrinsic vulnerability (e.g., a compromised immune state and a depleted marrow niche) could be modified by the physiological robustness of the donor graft beyond standard HLA matching. Single-center studies from unique real-world settings are valuable for generating new hypotheses in this area, particularly in under-investigated patient populations [15].

Therefore, this single-center, retrospective study was designed as a hypothesis-generating analysis to explore pre-transplant risk factors in our cohort. Our primary objective was to investigate the potential interaction between pre-transplant recipient neutropenia, a marker of host immunocompromise and a hostile microenvironment, and donor hemoglobin, representing graft vigour, levels on 5-year overall survival. Secondary objectives included identifying other independent predictors of outcome in our specific patient population.

Materials and methods

Results

Cohort derivation and baseline patient characteristics

From an initial 116 consecutive patients assessed for eligibility at our center between September 2015 and March 2024, a final cohort of 94 patients was included for analysis. Twenty-two patients were excluded due to incomplete records with over 30% missing data (n = 15) or unavailable survival information (n = 7). The complete patient selection process is detailed in the STROBE flow diagram (Fig 1).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. STROBE flow diagram of patient selection.

Flow diagram illustrating the process of patient inclusion and exclusion. The chart details the initial number of consecutive patients assessed for eligibility, the number of patients excluded, the specific reasons for exclusion with corresponding counts, and the final number of patients included in the definitive analysis.

https://doi.org/10.1371/journal.pone.0349615.g001

During a median follow-up of 33.6 months (IQR, 4.7–61.7), a total of 30 mortality events (31.9% of the cohort) were recorded. The median overall survival for the entire cohort was not reached. The estimated 1-year, 3-year, and 5-year overall survival rates were 71.3%, 67.3%, and 67.3%, respectively. The baseline demographic, clinical, transplant, and key laboratory characteristics of the study population are summarized in Table 1. The cohort had a median age of 43 years (IQR, 31–54), with a majority of male patients (58.5%). Acute Myeloid Leukemia (AML) was the most frequent primary diagnosis (61.7%). The transplant procedures were relatively homogeneous; most patients received a myeloablative conditioning (MAC) regimen (85.1%) and a graft from a fully matched donor (89.4%). A critical pre-transplant vulnerability, neutropenia, was present at the time of assessment in 41 patients (43.6%).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Baseline characteristics of patients undergoing allogeneic stem cell transplantation (N = 94).

https://doi.org/10.1371/journal.pone.0349615.t001

Initial univariate survival analyses and identification of prognostic signals

To identify potential prognostic factors, we first conducted univariate survival analyses using two complementary methods. The results from these initial screenings are fully presented in Table 2.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Univariate analysis of pre-transplant factors associated with overall survival.

https://doi.org/10.1371/journal.pone.0349615.t002

First, log-rank tests were used to assess survival differences across categorical subgroups. This analysis revealed several trends toward an association with poorer overall survival (OS) at a p < 0.10 significance threshold. The most notable associations were observed for pre-transplant recipient neutropenia (p = 0.052), a high Neutrophil-to-Lymphocyte Ratio (NLR > 3, p = 0.093), and high serum uric acid levels (>7.3 mg/dL, p = 0.078). Kaplan-Meier survival curves for these key initial comparisons are illustrated in Fig 2.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Kaplan-Meier curves for overall survival based on initial univariable analysis.

Survival estimates for patient subgroups stratified by pre-transplant variables identified in initial univariable screening. Panels display comparisons for (A) Recipient Neutropenia status, (B) Recipient Neutrophil-to-Lymphocyte Ratio (NLR), categorized at a cutoff of >3, and (C) Recipient Uric Acid level, categorized at a cutoff of >7.3 mg/dL. P-values were calculated using the log-rank test. Abbreviation: NLR, Neutrophil-to-Lymphocyte Ratio.

https://doi.org/10.1371/journal.pone.0349615.g002

Second, to further investigate these and other continuous variables, univariate Gompertz proportional hazards regression was performed. This analysis corroborated the log-rank test findings and identified several additional significant predictors of increased mortality. These included higher recipient age (≥50 years; HR 1.87, p = 0.093), elevated liver enzymes (ALT and AST, p < 0.07 for both), and importantly, higher continuous donor hemoglobin (HR per 1 g/dL increase = 1.017; 90% CI, 1.003–1.031; p = 0.045). Conversely, factors trending toward improved survival included a normal recipient white blood cell count and a normal NLR.

