Beruflich Dokumente
Kultur Dokumente
30
NUMBER
19
JULY
2012
O R I G I N A L
R E P O R T
Purpose
Our study aimed to develop a model to predict distant recurrence in locally advanced cervical
cancer, which can be used to select high-risk patients in enriched clinical trials.
Patients and Methods
Our study was a retrospective analysis of a multi-institutional cohort of patients treated between 2001 and 2009. According to the order of data submission, data from three institutions
were allocated to a model development cohort (n 434), and data from the remaining two
institutions were allocated to an external validation cohort (n 115). Patient information including
[18F]fluorodeoxyglucose positron emission tomography (FDG-PET) data and clinical outcome was
modeled using competing risk regression analysis to predict 5-year cumulative incidence of
distant recurrence.
Results
The competing risk analysis revealed that the following four parameters were significantly
associated with distant recurrence: pelvic and para-aortic nodal positivity on FDG-PET, nonsquamous cell histology, and pretreatment serum squamous cell carcinoma antigen levels. This
four-parameter model showed good discrimination and calibration, with a bootstrap-adjusted
concordance index of 0.70. Also, the validation set showed good discrimination with a bootstrapadjusted concordance index of 0.73. A user-friendly Web-based nomogram predicting 5-year
probability of distant recurrence was developed.
Conclusion
We have developed a robust model to predict the risk of distant recurrence in patients with locally advanced cervical cancer. Further, we discussed how the selective enrichment of the
patient population could facilitate clinical trials of systemic chemotherapy in locally advanced
cervical cancer.
J Clin Oncol 30:2369-2374. 2012 by American Society of Clinical Oncology
DOI: 10.1200/JCO.2011.37.5923
INTRODUCTION
thors showed that gemcitabine plus cisplatin chemoradiotherapy followed by brachytherapy and
adjuvant gemcitabine/cisplatin chemotherapy improved survival outcomes. The possible mechanism
of this benefit was explained by a 50% reduction in
distant failure, as the rates of local recurrence were
not significantly different in both arms. This explanation corresponds to meta-analyses indicating that
the addition of systemic chemotherapy has a benefit
in reducing the risk of distant failure.6,9 However,
this survival gain was not free from considerable
costs. Not only were more grade 3 and 4 toxicities
observed in the study arm but there were also two
treatment-related deaths. Thus, this trade-off raised
the following questions. Can we identify patients
2012 by American Society of Clinical Oncology
2369
Kang et al
Characteristic
Age, years
Mean
Range
Stage
Bulky IB2-IIA
IIB
III
IVA
Histology
Squamous cell
Adeno
Adenosquamous
Unknown
Tumor size, measured by MRI
Median
Range
Unknown
Serum SCC Ag level, ng/mL
Median
Range
Unknown
Pelvic node status by PET
Negative
Positive
Para-aortic node status by PET
Negative
Positive
Extended-field radiation
Not done
Done
Unknown
Recurrence
Locoregional only
Distant only
Locoregional plus Distant
Death
No. of
Patients
Validation
Cohort
(n 115)
No. of
Patients
P
.68
54
27-78
54
32-77
72
274
70
18
16.6
63.1
16.1
4.2
8
71
26
10
7.0
62.8
22.6
8.7
.009
385
29
17
3
88.7
6.7
3.9
0.7
106
4
2
3
92.2
3.5
1.7
2.6
.005
5.5
1.5-9.5
10
8.7
.001
4.5
1.6-9.0
16
3.7
4.9
0-395
5
1.2
.001
8.3
0-402
0
174
260
40.1
59.9
56
59
48.7
51.3
.096
346
88
79.7
20.3
90
25
78.3
21.7
.73
283
150
1
65.2
34.6
0.2
74
39
2
64.4
33.9
1.7
.15
29
66
26
74
6.7
15.2
6.0
17.1
7
26
6
21
6.1
22.6
5.2
18.3
.31
Patient Cohort
Five institutions participated in our retrospective study. The method of
patient allocation was predetermined before the analysis. According to the
order of data submission, the first three institutions were allocated to a development cohort and the final two institutions were allocated to a validation
cohort. Between November 2001 and August 2009, 748 patients from three
institutions (Asan Medical Center, Seoul, Korea; the National Cancer Center,
Goyang, Korea; and Ehwa Womans University Hospital, Seoul, Korea) underwent definitive platinum-based chemoradiotherapy for the primary treatment of locally advanced cervical cancer. Among these patients, 434 patients
(58%) underwent FDG-PET before chemoradiotherapy and were included in
the development cohort. During the same period, 167 patients underwent
chemoradiotherapy at two institutions (Seoul National University Hospital,
