Cost effectiveness analysis of PEG-rhG-CSF versus rhG-CSF for primary prophylaxis of chemotherapy induced neutropenia in Chinese patients with non-Hodgkin lymphoma using real world data

Abstract

Objectives

To evaluate the cost-effectiveness of pegylated recombinant human granulocyte colony-stimulating factor (PEG-rhG-CSF) versus recombinant human granulocyte colony-stimulating factor (rhG-CSF) for primary prophylaxis of chemotherapy-induced neutropenia (CIN) among patients with non-Hodgkin lymphoma (NHL) treated with R-CHOP.

Methods

We developed two linked state-transition (Markov) models to estimate clinical and economic outcomes: a chemotherapy-phase model (18 weeks) and a post-chemotherapy model with a 20-year time horizon. Costs and selected transition probabilities were informed by real-world data from the Jiangsu Provincial Public Health Information Platform (a regional multicenter database); remaining inputs were obtained from the published literature. Parameter uncertainty was explored using deterministic one-way sensitivity analysis (DSA) and probabilistic sensitivity analysis (PSA).

Results

In the base-case analysis, rhG-CSF was associated with lower total costs ($17,299.41 vs. $17,947.88) and slightly higher effectiveness (7.75 vs. 7.74 QALYs) compared with PEG-rhG-CSF. At a willingness-to-pay (WTP) threshold of one times China’s per capita GDP in 2024 ($13,581.42 per QALY), PEG-rhG-CSF yielded a negative incremental net monetary benefit (iNMB: −$739.55), indicating that rhG-CSF was the economically preferred strategy. Deterministic and probabilistic sensitivity analyses demonstrated that the base-case findings were generally robust across plausible variations in model parameters. The probability that rhG-CSF was cost-effective relative to PEG-rhG-CSF remained above 60% across a range of WTP thresholds.

Conclusions

Based on real-world data and economic modeling, rhG-CSF showed a slightly higher probability of being cost-effective than PEG-rhG-CSF for primary prophylaxis of CIN among patients with NHL treated with R-CHOP.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-026-04954-w.

Introduction

According to GLOBOCAN 2022 [1], non-Hodgkin lymphoma (NHL) accounted for an estimated 553,389 new cases and 250,679 deaths worldwide in 2022. Epidemiological studies have demonstrated age-related variation in incidence: in Spain, overall NHL incidence peaks at ages 80–84 years [2]; in the United States, diffuse large B-cell lymphoma (DLBL) and peripheral T-cell lymphoma (PTCL) show peak incidence in the 70–74 age group [3]. Among individuals aged 60 years and older in China, the age-specific incidence rates were highest in men aged 85–89 years and in women aged 95 years and above [4].

China is experiencing a rapid demographic transition, with the proportion of older adults (defined as individuals aged ≥ 65 years) expected to rise from 10.92% in 2018 to 24.71% by 2044 [5]. Coupled with the relatively late age at onset of NHL in China, the national disease burden is likely to rise substantially over the coming decades. Increasing incidence and mortality will intensify demand for health care services and drive medical expenditures. In this context, evaluating the cost-effectiveness of NHL therapies is critical to inform resource allocation and optimize health outcomes in China.

Chemotherapy-induced neutropenia (CIN) and febrile neutropenia (FN) are common hematologic complications of cytotoxic chemotherapy. Greater depth and longer duration of neutropenia are associated with higher antimicrobial use, increased FN-related hospitalizations, and elevated mortality. To mitigate these risks, granulocyte colony-stimulating factors-recombinant human G-CSF (rhG-CSF) and its pegylated, long-acting formulation (PEG-rhG-CSF)-are widely used for prophylaxis. The National Comprehensive Cancer Network (NCCN) and European Society for Medical Oncology (ESMO) both recommend primary prophylaxis with G-CSF for patients receiving chemotherapy regimens that confer an intermediate-to-high risk of FN [6, 7].