Independent predictors of mortality in multivariable analysis

To disentangle the independent effects of these correlated factors, we constructed our primary multivariable Gompertz proportional hazards model, adjusting for recipient age and sex. The results, including adjusted hazard ratios (aHRs) and 95% confidence intervals, are presented in the forest plot in Fig 3.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Forest plot of adjusted hazard ratios for overall survival.

The plot displays the results from the final multivariable Cox proportional hazards model. Circles represent the adjusted Hazard Ratios (aHRs) for the primary pre-transplant variables, and horizontal lines indicate the 90% confidence intervals (CIs). All hazard ratios are adjusted for recipient age and gender. An aHR greater than 1.0 indicates an increased risk of mortality. Abbreviations: aHR, adjusted Hazard Ratio; CI, Confidence Interval.

https://doi.org/10.1371/journal.pone.0349615.g003

Several factors remained statistically significant predictors of mortality. Elevated recipient liver enzymes (ALT and AST) retained their strong association with poorer survival. Subsequently, to evaluate our core hypothesis, we constructed our primary interaction model. Within this specific model, recipient neutropenia also remained a powerful independent predictor of adverse outcomes (p = 0.045). The main effect of continuous donor hemoglobin did not reach statistical significance (aHR = 1.45; 95% CI, 0.88–2.39; p = 0.148).

The central finding: A qualitative host-graft interaction

The pre-specified primary analytical goal was to test for an interaction between recipient neutropenia and continuous donor hemoglobin. Our analysis revealed a statistically significant and clinically meaningful negative interaction term (aHR for interaction = 0.44; 95% CI, 0.23–0.84; p = 0.013). This indicates a qualitative or “crossover” interaction, signifying that the prognostic impact of donor hemoglobin is not uniform but is fundamentally altered by the recipient’s baseline immune state.

This complex relationship is powerfully visualized by the model-predicted survival curves presented in Fig 4. The plot clearly illustrates that patients with isolated neutropenia (Neutropenia, Low HGB; curve D) experience the poorest predicted survival. In contrast, the addition of a high-hemoglobin graft to neutropenic recipients (Neutropenia, High HGB; curve C) results in a dramatic mitigation of this risk, elevating their predicted survival substantially and demonstrating a clear “rescue effect.” This confirms that the prognostic meaning of donor hemoglobin is context-dependent.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Model-predicted survival curves illustrating the interaction between recipient neutropenia and donor hemoglobin.

The plot displays adjusted survival probabilities derived from the final multivariable Gompertz proportional hazards model, adjusted for recipient age and sex. Patients are stratified into four mutually exclusive subgroups based on pre-transplant neutropenia status (present/absent) and donor hemoglobin level (Low vs. High, categorized by the cohort median of 13.5 g/dL). The curves visually demonstrate the qualitative interaction: while high donor HGB is associated with worse survival in non-neutropenic recipients (curve B vs. A), it is associated with a marked improvement in survival in neutropenic recipients (curve C vs. D), visually suggestive of a potential ‘rescue effect’ that requires formal validation. Confidence intervals (95%) for the predicted survival are shown as shaded areas. The table below the plot indicates the number of patients at risk in each subgroup over time. Abbreviation: HGB, Hemoglobin.

https://doi.org/10.1371/journal.pone.0349615.g004

Model robustness and confirmatory analyses

To ensure the statistical validity of our central finding, a series of pre-planned confirmatory and sensitivity analyses were performed.

First, to confirm that the interaction was not an artifact of the parametric assumptions of the Gompertz model, we repeated the multivariable analysis using a standard semi-parametric Cox proportional hazards model. The result was highly consistent, again yielding a statistically significant interaction of similar magnitude and direction (p for interaction = 0.030). Second, to confirm the robustness of findings derived from our multiple imputation dataset, we compared the results for key continuous variables with those from a complete-case analysis. As shown in S1 Table, the protective association of lower uric acid was highly consistent across both methods (MI aHR = 0.768 vs. CCA aHR = 0.759), enhancing confidence in the primary analysis.