Seoul, Korea; and Korea Cancer Center, Seoul, Korea). Among these patients, 115
patients (69%) who underwent FDG-PET were allocated to an external validation
cohort. The inclusion criteria were as follows: a histologic diagnosis of primary
carcinoma of uterine cervix; International Federation of Gynecology and Obstetrics stage IIB to IVA or IB2 to IIA bulky (greater than 4 cm) disease; treatment with
curative chemoradiotherapy, with or without extended-field radiation; and FDGPET or positron emission tomography/computed tomography scan (PET/CT)
beforethestartofchemoradiotherapy,butnomorethan4weeksbefore.Exclusion
criteria were as follows: small-cell or neuroendocrine type histology; concomitant primary cancer; nonplatinum-based chemoradiotherapy; radiation dose
less than 40 Gy; or application of surgery or chemotherapy before chemoradiotherapy. Application of intracavitary brachytherapy, parametrial boost,
Modeling set
Validation set
1.00
0.75
0.50
0.25
12
24
36
60
Modeling set
Validation set
.3
.2
.1
0
.76
48
Time (months)
B
.4
2370
Cumulative Incidence of
Distant Recurrence
Model
Derivation
Cohort
(n 434)
Distant RecurrenceFree
Survival (probability)
20
40
60
Time (months)
Fig 1. (A) Kaplan-Meier estimate of distant-recurrence free survival probability of
the model development and the validation cohorts. (B) Cumulative incidence of
distant-recurrence estimated by competing risk analysis.
JOURNAL OF CLINICAL ONCOLOGY
according to the Fine and Gray method, was performed using Cox regression.15 The first observed distant metastasis was considered an event. Both
locoregional recurrence and death before recurrence were regarded as competing risks. To handle missing data, the complete-case method was applied.
Although many predictor-selection strategies exist, the strategy we chose
to use was as follows. We began developing a risk model by fitting the Cox
proportional hazards model, using all predictors that had P value less than .25
in single predictor analysis. A reduced model was also created from parameters
showing a P value of less than .25 in the full model. Only variables showing
significant P values ( .05) in the reduced model were regarded as viable
predictors. The significance of all predictors was internally validated using the
jackknife method. The possible interactions between variables were tested
within the variables selected for the final model.
The discrimination ability of the model was evaluated by Harrells concordance index.16-17 The discrimination ability of the final model was internally
validated using estimation of bootstrap-adjusted concordance index with 200
bootstrap resamples. Calibration of the model was assessed by plotting the observed incidence rate of distant recurrence estimated by the Kaplan-Meier method
and the predicted probability of distant recurrence in three risk groups partitioned
according to the distribution of the predicted risk. In addition, a nomogram was
constructed, based on the Cox proportional hazards model, and was converted
into a user-friendly Web-based program. All statistical analyses were performed using the STATA computer program, version 11.0 (STATA, College
Station, TX; Computing Resource Center, Santa Monica, CA).
and extended-field radiation to the para-aortic area were not considered selection criteria. All clinical information was investigated after obtaining the approval of the institutional review board of each of the participating institutions.
Clinical characteristics of modeling and validation cohorts are summarized in
Table 1 and Appendix Table 1 (online-only).
Pretreatment Assessment
Tumor size was measured by magnetic resonance imaging. Computed tomography(CT)-basedorclinicallymeasuredtumorsizewasnotconsideredinthe
analysis. If tumor size was reported in three axes, the largest diameter was included
as the tumor size. The instrumental setting and the condition of the process of PET
scanning are summarized according to the participating institutions in Appendix
Table 2 (online-only). The commonly applied interpretation criteria were the
rules proposed by the International Harmonization Project in Lymphoma.13
Briefly, a lymph node was considered positive if its FDG uptake was greater
than that of the blood pool activity or background tissues. The steering committee made the decision not to perform a central review of the data because of
the heterogeneous settings of the PET imaging, as well as because of concerns
regarding the generalization of the model. Thus, all pathologic and imaging
data were interpreted locally by specialists from each institution.
