Annals of Oncology 14: 248–257, 2003

Original article DOI: 10.1093/annonc/mdg073

Economic evaluation of antibiotic prophylaxis in small-cell lung

cancer patients receiving chemotherapy: an EORTC double-blind
placebo-controlled phase III study (08923)
V. C. G. Tjan-Heijnen1*, S. Caleo2, P. E. Postmus3, A. Ardizzoni4, J. T. M. Burghouts5, E. Buccholz6,
B. Biesma5, T. Gorlia2, R. Crott2, G. Giaccone3, C. Debruyne2 & C. Manegold6
On behalf of the European Organisation for Research Treatment of Cancer (EORTC)—Lung Cancer
Group and Health Economics Unit
1
University Medical Centre Nijmegen, Nijmegen, The Netherlands; 2EORTC Data Center, Brussels, Belgium; 3Vrije Universiteit Medical Center, Amsterdam,
The Netherlands; 4Istituto Nazionale per la Ricerca sul Cancro, Genova, Italy; 5Bosch Medicentrum Groot Ziekengasthuis, ’s-Hertogenbosch, The Netherlands;
6
Thoraxklinik Rohrbach, Heidelberg, Germany

Received 24 June 2002; revised 16 September 2002; accepted 22 October 2002

Background: To determine whether the cost of prophylactic antibiotics during chemotherapy is offset by
cost savings due to a decreased incidence of febrile leukopenia (FL).
Patients and methods: Small-cell lung cancer (SCLC) patients were randomised to standard or intensified
chemotherapy with granulocyte colony-stimulating factor to assess the impact on survival (n = 244). In
addition, patients were randomised to prophylactic ciprofloxacin and roxithromycin or placebo to assess the
impact on FL (n = 161). The economic evaluation examined the costs and effects of patients taking antibiotics
versus placebo. Medical resource utilisation was documented prospectively, including 33 patients from one
centre in The Netherlands (NL) and 49 patients from one centre in Germany (GE). The evaluation takes the
perspective of the health insurance systems and of the hospitals. Sensitivity analyses were performed.
Results: In the main trial, prophylactic antibiotics reduced the incidence of FL, hospitalisation due to FL and
use of therapeutic antibiotics by 50%. In GE, the incidence of FL was not reduced by prophylaxis. This
resulted in an average cost difference of only 35 Euros [95% confidence interval (CI) (–)1.713–2.263] in
favour of prophylaxis (not significant). In NL, prophylaxis reduced the incidence of FL by nearly 50%, com-
parable with the results of the main trial, resulting in a cost difference of 2706 Euros [95% CI 810–5948],
demonstrating savings in favour of prophylactic antibiotics of nearly 45%. Sensitivity analyses indicate that
with an efficacy of prophylaxis of 50%, and with expected costs of antibiotic prophylaxis of 500 Euros or less,
cost savings will incur over a broad range of baseline risks for FL; that is, a risk >10–20% for FL per cycle.
Conclusions: Giving oral prophylactic antibiotics to SCLC patients undergoing chemotherapy is the
dominant strategy in both GE and NL, demonstrating both cost-savings and superior efficacy. The sensitivity
analyses demonstrate that, due to the efficacy of prophylactic antibiotics and their low unit cost, cost savings
will incur over a broad range of baseline risks for FL. We recommend the use of prophylactic antibiotics in
patients at risk for FL during chemotherapy.
Key words: antibiotic prophylaxis, chemotherapy, economic evaluation, small-cell lung cancer

Introduction mainstay of treatment for FL is hospitalisation and intravenous

(i.v.) antibiotics. The administration of oral prophylactic anti-
The most frequent dose-limiting toxicity of chemotherapy is
biotics may have clinical benefits for patients by reducing the
febrile leukopenia (FL). Cyclophosphamide, doxorubicin and
incidence of FL and at the same time produce cost savings.
etoposide (CDE), one of the standard chemotherapy regimens in
Economic evaluation is a useful tool to help determine whether it
the treatment of small-cell lung cancer (SCLC), is associated
is worthwhile to give prophylactic antibiotics in terms of cost,
with an incidence of FL varying from 43% to 77% [1–3]. The
and benefits to patients and to the health care system.
We have previously reported the results of a randomised trial in
patients with SCLC treated with CDE chemotherapy in which the
*Correspondence to: Dr V. C. G.Tjan-Heijnen, University Medical Centre role of prophylactic antibiotics was evaluated [1]. We demon-
Nijmegen, Department of Medical Oncology (550), PO Box 9101, 6500 HB
Nijmegen, The Netherlands. Tel: +31-24-3615215; Fax: +31-24-3540788; strated that with prophylactic antibiotics the incidence of FL,
E-mail: [email protected] number of documented infections, use of therapeutic antibiotics

