Accepted Manuscript Title: Prognostic value of circulating tumor cells’ reduction in patients with extensive small-cell lung cancer Author: Nicola Normanno Antonio Rossi Alessandro Morabito Simona Signoriello Simona Bevilacqua Massimo Di Maio Raffaele Costanzo Antonella De Luca Agnese Montanino Cesare Gridelli Gaetano Rocco Francesco Perrone Ciro Gallo PII: DOI: Reference:
S0169-5002(14)00211-6 http://dx.doi.org/doi:10.1016/j.lungcan.2014.05.002 LUNG 4599
To appear in:
Lung Cancer
Received date: Revised date: Accepted date:
3-8-2013 24-4-2014 1-5-2014
Please cite this article as: Normanno N, Rossi A, Morabito A, Signoriello S, Bevilacqua S, Di Maio M, Costanzo R, De Luca A, Montanino A, Gridelli C, Rocco G, Perrone F, Gallo C, Prognostic value of circulating tumor cells’ reduction in patients with extensive small-cell lung cancer, Lung Cancer (2014), http://dx.doi.org/10.1016/j.lungcan.2014.05.002 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Prognostic value of circulating tumor cells’ reduction in patients with extensive small-cell lung cancer
ip t
Nicola Normanno,1 Antonio Rossi,2* Alessandro Morabito,3* Simona Signoriello,4* Simona
Bevilacqua,1* Massimo Di Maio,5* Raffaele Costanzo,3 Antonella De Luca,1 Agnese Montanino,3
cr
Cesare Gridelli,2 Gaetano Rocco,6 Francesco Perrone 5 and Ciro Gallo 4 1
te
d
*these authors equally contributed to the paper
M
an
us
Cell Biology and Biotherapy Unit, Istituto Nazionale per lo Studio e la Cura dei Tumori “Fondazione Giovanni Pascale” - IRCCS, Napoli; 2Division of Medical Oncology, S.G. Moscati Hospital, Avellino; 3 Medical Oncology Unit, Department of Thoracic Surgical and Medical Oncology, Istituto Nazionale per lo Studio e la Cura dei Tumori “Fondazione Giovanni Pascale” - IRCCS, Napoli; 4Medical Statistics, Second University, Napoli; 5Clinical Trials Unit, Istituto Nazionale per lo Studio e la Cura dei Tumori “Fondazione Giovanni Pascale” - IRCCS, Napoli; 6Division of Thoracic Surgery, Department of Thoracic Surgical and Medical Oncology, Istituto Nazionale per lo Studio e la Cura dei Tumori “Fondazione Giovanni Pascale” - IRCCS, Napoli; Italy.
Ac ce p
Corresponding author: Nicola Normanno Cell Biology and Biotherapy Unit INT-Fondazione Pascale Via Mariano Semmola 80131 Napoli (Italy) Tel and Fax: +39/081/5903826 Email: [email protected]; [email protected] Running title: CTCs in SCLC
1 Page 1 of 29
Highlights
us
cr
ip t
We assessed circulating tumor cells (CTCs) with the CellSearch system in extensive SCLC patients The accuracy of prognostic role of different cutoffs was assessed by Harrell’s c-index CTC count at baseline or after chemotherapy cycle-1 marginally increased the prognostic accuracy Prognostic accuracy was significantly improved by CTC count reduction higher than 89% after cycle-1 CTC count reduction higher than 89% after cycle-1 was associated with longer survival
an
Abstract
Objectives: Circulating tumor cells (CTCs) have been hypothesized to be a prognostic factor in small-
M
cell lung cancer (SCLC), and different cutoffs have been proposed to identify patients at high risk. We
d
assessed the prognostic value of CTCs in patients with extensive SCLC.
te
Materials and Methods: CTCs were assessed with the CellSearch system in 60 extensive SCLC patients. CTC count at baseline or after one cycle of chemotherapy (cycle-1) or as change after
Ac ce p
chemotherapy were analyzed separately. Primary outcome was overall survival. The accuracy of prognostic role was assessed by Harrell’s c-index. “Optimal” cutoffs were derived by bootstrap resampling to reduce the overfitting bias; accuracy improvement was estimated by calculating the difference of c-indexes of models including clinical variables with or without CTCs. Results: CTCs were identified in 90% (54/60) of patients at baseline, in which CTC count ranged from 0 to 24281. CTC count was strongly associated with the number of organs involved. The prognostic accuracy was only marginally increased by the addition to clinical information of “optimal” CTC cutoffs at baseline and after cycle-1. Conversely, a reduction of CTC count higher than 89% following chemotherapy significantly improved prognostic accuracy (bootstrap p-value = 0.009) and was 2 Page 2 of 29
associated with a lower risk of death (HR 0.24, 95% CI 0.09 to 0.61). When previously proposed cutoffs were applied to our cohort, they showed only marginal improvement of the prognostic accuracy.
ip t
Conclusion: CTCs have useful prognostic role in extensive SCLC, but only the change of CTC count
cr
after the first cycle of chemotherapy provides clinically relevant information. Previously reported CTC
us
cutoffs were not prognostic in our cohort of patients.
Ac ce p
te
d
M
an
Key words: circulating tumor cells; small-cell lung cancer; prognosis; chemotherapy
3 Page 3 of 29
Introduction Small-cell lung cancer (SCLC) is characterized by rapid doubling time, propensity for early dissemination, significant sensitivity to chemotherapy and radiotherapy and development of drug
ip t
resistance during the course of disease [1]. Clinical and pathological characteristics, as extensive-stage
cr
disease, performance status (PS), gender, elevated lactate dehydrogenase (LDH) and alkaline
phosphatase (ALP) serum levels, are common prognostic markers [1, 2]. To date, no histopathological,
us
molecular or genetic prognostic tests are available.