Third, to address potential residual confounding by key transplant-specific covariates, we conducted targeted sensitivity analyses. Respecting EPV constraints, we sequentially and individually added primary diagnosis group, conditioning regimen, and relapse status to our primary adjusted Gompertz model. In each of these separate sensitivity models, the core interaction between recipient neutropenia and donor hemoglobin remained robust and its significance was not materially altered. Furthermore, none of these added clinical covariates emerged as significant independent predictors of mortality when adjusting for the primary exposures.Finally, exploratory subgroup analyses provided further insights into potential effect modification for other variables, as detailed in Fig 5A and 5B. These analyses suggested, for instance, that the adverse effect of neutropenia was most pronounced in older patients (aged ≥50 years). Notably, sex-based stratification revealed an intriguing divergence regarding donor hemoglobin: while higher donor hemoglobin showed a trend towards increased mortality risk in male recipients (aHR = 1.31; 95% CI, 0.86–2.00; p = 0.204), it demonstrated a potentially protective effect against mortality in female recipients (aHR = 0.69; 95% CI, 0.45–1.06; p = 0.092).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Forest plots of exploratory subgroup analyses for key prognostic factors.

(A) Adjusted Subgroup Analysis of Liver Enzymes by Sex. The plot shows adjusted hazard ratios (aHRs) for recipient Alanine Aminotransferase (ALT) and Aspartate Aminotransferase (AST), stratified by patient sex. Models were adjusted for recipient age. Note the significant association with mortality risk is present only in male patients. (B) Subgroup Analyses of Other Key Factors. The plot shows hazard ratios (HRs) for Recipient Neutropenia (stratified by age < 50 vs. ≥ 50 years) and Donor Hemoglobin (stratified by patient sex). Analyses for Recipient Neutropenia and Donor Hemoglobin are from multivariate models adjusted for age and gender. Abbreviations: aHR, adjusted Hazard Ratio; HR, Hazard Ratio; CI, Confidence Interval; ALT, Alanine Aminotransferase; AST, Aspartate Aminotransferase.

https://doi.org/10.1371/journal.pone.0349615.g005

Discussion

In this single-center, hypothesis-generating study of 94 allo-HSCT recipients, we identified a novel, statistically robust interaction between pre-transplant recipient neutropenia and donor hemoglobin. Our primary multivariable analysis confirmed two key findings: first, that the main effect of higher continuous donor hemoglobin was, paradoxically, associated with a trend toward increased mortality in the overall cohort (aHR 1.45, p = 0.148). Second, and more importantly, a significant qualitative interaction (aHR 0.44, p = 0.013) revealed that this association was fundamentally context-dependent. The deleterious effect of high donor hemoglobin was reversed in neutropenic recipients, in whom it was instead associated with a strong trend toward a protective “rescue effect.” In separate adjusted models, established factors, such as elevated pre-transplant liver enzymes, were also confirmed as independent predictors of poorer survival in our cohort.

The central objective of this exploratory analysis was to move beyond linear risk paradigms and investigate the interplay between host vulnerabilities and graft characteristics. Our findings directly address this objective by demonstrating that the prognostic meaning of a key donor characteristic is not fixed but is profoundly altered by the recipient’s pre-transplant immune context.

The observed interaction suggests that for a neutropenic host, where poor hematopoietic reconstitution is a primary threat [17], a “high-vigor” graft—for which high donor hemoglobin may be a proxy—could be uniquely beneficial. This “graft-rescues-host” framework, starkly visualized by the model-predicted survival curves in Fig 4, reframes the clinical challenge from a simple additive model of risk factors to a more dynamic interplay where the optimal graft is context-specific. This concept of dual, context-dependent roles is a recognized principle in transplant immunology, where alloimmune responses can mediate both detrimental GvHD and beneficial GvL effects [18].