Patient Follow-Up
All patients were followed up every 2 to 4 months for the first 2 years and
then subsequently every 2 to 6 months, which varied according to the policy of
each institution and individual patient risk. Pelvic examination and cytologic
tests were performed at every visit. Diagnostic imaging (CT or magnetic
resonance imaging) was performed every 6 or 12 months and when clinically
indicated. The end point was the time from the start of chemoradiotherapy to
the earliest diagnosis of distant recurrence. Distant recurrence was defined as
tumor recurrence at a site beyond the pelvic radiation field. Para-aortic nodal
recurrence above the L4-L5 interspace was regarded as a distant recurrence.
RESULTS
Statistical Analysis
Survival distributions were estimated by the Kaplan-Meier method, with
reverse meaning of the status indicator.14 Competing risk regression analysis,
Table 2. Competing Risk Analysis of Time to Distant Recurrence in Locally Advanced Cervical Cancer
Multiple Predictor Analysis
Single Predictor Analysis
Characteristic
Age, years (continuous)
40
70
Stage
Bulky IB-IIA
IIB
III
IVA
III-IVA v I-II
Histology
Squamous cell
Adeno
Adenosquamous
Squamous v adeno plus adenosquamous
Tumor size by MRI, cm (continuous)
4 cm v 4 cm
Serum SCC Ag, ng/mL
SUVmax cervix by PET
Pelvic node by PET
Para-aortic node by PET
Extended-field radiation
Full Model
Reduced Model
SHR
95% CI
SHR
95% CI
1.00
1.51
1.00
0.98 to 1.02
0.85 to 2.68
0.47 to 2.11
.76
.16
.99
1.50
0.71 to 3.13
.29
Reference
1.48
1.69
1.19
1.15
0.78 to 2.78
0.79 to 3.60
0.33 to 4.21
0.70 to 1.90
.23
.18
.79
.57
1.72
1.38
0.64
0.76 to 3.93
0.47 to 4.03
0.13 to 3.08
.20
.56
.56
Reference
1.82
1.53
1.71
1.19
1.87
1.01
1.02
2.54
2.98
1.78
0.94 to 3.53
0.60 to 3.90
0.97 to 3.00
1.06 to 1.33
1.15 to 3.04
1.00 to 1.01
0.99 to 1.05
1.56 to 4.13
1.95 to 4.56
1.17 to 2.69
.37
.062
.003
.012
.001
.21
.001
.001
.007
2.82
1.15
1.48 to 5.34
1.00 to 1.33
.002
.052
1.01
0.99
2.04
1.98
1.26
1.00 to 1.01
0.95 to 1.03
1.10 to 3.80
1.06 to 3.72
0.78 to 2.20
.005
.64
.024
.032
.34
SHR
95% CI
Jack-knifed P
2.30
1.08
1.31
1.00
1.28 to 4.14
0.94 to 1.23
0.78 to 2.22
1.00 to 1.01
.006
.276
.305
.026
.008
1.95
2.36
1.15 to 3.33
1.49 to 3.72
.014
.001
.017
.001
.024
Abbreviations: MRI, magnetic resonance imaging; PET, positron emission tomography; SCC Ag, squamous cell carcinoma antigen; SHR, subdistribution hazard
ratio; SUV, standardized uptake value.
www.jco.org
2371
Kang et al
Model Validation
Internal validation of predictive accuracy was performed using
200 times of bootstrap resampling. In the modeling cohort, the
bootstrap-adjusted corrected concordance index was 0.70 (95% CI,
0.64 to 0.75), which exceeded that of the International Federal of
Gynecology and Obstetrics stage (0.57; 95% CI, 0.52 to 0.62). In the
external validation set (n 113), the model showed a good discrimination performance (concordance index, 0.73) and the bootstrapadjusted concordance index was 0.73 (95% CI, 0.65 to 0.81). A
calibration curve for the validation set is illustrated in Figure 2B.