an
Detection and characterization of circulating tumor cells (CTCs) in the peripheral blood can provide information on cancer aggressiveness, might improve prediction of patient’s outcome and help to
M
develop novel therapeutic approaches [3, 4]. CTCs can be isolated from peripheral blood by different methods [3, 4]. The CellSearch system (Veridex, Raritan, NJ) allows automated immunomagnetic
d
enrichment and staining of CTCs. The CellSearch has been approved by US Food and Drug
te
Administration for breast, colorectal and prostate cancer in which CTC count has been proposed to be
Ac ce p
prognostic [5-9]. Several studies also suggested that a reduction in CTC count following the administration of therapy might provide prognostic information [6, 7, 9, 10]. A prognostic role of CellSearch-identified CTCs in non-small-cell lung cancer has been recently suggested [11]. Our group was the first to demonstrate that CTCs can be isolated from the peripheral blood of SCLC patients by using the CellSearch system [12]. Three studies previously explored the prognostic role of CTCs in SCLC and identified three different cutoffs (50, 8 and 2, respectively) that separate patients in favorable and unfavorable groups according to CTC count [13-15]. Hou et al. [13] and Naito et al. [14] found that baseline CTC count and change in CTC number after one cycle of chemotherapy were independent prognostic factors. Hiltermann et al. [15] confirmed the value of CTC count at baseline and after one cycle of chemotherapy. Of note, all these studies included patients with either limited or 4 Page 4 of 29
extensive disease. Objective of this study was to explore the role of CTCs in SCLC as prognostic marker in patients with extensive disease only, and to externally validate the findings of the previously
Ac ce p
te
d
M
an
us
cr
ip t
reported studies.
5 Page 5 of 29
Patients and Methods Patients
ip t
Patients with chemotherapy-naive, histologically proven SCLC, extensive disease according to Veterans Administration Lung Cancer Group classification, who gave informed consent, were eligible.
cr
Blood samples were collected for analysis within 7 days before commencing treatment (baseline) and
us
after one chemotherapy cycle (cycle-1).
an
Ethical considerations
In 2008, the Scientific Board of the National Cancer Institute of Napoli approved CTC count
M
assessment in a non-interventional study, as part of the clinical workout of patients, based on our previous observation that CTCs can be detected in SCLC patients by using the CellSearch system [12]
d
and that the CellSearch Circulating Tumor Cell kit was approved in Europe for in vitro diagnostics
te
(CE-IVD). Just oral consent was sought upon information about the exploratory nature of CTC count
Ac ce p
assessment and that the result would have not affected treatment. In 2012, after the publication of the Krebs paper in NSCLC [11] we realized that information coming from our data could be interesting for scientific community and we sought the approval of the Independent Ethical Committee (IEC) regarding the opportunity to analyse and publish our data. IEC supported our request. CTC analysis
Peripheral blood samples (7.5 ml) were collected into CellSave preservative tubes and processed within 72 hours of collection. The CellSearch (Veridex LLC, Raritan, NJ) platform was used for isolation and enumeration of CTCs [12, 16]. Samples were processed in the CellTracks Autoprep system by using the Circulating Tumor Cell kit and analysed with the CellTracks Analyzer II according to the manufacturer’s instructions. 6 Page 6 of 29
Statistical analysis We addressed three main questions: do (i) CTC count at baseline or (ii) CTC after cycle-1 or (iii) CTC change after cycle-1 increase the accuracy of prognostic evaluation beyond what is expected by using
ip t
clinical information only? For each of the three questions, we had to look for the "optimal" cutoff first,
cr
and to assess the prognostic role of CTC (categorized according the "optimal" cutoff) in the
multivariable analysis, as the second step. Due to the exploratory nature of the study no formal sample
us
size definition was planned before the study.
an
Primary outcome was overall survival (OS). When assessing the value of baseline CTC, OS was defined as the time from date of baseline blood collection to death or to last follow-up visit for living
M
patients; when the value of CTC count after cycle-1 or of CTC variation was evaluated, starting time
d
for OS was the date of the second blood sample collection.
te
As first step we had to look for the "optimal" cutoff . The “optimal” cutoff was defined as the CTC value that maximized the discriminating ability, according to the Harrell’s concordance index (c-
Ac ce p
index), a modified version of the area under ROC curve for survival data [17]. All the observed CTC values were screened as potentially "optimal" cutoff, i.e. cutoff levels were not chosen 'a priori', This approach invariably leads to overfitting bias, i.e. to results tailored just for the current data; therefore, to reduce overfitting and improve internal validity we applied a bootstrap procedure [18], randomly sampling with repetition a thousand samples of size 60 from the original data set. For each bootstrap sample the CTC cutoff that maximized the c-index was found, and the distribution of the "optimal" cutoffs observed in the 1,000 bootstrap samples was derived. The most frequent "optimal" cutoff was entered in the subsequent prognostic model. Cutoffs are always intended as greater than or equal to the stated value.
7 Page 7 of 29
Since CTC values were strongly associated with the number of organs involved by the disease, we repeated the analysis after adjustment by the number of involved organs in a Cox proportional hazard regression model. Therefore, for each study question, two CTC cutoffs (unadjusted and adjusted) were
ip t
derived, and were further assessed for prognostic accuracy.
cr
In the second step, the prognostic contribution of CTC count was assessed. CTC count, categorized by the unadjusted and adjusted cutoffs was added to a model with PS and gender as clinical covariates.
us
The improvement of prognostic accuracy due to the addition of CTCs was measured by the difference
an
of the c-indexes of the models with and without the CTC variable; statistical significance and 95% percentile confidence intervals (CI) of the difference of c-indexes were assessed by bootstrap 10,000
M
replicates. The effect of CTC on risk of death was estimated by the Hazard Ratio (HR) of death, using the higher CTC category as a reference category. Because of the strong association of treatment with
d
PS, we did not enter treatment in the multivariable bootstrap analyses.
te
When studying CTC change after cycle-1, the relative reduction was measured (rather than the absolute
Ac ce p
values) and baseline CTC count was entered as a covariate in the Cox model, because change is known to be positively correlated with baseline value. CTC change was measured as the ratio of the difference between CTC count after cycle-1 and CTC at baseline over CTC at baseline, by 100; thus, a negative value indicates reduction from baseline and a positive one indicates increase from baseline. Finally, we evaluated in our set of patients the cutoffs reported in three recently published papers [1315]. The difference of the c-indexes between the models with and without the addition of CTC count to clinical covariates was measured. Statistical significance and 95% CI of the difference of c-indexes were assessed by bootstrap resampling (10,000 replicates). HRs were also estimated. CTC change after chemotherapy was tested as suggested by Hou et al. [13] and Naito et al. [14], that created three subgroups: (i) patients whose baseline and post-treatment CTC count remained below the proposed 8 Page 8 of 29
cutoff, (ii) patients whose CTC values were above the cutoff at baseline and below the cutoff after chemotherapy and (iii) patients whose post-treatment CTC count was above the cutoff. In this paper we report these subgroups as -/-, +/- and any/+, respectively.