Our observation that higher donor hemoglobin is potentially associated with increased mortality in the absence of neutropenia is a provocative finding that runs counter to clinical intuition and lacks direct precedent in donor selection literature, which has primarily focused on canonical factors like age and sex [17]. However, several plausible, non-mutually exclusive hypotheses can be proposed to explain this main effect. First, high hemoglobin may not be a direct causal agent but rather a surrogate for an unmeasured systemic pro-inflammatory state in the donor. This is conceptually analogous to the situation in the recipient, where pre-transplant inflammation, often measured by C-reactive protein (CRP), is a powerful and established independent predictor of increased NRM and poorer OS [19,20]. It is therefore conceivable that a state of “stress erythropoiesis” in the donor, reflected by high hemoglobin [21], marks the transfer of a graft primed with inflammatory cytokines or effector cells, thereby exacerbating post-transplant inflammatory sequelae [22]. A second, alternative hypothesis historically involves direct iron toxicity and reactive oxygen species [23,24]. However, it is crucial to note that our entire cohort utilized peripheral blood stem cell (PBSC) grafts, which contain a negligible red blood cell mass compared to traditional bone marrow harvest. Therefore, direct systemic iron loading strictly from the graft itself is highly unlikely. Instead, we postulate that a constitutionally high donor hemoglobin may act as a systemic surrogate marker for an underlying donor metabolic state—such as altered iron homeostasis or subclinical pro-inflammatory pathways—which could indirectly influence graft composition, immune cell polarization, and early post-transplant inflammatory sequelae [10,25].

However, these general hypotheses alone fail to explain the centerpiece of our analysis: the paradoxical “rescue effect” observed specifically in neutropenic recipients. For this, we propose a more specific, multi-layered mechanism grounded in the fundamental biology of hematopoietic stem cell (HSC) regulation by erythropoietin (Epo). We hypothesize that a donor with a constitutionally high hemoglobin level has a correspondingly low systemic Epo tone. Based on emerging biological evidence, the graft from such a donor would be ‘doubly advantaged.’ Seminal studies have established that high systemic Epo levels act as a master regulator that not only (i) impedes the physical migration and bone marrow homing of hematopoietic stem/progenitor cells (HSPCs), but also (ii) actively reprograms their lineage fate by suppressing non-erythroid options, including the critical myeloid pathway [26]. This concept is further supported by recent clinical trials and mechanistic models from studies recent demonstrating that elevated EPO signaling in the stem cell niche actively impairs the engraftment and early homing of infused CD34 + cells, whereas attenuating this pathway accelerates neutrophil recovery [27,28]. Our finding is perfectly consistent with an inverse of this dual mechanism. A graft originating from a low-Epo environment would be released from both Epo-mediated migratory inhibition and myeloid suppression. This would render its HSPCs intrinsically permissive for more rapid marrow homing and robust myeloid differentiation—a decisive advantage for a neutropenic recipient in the critical race against early infection.

Furthermore, a third, complementary mechanism could involve the intrinsic epigenetic state of the HSCs themselves. Recent work in chronic stress erythropoiesis models (β-thalassemia) has demonstrated that the HSC pool can become epigenetically ‘primed’ for the erythroid lineage through complex signaling involving TGFβ and autophagy [29]. While this inherent erythroid priming might seem counter-intuitive to a myeloid rescue, it suggests that HSCs from a donor in a state of chronic erythropoietic demand exist in a hyper-responsive or ‘poised’ state. It is plausible that when transplanted into the powerful inflammatory, cytokine-rich milieu of a neutropenic host, these poised stem cells are uniquely capable of rapidly responding to the overriding signals demanding urgent myeloid regeneration. Taken together, these interconnected mechanisms transform our statistical observation into a tangible, testable biological hypothesis rooted in established principles of hematopoiesis.

The primary theoretical implication of our research is its challenge to simplistic, linear risk-assessment paradigms in allo-HSCT. Our data, supported by recent biological findings [26,30], suggest that the interplay between the host environment and graft potential is a critical, underappreciated axis in determining transplant outcomes. The practical implication, though strictly reserved for the future, is the potential for more nuanced and personalized donor selection strategies. For a profoundly immunocompromised patient—a group known to suffer high mortality from poor graft function [17]—the ideal graft may not simply be the one with the fewest intrinsic risk factors, but rather the one with the highest intrinsic potential for rapid myeloid reconstitution. However, it must be unequivocally stated that these findings are strictly hypothesis-generating and should not, under any circumstances, be used to guide current clinical practice.