© 2003 European Society for Medical Oncology

 249

and hospitalisations were reduced by approximately 50%, with a In addition, we looked at the incidence of FL per cycle of delivered chemo-
reduced number of infectious deaths. We report here the results therapy.
of the economic evaluation of this prospectively randomised
trial. Importantly, we also performed sensitivity analyses to place Uncertainty analysis. The skewness of cost data means that producing con-
fidence intervals by parametric methods is inappropriate. Therefore, the cost
the outcome in a broader perspective.
data were analysed using the non-parametric bootstrap, specifically employ-
ing the bias corrected and accelerated bootstrapping method [5]. The number
Patients and methods of bootstrap replications for each sample was 5000. These calculations
yielded an average cost per patient and cycle for both arms with their
A more detailed report of the clinical trial has been reported elsewhere [1, 4]. corresponding 95% confidence interval (CI). The difference in average cost
In brief, patients with chemo-naïve SCLC with a European Cooperative between the antibiotic and placebo arms for both sites was also calculated in
Oncology Group (ECOG) performance status of 0 or 1 were randomised to this manner.
standard-dose CDE (cyclophosphamide 1000 mg/m2, day 1; doxorubicin
45 mg/m2, day 1; etoposide 100 mg/m2, days 1–3, i.v., q 3 weeks, five times) Resource utilisation. Resource use included those items that were associated
or to intensified CDE chemotherapy (125% dose, q 2 weeks, four times; with with direct medical treatment costs and did not include patient out-of-pocket
filgrastim 5 µg/kg/day, days 4–13) to assess the impact on survival (n = 244). costs, non-medical costs, indirect costs or quality-of-life issues.
Patients were also randomised to prophylactic antibiotics (ciprofloxacin
750 mg plus roxithromycin 150 mg, b.i.d., days 4–13) or to placebo in a 2 × 2 Unit cost data. It was assumed that patients were treated as public patients.
factorial design (n = 161) with as primary end point the incidence of FL Unit costs were applied from the health insurance and hospital perspectives
during the first cycle. In cases of FL, prophylaxis was interrupted and (Table 1). Where it was not possible to obtain the pertinent hospital prices for
replaced by i.v. broad-spectrum antibiotics. After inclusion of 161 eligible some items, such as pharmaceuticals, which are subject to negotiated prices
patients, the antibiotic/placebo part of the trial was prematurely terminated on and considered commercially sensitive information, the listed tariff prices
advice from an independent data monitoring committee. Randomisation was had to be used. The costs were expressed in Euros (1 Euro = 1.96 DM = 2.20 f
done using the minimisation technique stratifying patients according to their = 0.86 US $ as of 28 June 2001).
institution, age (>60 versus ≤60 years) and stage of disease [limited disease
(LD) versus extensive disease].
Hospitalisations. The per diem rates in both countries are hospital specific
and based on their annual budgets. Hospital budgets are influenced by their
Economic evaluation relative size, whether it is a university hospital and also whether the hospital
Prospective economic evaluation was conducted alongside the trial. This is specialised in certain diagnostic areas (e.g. thoracic clinic). In Heidelberg
evaluation concerned only the randomisation of prophylactic antibiotics versus (GE), the rate for 1 day (no overnight stay) and also per diem (including over-
placebo. All prices were adjusted to 1998 prices. No discounting of costs was night stay) was 280 Euros, while the per diem rate of intensive care was 765
necessary because treatments were given over a period of 4–6 months. Euros. The figures in ’s-Hertogenbosch (NL) were 170, 625 and 1416 Euros,
Although the clinical trial was conducted in 13 centres throughout Europe, respectively. Hospitalisation for chemotherapy administration was not taken
Heidelberg in Germany (GE) and ’s-Hertogenbosch in The Netherlands (NL) into account, as this was considered to be comparable for both arms.
were both expected to recruit the highest number of patients and thus were
chosen for the cost assessments (n = 82, 51% of total). Pharmaceuticals. The unit costs of the prophylactic antibiotics were applied
The objective of the analysis was to determine whether the costs of prophy- as if the patient had bought the full box of antibiotics from their local
lactic antibiotics were offset by cost savings associated with the expected pharmacy. A course of prophylactic antibiotics costs 180 Euros in GE and
decrease in incidence of FL, fever, documented infections, days of i.v. anti- 110 Euros in NL. National public tariff prices were applied to all medications.
biotics and days of hospitalisation. Protocol-driven costs were not included in
the analysis. Transfusions. In GE, the cost of transfusion with either one unit of red blood
The average clinical effect and costs were determined for each arm. The cells or one unit of platelets (six donors) was 50 Euros, while in NL this was
economic evaluation was primarily based on the incidence of FL per patient. 86 and 273 Euros, respectively.

Table 1. Sources of unit costs

Heidelberg (GE) ’s-Hertogenbosch (NL)

From perspective of health care system
Hospital per diem cost Reimbursed amount from health insurancesa Reimbursed amount from health insurancesa
From perspective of hospital
Pharmaceuticals Rote List, Service Gmbh, 1998 Koninklijke Nederlandse Maatschappij ter bevordering der
Pharmacie (KNMP), 1998
Transfusions Local blood bank prices paid by the hospital Local blood bank prices paid by the hospital
Diagnostic tests Tarif der Deutschen Krankenhausgesellschaft Centraal Orgaan Tarieven Gezondheidszorg (COTG), 1998
(DKG-NT), 1997; Band 1b
Antimicrobial cultures Hospital’s own records Hospital’s own records

a
Centre specific.
b
These 1997 prices were adjusted by an additional 3% to convert them to 1998 prices.
250

