Autologous hematopoietic stem cell transplantation (aHSCT) is a therapeutic intervention for patients with lymphoma and multiple myeloma (MM) [1,2]. The procedure is typically performed using peripheral blood stem cells (PBSCs). Because hematopoietic stem cells (HSCs) normally circulate in the peripheral blood in very low numbers, they must first be mobilized into the circulation before collection by apheresis [3].

Poor mobilization can result in insufficient PBSC yield for aHSCT [4,5]. This represents a significant challenge, as a minimum of 2 × 10⁶ CD34+ cells/kg is required to ensure successful transplantation. This threshold is necessary to achieve rapid and sustained hematopoietic recovery, while also reducing the risks of prolonged hospitalization, blood transfusion requirements, and infection [6,7]. The optimal HSC dose is approximately 5 × 10⁶ CD34+ cells/kg [8]. Failure to achieve adequate mobilization may necessitate re-mobilization or, in some cases, prevent transplantation altogether.

Several factors have been associated with an increased risk of mobilization failure, including bone marrow infiltration, older age, and myelodysplastic changes [9–11]. However, the predictive accuracy of these parameters for identifying suboptimal mobilization remains uncertain. Although predictive models have been proposed to address this limitation [12], their applicability across different patient populations has not been fully validated. Early identification of patients at high risk for poor mobilization is essential, as it could minimize resource use, facilitate timely consideration of alternative strategies, and ultimately improve patient outcomes.

This study aimed to identify predictive factors and develop a new risk model for poor mobilization in aHSCT candidates. Additionally, the study validated an existing predictive model [12], developed a novel scoring system based on specific criteria, and compared their performance to optimize clinical decision-making.

This retrospective study was conducted at the Hematology, Oncology, and Stem Cell Transplantation Research Center, Shariati Hospital in Tehran, Iran. Patients eligible for aHSCT with a diagnosis of MM, non-Hodgkin lymphoma (NHL), or Hodgkin’s disease (HD) between January 2013 and December 2021 were included. The study protocol was approved by the Ethics Committee of Tehran University of Medical Sciences, and all data were collected and analyzed anonymously. Diagnoses of lymphoma were confirmed by pathology review, and diagnoses of MM were established jointly by hematologists and pathologists.

The inclusion criterion was age ≥18 years. Patients with acute or chronic leukemia or those previously mobilized with plerixafor were excluded, as the latter group typically represents poor mobilizers and could bias the predictive analysis. Clinical data, including disease characteristics, prior therapies, mobilization regimen, and apheresis outcomes, were recorded.

Mobilization failure was defined as a total CD34+ cell yield of <2 × 10⁶ cells/kg or the requirement for large-volume apheresis involving multiple collection cycles. Stem cell collections were performed using a continuous-flow cell separator following the institutional protocol.

The primary dependent variable was poor mobilization. Independent variables were defined as follows:

• •

  Extensive bone marrow (BM) infiltration was defined based on the most recent BM assessment performed prior to stem cell mobilization, and not at the time of initial diagnosis. Extensive BM infiltration was considered as >60% plasma cells in patients with MM or >30% lymphomatous infiltration in patients with lymphoma.

• •

  Advanced-phase disease: prior treatment with at least two lines of cytotoxic chemotherapy.

• •

  Prior marrow-toxic exposure: previous therapy with lenalidomide, melphalan, or fludarabine.

• •

  Extensive radiotherapy: prior radiation involving marrow-bearing sites.

A total of 430 patients, including Hodgkin’s lymphoma (n = 134), non-Hodgkin’s lymphoma (n = 113), and MM (n = 183), were enrolled in this study. The median age was 47 years. The median number of chemotherapy courses prior to mobilization was nine. Overall, 15.5% of patients had received prior marrow-toxic exposure, including melphalan or fludarabine in lymphoma patients and maintenance lenalidomide in MM patients. At the time of mobilization, 17.5% of patients were in partial remission, while the remainder had achieved complete remission. BM biopsy was performed in all patients prior to mobilization. Median BM infiltration was 35% in patients with mobilization failure compared with 45% in patients with successful mobilization. Baseline demographic and clinical characteristics by disease group are summarized in Tables 1 and 2.