In China, basic medical insurance coverage for granulocyte colony-stimulating factors (G-CSFs) has broadened as reimbursement policies have evolved, but criteria remain product-specific. Short-acting recombinant human G-CSF (rhG-CSF) is generally reimbursed with minimal utilization restrictions. In contrast, PEG-rhG-CSF reimbursement commonly requires prior severe chemotherapy-related neutropenia or other clinical prerequisites, reflecting cost-containment policies of the basic medical insurance system rather than differential clinical effectiveness. Although PEG-rhG-CSF has a higher acquisition cost per cycle, its extended duration of action enables once-per-cycle dosing. It may reduce clinic visits, administration costs, and overall resource use. By contrast, rhG-CSF has a lower per-dose price but typically requires repeated dosing within each chemotherapy cycle, potentially increasing cumulative treatment and administration costs. From a pharmacoeconomic perspective, formulary decisions should weigh the total cost of care, effectiveness in preventing FN, and treatment adherence within the prevailing reimbursement framework to identify the most cost-effective option for Chinese payers and patients.

Methods

General information

Literature sources

We conducted a systematic literature review to inform model inputs. PubMed and Embase were searched to identify studies relevant to CIN and FN. The primary objective was to extract health state-specific parameters for the cost-effectiveness model, notably utility weights and mortality rates. Where real-world data were unavailable, these literature-derived estimates were used to populate the model.

Patient data sources

Data were obtained from the Jiangsu Provincial Public Health Information Platform, a regional multicenter medical database. We extracted electronic health records for patients with non-Hodgkin lymphoma (NHL) admitted to tertiary hospitals in Jiangsu Province from January 1, 2019, through December 31, 2023. The analysis evaluated the cost-effectiveness of pegylated recombinant human granulocyte colony-stimulating factor (PEG-rhG-CSF) versus recombinant human granulocyte colony-stimulating factor (rhG-CSF) for primary prophylaxis of CIN.

Inclusion criteria:

• Complete demographic information (sex and age).

• Confirmed diagnosis of NHL.

• Receipt of at least one chemotherapy cycle and administration of PEG-rhG-CSF or rhG-CSF within 7 days after chemotherapy.

• Treatment with the R-CHOP regimen (rituximab plus cyclophosphamide, doxorubicin, vincristine, and prednisone).

Exclusion criteria:

• Bone marrow transplantation (BMT) or hematopoietic stem cell transplantation (HSCT) within 30 days before or after the first chemotherapy, or a diagnosis of acute myeloid leukemia (AML).

• Chemotherapy within 180 days prior to the index (first) chemotherapy.

• Laboratory evidence or ICD-10 diagnosis of neutropenia—absolute neutrophil count (ANC) < 2 × 10⁹/L—before the first chemotherapy.

• Weekly chemotherapy schedules.

• Febrile neutropenia (FN) occurring between the first chemotherapy and the first G-CSF administration.

• Use of G-CSF within 30 days prior to the first chemotherapy.

Propensity score matching (PSM) was applied to minimize confounding between treatment groups. Patients receiving PEG-rhG-CSF were matched 1:1 with those receiving rhG-CSF using predefined baseline covariates. Details of the matching procedure are described below. Propensity scores were estimated using a logistic regression model, which incorporated a comprehensive set of covariates: demographic variables (sex, age), baseline comorbidities (diabetes mellitus, renal disease, liver disease, human immunodeficiency virus (HIV), infection, chronic obstructive pulmonary disease (COPD)), and laboratory measures (platelet count, white blood cell count, neutrophil count, and hemoglobin level). Subsequently, 1:1 nearest neighbor matching was conducted to balance the two groups, with a caliper width of 0.01 and no replacement of control subjects post-matching. The balance of covariates between the matched groups was evaluated using the standardized mean difference (SMD), where an SMD < 0.1 was considered indicative of adequate balance.

Model structure

A pharmacoeconomic model was developed in Microsoft Excel to evaluate the cost-effectiveness of PEG-rhG-CSF versus rhG-CSF for the primary prophylaxis of CIN in patients with NHL. The model comprised two components. A Markov model was constructed, simulating a hypothetical cohort of 1,000 patients, with baseline demographic characteristics derived from the retrospective patient data collected in this study.