ip t
Comparison of CTC values among groups was performed by non parametric test (exact Wilcoxon-
cr
Mann-Whitney (WMW) test and asymptotic Kruskal-Wallis non parametric ANOVA). OS curves were drawn according to the Kaplan-Meier product limit method. All analyses were performed with R
us
software, version 2.11.1 (The R Foundation for Statistical Computing. 2010).
an
On request of a reviewer an external validation of our result was further performed in a small data set previously published [15]. The cutoff of CTC relative reduction that had the best accuracy in
M
discriminating prognosis in our study was applied to this validation sample. According to the available information in the validation data set, OS was calculated from the date of blood sampling at baseline.
te
Ac ce p
the small number of cases.
d
Groups were compared by means of log-rank test; no multivariate analysis was performed because of
9 Page 9 of 29
Results Patients’ characteristics
ip t
Sixty patients with extensive SCLC, consecutively evaluated between April 2008 and December 2010, were included in the analysis (Table 1). Median age was 66 years (range 49-81). Most patients were
cr
males (80%), with a good PS (ECOG 0 or 1 in 73.3%). Common sites of disease were lung (98.3%),
us
lymph-nodes (81.7%), adrenal glands (45%) and liver (40%). Three patients did never start chemotherapy, due to rapid worsening of general conditions. Out of the remaining patients, 49 (81.7%)
an
were treated with a platinum-containing doublet, 7 (11.7%) with single-agent carboplatin and 1 (1.6%) with single-agent etoposide. Treatment was strongly associated with PS: 41/44 (93%) patients with PS
M
0-1, indeed, received platinum-based doublet regimens, while 8/16 PS2 patients (50%) received single
te
Baseline CTC count
d
agent regimens. After a median follow-up of 15 months, 49 deaths (81.7%) had been recorded.
Ac ce p
Baseline CTC count was available for all enrolled patients (Supplementary Figure 1). At baseline median CTC count was 47 (Figure 1), and 54 patients (90%) had at least 1 CTC (range, 1 – 24281). Baseline CTC count was strongly associated with the number of organs involved by disease (p=0.005), PS (p=0.037) and type of treatment (p=0.021) (Table 1). After stratification for the number of organs, the latter two associations disappeared; therefore, only the number of involved organs was used to adjust bootstrap search for the most frequent "optimal" cutoff. At unadjusted bootstrap analysis, the most frequent “optimal” cutoff value was 282, that was the best in 29,7% of cases (Supplementary Figure 2, panel A). After adjustment by number of involved organs (Supplementary Figure 2, panel B), the "optimal" cutoffs were definitely lower and no evident peak was observed. However, the value of 25 was the most frequent, resulting the best in 10.2% of cases. 10 Page 10 of 29
The accuracy of the prognostic model was only marginally increased by the addition of baseline CTC value to gender and PS, although the risk of death was significantly reduced in the category with fewer
ip t
CTCs with both cutoffs (Table 2). CTC count after cycle-1
cr
CTCs after cycle-1 were determined in 45 patients (75%) (Supplementary Figure 1). Median time from
us
baseline sample was 22 days. CTC values were definitely lower than at baseline (Figure 1). Median CTC count was 5 (range, 0-1236); 12 patients had no CTC, including 5 with a zero CTC count also at
an
baseline; 33 patients (73.3%) had at least 1 CTC (range, 1 - 1236).
At unadjusted bootstrap analysis, the most frequent “optimal” cutoff value was 36, that was the best in
M
35.5% of cases (Supplementary Figure 2, panel C). After adjustment (Supplementary Figure 2, panel
d
D), the distribution shape was very similar and the most frequent "optimal" cutoff was 40, which was
te
the best in 26.6% of the 1,000 bootstrap samples. Similarly to findings at baseline, the accuracy of the prognostic model was only marginally increased by the addition of CTC information to gender and PS,
(Table 3).
Ac ce p
although the risk of death was significantly reduced in the category with fewer CTCs with both cutoffs
CTC variation after cycle-1
Forty patients (66.7%) were evaluable for analysis of CTC change between baseline and cycle-1, since 5 of those with both CTC counts available had baseline value equal to 0 and CTC change could not be calculated (Supplementary Figure 1). Median change was -86.5% (ranging from -100% to +780%). The most frequent “optimal” cutoff value of CTC change was -89% both at unadjusted (30.1%) and adjusted (30.1%) bootstrap analyses (Supplementary Figure 2, panels E and F). The accuracy of the prognostic model was significantly increased by the addition of CTC change (bootstrap p-value = 11 Page 11 of 29
0.009). Further, a strong reduction of CTC count after one cycle of chemotherapy was deeply associated to lower risk of death (HR 0.24, 95% CI 0.09 to 0.61, Table 4). For descriptive purposes only, univariate OS curves scattered by CTC change (cutoff -89%) are depicted in Figure 2. In the low-
ip t
risk group we also included the five patients with zero CTC count both at baseline and after cycle-1. Median OS (starting from the date of blood sample after cycle-1) was 7.2 months (95% CI 5.7 – n.a.)