Diagnostic tests. For GE, we used the tariffs from ‘DKG-NT Band 1, Tarif The baseline assumptions for the threshold and sensitivity analyses were
der Deutschen Krankenhausgesellschaft’. For NL, we used the listed based on the incidence of FL, the efficacy of prophylaxis (relative reduction
‘Centraal Orgaan Tarieven Gezondheidszorg’ (COTG) tariffs, with an in FL) and the duration of hospitalisation due to FL as seen in the main trial,
additional physician’s fee when these tests were carried out in the outpatient and not those of only one or two subgroups (GE and/or NL), as such sub-
setting. groups may not be representative of the whole patient population [6].
Thresholds were calculated for both GE and NL separately, as the unit
Microbial cultures. In GE, the cost of a single culture depends upon the number costs were not the same for both countries. The costs used were those that
of tests that need to be carried out (ranging from 20 Euros to 42 Euros). In NL, determined the outcome from a health insurance perspective. Cost variations
the costs of the tests depend upon whether the sample test was positive in cultures and diagnostic tests were not included for NL, as these accounted
(52 Euros) or negative (31 Euros), regardless of the source of the culture. for <5% of total costs.

Three-way sensitivity analysis. A three-way sensitivity analysis was con-

Sensitivity analyses ducted by varying the average baseline risk for hospitalisation due to FL per
Threshold and three-way sensitivity analyses were performed. cycle, the average cost of a hospitalisation due to FL and the cost of antibiotic
prophylaxis per cycle. The baseline risk for hospitalisation due to FL was
Threshold analysis. The following formula was used to calculate cost neutrality varied from 10% to 50%, by step increments of 10%. The average cost of
[6–8]: hospitalisation due to FL was varied from 2000 to 20 000 Euros, with step
increments of 2000 Euros. The cost of prophylaxis per cycle was varied from
(FL risk × relative reduction) × (unit cost × days hospitalisation) = (unit cost × 100, 250, 500 to 1000 Euros.
days prophylaxis) The hospitalisation cost for which there is cost neutrality can be calculated
On the left-hand side of the equation, the cost savings of prophylactic anti- per cost of prophylaxis for one baseline risk for hospitalisation due to FL
biotics are determined by taking ‘the avoided number of FL episodes’ times (threshold point), but also over a range of baseline risks for FL (threshold
‘the average cost of hospitalisation per FL episode’. On the right-hand side of line). Several threshold lines can be calculated for different unit costs of
the equation, the costs of prophylaxis are calculated. There is cost neutrality prophylaxis. Any combination in the area above a given threshold line
in cases where the cost savings are equal to the cost of prophylaxis itself. In a favours prophylaxis on a cost basis.
threshold analysis, all parameters but one are kept unchanged. The threshold Such three-way sensitivity analysis enables extrapolation of results to
of this parameter is the value at which point there is cost neutrality. Prophy- different countries and to different chemotherapy regimens with different
lactic antibiotics will produce cost savings, in cases where the parameters on risks for FL.
the left-hand side of the equation do in reality have values above their thresh-
old value, or in cases where the values on the right-hand side of the equation
have values below their threshold value. Results
The numbers in the threshold and sensitivity analyses refer to costs and Efficacy
incidences per cycle, in contrast to the efficacy and economic evaluation in
which primarily incidences per patient were used as was specified in the The baseline patient characteristics were comparable for both
prospective trial. sites, except for more LD in patients from GE (Table 2).

Table 2. Patient characteristics at baseline

Characteristics Heidelberg (GE) ’s-Hertogenbosch (NL) P value

[n = 49 (%)] [n = 33 (%)]
Median age, years (range) 59 (35–69) 57 (33–70) 0.238a
Sex
Male 37 (76) 26 (79) 0.795b
Female 12 (24) 7 (21)
ECOG PS
0 25 (51) 16 (48) 0.174b
1 24 (49) 17 (52)
Stage
Limited 35 (71) 16 (48) 0.021b
Extensive 14 (29) 17 (52)
Median weight loss over last 3 months (%) 2.5 4.8 0.244a
Range –7 to 12c –8 to 22c

a
Wilcoxon two-sample test.
b
Fisher’s exact test.
c
Negative weight loss means weight gain.
ECOG, European Cooperative Oncology Group.
 251

Table 3. Clinical outcome comparisons for the whole trial minus Heidelberg (All-GE) and Heidelberg (GE)

Heidelberg (GE) Placebo Antibiotics

All-GE GE P value All-GE GE P value
(n = 53) (n = 26) (n = 59) (n = 23)
Incidence of FL in first cyclea 16 4 0.180b 7 2 1.000b
Incidence of FL in subsequent cyclesc 17 3 0.054b 8 4 0.490b
b
Overall incidence of FL 27 7 0.055 15 5 1.000b
Hospitalisation due to FL 23 8 0.333b 12 5 1.000b
b
Incidence of fever 36 13 0.144 26 13 0.336b
d b
i.v. antibiotics 32 11 0.154 22 9 1.000b
No. of FL episodes
0 26 19 44 18
1 19 7 14 3
2 4 0 1 2
3 2 0 0 0
4 2 0 0.025e 0 0 0.869e
f
No. of hospitalisations
0 18 11 33 5
1 19 8 16 13
2 8 6 8 4
3 6 1 1 0
e
4 2 0 0.410 1 1 0.023e

a
For one patient, incidence of febrile leukopenia (FL) is missing at cycle 1.
b
Fisher’s exact test.
c
Incidence of FL is missing for 10 patients; not including eight patients, who only received the first cycle of CDE chemotherapy.
d
Not only for FL.
e
Wilcoxon two-sample test.
f
For FL, fever, transfusion and for treatment of adverse events, but not for chemotherapy; missing data for one patient.