Short-term Markov model (chemotherapy phase)

The short-term model included five health states: Prevent CIN after chemotherapy, severe neutropenia (SN), FN, reduced relative dose intensity (RDI ≤ 90%), and death (Figure 1). SN was defined as an absolute neutrophil count (ANC) below 1 × 10⁹/L. FN was defined as meeting both of the following criteria: (1) ANC below 1 × 10⁹/L or confirmed granulocytopenia; and (2) prior use of antibiotics or antipyretics, or confirmed infection or fever. Patients transitioning to the RDI ≤ 90% state were defined as those who, after experiencing SN or FN, received chemotherapy at a reduced dose intensity (< 90% of the planned chemotherapy regimen). Previous studies have commonly applied RDI thresholds of 80% or 85% to assess the association between RDI and clinical outcomes [8, 9]. In the present study, an RDI threshold of 90% was used as a more conservative cutoff, closer to the planned therapeutic dose, to evaluate whether modest reductions in dose intensity may affect treatment outcomes [10]. To avoid double counting of direct medical costs associated with SN and FN, the SN state in this model included only patients who experienced SN without concurrent FN. The model allows a transition from SN to FN to reflect clinical progression within a chemotherapy cycle, whereas FN was treated as the most severe neutropenia outcome within a cycle and thus not followed by a transition back to SN. Once patients entered the death state, no further transitions were permitted. Prophylactic administration of G-CSF was assumed to begin 24 h after completion of chemotherapy. Based on state transition probabilities, patients could either remain in their current health state or progress to another state within each model cycle. SN and FN were modeled as cycle-specific acute event states, indicating the occurrence of neutropenia or febrile neutropenia during a given chemotherapy cycle rather than persistent conditions. Therefore, patients could experience SN or FN in multiple cycles, but these states were not allowed to persist across consecutive cycles. The model simulated a 21-day chemotherapy-induced leukopenia cycle, repeated for a total of six cycles.

Fig. 1.

Markov model during chemotherapy

Long-term Markov model (post-chemotherapy phase)

The long-term model comprised three health states: RDI, non-RDI, and death (Figure 2). After completion of chemotherapy for non-Hodgkin lymphoma, patients who achieved an RDI ≤ 90% in the short-term model entered the RDI state, while those who did not achieve an RDI ≤ 90% entered the non-RDI state. Patients in both states could transition to the death state, with differing mortality risks. All patients entered the model at an average age of 60 years. The model adopted a one-year cycle length and a total time horizon of 20 years. During this period, patients were assumed to transition to the death state annually according to state-specific transition probabilities.

Fig. 2.

Markov model after chemotherapy

Model parameters

All model input parameters, including costs, utilities, and transition probabilities, were obtained from published literature and Real-world data. Base-case values and plausible ranges were determined according to the most relevant and high-quality sources.

For parameters included in the probabilistic sensitivity analysis, appropriate probability distributions were assigned following standard practice: beta distributions for probabilities and utility values (bounded between 0 and 1), gamma distributions for cost parameters (non-negative and right-skewed), and log-normal distributions for relative risks. Costs and outcomes were discounted at an annual rate of 5%. This approach is consistent with recommendations from the Chinese Guidelines for Pharmacoeconomic Evaluations (2020) [11].

Cost parameters

The analysis was conducted from the payer’s perspective, considering only direct medical costs. These included the costs of leukocyte-stimulating agents, chemotherapy, and treating infections related to SN or FN. Given that PEG-rhG-CSF and rhG-CSF have comparable incidence and severity of adverse events, and that bone pain-a common adverse reaction-is generally managed with low-cost nonsteroidal anti-inflammatory drugs (NSAIDs), costs associated with adverse event management were not included in the model. Drug prices were obtained from a publicly available provincial procurement database in Jiangsu, China [12]. All costs were converted to 2024 US dollars (USD) using the average exchange rate, and presented as USD thereafter.

Cost of G-CSF

The study included 215 patients in each of the PEG-rhG-CSF and rhG-CSF groups. G-CSF products were identified based on real-world prescribing patterns observed in the study cohort. For the economic model, a representative PEG-rhG-CSF product and a representative rhG-CSF product were selected according to their frequency of use. Drug specifications, standard dosing regimens, and administration schedules were obtained from approved product labels and real-world prescribing records in the database. Individual patient weight data were not available in the database; therefore, average dosing and vial usage were derived from observed real-world prescribing patterns rather than weight-based calculations.

