Cost-effectiveness of prophylactic granulocyte colony-stimulating factor for febrile neutropenia in breast cancer patients receiving FEC-D

Abstract

5-fluorouracil, epirubicin, cyclophosphamide → docetaxel (FEC-D) has been associated with higher-than-expected rates of febrile neutropenia (FN) that meet the current guideline threshold of 20 % for primary prophylaxis (PP) with granulocyte colony-stimulating factor (G-CSF). We examined the cost-effectiveness of FEC-D with varying strategies of G-CSF prophylaxis from the perspective of the public payer in Ontario, Canada. A state-transition model was developed to compare three strategies: FEC-D with secondary prophylaxis (SP) only, PP starting with the first cycle of D, and PP starting with the first cycle of FEC. Analysis was conducted for a hypothetical cohort of 50-year-old early-stage breast cancer patients undergoing adjuvant chemotherapy, at a 10-year horizon. Results were expressed in quality-adjusted life-years (QALYs) and 2013 Canadian dollars. Costs and benefits were discounted at 5 %. Event rates, costs, and utilities were derived from the literature. One-way and probabilistic sensitivity analyses were conducted. Using filgrastim, the incremental cost-effectiveness ratios (ICERs) for starting PP with the first cycle of D and starting PP with the first cycle of FEC, compared to using SP only, were $57,886/QALY and $116,186/QALY, respectively. With pegfilgrastim, the ICERs for the same strategies were $90,735/QALY and $149,483/QALY. Compared to using filgrastim SP only, starting PP with D had a 24 % chance of being cost-effective at a willingness-to-pay (WTP) threshold of $50,000/QALY, and a 99 % chance at a WTP threshold of $100,000/QALY. Results were sensitive to FN-related parameters, such as the risk of FN per cycle with D and the associated mortality, but were robust to uncertainty in parameters related to breast cancer, such as the utilities and hazard of relapse. FEC-D with PP starting with the first cycle of D is most likely to be cost-effective, especially with increased risk of FN and mortality from FN.

Appendix: Detailed methodological section

Cohort

We developed a state-transition model of breast cancer to assess the cost-effectiveness of various strategies of prophylactic G-CSF use for FEC-D in Canada. Our baseline analysis considered a hypothetical cohort of 50-year-old female patients with early-stage breast cancer undergoing adjuvant chemotherapy. The analysis was conducted at a 10-year time horizon.

Strategies

In our baseline analysis, we considered the cost-effectiveness of three different treatment strategies:

• (1) “FEC × 3 (SP) → D × 3 (SP)”: Patients are offered three cycles of FEC-100 using secondary G-CSF prophylaxis (SP), followed by three cycles of docetaxel using SP.

• (2) “FEC × 3 (SP) → D × 3 (PP)”: Patients are offered three cycles of FEC-100 using SP, followed by three cycles of docetaxel using primary G-CSF prophylaxis (PP).

• (3) “FEC × 3 (PP) → D × 3 (PP)”: Patients are offered three cycles of FEC-100 using PP, followed by three cycles of docetaxel using PP.

Decision model

A cohort-based state-transition model was implemented using TreeAge Pro 2013 software [13]. Our model included 16 health states related to breast cancer, including chemotherapy, disease-free, local relapse, treated relapse, and distant relapse. The model was constructed in two stages—the first module (chemotherapy module) was adapted from Chan et al. [62] and modeled the FN-related events occurring within the six cycles of adjuvant FEC-D. The second module (breast cancer natural history module) was adapted from Younis et al. [11] and modeled the breast cancer-related events occurring over a 10-year time horizon.

In our simulations, patients moved between predefined health states in monthly cycles for 10 years. At the time of treatment, a patient first enters the chemotherapy module, and goes through six cycles of the predefined treatment strategies. During this module, the patient may (1) die from non-FN/infectious causes; (2) develop severe musculoskeletal (MSK) pain from the G-CSF; (3) develop severe neutropenia (SN); (4) develop FN; or (5) die from FN. If the patient survives the chemotherapy module, she then enters the breast cancer natural history module, and may be in any of the following health states: disease-free, local recurrence, disease-free after local recurrence, or distant relapse. Health states and allowed transitions among health states are shown in Fig. 1.