In almost all instances, each of the subgroups, when compared of these patients (in the placebo arm) spent 30 days in intensive
separately to the main trial results, had a similar rate of outcome, care at a cost of 22 959 Euros.
although there were some differences, especially in GE (Tables 3 The cost of treating an episode of FL in either GE or NL with i.v.
and 4). antibiotics was about the same on a daily basis (44 and 45 Euros,
In the main clinical trial (n = 161), 48 episodes of FL in total respectively), although different antibiotics were used. Genta-
occurred in the placebo arm (15% of 320 chemotherapy cycles) micin and cefotiam were often used to treat FL in Heidelberg. In
versus 23 in the antibiotics arm (7% of 335 cycles). This reflects contrast, at ’s-Hertogenbosch, cefuroxime and tobramycin were
a 55% relative decrease in risk for FL (taking all cycles into most often given.
account). In NL, the average incidence of FL per cycle was 16% Relatively more patients underwent cultures and diagnostic
versus 5% in the antibiotics arm, which is comparable with the tests in the placebo and antibiotics arms in NL compared with the
main trial. However, in GE the outcome was quite different with placebo and antibiotics arms in GE (Table 6).
incidences of 8% and 9%. From a health insurance perspective, there was an average cost
For GE, there was a prolonged duration of hospitalisation due saving of 35 Euros [95% CI (–)1713–2263] per patient in favour
to FL and longer i.v. antibiotic treatment in the prophylactic arm of giving prophylactic antibiotics in GE (not significant) (Table 7).
(Table 5). In NL, the average duration of i.v. antibiotics in the The cost difference in NL was 2706 Euros (95% CI 810–5948)
placebo arm was higher than in the main trial. per patient, demonstrating savings in favour of prophylactic anti-
Finally, the average number of chemotherapy cycles for patients biotics of nearly 45% (Table 7). This correlated with cost savings
treated in GE (regardless of treatment arm) was 3.71 [standard devi- in NL of 605 Euros per cycle.
ation (SD) = 1.32], whilst in NL it was 4.33 (SD = 0.69); P <0.01.
From a hospital perspective, there is a similar overall picture
with again a cost reduction of approximately 45% per patient in
Costs NL (Table 7), and cost neutrality for GE.
In NL, there were no patients who required intensive care, In GE, the average cost of treating an episode of FL was higher
whereas two patients in GE had spent time in intensive care. One in patients in the antibiotics arm compared with the placebo arm
252

Table 4. Clinical outcome comparisons for the whole trial minus ’s-Hertogenbosch (All-NL) and ’s-Hertogenbosch (NL)

’s-Hertogenbosch (NL) Placebo Antibiotics

All-NL NL P value All-NL NL P value
(n = 63) (n = 16) (n = 65) (n = 17)
Incidence of FL in first cyclea 17 3 0.749b 7 2 1.000b
c b
Incidence of FL in subsequent cycles 14 6 0.341 10 2 1.000b
Overall incidence of FL 28 6 0.779b 16 4 1.000b
b
Hospitalisation due to FL 23 8 0.333 12 5 1.000b
Incidence of fever 37 12 0.265b 28 11 0.172b
d b
i.v. antibiotics 34 9 1.000 21 10 0.054b
Nο. of FL episodes
0 35 10 49 13
1 24 2 13 4
2 1 3 3 0
3 1 1 0 0
e
4 2 0 0.962 0 0 0.861e
No. of hospitalisationsf
0 24 5 32 6
1 23 4 23 6
2 11 3 7 5
3 4 3 1 0
e
4 1 1 0.215 2 0 0.236e

a
For one patient, incidence of febrile leukopenia (FL) is missing at cycle 1.
b
Fisher’s exact test.
c
Incidence of FL is missing for 10 patients; not including eight patients who only received the first cycle of CDE chemotherapy.
d
Not only for FL.
e
Wilcoxon two-sample test.
f
For FL, fever, transfusion and for treatment of adverse events, but not for chemotherapy; missing data for one patient.

Table 5. Comparison of duration of FL and of the main resource utilisations for the whole trial minus Heidelberg (All-GE) and
Heidelberg (GE) and for the whole trial minus ’s-Hertogenbosch (All-NL) and ’s-Hertogenbosch (NL)

Heidelberg (GE) Duration (days, average ± SD)

Placebo Antibiotics
b
All-GE GE P value All-GE GE P valueb
(n = 53) (n = 26) (n = 59) (n = 23)
i.v. antibiotics a
16 ± 10.4 13.1 ± 4.3 0.521 10.5 ± 6.5 17.4 ± 10.2 0.036
Hospitalisation due to FL 5 ± 6.7 6.4 ± 5.0 0.178 4.6 ± 2.3 9.8 ± 5.1 0.014
Hospitalisation c
15.1 ± 14.1 12.3 ± 11.2 0.345 11.5 ± 10.0 11.2 ± 11.3 0.565
’s-Hertogenbosch (NL) Duration (days, average ± SD)
Placebo Antibiotics
All-NL NL P valueb All-NL NL P valueb
(n = 63) (n = 16) (n = 65) (n = 17)
i.v. antibioticsa 13.5 ± 7.4 21.9 ± 12.8 0.021 14.3 ± 9.6 9.1 ± 2.7 0.144
Hospitalisation due to FL 5.6 ± 6.8 4.5 ± 3.5 0.919 6.8 ± 4.3 3.8 ± 1.5 0.123
Hospitalisationc 13.4 ± 13.1 17.3 ± 14.0 0.201 10.8 ± 10.2 13.1 ± 11.5 0.271

a
Not only for febrile leukopenia (FL).
b
Kruskal–Wallis test.
c
For FL, fever, transfusions, treatment of adverse events, but not for chemotherapy.
SD, standard deviation.
 253