It is essential to interpret our findings within the context of several significant limitations, in accordance with the STROBE guidelines. First and foremost, our study’s primary limitation is its modest sample size (n = 94) and the correspondingly limited number of events (n = 30). This constrains statistical power and elevates the risk of a Type I error. While our central interaction finding was statistically robust —persisting even after sensitivity analyses adjusting for primary diagnosis, conditioning regimen, and relapse status—it must be interpreted with significant caution until independently replicated.. Additionally, the absence of an external validation cohort is a key limitation; therefore, our findings remain strictly exploratory and must be validated in independent, larger databases.

Second, the retrospective, single-center design inherently limits generalizability. Findings from our unique regional cohort in Western Iran may not be applicable to other populations. Specifically, our region is characterized by a high prevalence of red blood cell disorders, such as thalassemia minor. Although our institutional protocol strictly prioritizes donors without any hematological conditions—utilizing them only as an absolute last resort when no alternative donor is available—the unique genetic background of our local population could still theoretically influence baseline donor hemoglobin distributions and limits direct generalizability to populations with different genetic backgrounds. Additionally, the notably high prevalence of pre-transplant neutropenia (43.6%) in our cohort requires consideration. This high rate likely reflects a heavily pre-treated patient population, with underlying factors such as advanced refractory disease or repeated cycles of salvage chemotherapy acting as potential unmeasured confounders that could independently influence both neutropenia and survival. Furthermore, this design is susceptible to unmeasured confounding. For instance, our retrospective dataset lacked granular data on established pre-transplant comorbidity indices (such as HCT-CI) and baseline recipient performance status (e.g., ECOG or Karnofsky scores). The absence of these standard variables prevents us from fully adjusting for the baseline physiological frailty of the recipients, which is a known independent predictor of mortality. Additionally, we lacked data on pre-transplant inflammatory markers in donors, such as C-reactive protein (CRP), which has been established as a potent prognostic factor in recipients [19,20]. The absence of such data prevents us from dissecting whether high hemoglobin is a direct biological effector or merely a surrogate for donor inflammation.

Third, while we analyzed engraftment kinetics—demonstrating a uniform median time to engraftment of 10 days across all patient subgroups with no significant influence from donor hemoglobin, recipient neutropenia, or their interaction (p = 0.812)—our study lacks the granular data required to directly test other downstream aspects of our biological hypotheses. This uniform engraftment suggests that the observed survival differences are likely driven by post-engraftment events, such as qualitative immune reconstitution, GVL effects, or susceptibility to late infections, rather than initial homing or primary graft failure. Without systematic data on these post-engraftment variables or documented rates of early infection, our proposed mechanism remains a plausible but unproven hypothesis.

While significant, the limitations of this study delineate a clear and compelling roadmap for future research. The primary contribution of this work is to catalyze a new line of inquiry. We propose a two-stage approach:

1. Validation: The essential first step is the validation of our statistical finding. Analysis of large, multi-center registry databases (e.g., CIBMTR, EBMT) is required to ascertain whether the interaction between recipient neutropenia and donor hemoglobin is a robust and generalizable phenomenon.2. Mechanistic Elucidation: Following validation, prospective studies are warranted. Drawing on established links between inflammation, EPO signaling, and graft function [30], these studies must be designed to co-collect standardized donor hemoglobin levels along with donor EPO levels, inflammatory markers (e.g., CRP), detailed graft cellular composition, and meticulously track qualitative immune reconstitution and rates of early infectious complications.

Only through such a systematic, multi-pronged approach can we move this observation from a novel statistical signal to a potential clinically actionable biomarker, ultimately refining the art of donor selection for our most vulnerable patients.

Conclusion

This single-center, hypothesis-generating study of 94 allo-HSCT recipients identified a statistically robust context-dependent interaction between recipient neutropenia and donor hemoglobin. While we propose a mechanistic framework involving EPO-mediated effects on HSPC behavior, this remains speculative without direct validation. The primary contribution of this work is to generate a testable hypothesis for multi-center validation studies that should prospectively collect donor EPO levels, graft composition, and engraftment kinetics. Only through such validation can this observation be considered for potential clinical application in future donor selection algorithms.

Supporting information

Acknowledgments

The authors would like to thank the staff of the Kermanshah Bone Marrow Transplantation Centre for their assistance in the course of this study. Furthermore, we extend our sincere gratitude to the Aamar Afzar Institute, and particularly Dr. Abbas Keshtkar, for their invaluable training courses on statistics and research methodology in medical and humanities sciences.