Overall, 123 patients (28.6%) were classified as poor mobilizers, while 307 patients (71.4%) achieved adequate stem cell collection for transplantation. The causes of suboptimal mobilization were an unsatisfactory peak CD34+ cell count (<20 cells/µL) in 19 (15.4%) patients, insufficient total CD34+ yield (≤2 × 10⁶ cells/kg) in ten (8.1%) patients, and both criteria in 94 (76.4%).

Patients with failed mobilization demonstrated significantly greater BM infiltration and lower pre-mobilization CD34+ cell counts compared with those who mobilized successfully. Quantitative parameters associated with mobilization outcome are detailed in Table 3.

Modeling

Given the modest performance of the pPM and GITMO scores in the current dataset, further analyses were conducted to develop a more robust predictive model. Adjusting cut-off values within these models resulted in reduced sensitivity and specificity with positive predictive values (PPVs) ranging only from 40.3% to 56.7%.

The predictive role of the CD34+ cell count in the mobilization outcome is well established [4–8]. In accordance with the primary objective of this study to develop a model capable of identifying poor mobilizers prior to apheresis, two distinct models were constructed: one including the pre-mobilization CD34+ cell count and one excluding this variable.

Multivariate logistic regression was performed using the variables found significant in the univariate analysis. In the model excluding the CD34+ cell count, the final regression model was formulated after five steps using forward stepwise selection. The improvement employing the chi-square test was 344.64 (p-value <0.001), indicating strong statistical significance. The Cox-Snell and Nagelkerke R² values demonstrated that the model explained between 55% and 79% of the variance in poor mobilization. This model achieved a sensitivity of 89.4%, specificity of 94.1%, PPV of 89.4%, NPV of 94.1%, and an overall accuracy of 92.8%.

When the CD34+ cell count was incorporated into the multivariate regression, the final model was achieved after four steps. The improvement employing the chi-square test was 415.28 (p-value <0.001), with Cox–Snell and Nagelkerke R² values showing that 62% to 89% of the variance in outcome could be explained.

Independent predictors identified in the two models were as follows:

• −

  Model excluding the pre-mobilization CD34+ cell count: platelet count, BM infiltration, advanced disease status, full-course chemotherapy, and the use of more than two alkylating agents.

• −

  Model including the pre-mobilization CD34+ cell count: platelet count, BM infiltration, extensive radiotherapy, and the CD34+ cell count.

Cut-off optimization yielded excellent discriminative performance. In the CD34+ model, a threshold >0.219 produced an AUC of 0.97, with sensitivity 97.6% and specificity 88.3%. In the model excluding the CD34+ count, a cut-off >0.502 resulted in an AUC of 0.98, with sensitivity 94.3% and specificity 97.7% (Figure 1; Tables 5 and 6).

Figure 1.

Area under the receiver operating characteristic curves for poor mobilization using (a) the model including the pre-mobilization CD34+ count and (b) the model excluding the pre-mobilization CD34+ count.

Table 5.

Statistical measures of predicting models for poor mobilization.

Measure	Value	95% CI
Model excluding the pre-mobilization CD34+ cell count
Cut-off	0.71
Sensitivity	0.911	0.846 - 0.954
Specificity	0.987	0.967 - 0.996
Positive Likelihood Ratio	69.89	26.36 −185.31
Negative Likelihood Ratio	0.09	0.05 - 0.16
Disease prevalence*	28.60%	24.38% - 33.13%
Positive Predictive Value*	0.965	0.913 - 0.987
Negative Predictive Value*	0.965	0.940 - 0.979
Accuracy*	0.965	0.943 - 0.980
Model including the pre-mobilization CD34+ cell count
Cut-off	0.43
Sensitivity	0.911	0.846 - 0.954
Specificity	0.935	0.901 - 0.959
Positive Likelihood Ratio	13.98	9.12 - 21.43
Negative Likelihood Ratio	0.10	0.05 - 0.17
Disease prevalence*	28.60%	24.38% - 33.13%
Positive Predictive Value*	0.848	0.785 - 0.896
Negative Predictive Value*	0.963	0.937 - 0.978
Accuracy*	0.928	0.899 - 0.950

95% CI: 95% confidence interval.

⁎

These values are dependent on the disease prevalence.