Cost of chemotherapy drug

The R-CHOP regimen is the standard first-line chemoimmunotherapy protocol for patients with NHL, particularly those with diffuse large B-cell lymphoma (DLBCL). It is associated with a high cure rate. In the decision-analytic model, this study selected the R-CHOP regimen as the representative chemotherapy protocol for NHL patients. The regimen comprises rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone. Details regarding dosage and administration are presented in Table 1.

Table 1.

R-CHOP regimen dosage and administration

Medicines	Dosage and administration (Adults)	Time of administration
Rituximab	375 mg/m² intravenous infusion	Day 1
Cyclophosphamide	750 mg/m² intravenous infusion	Day 1
Doxorubicin liposome	20 mg/m² intravenous bolus injection	Day 1
Vincristine	1.4 mg/m² (maximum 2 mg) intravenous infusion	Day 1
Prednisone	100 mg orally	Day 1–5

Chemotherapy drug dosages were derived from the matched real-world dataset. Average per-cycle dosages were calculated based on observed prescribing records. The required number of vials per cycle was estimated according to routine clinical practice, assuming no vial sharing or repackaging.

Direct Medical costs associated with SN and FN

Direct medical costs associated with SN and FN were derived from a published real-world cost analysis conducted in China [13]. Relevant cost items were extracted and converted to 2024 US dollars for model input.

Transition probability

Transition probabilities of short-term Markov model

Transition probabilities were defined to represent the likelihood of patients moving between health states in each cycle. Specifically, p-c-sn denotes the probability of transitioning from the “Prevent CIN after chemotherapy” state to the SN state, p-c-fn represents the probability of transitioning from the “Prevent CIN after chemotherapy” state to the FN state, and so forth.

Transition probabilities of long-term Markov models

Following completion of chemotherapy, patients were classified into two long-term health states based on whether the actual RDI during treatment was ≤ 90%: the RDI ≤ 90% group and the RDI > 90% group. These two states were subsequently incorporated into the Markov model for long-term survival simulation. The hazard ratio for mortality between patients with RDI > 90% and those with RDI ≤ 90% was 0.48 [10]. The overall mortality rate was estimated as a weighted average of the mortality rates for the two RDI groups [14]. Mortality probabilities for patients with RDI ≤ 90% and RDI > 90% during chemotherapy were further derived.

Utility value

The utility values for each health state in the study were obtained from published literature.

Measurement of outcomes

Health outcomes in the model were measured in quality-adjusted life years (QALYs), which capture both the quantity and quality of life by weighting survival time according to health state–specific utility values.

Cost-effectiveness was evaluated using incremental net monetary benefit (iNMB) in the base-case and one-way sensitivity analyses, and incremental cost-effectiveness ratio (ICER) in the probabilistic sensitivity analysis, following internationally accepted health economic evaluation practices [15].

A willingness-to-pay (WTP) threshold of one times China’s per capita gross domestic product in 2024 ($13,581.42 per QALY) was applied, in accordance with the Chinese Guidelines for Pharmacoeconomic Evaluations (2020).

Sensitivity analysis

One-way sensitivity analysis

Model robustness was assessed by varying parameters individually over prespecified plausible ranges. Transition probabilities were varied by ± 10%, cost parameters by ± 20%, utility values by ± 15%, and hazard ratio for mortality by ± 20% of their base-case values. The annual discount rate was varied from 0 to 8% based on commonly used assumptions in pharmacoeconomic evaluations [11]. The study covered four cost parameters, six utility parameters, ten clinical outcome parameters, the mortality hazard ratio associated with RDI status, and the discount rate, with their impacts on the iNMB summarized in a tornado diagram. The parameter-specific ranges and their justifications are summarized in Supplementary Table S1.

Probabilistic sensitivity analysis

A Monte Carlo simulation with 5,000 iterations was conducted to propagate uncertainty across all parameters simultaneously. Results are presented as a scatter plot on the cost-effectiveness plane (ICER distribution) and a cost-effectiveness acceptability curve (CEAC).

Together, these analyses comprehensively characterize parameter uncertainty and its potential influence on the pharmacoeconomic conclusions.

Result

Study population and baseline characteristics

After applying predefined inclusion and exclusion criteria, a total of 695 patient records were included in the analysis. Following propensity score matching, 430 patients were retained in the matched cohort, including 215 patients in the PEG-rhG-CSF group and 215 patients in the rhG-CSF group. Baseline characteristics of the matched cohort are summarized in Table 2, demonstrating good balance between the two treatment groups.