cr
vs. 4.2 months (95% CI 3.7 – 8.5) in patients with CTC change 65 Gender Male Female ECOG performance status 0 1 2 No of organs involved 2 3 4 or more Treatment received Single agent or none Platinum-based doublet
No (%)
CTC median (IQR)
28 (46.7%) 32 (53.3%)
55 (12.75-240) 34.5 (6.75-2317)
48 (80%) 12 (20%)
55 (11.25-671.20) 22 (0.75-400)
P-value* 0.794
ip t
0.203
0.037
cr 0.021
662 (149-13230) 30 (6-226)
M
11 (18.3%) 49 (81.7%)
0.005
27 (6-52) 19 (5-203) 1001 (123-3022)
an
13 (21.7%) 29 (48.3%) 18 (30.0%)
13 (1-30) 55 (14-597.5) 630 (8.5-3393)
us
13 (21.7) 31(51.7) 16 (26.6)
Ac ce p
te
d
* Wilcoxon-Mann-Whitney exact test or Asymptotic Kruskal-Wallis Test
20 Page 20 of 29
Cindex 0.703
C-index difference* (95% bootstrap CI)
P-value (bootstrap)
HR (95%CI)
PS, Gender, baseline CTC (cutoff 25)
0.715
0.012 (-0.028 to 0.059)
0.60
0.45 (0.23 to 0.88)
PS, Gender, baseline CTC (cutoff 282) 0.717
0.014 (-0.029 to 0.042)
0.46
0.50 (0.26 to 0.95)
ip t
Covariates included in the Cox model PS, Gender
cr
Table 2. Improvement of discriminating accuracy when baseline CTC count, categorized according to the findings of unadjusted and adjusted bootstrap analyses, are added to a Cox model with clinical covariates only.
Ac ce p
te
d
M
an
us
* Bootstrap percentile confidence interval on 10,000 bootstrap samples
21 Page 21 of 29
Cindex 0.635
C-index difference* (95% bootstrap CI)
P-value (bootstrap)
HR (95%CI)
PS, Gender, cycle-1 CTC (cutoff 36)
0.675
0.040 (-0.258 to 0.014)
0.55
0.40 (0.18 to 0.90)
PS, Gender, cycle 1 CTC (cutoff 40)
0.671
0.036 (-0.009 to 0.108)
0.43 (0.19 to 0.98)
ip t
Covariates included in the Cox model PS, Gender
cr
Table 3. Improvement of discriminating accuracy when CTC count after one cycle of chemotherapy, categorized according to the findings of unadjusted and adjusted bootstrap analyses, are added to a Cox model with clinical covariates only.
us
0.22
Ac ce p
te
d
M
an
* Bootstrap percentile confidence interval on 10,000 bootstrap samples
22 Page 22 of 29
Table 4. Improvement of discriminating accuracy when % change of CTC counts after chemotherapy, categorized according to the findings of unadjusted and adjusted bootstrap analyses, are added to a Cox model with clinical covariates only.
0.756
0.137 (0.005 – 0.206)
P-value (bootstrap)
HR (95%CI)
0.009
0.24 (0.09-0.61)
cr
PS, Gender, baseline CTC, CTC change (cutoff -89%)
C-index difference* (95% bootstrap CI)
ip t
PS, Gender, baseline CTC
Cindex 0.619
Ac ce p
te
d
M
an
us
* Bootstrap percentile confidence interval on 10,000 bootstrap samples
23 Page 23 of 29
Ac
ce
pt
ed
M
an
us
cr
i
Figure
Page 24 of 29
Ac
ce
pt
ed
M
an
us
cr
i
Figure 2
Page 25 of 29
ip t
Table 1
CTC median (IQR)
28 (46.7%) 32 (53.3%)
55 (12.75-240) 34.5 (6.75-2317)
us
No (%)
55 (11.25-671.20) 22 (0.75-400)
13 (21.7) 31(51.7) 16 (26.6)
13 (1-30) 55 (14-597.5) 630 (8.5-3393)
M an
48 (80%) 12 (20%)
ed
13 (21.7%) 29 (48.3%) 18 (30.0%) 11 (18.3%) 49 (81.7%)
ce pt
Variable Age ≤ 65 > 65 Gender Male Female ECOG performance status 0 1 2 No of organs involved 2 3 4 or more Treatment received Single agent or none Platinum-based doublet
cr
Table 1. Patient characteristics (n = 60)
P-value* 0.794
0.203
0.037
0.005
27 (6-52) 19 (5-203) 1001 (123-3022) 0.021 662 (149-13230) 30 (6-226)
Ac
* Wilcoxon-Mann-Whitney exact test or Asymptotic Kruskal-Wallis Test
Page 26 of 29
ip t
Table 2
C- index
C-index difference* (95% bootstrap CI)
P-value (bootstrap)
HR (95%CI)
PS, Gender, baseline CTC (cutoff 25)
0.715
0.012 (-0.028 to 0.059)
0.60
0.45 (0.23 to 0.88)
PS, Gender, baseline CTC (cutoff 282)
0.717
0.014 (-0.029 to 0.042)
0.46
0.50 (0.26 to 0.95)
us
Covariates included in the Cox model PS, Gender
cr
Table 2. Improvement of discriminating accuracy when baseline CTC count, categorized according to the findings of unadjusted and adjusted bootstrap analyses, are added to a Cox model with clinical covariates only.
M an
0.703
Ac
ce pt
ed
* Bootstrap percentile confidence interval on 10,000 bootstrap samples
Page 27 of 29
Table 3
Table 3. Improvement of discriminating accuracy when CTC count after one cycle of chemotherapy, categorized according to the findings of unadjusted and adjusted bootstrap analyses, are added to a Cox model with clinical covariates only. Cindex 0.635
C-index difference* (95% bootstrap CI)
P-value (bootstrap)
HR (95%CI)
PS, Gender, cycle-1 CTC (cutoff 36)
0.675
0.040 (-0.258 to 0.014)
0.55
0.40 (0.18 to 0.90)
PS, Gender, cycle 1 CTC (cutoff 40)
0.671
0.036 (-0.009 to 0.108)
0.22
0.43 (0.19 to 0.98)
us
cr
ip t
Covariates included in the Cox model PS, Gender
Ac ce p
te
d
M
an
* Bootstrap percentile confidence interval on 10,000 bootstrap samples
1 Page 28 of 29
ip t
Table 4
0.619
PS, Gender, baseline CTC, CTC change (cutoff 89%)
0.756
M an
PS, Gender, baseline CTC
C-index difference* (95% bootstrap CI)
us
C- index
cr
Table 4. Improvement of discriminating accuracy when % change of CTC counts after chemotherapy, categorized according to the findings of unadjusted and adjusted bootstrap analyses, are added to a Cox model with clinical covariates only.