Table 6.

Independent predictive factors for mobilization failure identified by forward stepwise method with multiple logistic regression.

Model excluding the pre-mobilization CD34+ cell count
Variables in the Equation	β	p-value	Odds ratio	95% CI
				Lower	Upper
Advanced phase disease	1.086	0.010	2.961	1.292	6.788
Bone marrow infiltration	−0.156	<0.0001	0.855	0.810	0.904
Platelet count	−0.043	<0.0001	0.958	0.946	0.970
Full courses of previous therapy	2.443	0.001	11.512	3.660	36.204
More than two Alkylating	1.615	0.001	5.027	2.000	12.637

Model including the pre-mobilization CD34+ cell count
Variables in the Equation	β	p-value	Odds ratio	95% CI
				Lower	Upper
Previous extensive radiotherapy	1.933	0.012	6.912	1.525	31.333
Bone marrow infiltration	−0.118	0.001	0.889	0.830	0.952
Pre-mobilization CD34+ cell count	−0.518	<0.0001	0.596	0.502	0.708
Platelet count	−0.033	<0.0001	0.967	0.954	0.981

95% CI: 95% confidence interval.

The probability of mobilization failure according to the model score without the CD34+ cell count, can be calculated as below:

The AUCs of the GITMO score, pPM score and the scores of the models with and without CD34+ inclusion were compared. The AUC of previous models were significantly inferior to both the models constructed in this study (p-value <0.0001). Figure 2 shows the result.

Figure 2.

Comparing the area under the receiver operating characteristic curve of the GITMO (Gruppo Italiano Trapianto Midollo Osseo) score with (a) the model including the pre-mobilization CD34+ count and (b) the model excluding the pre-mobilization CD34+ count.

This study evaluated mobilization outcomes in patients undergoing aHSCT, with the primary aim of identifying prognostic risk factors and assessing the applicability of previously published predictive scores (GITMO and pPM). A secondary objective was to develop a practical predictive model capable of identifying patients at high risk of mobilization failure, ideally independent of the well-established predictor: pre-apheresis CD34+ cell count. To this end, four models were examined: the GITMO score, the pPM score, and two newly developed models (one including and one excluding the pre-mobilization CD34+ cell count).

The pPM score demonstrated poor performance in the current dataset, essentially failing to predict poor mobilization. The GITMO score also showed limited accuracy, with modest sensitivity (70%), specificity (69%), and an AUC of 0.70. In contrast, the new predictive models performed substantially better, each achieving an AUC greater than 0.9, with excellent sensitivity, specificity, and PPVs. Using optimized cut-offs of 0.71 (model excluding the pre-mobilization CD34+ cell count) and 0.43 (model including the pre-mobilization CD34+ cell count), both models demonstrated strong clinical utility in predicting mobilization failure.

Several factors may explain the discrepancy between the present findings and those of prior studies. The pPM score developed by Olivieri et al. [14], was based on variables markedly different from those of this patient population. For example, in this study the median number of prior chemotherapy courses was nine (range: 2–26), compared with only 1–6 in the Italian study. Likewise, conditioning regimens differed substantially: Olivieri et al. identified granulocyte colony-stimulating factor (G-CSF) use without chemotherapy as a strong negative predictor (β = 2.25), whereas in this study all patients received G-CSF alone. Age distribution also varied, with the present median patient age being 46 years, while 26% of patients in the Italian cohort were younger than 45 years. Moreover, Italian patients tended to undergo aHSCT earlier in their disease course, resulting in lower cumulative exposure to marrow-toxic therapy compared to this study.

These contextual differences highlight why previously published models may lack external validity across diverse patient populations. Despite recent advances in peripheral blood stem cell collection including apheresis instrumentation, mobilization failure remains a challenge, with reported rates ranging from 5% to 40% [15–17]. The transfused stem cell dose, directly linked to the number of CD34+ cells collected, is a key determinant of engraftment success [14,18]. While circulating CD34+ cell counts remain the most reliable predictor of mobilization success, their utility is limited by the fact that they are only available immediately prior to apheresis. This delay prevents early identification and stratification of patients who might benefit from alternative mobilization strategies.