Table 2.

Baseline characteristics of the PEG-rhG-CSF group and the rhG-CSF group

Variable	PEG-rhG-CSF (215)	rhG-CSF (215)	SMDa
Basic information
Gender	Male: 117 (54.4%) Female: 98 (45.6%)	Male: 117 (54.4%) Female: 98 (45.6%)	< 0.001
Age	61.46 ± 13.42 years	61.25 ± 11.77 years	0.017
Comorbidities
Diabetes	19 (8.8%)	16 (7.4%)	0.051
Liver disease	23 (10.7%)	19 (8.8%)	0.063
Infection	7 (3.3%)	4 (1.9%)	0.088
HIVb	1 (0.5%)	1 (0.5%)	< 0.001
COPDc	4 (1.9%)	5 (2.3%)	0.032
Laboratory tests
White blood cells	6.69 ± 2.51 × 10⁹/L	6.83 ± 2.71 × 10⁹/L	0.051
Neutrophils	4.23 ± 1.49 × 10⁹/L	4.34 ± 1.93 × 10⁹/L	0.067
Platelets	219.40 ± 81.41 × 10⁹/L	223.84 ± 79.76 × 10⁹/L	0.055
Hemoglobin	125.44 ± 18.21 g/L	127.42 ± 20.04 g/L	0.101

^aSMD: standardized mean difference; ^bHIV: human immunodeficiency virus; ^cCOPD: chronic obstructive pulmonary disease

In the PEG-rhG-CSF group, 215 patients were included, of whom 130 received Jinyouli (manufactured by CSPC Pharmaceutical Group Co., Ltd.) for primary prophylaxis. Jinyouli was the most frequently used PEG-rhG-CSF product in this cohort and was therefore selected as the representative product for cost estimation. The drug specification was uniformly 3 mg/1 mL per vial, with a single-vial price of $93.62. The standard adult dose was 6 mg per cycle, administered once ≥ 24 h after chemotherapy, resulting in a per-cycle drug cost of $187.24.

In the rhG-CSF group, 215 patients were included, among whom 122 received Lishengsu (Beijing SL Pharmaceutical Co., Ltd.). Lishengsu accounted for more than half of rhG-CSF use and was therefore selected as the representative short-acting G-CSF product. The drug specification was 250 µg per vial, with a unit price of $16.93. Based on observed prescribing data, patients received an average dose of 970 µg per chemotherapy cycle. Given the packaging specification, four vials were administered per cycle in routine practice, resulting in an average drug cost of approximately $67.72 per cycle.

Model population data indicated that the average per-patient dosage per chemotherapy cycle was 588.8 mg for rituximab, 1,253 mg for cyclophosphamide, 37.5 mg for doxorubicin liposome, 2 mg for vincristine, and 500 mg for prednisone. When administered according to routine clinical practice without vial sharing or repackaging, the corresponding number of packages required per cycle was estimated to be six vials, seven vials, two vials, two vials, and 100 tablets, respectively. Detailed information is presented in Table 3.

Table 3.

Specifications and prices of G-CSF and chemotherapy drugs

Medicines	Brand name	Specification	Unit price (USD)	Single-cycle dosage (mg)	Number of medicine packages	Cycle cost (USD)
PEG-rhG-CSF	Jinyouli	3 mg/vial	93.62	6	2	187.24
rhG-CSF	Lishengsu	250 ug/vial	16.93	0.97	4	67.72
Rituximab	Hanlikang	100 mg/bottle	193.79	588.8	6	1162.74
Cyclophosphamide	Endoxan	200 mg/vial	3.43	1253	7	24.01
Doxorubicin liposome	Libord	20 mg/bottle	560.56	37.5	2	1121.12
Vincristine	Wanle	1 mg/bottle	21.73	2	2	43.46
Prednisone	Pengyao	5 mg * 100 tablets/bottle	1.06	500	1	1.06

Drug prices were obtained from the Jiangsu Provincial Drug (Medical Consumables) Sunshine Procurement and Comprehensive Supervision Platform

Based on the referenced real-world cost study, the estimated direct medical cost per SN event was $3,546.10, while the corresponding cost per FN event was $7,092.20. These costs primarily reflected increased hospitalization duration, antibiotic treatment, laboratory testing, and imaging examinations associated with neutropenia-related complications.