0.137 (0.005 – 0.206)
P-value (bootstrap)
HR (95%CI)
0.009
0.24 (0.09-0.61)
Ac
ce pt
ed
* Bootstrap percentile confidence interval on 10,000 bootstrap samples
Page 29 of 29
S0169-5002(14)00211-6 http://dx.doi.org/doi:10.1016/j.lungcan.2014.05.002 LUNG 4599
To appear in:
Lung Cancer
Received date: Revised date: Accepted date:
3-8-2013 24-4-2014 1-5-2014
Please cite this article as: Normanno N, Rossi A, Morabito A, Signoriello S, Bevilacqua S, Di Maio M, Costanzo R, De Luca A, Montanino A, Gridelli C, Rocco G, Perrone F, Gallo C, Prognostic value of circulating tumor cells’ reduction in patients with extensive small-cell lung cancer, Lung Cancer (2014), http://dx.doi.org/10.1016/j.lungcan.2014.05.002 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Prognostic value of circulating tumor cells’ reduction in patients with extensive small-cell lung cancer
ip t
Nicola Normanno,1 Antonio Rossi,2* Alessandro Morabito,3* Simona Signoriello,4* Simona
Bevilacqua,1* Massimo Di Maio,5* Raffaele Costanzo,3 Antonella De Luca,1 Agnese Montanino,3
cr
Cesare Gridelli,2 Gaetano Rocco,6 Francesco Perrone 5 and Ciro Gallo 4 1
te
d
*these authors equally contributed to the paper
M
an
us
Cell Biology and Biotherapy Unit, Istituto Nazionale per lo Studio e la Cura dei Tumori “Fondazione Giovanni Pascale” - IRCCS, Napoli; 2Division of Medical Oncology, S.G. Moscati Hospital, Avellino; 3 Medical Oncology Unit, Department of Thoracic Surgical and Medical Oncology, Istituto Nazionale per lo Studio e la Cura dei Tumori “Fondazione Giovanni Pascale” - IRCCS, Napoli; 4Medical Statistics, Second University, Napoli; 5Clinical Trials Unit, Istituto Nazionale per lo Studio e la Cura dei Tumori “Fondazione Giovanni Pascale” - IRCCS, Napoli; 6Division of Thoracic Surgery, Department of Thoracic Surgical and Medical Oncology, Istituto Nazionale per lo Studio e la Cura dei Tumori “Fondazione Giovanni Pascale” - IRCCS, Napoli; Italy.
Ac ce p
Corresponding author: Nicola Normanno Cell Biology and Biotherapy Unit INT-Fondazione Pascale Via Mariano Semmola 80131 Napoli (Italy) Tel and Fax: +39/081/5903826 Email: [email protected]; [email protected] Running title: CTCs in SCLC
1 Page 1 of 29
Highlights
us
cr
ip t
We assessed circulating tumor cells (CTCs) with the CellSearch system in extensive SCLC patients The accuracy of prognostic role of different cutoffs was assessed by Harrell’s c-index CTC count at baseline or after chemotherapy cycle-1 marginally increased the prognostic accuracy Prognostic accuracy was significantly improved by CTC count reduction higher than 89% after cycle-1 CTC count reduction higher than 89% after cycle-1 was associated with longer survival
an
Abstract
Objectives: Circulating tumor cells (CTCs) have been hypothesized to be a prognostic factor in small-
M
cell lung cancer (SCLC), and different cutoffs have been proposed to identify patients at high risk. We
d
assessed the prognostic value of CTCs in patients with extensive SCLC.
te
Materials and Methods: CTCs were assessed with the CellSearch system in 60 extensive SCLC patients. CTC count at baseline or after one cycle of chemotherapy (cycle-1) or as change after
Ac ce p
chemotherapy were analyzed separately. Primary outcome was overall survival. The accuracy of prognostic role was assessed by Harrell’s c-index. “Optimal” cutoffs were derived by bootstrap resampling to reduce the overfitting bias; accuracy improvement was estimated by calculating the difference of c-indexes of models including clinical variables with or without CTCs. Results: CTCs were identified in 90% (54/60) of patients at baseline, in which CTC count ranged from 0 to 24281. CTC count was strongly associated with the number of organs involved. The prognostic accuracy was only marginally increased by the addition to clinical information of “optimal” CTC cutoffs at baseline and after cycle-1. Conversely, a reduction of CTC count higher than 89% following chemotherapy significantly improved prognostic accuracy (bootstrap p-value = 0.009) and was 2 Page 2 of 29
associated with a lower risk of death (HR 0.24, 95% CI 0.09 to 0.61). When previously proposed cutoffs were applied to our cohort, they showed only marginal improvement of the prognostic accuracy.
ip t
Conclusion: CTCs have useful prognostic role in extensive SCLC, but only the change of CTC count
cr
after the first cycle of chemotherapy provides clinically relevant information. Previously reported CTC
us
cutoffs were not prognostic in our cohort of patients.
Ac ce p
te
d
M
an
Key words: circulating tumor cells; small-cell lung cancer; prognosis; chemotherapy
3 Page 3 of 29
Introduction Small-cell lung cancer (SCLC) is characterized by rapid doubling time, propensity for early dissemination, significant sensitivity to chemotherapy and radiotherapy and development of drug
ip t
resistance during the course of disease [1]. Clinical and pathological characteristics, as extensive-stage
cr
disease, performance status (PS), gender, elevated lactate dehydrogenase (LDH) and alkaline
phosphatase (ALP) serum levels, are common prognostic markers [1, 2]. To date, no histopathological,
us
molecular or genetic prognostic tests are available.