Table 6. Total resource use in Heidelberg (GE) and ’s-Hertogenbosch (NL)

Resource Heidelberg (GE) ’s-Hertogenbosch (NL)

Placebo Antibiotics Placebo Antibiotics
(n =26) (n =23) (n =16) (n =17)
Total no.a No. of patients Total no.a No. of patients Total no.a No. of patients Total no.a No. of patients
RBC transfusions 32 9 46 12 21 8 18 6
PLT transfusions 14 3 37b 4 3 3 7 5
Chest X-rays 28c 4 3 3 15 6 3 3
Abd. ultrasounds 0 0 2 2 1 1 0 0
Blood cultures 15 10 12 8 34 8 22 9
Urine cultures 7 4 4 2 28 9 11 8
Faeces cultures 0 0 0 0 6 4 3 2

a
Number of transfusions or investigations.
b
One patient had 16 platelet (PLT) transfusions.
c
One patient had 24 X-rays.
RBC, red blood cell; abd., abdominal.

Table 7. Average costs (Euro) of all patients included for Heidelberg (GE) and ’s-Hertogenbosch (NL), from both a health insurance
perspective and the hospital perspective

Average costs (Euros)

Placebo (95% CI)a Antibiotics (95% CI)a Difference (95% CI)a
Heidelberg (GE): health insurance perspective, n = 49
Hospital 2405 (1112–5114) 1750 (1037–2665) 655 [(–)990–3100]
Antibiotics – 619 (517–713) –619 [(–)517–(–)713]
Total 2405 (1112–5114) 2369 (1550–3340) 35 [(–)1713–2263]

’s-Hertogenbosch (NL): health insurance perspective, n = 33

Hospital 5672 (2763–9101) 2669 (1659–3723) 3003 (8–6579)
Testsb 272 (122–458) 95 (49–145) 177 (24–374)
Antibiotics – 474 (428–500) –474 [(–)428–(–)500]
Total 5944 (3359–9655) 3238 (2195–4311) 2706 (810–5948)

Heidelberg (GE): hospital perspective, n = 49

Pharmaceuticals 224 (121–420) 352 (126–624) –128 [(–)444–155]
b
Transfusions/tests 240 (85–616) 190 (105–277) 50 [(–)137–418]
Total 465 (227–1008) 542 (284–867) –77 [(–)514–457]

’s-Hertogenbosch (NL): hospital perspective, n = 33

Pharmaceuticals 394 (209–671) 134 (80–192) 260 (64–534)
Transfusions 165 (88–232) 204 (77–319) –39 [(–)188–101]
Testsb 272 (122–458) 95 (49–145) 177 (24–374)
Total 831 (410–1259) 432 (252–631) 399 [(–)44–878]

a
Bootstrapped 95% confidence interval (CI).
b
Cultures and diagnostic tests.
Note: 1 Euro = 2.20f = 1.96 DM.

(Table 8), largely due an unexpected longer duration of FL and Cost-effectiveness

hospitalisation due to FL in the antibiotics arm (Table 5). In con-
trast, in NL the cost difference incurred per patient who experi- For the first cycle in GE, prophylactic antibiotics reduced the abso-
enced FL was 37% [1 – (2485/3970)] in favour of the antibiotics lute risk of FL by 6% (down from 15%). It also saved 274 Euros
arm (Table 8). per patient. In NL, prophylactic antibiotics reduced the absolute
254

Table 8. Cost of treating FL for Heidelberg (GE) and ’s-Hertogenbosch (NL), health insurance perspective

Average cost ± SD (Euros)

Heidelberg (GE) ’s-Hertogenbosch (NL)
Placebo Antibiotics Placebo Antibiotics
Patients who experienced FL 1789 ± 1407 (n = 8) 3367 ± 1775 (n = 5) 3970 ± 2452 (n = 6) 2485 ± 964 (n = 4)
All patients 551 ± 1124 (n = 26) 732 ± 1609 (n = 23) 1489 ± 2438 (n = 16) 585 ± 1164 (n = 17)

Note: 1 Euro = 2.20f = 1.96 DM.

FL, febrile leukopenia; SD, standard deviation.