Finally, a nomogram was constructed from the coefficients of
this model. For easier use, a computerized version of the nomogram was created. This Web-based software tool is available for use
at the Korean Gynecologic Oncology Group Web site (Appendix
Fig 2, online-only).
Utility of Our Prediction Model
Figure 3A shows how the model could be used to design a clinical
trial to test the efficacy of systemic chemotherapy to reduce the risk of
distant failure after chemoradiotherapy. We assumed a 50% risk reduction by systemic chemotherapy on the basis of a recent report. If a
trial included all patients without any selection criteria, then 86% of
patients would receive unnecessary treatment and suffer from severe
complications. However, a scenario can be suggested based on our risk
model. If we enrolled cases with an estimated risk of more than 25%,
then the percentage of unnecessarily treated patients and the required
size of the study population would both be reduced by 60%, whereas
the proportion of benefited patients among enrolled patients would
increase by 43%.
DISCUSSION
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
B
Observed 5-Year Probability
of Distant Recurrence
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
advanced cervical cancer who underwent definitive chemoradiotherapy. Furthermore, we built a Web-based nomogram for easier access.
We believe these tools can help select candidates for trials designed to
evaluate the efficacy of systemic therapy. The relevance of our predictors corresponds well with previous reports.10-12,18-21
Survival outcomes in locally advanced cervical cancer have
been improved with the inclusion of systemic chemotherapy as a
multimodal treatment.6,22 Obviously, one of the main mechanisms
of improved outcome is the reduction of distant recurrence.6,9 A
recent randomized trial successfully indicated that aggressive systemic therapy may improve outcomes by reducing the rate of
distant failure. It is not difficult to expect that currently ongoing
international trials, such as the ANZGOG (Australian New Zealand Gynaelogical Oncology Group) -0902 study, will be stimulated by this success, and more trials will be launched to test the
efficacy of systemic treatment in locally advanced cervical cancer.23
However, not all patients with locally advanced disease bear the
same risk of distant recurrence. It has been argued that not all
patients with locally advanced cervical cancer are candidates for
systemic chemotherapy.11,24-25 Moreover, although the investigators in a recent randomized trial claimed that increased toxicity
JOURNAL OF CLINICAL ONCOLOGY
Cumulative Incidence of
Distant Recurrence (%)
100
90
80
70
60
50
40
30
20
10
40
%
35
>
>
>
30
%
25
>
20
%
>
15
%
>
10
>
5%
Tr
ea
ta
ll
c
>
om
er
s
500
No. of Patients
450
400
350
300
250
200
150
100
50
>
40
%
35
>
%
30
>
>
25
%
20
>
%
15
>
%
10
>
>
Tr
ea
ta
ll
co
er
5%
Fig 3. Graph showing how the prediction model could be used to design a
clinical trial testing the efficacy of systemic chemotherapy in locally advanced
cervical cancer. (A) Using the risk of subsequent distant recurrence as a cutoff
(x-axis), patients at high risk can be selected and enrolled onto the clinical trial. As
the threshold for enrollment increases, the number of patients enrolled (gray line)
and unnecessarily treated (red dash) would be dramatically decreased, whereas
the proportion of benefited patients among the enrolled study population would
be increased. Note that the reduced risk of distant failure also gradually
decreases (blue dash). A 50% risk reduction by systemic chemotherapy was
assumed. (B) Graph showing the required sample size (blue dash) and the size of
the population that needs to be screened to obtain the required sample size (red
line) at the error of .05 and the power of 90% (for each arm).
was manageable, the physical and economical burden of the systemic chemotherapy might be unnecessary in the patients with a
low risk of distant failure.