Model input parameters

Tables 4, 5 and 6 summarize the transition probabilities and utility values used as inputs in the short-term and long-term Markov models.

Table 4.

Transition probabilities of the short-term Markov model

	PEG-rhG-CSF group	rhG-CSF group	Note
p-c-sn	0.88	0.13	Real-world data
p-c-fn	0.0001	0.0001	As above
p-c-c	0.9119	0.8699	As above
p-sn-fn	0.023	0.023	As above
p-sn-rdi	0.36	0.36	Cited from literature [16]
p-sn-dea	0.0005	0.0005	Cited from literature [17]
p-fn-rdi	0.45	0.45	Cited from literature [16]
p-fn-dea	0.089	0.089	Cited from literature [18]
p-rdi-dea	0.001	0.001	Cited from literature [17]
p-sn-c	0.6165	0.6165	Calculated based on existing data
p-fn-c	0.461	0.461	As above
p-rdi-c	0.999	0.999	As above

Table 5.

Transition probabilities of long-term Markov model

Age	p-rdi-dea		p-nonrdi-dea
	PEG-rhG-CSF group	rhG-CSF group	PEG-rhG-CSF group	rhG-CSF group
55 ~ 64	9.51%	9.40%	4.80%	4.74%
65 ~ 74	11.46%	11.33%	5.84%	5.77%
≥ 75	19.81%	19.60%	10.60%	10.47%

Table 6.

Utility values for each state in the model [19, 20]

State	Utility value
Prevent CIN after chemotherapy	0.61
SN	0.42
FN	0.33
RDI	0.5
≤ 1 year after completion of chemotherapy	0.79
> 1 year after completion of chemotherapy	0.89

Results of Markov model analysis

A Markov model simulation was performed with a hypothetical cohort of 1,000 patients during chemotherapy. The short-term simulation results indicated that, following treatment, the survival rate was 99.86% in the PEG-rhG-CSF group and 99.81% in the rhG-CSF group. Long-term survival beyond the observed follow-up period was extrapolated using model-based assumptions. Under the base-case analysis, the projected 1-year survival rate was 92.72% in the PEG-rhG-CSF group and 92.73% in the rhG-CSF group, declining to 23.92% and 24.12% at 20 years, respectively.

Base-case analysis results

As shown in Table 7, the average treatment cost in the PEG-rhG-CSF group was $17,947.88, yielding a total of 7.74 QALYs, whereas the rhG-CSF group incurred an average cost of $17,299.41 and achieved 7.75 QALYs. Under the base-case assumptions of the model, PEG-rhG-CSF was associated with higher costs and marginally lower health benefits and was therefore considered dominated by rhG-CSF. The incremental net monetary benefit (iNMB) was –$739.46 at a willingness-to-pay threshold of one time China’s per capita GDP for 2024, indicating that rhG-CSF was the preferred strategy in the base-case analysis.

Table 7.

Base-case analysis results

Treatment Plan	Cost, C(USD)	Utility, QALYs	∆C	∆QALY	iNMB (USD)
rhG-CSF	17299.41	7.75
PEG-rhG-CSF	17947.88	7.74	648.47	-0.01	–739.55

iNMB calculated at a WTP threshold of one times China’s per capita GDP in 2024 ($13,581.42 per QALY)

However, given the very small absolute differences in both costs and QALYs between the two strategies, their long-term cost-effectiveness was largely comparable. In this context, patient preference—particularly with respect to dosing frequency (single administration per cycle versus repeated injections)—as well as other practical considerations may play an important role in clinical decision-making.

Sensitivity analysis results

One-way sensitivity analysis

The one-way sensitivity analysis results are presented in the Figure 3. The model was most sensitive to the direct medical cost associated with SN and the cost of PEG-rhG-CSF. Variations in these two parameters produced the largest changes in iNMB.

Fig. 3.

One-way sensitivity analysis chart

However, across all tested parameter ranges, the iNMB remained consistently below zero, indicating that the base-case conclusion was robust.