an
Detection and characterization of circulating tumor cells (CTCs) in the peripheral blood can provide information on cancer aggressiveness, might improve prediction of patient’s outcome and help to
M
develop novel therapeutic approaches [3, 4]. CTCs can be isolated from peripheral blood by different methods [3, 4]. The CellSearch system (Veridex, Raritan, NJ) allows automated immunomagnetic
d
enrichment and staining of CTCs. The CellSearch has been approved by US Food and Drug
te
Administration for breast, colorectal and prostate cancer in which CTC count has been proposed to be
Ac ce p
prognostic [5-9]. Several studies also suggested that a reduction in CTC count following the administration of therapy might provide prognostic information [6, 7, 9, 10]. A prognostic role of CellSearch-identified CTCs in non-small-cell lung cancer has been recently suggested [11]. Our group was the first to demonstrate that CTCs can be isolated from the peripheral blood of SCLC patients by using the CellSearch system [12]. Three studies previously explored the prognostic role of CTCs in SCLC and identified three different cutoffs (50, 8 and 2, respectively) that separate patients in favorable and unfavorable groups according to CTC count [13-15]. Hou et al. [13] and Naito et al. [14] found that baseline CTC count and change in CTC number after one cycle of chemotherapy were independent prognostic factors. Hiltermann et al. [15] confirmed the value of CTC count at baseline and after one cycle of chemotherapy. Of note, all these studies included patients with either limited or 4 Page 4 of 29
extensive disease. Objective of this study was to explore the role of CTCs in SCLC as prognostic marker in patients with extensive disease only, and to externally validate the findings of the previously
Ac ce p
te
d
M
an
us
cr
ip t
reported studies.
5 Page 5 of 29
Patients and Methods Patients
ip t
Patients with chemotherapy-naive, histologically proven SCLC, extensive disease according to Veterans Administration Lung Cancer Group classification, who gave informed consent, were eligible.
cr
Blood samples were collected for analysis within 7 days before commencing treatment (baseline) and
us
after one chemotherapy cycle (cycle-1).
an
Ethical considerations
In 2008, the Scientific Board of the National Cancer Institute of Napoli approved CTC count
M
assessment in a non-interventional study, as part of the clinical workout of patients, based on our previous observation that CTCs can be detected in SCLC patients by using the CellSearch system [12]
d
and that the CellSearch Circulating Tumor Cell kit was approved in Europe for in vitro diagnostics
te
(CE-IVD). Just oral consent was sought upon information about the exploratory nature of CTC count
Ac ce p
assessment and that the result would have not affected treatment. In 2012, after the publication of the Krebs paper in NSCLC [11] we realized that information coming from our data could be interesting for scientific community and we sought the approval of the Independent Ethical Committee (IEC) regarding the opportunity to analyse and publish our data. IEC supported our request. CTC analysis
Peripheral blood samples (7.5 ml) were collected into CellSave preservative tubes and processed within 72 hours of collection. The CellSearch (Veridex LLC, Raritan, NJ) platform was used for isolation and enumeration of CTCs [12, 16]. Samples were processed in the CellTracks Autoprep system by using the Circulating Tumor Cell kit and analysed with the CellTracks Analyzer II according to the manufacturer’s instructions. 6 Page 6 of 29
Statistical analysis We addressed three main questions: do (i) CTC count at baseline or (ii) CTC after cycle-1 or (iii) CTC change after cycle-1 increase the accuracy of prognostic evaluation beyond what is expected by using
ip t
clinical information only? For each of the three questions, we had to look for the "optimal" cutoff first,
cr
and to assess the prognostic role of CTC (categorized according the "optimal" cutoff) in the
multivariable analysis, as the second step. Due to the exploratory nature of the study no formal sample
us
size definition was planned before the study.
an
Primary outcome was overall survival (OS). When assessing the value of baseline CTC, OS was defined as the time from date of baseline blood collection to death or to last follow-up visit for living
M
patients; when the value of CTC count after cycle-1 or of CTC variation was evaluated, starting time
d
for OS was the date of the second blood sample collection.
te
As first step we had to look for the "optimal" cutoff . The “optimal” cutoff was defined as the CTC value that maximized the discriminating ability, according to the Harrell’s concordance index (c-
Ac ce p
index), a modified version of the area under ROC curve for survival data [17]. All the observed CTC values were screened as potentially "optimal" cutoff, i.e. cutoff levels were not chosen 'a priori', This approach invariably leads to overfitting bias, i.e. to results tailored just for the current data; therefore, to reduce overfitting and improve internal validity we applied a bootstrap procedure [18], randomly sampling with repetition a thousand samples of size 60 from the original data set. For each bootstrap sample the CTC cutoff that maximized the c-index was found, and the distribution of the "optimal" cutoffs observed in the 1,000 bootstrap samples was derived. The most frequent "optimal" cutoff was entered in the subsequent prognostic model. Cutoffs are always intended as greater than or equal to the stated value.
7 Page 7 of 29
Since CTC values were strongly associated with the number of organs involved by the disease, we repeated the analysis after adjustment by the number of involved organs in a Cox proportional hazard regression model. Therefore, for each study question, two CTC cutoffs (unadjusted and adjusted) were
ip t
derived, and were further assessed for prognostic accuracy.
cr
In the second step, the prognostic contribution of CTC count was assessed. CTC count, categorized by the unadjusted and adjusted cutoffs was added to a model with PS and gender as clinical covariates.
us
The improvement of prognostic accuracy due to the addition of CTCs was measured by the difference
an
of the c-indexes of the models with and without the CTC variable; statistical significance and 95% percentile confidence intervals (CI) of the difference of c-indexes were assessed by bootstrap 10,000
M
replicates. The effect of CTC on risk of death was estimated by the Hazard Ratio (HR) of death, using the higher CTC category as a reference category. Because of the strong association of treatment with
d
PS, we did not enter treatment in the multivariable bootstrap analyses.
te
When studying CTC change after cycle-1, the relative reduction was measured (rather than the absolute
Ac ce p
values) and baseline CTC count was entered as a covariate in the Cox model, because change is known to be positively correlated with baseline value. CTC change was measured as the ratio of the difference between CTC count after cycle-1 and CTC at baseline over CTC at baseline, by 100; thus, a negative value indicates reduction from baseline and a positive one indicates increase from baseline. Finally, we evaluated in our set of patients the cutoffs reported in three recently published papers [1315]. The difference of the c-indexes between the models with and without the addition of CTC count to clinical covariates was measured. Statistical significance and 95% CI of the difference of c-indexes were assessed by bootstrap resampling (10,000 replicates). HRs were also estimated. CTC change after chemotherapy was tested as suggested by Hou et al. [13] and Naito et al. [14], that created three subgroups: (i) patients whose baseline and post-treatment CTC count remained below the proposed 8 Page 8 of 29
cutoff, (ii) patients whose CTC values were above the cutoff at baseline and below the cutoff after chemotherapy and (iii) patients whose post-treatment CTC count was above the cutoff. In this paper we report these subgroups as -/-, +/- and any/+, respectively.