Thus, with the current prediction model, we suggested the risk
prediction model for distant failure, and this model may be useful
in candidate selection for clinical trials. The risk-based enrollment
of candidates is not a new concept and has been used successfully in
other trials.26 Our illustrations show how our prediction models
can be applied in clinical trials (Fig 3A). In our scenario, we could
reduce unnecessary treatment by 60% using a probability of 25% as
a cutoff for study enrollment. Moreover, this selective enrollment
strategy could also reduce the number of patients to be enrolled by
60%. In addition, we can estimate the changes in the required
sample size and in the population size to be screened (Fig 3B). If we
www.jco.org
AUTHOR CONTRIBUTIONS
Conception and design: Sokbom Kang
Financial support: Sokbom Kang
Administrative support: Sokbom Kang
Provision of study materials or patients: Sokbom Kang, Sang-Soo Seo,
Sang-Young Ryu, Jae Weon Kim, Seung-Cheol Kim, Sang-Yoon Park,
Joo-Hyun Nam
Collection and assembly of data: Sokbom Kang, Jeong-Yeol Park,
Sang-Soo Seo, Sang-Young Ryu, Jae Weon Kim, Seung-Cheol Kim,
Sang-Yoon Park, Joo-Hyun Nam
Data analysis and interpretation: Sokbom Kang, Byung-Ho Nam
Manuscript writing: All authors
Final approval of manuscript: All authors
2012 by American Society of Clinical Oncology
2373
Kang et al
REFERENCES
1. Keys HM, Bundy BN, Stehman FB, et al:
Cisplatin, radiation, and adjuvant hysterectomy compared with radiation and adjuvant hysterectomy for
bulky stage IB cervical carcinoma. N Engl J Med
340:1154-1161, 1999
2. Morris M, Eifel PJ, Lu J, et al: Pelvic radiation
with concurrent chemotherapy compared with pelvic and para-aortic radiation for high-risk cervical
cancer. N Engl J Med 340:1137-1143, 1999
3. Peters WA III, Liu PY, Barrett RJ II, et al:
Concurrent chemotherapy and pelvic radiation therapy compared with pelvic radiation therapy alone as
adjuvant therapy after radical surgery in high-risk
early-stage cancer of the cervix. J Clin Oncol 18:
1606-1613, 2000
4. Rose PG, Bundy BN, Watkins EB, et al: Concurrent cisplatin-based radiotherapy and chemotherapy for locally advanced cervical cancer. N Engl
J Med 340:1144-1153, 1999
5. Whitney CW, Sause W, Bundy BN, et al:
Randomized comparison of fluorouracil plus cisplatin versus hydroxyurea as an adjunct to radiation
therapy in stage IIB-IVA carcinoma of the cervix with
negative para-aortic lymph nodes: A Gynecologic
Oncology Group and Southwest Oncology Group
study. J Clin Oncol 17:1339-1348, 1999
6. Collaboration Chemoradiotherapy for Cervical
Cancer Meta-Analysis Collaboration: Reducing uncertainties about the effects of chemoradiotherapy
for cervical cancer: A systematic review and metaanalysis of individual patient data from 18 randomized trials. J Clin Oncol 26:5802-5812, 2008
7. Pearcey R, Miao Q, Kong W, et al: Impact of
adoption of chemoradiotherapy on the outcome of
cervical cancer in Ontario: Results of a populationbased cohort study. J Clin Oncol 25:2383-2388,
2007
8. Duenas-Gonzalez A, Zarba JJ, Patel F, et al:
Phase III, open-label, randomized study comparing
concurrent gemcitabine plus cisplatin and radiation
followed by adjuvant gemcitabine and cisplatin versus concurrent cisplatin and radiation in patients
with stage IIB to IVA carcinoma of the cervix. J Clin
Oncol 29:1678-1685, 2011
9. Green JA, Kirwan JM, Tierney JF, et al: Survival and recurrence after concomitant chemotherapy and radiotherapy for cancer of the uterine cervix:
A systematic review and meta-analysis. Lancet 358:
781-786, 2001
10. Kang S, Park JY, Lim MC, et al: Pelvic lymph
node status assessed by 18F-fluorodeoxyglucose
positron emission tomography predicts low-risk
group for distant recurrence in locally advanced
cervical cancer: A prospective study. Int J Radiat
Oncol Biol Phys 79:788-793, 2011
2374
Copyright of Journal of Clinical Oncology is the property of American Society of Clinical Oncology and its
content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for individual use.