Probabilistic sensitivity analysis

Using Monte Carlo simulation with 5,000 iterations for the probabilistic sensitivity analysis, a scatter plot illustrating the distribution of ICER values for PEG-rhG-CSF was generated (Figure 4). The majority of the scatter points are clustered in the north-west (NW) Quadrant, where incremental costs are positive and incremental utilities are negative, indicating that in most simulations, PEG-rhG-CSF is more costly and slightly less effective than rhG-CSF, thereby showing an economic disadvantage. Moreover, a considerable proportion of data points were located above the thresholds corresponding to one and three times China’s 2024 per capita GDP, suggesting that rhG-CSF may provide greater cost-effectiveness compared with PEG-rhG-CSF. The relatively wide dispersion of incremental costs reflects the cumulative nature of chemotherapy-related and FN management costs over multiple cycles, whereas incremental QALYs remain small and concentrated around zero.

Fig. 4.

Scatter plot of probability sensitivity distribution for PEG-rhG-CSF

The cost-effectiveness acceptability curve (Figure 5) indicates that when the WTP threshold is set at one time China’s per capita GDP for 2024, rhG-CSF has a 63.5% probability of being cost-effective. When the threshold increases to three times the per capita GDP, this probability rises to 67.1%. Across all WTP thresholds, rhG-CSF consistently shows a higher probability of cost-effectiveness compared with PEG-rhG-CSF.

Fig. 5.

Cost-effectiveness acceptability curves

Discussion

PEG-rhG-CSF is produced by covalently attaching polyethylene glycol (PEG) chains (molecular weight 20 × 10³) to the rhG-CSF protein through terminal cross-linking and purification. Several studies have compared the efficacy and safety profiles of PEG-rhG-CSF and rhG-CSF [21–23],

Compared with rhG-CSF, PEG-rhG-CSF demonstrates a longer duration of action and a more sustained pharmacological effect; however, this advantage does not necessarily translate into improved long-term survival outcomes. Direct economic evaluations comparing PEG-rhG-CSF with rhG-CSF remain limited. Evidence from breast cancer populations suggests that although PEG-rhG-CSF is associated with higher upfront costs, it may achieve superior cost-effectiveness due to additional clinical benefits [24, 25]. Similar findings have also been reported in pediatric patients receiving intensive chemotherapy, where PEG-rhG-CSF was shown to be more cost-effective than rhG-CSF [26].

In contrast to these studies, our real-world analysis observed slightly lower QALYs in the PEG-rhG-CSF group compared with the rhG-CSF group. This discrepancy is likely attributable to differences in treatment allocation under routine clinical practice. In China, reimbursement policies restrict PEG-rhG-CSF use to patients with prior severe chemotherapy-related neutropenia or other predefined clinical prerequisites, resulting in a systematically higher baseline risk profile among patients receiving PEG-rhG-CSF. Consequently, effectiveness estimates derived from real-world data may differ from those reported in randomized trials or economic evaluations conducted in broader or less selected populations.

This study found that rhG-CSF was associated with slightly more favorable long-term economic outcomes than PEG-rhG-CSF in patients with NHL treated with R-CHOP, although the magnitude of the difference was small.

Specifically, the base-case analysis showed that rhG-CSF resulted in lower overall costs (a reduction of $648.47) and a marginally higher health benefit (an incremental gain of 0.0064 QALYs) compared with PEG-rhG-CSF. While these results indicate a numerical economic advantage for rhG-CSF, the limited size of the difference suggests that the interpretation of cost-effectiveness results should be made with caution. In clinical practice, additional considerations—such as patient preference, treatment convenience, and adherence associated with single versus repeated injections—may meaningfully influence treatment choice and should be taken into account alongside economic outcomes. The univariate sensitivity analysis examined the effects of four cost parameters, six utility parameters, ten clinical outcome parameters, the mortality hazard ratio associated with RDI status, and the discount rate on iNMB. The results indicated that the model results were generally robust to plausible variation in key parameters. The model outcomes were most sensitive to the direct medical cost associated with SN and the cost of PEG-rhG-CSF. This is expected, as these cost inputs represent major contributors to total expenditures and directly determine the incremental cost difference between the two strategies. Importantly, variation in other clinical, utility, and cost parameters had comparatively limited impact on the economic conclusions. Although the base-case analysis adopted a 5% annual discount rate as recommended by the Chinese Pharmacoeconomic Guidelines (2020), this rate is higher than those commonly applied in international economic evaluations (typically 1.5-3%), as summarized in cross-country guideline comparisons [27]. However, the results of the one-way sensitivity analysis, in which the discount rate was varied between 0% and 8%, indicate that the main conclusions of the study are robust to alternative discounting assumptions. Probabilistic sensitivity analysis using Monte Carlo simulations and cost-effectiveness acceptability curves confirmed that rhG-CSF consistently had a higher probability of being cost-effective than PEG-rhG-CSF across different WTP thresholds, with this probability remaining above 60%. In addition, the wide spread of incremental costs observed in the probabilistic sensitivity analysis reflects the propagation of uncertainty in high-magnitude cost inputs rather than large structural differences between strategies. These results underscore the importance of accurately characterizing background treatment costs when evaluating the cost-effectiveness of supportive care interventions.