The new models address this limitation by incorporating additional predictive variables that are assessable before apheresis. In the model excluding the CD34+ cell count, independent predictors of mobilization failure included platelet count, BM infiltration, advanced disease status, history of full-course chemotherapy, and prior exposure to more than two alkylating agents. When the CD34+ cell count was included, the independent predictors were platelet count, BM infiltration, pre-mobilization CD34+ cell count, and prior extensive radiotherapy. These findings suggest that predictive models can be tailored to different clinical contexts, either allowing earlier patient stratification without pre-mobilization CD34+ cell data or achieving maximum predictive accuracy including the pre-mobilization CD34+ count.

A 28.6% rate of mobilization failure was observed in older patients in this study, whereas other investigators such as Rossi et al. [19] and Wuchter et al. [20] reported rates as high as 46%. This discrepancy is likely attributable to differences in mobilization techniques between studies, but it warrants further investigation. Demiroğlu et al. reported that higher pre-mobilization platelet counts, and the use of chemotherapy-based mobilization regimens were associated with greater mobilization success. While the platelet count was also identified as a significant predictor in the present study, the mobilization regimen could not be examined since our protocol was chemotherapy-free [21].

Although no single measure of accuracy can fully capture the clinical value of a predictive tool, the AUC is widely considered a reliable indicator of discriminative performance, as it is independent of cut-off values or disease prevalence [22]. In this respect, the new models demonstrated excellent predictive capability, each achieving an AUC greater than 0.95.

One important consideration in evaluating predictive scores across studies is the discrepancy between the minimal threshold and the optimal target dose of CD34+ cells. Defining poor mobilization solely by the minimum cut-off of 2 × 10⁶ CD34+ cells/kg may underestimate clinical requirements, as higher doses are often needed depending on the patient’s condition. This variability may affect the accuracy and applicability of predictive models.

The primary goal of this study was to establish a practical tool for decision-making before mobilization. These new models incorporate easily obtainable variables available prior to apheresis and classify patients as ‘poor mobilizers’ or ‘successful mobilizers.’ Such information can guide clinical planning, including reconsideration of mobilization targets, early use of CXCR4 antagonists, cytokine therapy, or large-volume apheresis. The strong positive likelihood ratio (>10), combined with sensitivity and specificity above 85%, supports the reliability and potential generalizability of these models across different settings.

In the present cohort, plerixafor was used only on-demand, limiting the ability to evaluate or recommend its upfront use. Nevertheless, based on this model, it would be reasonable to consider early administration of plerixafor in patients classified as poor mobilizers, in line with recent reports [36]. Similarly, because G-CSF–based mobilization was employed exclusively, conclusions regarding chemotherapy-based mobilization strategies cannot be provided.

In addition to cytokine-based approaches, chemotherapy-based mobilization strategies, particularly vinorelbine combined with G-CSF, have been reported as effective alternatives for poor mobilizers, leading to higher pre-apheresis CD34⁺ cell counts, predictable timing of collection (typically on days 7–8), and acceptable toxicity with a relatively low risk of infectious complications, while remaining substantially more cost-effective than plerixafor [4,6,10].

As a future direction, prospective studies could apply this predictive framework prior to mobilization, without reliance on circulating CD34⁺ counts, to identify poor mobilizers and randomize them between G-CSF–based mobilization with early or on-demand plerixafor versus chemotherapy-based mobilization with vinorelbine. In addition to apheresis success rates, such studies should incorporate pharmacoeconomic analyses to assess the cost-effectiveness of these competing strategies.

Finally, although the findings of this study suggest that adapting mobilization approaches according to the predictive model may be clinically beneficial, external validation in an independent cohort is required. Furthermore, prospective clinical trials are essential to confirm the effectiveness and safety of implementing this model in routine practice.

This study developed a predictive model for inadequate mobilization that excludes the pre-mobilization CD34+ cell count, making it applicable prior to the initiation of the mobilization process. Both versions of the model demonstrated sensitivity, specificity, positive predictive value, and positive likelihood ratio exceeding 90%. Early identification of poor mobilizers enables individualized monitoring, supports optimized decision-making, and ultimately enhances patient outcomes.

This research received no specific grant from funding agencies in the public, commercial, or non-profit sectors.

The data that support the findings of this study are available from the corresponding author upon reasonable request.