Table 9. Threshold analysis for The Netherlands (NL) and Germany (GE) based on rates from main clinical trial

Parameter Base case Thresholds Base case Thresholds

GEa GEa NLb NLb
Probability FL
Baseline risk hospitalisation 0.15 0.20 0.15 0.06
Relative reduction in FL 0.53 0.71 0.53 0.20
Hospitalisation due to FL
Cost/day (Euro) 280 377 625 231
Average duration (days per cycle) 6 8 6 2
Prophylaxis
Cost/day (Euros) 18 13 11 30
Duration (days per cycle) 10 7 10 27

a
Based on incidence/duration FL per cycle seen in the main clinical trial (in bold) (n = 161), while costs are based
on prices seen in Heidelberg (Euros).
b
Based on incidence/duration FL per cycle seen in the main clinical trial (in bold) (n = 161), while costs are based
on prices seen in ’s-Hertogenbosch (Euros).
Formula cost-neutrality: (FL baseline risk × relative reduction) × (unit cost × days hosp.) = (unit cost × days
prophylaxis).

risk of FL by 7% (down from 19%) in the first cycle. It also saved In addition, a three-way sensitivity analysis was conducted
676 Euros per patient. (Figure 1). Any combination in the area above a given threshold
In subsequent cycles in GE, prophylactic antibiotics did not line favours prophylaxis on a cost basis. For example, with an
reduce the risk of FL. With effectiveness the same, a cost mini- average hospitalisation cost for FL of 4000 Euros, prophylaxis at
misation analysis yields a saving of 28 Euros per patient. In NL, the cost of 100 Euros would be cost saving for all baseline risks,
prophylactic antibiotics reduced the absolute risk of FL by 21% whereas with prophylaxis at a cost of 1000 Euros cost savings
(down from 38%). It also saved 892 Euros per patient. occur only at a baseline risk for FL of >47%.
In both cases treatment with prophylactic antibiotics was the
dominant strategy as it results, at the same time, in a risk reduc-
tion and lower costs. Discussion
To our knowledge, this is the first economic evaluation of the cost
Threshold and sensitivity analysis
savings of prophylactic antibiotics. This prospective evaluation
In the sensitivity and threshold analyses the numbers refer to was performed at two centres, in GE and NL, which accrued
incidences per cycle. together 51% of the patients of a randomised trial evaluating
In GE, prophylaxis produces cost savings in the following situ- prophylactic antibiotics during CDE chemotherapy in SCLC.
ations: (i) a baseline risk for FL of >20% per cycle; (ii) a relative Determination of resource utilisation and the associated unit
reduction in risk of FL of >71%; (iii) costs of hospitalisation due costs required site visits; thus, it was not feasible both from a
to FL of >377 Euros per day; (iv) a hospital stay >8 days; (v) a practical and funding perspective to visit all the centres included
reduced cost of prophylaxis due to a unit cost price of <13 Euros in the trial. It was demonstrated that with prophylactic antibiotics
per day or (vi) a shorter need for prophylaxis of <7 days (Table 9). the incidence of FL, number of documented infections, use of
The figures for NL are 6%, 20%, 231 Euros, 2 days, 30 Euros and therapeutic antibiotics and hospitalisations due to FL were
27 days, respectively (Table 9), with better threshold levels. decreased by approximately 50%, along with a reduced number
 255