Several limitations should be acknowledged. First, some transition probabilities and utility values were derived from international literature and may not fully reflect the clinical characteristics of Chinese patients with NHL. Second, adverse event-related costs were not included due to limited data availability and the generally comparable safety profiles of PEG-rhG-CSF and rhG-CSF; this may have led to a slight underestimation of total costs. Third, despite the use of propensity score matching, residual confounding and selection bias cannot be fully excluded in this real-world observational study. Patients perceived to be at higher risk of neutropenia or with poorer expected outcomes may have been preferentially prescribed PEG-rhG-CSF, partly influenced by reimbursement-related practices. In addition, long-term outcomes were driven primarily by RDI-based assumptions rather than neutropenia events alone. The mortality hazard ratio distinguishing RDI ≤ 90% from RDI > 90% was derived from limited published evidence and applied as a constant over a 20-year horizon, representing a simplifying assumption. Although sensitivity analyses suggested that the results were robust to plausible variations in this parameter, long-term survival estimates should be interpreted with caution and regarded as supportive inputs for economic modeling rather than definitive clinical predictions.

Using established pharmacoeconomic methods, this study developed an integrated model incorporating real-world clinical outcomes, cost data, and health utility parameters. Under the assumptions of the base-case analysis, rhG-CSF was more likely to be cost-effective than PEG-rhG-CSF for the primary prophylaxis of CIN among patients with NHL treated with R-CHOP. However, given the small absolute differences in costs and health outcomes and the model-based nature of long-term projections, these findings should be interpreted in conjunction with clinical considerations and patient preferences, particularly regarding dosing convenience and treatment adherence.

Conclusions

Based on real-world data, rhG-CSF appears to be more cost-effective than PEG-rhG-CSF for the primary prophylaxis of CIN among patients with NHL treated with R-CHOP in China, under the assumptions of the present model. Overall, these findings provide supportive economic evidence for the use of rhG-CSF in this setting. Further real-world studies with longer follow-up and broader patient populations are warranted to confirm these results.

Author contributions

All authors made substantial contributions to the conception, design, analysis, and interpretation of the paper. All authors read and approved the final manuscript.

Funding

This study was supported by the Clinical Comprehensive Drugs Evaluation Program of Jiangsu Province (grant No. KY20231613), the Changzhou Science and Technology Program (grant No. CM20223005), and the Changzhou Health Top Talent Program (grant No. KY201770).

Data availability

The data supporting the findings of this study were obtained from the Jiangsu Provincial Public Health Information Platform, a regional multicenter medical database containing confidential patient information. Therefore, the datasets are not publicly available due to privacy restrictions. De-identified data may be available from the corresponding author upon reasonable request and with permission from the data provider.

Declarations

Ethics approval and consent to participate

The study protocol was reviewed and approved by the Ethics Committee of Changzhou First People’s Hospital (Approval No. 2024121). All research procedures were conducted in accordance with the ethical standards of the institutional ethics committee and complied with the principles of the Declaration of Helsinki. The study was conducted using de-identified data from the Jiangsu Provincial Public Health Information Platform. Therefore, the requirement for informed consent was waived by the Ethics Committee of Changzhou First People’s Hospital, as the study involved retrospective analysis of anonymized data.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.