Model probabilities (Tables 1 and 2)

Breast cancer disease progression parameters were adapted from previously published models [2–4, 11]. Risk ratio of FN with G-CSF, relative risk reduction of SN with G-CSF, and relative risk reduction of death from FN with G-CSF were collected from the literature [14–17]. Probability of severe MSK pain from G-CSF was obtained from a systematic review of CSF use in lymphoma [18]. All-cause mortality rates were obtained from Statistics Canada [19].

Probabilities related to FN, such as the probability of FN for D or FEC per cycle and the probability of death from FN were obtained using data by linking administrative databases, including the Ontario Cancer Registry, Ontario Health Insurance Plan, New Drug Funding Program, Canadian Institutes of Health Information Discharge Abstract Database, CIHI National Ambulatory Care Reporting System, and Registered Persons Database, as per the method developed by Torres et al. [20]. FN was identified using the ICD codes “D70″ for agranulocytosis and “R50″ for fever in patients 18 years old and older who were diagnosed with stage I–III breast cancer between 2003 and 2009 and commenced FEC-D chemotherapy after surgery. It should be noted that the probabilities obtained from these databases, which are summarized in Table 2, are reflective of the standard of practice in Ontario during that period, which involves G-CSF SP. Per-cycle rate of severe neutropenia (SN) was calculated using the odds ratio of SN to FN (4.315) [4] based on the per-cycle rate of FN. We used the per-cycle rate of FN as found in the first and fourth cycles in the databases, and assumed that all the FEC cycles would have the same rate as the first cycle (4.5 %) and that the D cycles would have the same rate as the fourth cycle (11.23 %) if SP with G-CSF was not available.

Direct medical costs and utilities (Table 3)

Costs of G-CSF were obtained from the Ontario Drug Benefit Formulary with additional standard pharmacy mark-up costs and dispensing fees, for a 300 mcg vial and assuming a patient weight of 60 kg. The cost of administration of G-CSF by home care was obtained from the Community Care Access Centre in Ontario [21], under the assumption that home care would teach the patient to self-inject, thus, it was a one-time cost at the initiation of G-CSF. The cost of hospitalization with FN was obtained from a Canadian study [22]. The professional fees of physicians managing FN during hospitalization were obtained from the Ontario Ministry of Health and Long-term Care (MOHLTC) Schedule of Benefits [23]. Costs of breast cancer-related health states were extracted from the literature [11, 24, 25].

Utility data of breast cancer health states, hospitalization, with FN, and MSK pain were obtained from the literature [26–37]. Table 2 summarizes the costs and utilities used in the model.

Economic assumptions

This analysis was conducted from a public payer perspective in Ontario, which is the most populous province in Canada, and was structured as a cost-utility analysis, with outcomes expressed in quality-adjusted life-years (QALYs) and costs in 2013 Canadian dollars. Future costs and health benefits were discounted at 5 % annually [38]. Non-Canadian cost data were converted to Canadian dollars at the purchasing power parity conversion rate [63]. All cost data were inflated to 2013 dollars using the Statistics Canada Consumer Price Index for health care and personal items [64].

Analytic strategy

We first conducted the base-case analysis (the state-transition model) to estimate the expected value using deterministic calculations with filgrastim as the G-CSF treatment. We then repeated the base-case analysis with pegfilgrastim. A full deterministic one-way sensitivity analysis was then conducted on all model parameters over the plausible ranges using the reported 95 % confidence interval (CI) ranges (Tables 1 and 2). Finally, probabilistic sensitivity analyses (PSA) were conducted using Monte Carlo simulation for 5,000 iterations for all three treatment strategies.