threshold values are hereby underestimated, which is also

stressed by others [9].
Recently, it was reported that (out-patient) treatment of FL by
oral broad-spectrum antibiotics may be feasible [10, 11]. How-
ever, for patients already receiving prophylactic antibiotics the
optimal choice of antibiotic treatment in the case of FL has not
yet been clarified. Therefore, we did not implement this variable
in our sensitivity analyses.
From a health insurance perspective, cost savings largely
depend on the costs of hospitalisation versus those of prophy-
laxis. In the threshold analysis it was demonstrated that for the
two countries, with the same baseline assumptions, the thresholds
reflecting cost neutrality are quite different. In GE, the unit cost
price of prophylaxis was higher, while the unit cost price of hos-
Figure 1. Three-way sensitivity analysis based on rates of the main clinical
pitalisation was lower compared with NL, making prophylaxis
trial. The effectiveness of prophylaxis in reducing FL is assumed to be 53%
for NL more cost saving.
per cycle. Threshold lines are displayed for different costs of prophylaxis
(100–1000 Euros). Three-way sensitivity analysis clearly demonstrates that for
antibiotic prophylaxis costs of ≤500 Euros, cost savings will
incur over a broad range of baseline risks for FL, i.e. a risk above
of infectious deaths (6% versus 0%) [1]. In GE, the use of 10–20% for FL per cycle. Costs of antibiotic prophylaxis and
prophylactic antibiotics was cost neutral, while in NL prophy- hospitalisations will fall in this range in most countries.
laxis incurred cost savings of 2706 Euros (95% CI 810–5948) per Importantly, these sensitivity analyses indicate that the strat-
patient, i.e. 605 Euros per cycle, demonstrating savings in favour egy of antibiotic prophylaxis may be useful, both from a clinical
of prophylactic antibiotics of nearly 45% of average costs in the and a cost–benefit viewpoint, to other regimens and other tumour
placebo arm. types with an increased risk for FL.
The different outcome according to centre indicates that dif- Prophylactic G-CSF is an alternative way of preventing FL or
ferences in clinical parameters (more LD in GE), change due to febrile neutropenia (FN). The efficacy of G-CSF and antibiotic
the relatively small sample size per centre, patient management prophylaxis appears to be comparable [1–3, 12, 13], although for
and financing mechanisms of health care systems influence cost G-CSF a reduction in infectious mortality has never been re-
results. In NL, the clinical outcome was similar to that of the main ported [2, 3]. The American Society of Clinical Oncology (ASCO)
trial. However, in GE the incidence of FL was more or less the recommended that primary administration of CSFs should be
same for both treatment arms. Consequently, no difference in reserved for patients with an expected incidence of FN ≥40%
costs was observed in GE. [14]. In our trial the overall incidence of FL in the placebo arm
For straightforward economic evaluations, it is not uncommon was not that high (25% in the first cycle and 15% over all cycles).
to evaluate only one or two subgroups of patients because it is This lower incidence is in line with what was reported by others
not feasible from a practical viewpoint to evaluate all patients, [15, 16], implying that for the majority of (SCLC) patients treated
especially not in a multicentre international trial. Of note, the
by standard-dose chemotherapy primary G-CSF prophylaxis
most important reason for performing an economic evaluation is
may not be indicated.
to obtain the costs of relevant variables (in this case, of prophy-
In fact, this 40% threshold remains somewhat confusing, as it
laxis and of treating FL). These cost data are essential in order
was not based on clinical grounds [17]. This threshold was based
to perform sensitivity analyses, of which the outcome is more
on a sensitivity analysis, which used as its parameters a 50%
important than the outcome of the economic evaluation, per-
formed in a subgroup of patients. To estimate the variations in efficacy of G-CSF in preventing FN in the first cycle, an average
costs between various countries, one should preferably consider cost of $10 000 (11 628 Euros) per hospitalisation for FN and an
the costs of more than one subgroup. For that reason, we evalu- average cost of $2000 (2326 Euros) per cycle for G-CSF prophy-
ated the costs in two different countries. laxis [6].
It should be noted that the clinical parameters of the main trial This threshold of 40% only holds true for the cost prices used
were used for the sensitivity analyses. The numbers from the main above. However, the actual cost prices may vary enormously.
trial were considered to be more accurate and representative of The Canadian Coordinating Office for Health Technology
the real efficacy of prophylactic antibiotics. The same method Assessment estimated the costs of treating an episode of FN at
was used for the pivotal ‘granulocyte colony-stimulating factor US $6000 (6980 Euros) [18]. With this lower hospitalisation
(G-CSF) versus placebo’ trial [2], with an economic evaluation in price, the use of G-CSF would be cost neutral when the risk of FN
a subgroup of patients [7], but with (frequently quoted) sensitiv- is 50.9%. Other studies reported hospitalisation costs of $4000
ity analyses based on the results of the main clinical trial [6, 7]. (3440 Euros) [16] and $7464 (8680 Euros) for 6 days of hospital-
For the sake of simplicity, in the sensitivity analyses only isation due to FL [15]. In our study, we demonstrated also that
hospitalisations for FL were included [5]. We accept that the applying the cost of one country (1680 Euros for 6 days hospital-
256