ip t
Comparison of CTC values among groups was performed by non parametric test (exact Wilcoxon-
cr
Mann-Whitney (WMW) test and asymptotic Kruskal-Wallis non parametric ANOVA). OS curves were drawn according to the Kaplan-Meier product limit method. All analyses were performed with R
us
software, version 2.11.1 (The R Foundation for Statistical Computing. 2010).
an
On request of a reviewer an external validation of our result was further performed in a small data set previously published [15]. The cutoff of CTC relative reduction that had the best accuracy in
M
discriminating prognosis in our study was applied to this validation sample. According to the available information in the validation data set, OS was calculated from the date of blood sampling at baseline.
te
Ac ce p
the small number of cases.
d
Groups were compared by means of log-rank test; no multivariate analysis was performed because of
9 Page 9 of 29
Results Patients’ characteristics
ip t
Sixty patients with extensive SCLC, consecutively evaluated between April 2008 and December 2010, were included in the analysis (Table 1). Median age was 66 years (range 49-81). Most patients were
cr
males (80%), with a good PS (ECOG 0 or 1 in 73.3%). Common sites of disease were lung (98.3%),
us
lymph-nodes (81.7%), adrenal glands (45%) and liver (40%). Three patients did never start chemotherapy, due to rapid worsening of general conditions. Out of the remaining patients, 49 (81.7%)
an
were treated with a platinum-containing doublet, 7 (11.7%) with single-agent carboplatin and 1 (1.6%) with single-agent etoposide. Treatment was strongly associated with PS: 41/44 (93%) patients with PS
M
0-1, indeed, received platinum-based doublet regimens, while 8/16 PS2 patients (50%) received single
te
Baseline CTC count
d
agent regimens. After a median follow-up of 15 months, 49 deaths (81.7%) had been recorded.
Ac ce p
Baseline CTC count was available for all enrolled patients (Supplementary Figure 1). At baseline median CTC count was 47 (Figure 1), and 54 patients (90%) had at least 1 CTC (range, 1 – 24281). Baseline CTC count was strongly associated with the number of organs involved by disease (p=0.005), PS (p=0.037) and type of treatment (p=0.021) (Table 1). After stratification for the number of organs, the latter two associations disappeared; therefore, only the number of involved organs was used to adjust bootstrap search for the most frequent "optimal" cutoff. At unadjusted bootstrap analysis, the most frequent “optimal” cutoff value was 282, that was the best in 29,7% of cases (Supplementary Figure 2, panel A). After adjustment by number of involved organs (Supplementary Figure 2, panel B), the "optimal" cutoffs were definitely lower and no evident peak was observed. However, the value of 25 was the most frequent, resulting the best in 10.2% of cases. 10 Page 10 of 29
The accuracy of the prognostic model was only marginally increased by the addition of baseline CTC value to gender and PS, although the risk of death was significantly reduced in the category with fewer
ip t
CTCs with both cutoffs (Table 2). CTC count after cycle-1
cr
CTCs after cycle-1 were determined in 45 patients (75%) (Supplementary Figure 1). Median time from
us
baseline sample was 22 days. CTC values were definitely lower than at baseline (Figure 1). Median CTC count was 5 (range, 0-1236); 12 patients had no CTC, including 5 with a zero CTC count also at
an
baseline; 33 patients (73.3%) had at least 1 CTC (range, 1 - 1236).
At unadjusted bootstrap analysis, the most frequent “optimal” cutoff value was 36, that was the best in
M
35.5% of cases (Supplementary Figure 2, panel C). After adjustment (Supplementary Figure 2, panel
d
D), the distribution shape was very similar and the most frequent "optimal" cutoff was 40, which was
te
the best in 26.6% of the 1,000 bootstrap samples. Similarly to findings at baseline, the accuracy of the prognostic model was only marginally increased by the addition of CTC information to gender and PS,
(Table 3).
Ac ce p
although the risk of death was significantly reduced in the category with fewer CTCs with both cutoffs
CTC variation after cycle-1
Forty patients (66.7%) were evaluable for analysis of CTC change between baseline and cycle-1, since 5 of those with both CTC counts available had baseline value equal to 0 and CTC change could not be calculated (Supplementary Figure 1). Median change was -86.5% (ranging from -100% to +780%). The most frequent “optimal” cutoff value of CTC change was -89% both at unadjusted (30.1%) and adjusted (30.1%) bootstrap analyses (Supplementary Figure 2, panels E and F). The accuracy of the prognostic model was significantly increased by the addition of CTC change (bootstrap p-value = 11 Page 11 of 29
0.009). Further, a strong reduction of CTC count after one cycle of chemotherapy was deeply associated to lower risk of death (HR 0.24, 95% CI 0.09 to 0.61, Table 4). For descriptive purposes only, univariate OS curves scattered by CTC change (cutoff -89%) are depicted in Figure 2. In the low-
ip t
risk group we also included the five patients with zero CTC count both at baseline and after cycle-1. Median OS (starting from the date of blood sample after cycle-1) was 7.2 months (95% CI 5.7 – n.a.)
cr
vs. 4.2 months (95% CI 3.7 – 8.5) in patients with CTC change 65 Gender Male Female ECOG performance status 0 1 2 No of organs involved 2 3 4 or more Treatment received Single agent or none Platinum-based doublet
No (%)
CTC median (IQR)
28 (46.7%) 32 (53.3%)
55 (12.75-240) 34.5 (6.75-2317)
48 (80%) 12 (20%)
55 (11.25-671.20) 22 (0.75-400)
P-value* 0.794
ip t
0.203
0.037
cr 0.021
662 (149-13230) 30 (6-226)
M
11 (18.3%) 49 (81.7%)
0.005
27 (6-52) 19 (5-203) 1001 (123-3022)
an
13 (21.7%) 29 (48.3%) 18 (30.0%)
13 (1-30) 55 (14-597.5) 630 (8.5-3393)
us
13 (21.7) 31(51.7) 16 (26.6)
Ac ce p
te
d
* Wilcoxon-Mann-Whitney exact test or Asymptotic Kruskal-Wallis Test
20 Page 20 of 29
Cindex 0.703
C-index difference* (95% bootstrap CI)
P-value (bootstrap)
HR (95%CI)
PS, Gender, baseline CTC (cutoff 25)
0.715
0.012 (-0.028 to 0.059)
0.60
0.45 (0.23 to 0.88)
PS, Gender, baseline CTC (cutoff 282) 0.717
0.014 (-0.029 to 0.042)
0.46
0.50 (0.26 to 0.95)
ip t
Covariates included in the Cox model PS, Gender
cr
Table 2. Improvement of discriminating accuracy when baseline CTC count, categorized according to the findings of unadjusted and adjusted bootstrap analyses, are added to a Cox model with clinical covariates only.