isation in GE) to another (3750 Euros for NL) underestimates the Cura dei Tumori, Napoli, Italy; Spaarne Ziekenhuis, Haarlem, The Nether-
complexity of what determines such thresholds. lands; Ospedale Civile di Asti, Asti, Italy; EORTC Data Center, Brussels,
Belgium.
In the reported economic analyses of G-CSF [6–8, 13, 15, 16],
the cost of G-CSF was >1000 Euros per cycle. In general, with
cost of prophylaxis of >1000 Euros, cost savings will not occur in References
the range of baseline risks for FL as seen during standard-dose
chemotherapy (Figure 1). 1. Tjan-Heijnen VCG, Postmus PE, Ardizzoni A et al. Reduction of chemo-
therapy-induced febrile leucopenia by prophylactic use of ciprofloxacin
Another important issue is that the economic and sensitivity
and roxithromycin in small cell lung cancer patients: an EORTC double-
analyses concerning G-CSF are based on the risk of FN in the blind placebo controlled phase III study. Ann Oncol 2001; 12: 1359–1368.
first cycle only, while prophylaxis will generally be continued to 2. Crawford J, Ozer H, Stoller R et al. Reduction by granulocyte colony-
the end of chemotherapy. In SCLC, the incidence of FN/FL stimulating factor of fever and neutropenia induced by chemotherapy in
during the first cycle is in fact much higher than in subsequent patients with small-cell lung cancer. N Engl J Med 1991; 325: 164–170.
cycles [1–3]. In our trial, the risk for FL in the placebo arm was 3. Trillet-Lenoir V, Green J, Manegold C et al. Recombinant granulocyte
25% for the first cycle, but only 15% over all cycles [1]. This colony stimulating factor reduces the infectious complications of cyto-
declining risk through later cycles may be due to improved toxic chemotherapy. Eur J Cancer 1993; 29A: 319–324.
performance status, patient selection and tumour response with 4. Tjan-Heijnen VCG, Ardizzoni A, Postmus PE et al. Dose-intensification
less risk for post-obstructive pneumonia. This may indicate that of cyclophosphamide, doxorubicin and etoposide (CDE) chemotherapy
primary G-CSF or antibiotic prophylaxis in certain circum- does not improve survival in small cell lung cancer (SCLC): final results
of a randomized phase III EORTC study. Proc Am Soc Clin Oncol 2001;
stances and with certain unit cost prices may be cost-effective for
20: 346 (Abstr 1379).
the first cycle but not necessarily for subsequent cycles. One may
5. Efron B. Better bootstrap confidence intervals (with discussion). J Am
also hypothesise that for lung carcinoma with an increased risk
Stat Assoc 1987; 82: 171–200.
for pneumonia in the first cycle, the use of prophylaxis may be 6. Lyman GH, Lyman CG, Sanderson RA et al. Decision analysis of
indicated for the first cycle only, while in other tumours treated hematopoietic growth factor use in patients receiving cancer chemo-
by other chemotherapy regimens, prophylaxis may only be cost- therapy. J Natl Cancer Inst 1993; 85: 488–493.
effective during later cycles due to cumulative myelotoxicity. 7. Glaspy JA, Bleecker G, Crawford J et al. The impact of therapy with
In conclusion, this is the first economic evaluation concerning filgrastim (recombination granulocyte colony-stimulating factor) on
prophylactic antibiotics versus placebo during chemotherapy, health care costs associated with cancer chemotherapy. Eur J Cancer
prospectively conducted alongside a multi-national phase III 1993; 29 (Suppl 7): 23–30.
trial. We demonstrated that prophylactic antibiotics during 8. Uyl-de Groot CA, Vellenga E, Rutten FFH. An economic model to
assess the savings from a clinical application of haematopoietic growth
chemotherapy is the dominant strategy in both GE and NL,
factors. Eur J Cancer 1996; 32A: 57–62.
demonstrating both cost savings and superior efficacy. More-
9. Lyman GH, Kudere N, Green J et al. The economics of febrile neutro-
over, we showed that pooling of resource use in multi-centre
penia: implications for the use of the colony-stimulating factors. Eur
multinational clinical trials cannot be recommended. We demon- J Cancer 1998; 34: 1857–1864.
strated that a sensitivity analysis is a valuable tool to place results 10. Freifeld A, Marchigiani D, Walsh Th et al. A double-blind comparison of
in a broader perspective, enabling extrapolation to other institu- empirical oral and intravenous antibiotic therapy for low-risk febrile
tions or other reimbursement systems. Sensitivity analysis may patients with neutropenia during cancer chemotherapy. N Engl J Med
also prevent the misinterpretation of the results of economic 1999; 341: 305–311.
evaluations based on the results of a subgroup analysis. In this 11. Del Prete SA, Ryan SP, Jacobson JS et al. Safety and costs of treating
trial, the sensitivity analyses demonstrate that, due to the efficacy neutropenic fever in an outpatient setting. Conn Med 1999; 63: 713–716.
of prophylactic antibiotics and their low unit cost, cost savings 12. Carlson JW, Fowler JM, Saltzman AK et al. Chemoprophylaxis with oral
will incur over a broad range of baseline risks for FL—a risk ciprofloxacin in ovarian cancer patients receiving taxol. Gynecol Oncol
1994; 55: 415–420.
range relevant to routine clinical practice. Therefore, we recom-
13. Schröder CP, de Vries EGE, Mulder NH et al. Prevention of febrile
mend the use of prophylactic antibiotics in patients at risk for FL
leucopenia after chemotherapy in high-risk breast cancer patients: no
during chemotherapy.
significant difference between granulocyte-colony stimulating growth
factor or ciprofloxacin plus amphotericin B. J Antimicrob Chemother
1999; 43: 741–743.
Acknowledgements
14. American Society of Clinical Oncology. Recommendations for the use
We would like to thank Bayer B.V. for sponsoring this study. The following of hematopoietic colony-stimulating factors: evidence-based, clinical
institutions participated in the main trial: University Medical Centre Nijmegen, practice guidelines. J Clin Oncol 1994; 12: 2471–2508.
Nijmegen, The Netherlands; Istituto Nazionale per la Ricerca sul Cancro, 15. Nichols CR, Fox EP, Roth BJ et al. Incidence of neutropenic fever in
Genova, Italy; Vrije Universiteit Medical Center, Amsterdam, The Nether- patients treated with standard-dose combination chemotherapy for
lands; Thoraxklinik Rohrbach, Heidelberg, Germany; Bosch Medicentrum small-cell lung cancer and the cost impact of treatment with granulocyte
Groot Ziekengasthuis, ’s-Hertogenbosch, The Netherlands; University colony-stimulating factor. J Clin Oncol 1994; 12: 1245–1250.
Hospital Antwerp, Edegem, Belgium; Ziekenhuis St Jansdal, Harderwijk, 16. Chouiaid C, Bassinet L, Fuhrman C et al. Routine use of granulocyte
The Netherlands; Lukas Ziekenhuis, Apeldoorn, The Netherlands; Academic colony-stimulating factor is not cost-effective and does not increase
Medical Center, Amsterdam, The Netherlands; CHU de Marseille, Hopital patients comfort in the treatment of small-cell lung cancer: an analysis
Sainte-Marguerite, Marseille, France; Istituto Nazionale per lo Studio e la using a Markov model. J Clin Oncol 1998; 16: 2700–2707.
 257

17. Ozer H, Armitage JO, Bennett CL et al. 2000 update of recommendations 18. Canadian Coordinating Office for Health Technology Assessment: the
for the use of hematopoietic colony-stimulating factors: evidence-based, use of G-CSF in the prevention of febrile neutropenia. Ottawa, Canada:
clinical practice guidelines. American Society of Clinical Oncology Canadian Coordinating Office for Health Technology Assessment
Growth Factors Expert Panel. J Clin Oncol 2000; 18: 3558–3585. (CCOHTA) 1997; 1203–9012.