Ac ce p
te
d
M
an
us
* Bootstrap percentile confidence interval on 10,000 bootstrap samples
21 Page 21 of 29
Cindex 0.635
C-index difference* (95% bootstrap CI)
P-value (bootstrap)
HR (95%CI)
PS, Gender, cycle-1 CTC (cutoff 36)
0.675
0.040 (-0.258 to 0.014)
0.55
0.40 (0.18 to 0.90)
PS, Gender, cycle 1 CTC (cutoff 40)
0.671
0.036 (-0.009 to 0.108)
0.43 (0.19 to 0.98)
ip t
Covariates included in the Cox model PS, Gender
cr
Table 3. Improvement of discriminating accuracy when CTC count after one cycle of chemotherapy, categorized according to the findings of unadjusted and adjusted bootstrap analyses, are added to a Cox model with clinical covariates only.
us
0.22
Ac ce p
te
d
M
an
* Bootstrap percentile confidence interval on 10,000 bootstrap samples
22 Page 22 of 29
Table 4. Improvement of discriminating accuracy when % change of CTC counts after chemotherapy, categorized according to the findings of unadjusted and adjusted bootstrap analyses, are added to a Cox model with clinical covariates only.
0.756
0.137 (0.005 – 0.206)
P-value (bootstrap)
HR (95%CI)
0.009
0.24 (0.09-0.61)
cr
PS, Gender, baseline CTC, CTC change (cutoff -89%)
C-index difference* (95% bootstrap CI)
ip t
PS, Gender, baseline CTC
Cindex 0.619
Ac ce p
te
d
M
an
us
* Bootstrap percentile confidence interval on 10,000 bootstrap samples
23 Page 23 of 29
Ac
ce
pt
ed
M
an
us
cr
i
Figure
Page 24 of 29
Ac
ce
pt
ed
M
an
us
cr
i
Figure 2
Page 25 of 29
ip t
Table 1
CTC median (IQR)
28 (46.7%) 32 (53.3%)
55 (12.75-240) 34.5 (6.75-2317)
us
No (%)
55 (11.25-671.20) 22 (0.75-400)
13 (21.7) 31(51.7) 16 (26.6)
13 (1-30) 55 (14-597.5) 630 (8.5-3393)
M an
48 (80%) 12 (20%)
ed
13 (21.7%) 29 (48.3%) 18 (30.0%) 11 (18.3%) 49 (81.7%)
ce pt
Variable Age ≤ 65 > 65 Gender Male Female ECOG performance status 0 1 2 No of organs involved 2 3 4 or more Treatment received Single agent or none Platinum-based doublet
cr
Table 1. Patient characteristics (n = 60)
P-value* 0.794
0.203
0.037
0.005
27 (6-52) 19 (5-203) 1001 (123-3022) 0.021 662 (149-13230) 30 (6-226)
Ac
* Wilcoxon-Mann-Whitney exact test or Asymptotic Kruskal-Wallis Test
Page 26 of 29
ip t
Table 2
C- index
C-index difference* (95% bootstrap CI)
P-value (bootstrap)
HR (95%CI)
PS, Gender, baseline CTC (cutoff 25)
0.715
0.012 (-0.028 to 0.059)
0.60
0.45 (0.23 to 0.88)
PS, Gender, baseline CTC (cutoff 282)
0.717
0.014 (-0.029 to 0.042)
0.46
0.50 (0.26 to 0.95)
us
Covariates included in the Cox model PS, Gender
cr
Table 2. Improvement of discriminating accuracy when baseline CTC count, categorized according to the findings of unadjusted and adjusted bootstrap analyses, are added to a Cox model with clinical covariates only.
M an
0.703
Ac
ce pt
ed
* Bootstrap percentile confidence interval on 10,000 bootstrap samples
Page 27 of 29
Table 3
Table 3. Improvement of discriminating accuracy when CTC count after one cycle of chemotherapy, categorized according to the findings of unadjusted and adjusted bootstrap analyses, are added to a Cox model with clinical covariates only. Cindex 0.635
C-index difference* (95% bootstrap CI)
P-value (bootstrap)
HR (95%CI)
PS, Gender, cycle-1 CTC (cutoff 36)
0.675
0.040 (-0.258 to 0.014)
0.55
0.40 (0.18 to 0.90)
PS, Gender, cycle 1 CTC (cutoff 40)
0.671
0.036 (-0.009 to 0.108)
0.22
0.43 (0.19 to 0.98)
us
cr
ip t
Covariates included in the Cox model PS, Gender
Ac ce p
te
d
M
an
* Bootstrap percentile confidence interval on 10,000 bootstrap samples
1 Page 28 of 29
ip t
Table 4
0.619
PS, Gender, baseline CTC, CTC change (cutoff 89%)
0.756
M an
PS, Gender, baseline CTC
C-index difference* (95% bootstrap CI)
us
C- index
cr
Table 4. Improvement of discriminating accuracy when % change of CTC counts after chemotherapy, categorized according to the findings of unadjusted and adjusted bootstrap analyses, are added to a Cox model with clinical covariates only.
0.137 (0.005 – 0.206)
P-value (bootstrap)
HR (95%CI)
0.009
0.24 (0.09-0.61)
Ac
ce pt
ed
* Bootstrap percentile confidence interval on 10,000 bootstrap samples
Page 29 of 29
