- Email: [email protected]
Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer
YGYNO-976526; No. of pages: 6; 4C: Gynecologic Oncology xxx (2016) xxx–xxx
Contents lists available at ScienceDirect
Gynecologic Oncology journal homepage: www.elsevier.com/locate/ygyno
Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer Seung-Hyuk Shim, MD a, Dae-Yeon Kim, MD, PhD b,⁎,1, Sun Joo Lee, MD, PhD a, Soo-Nyung Kim, MD, PhD a, Soon-Beom Kang, MD, PhD a,⁎⁎,1, Shin-Wha Lee, MD, PhD b, Jeong-Yeol Park, MD, PhD b, Dae-Shik Suh, MD, PhD b, Jong-Hyeok Kim, MD, PhD b, Yong-Man Kim, MD, PhD b, Young-Tak Kim, MD, PhD b, Joo-Hyun Nam, MD, PhD b a b
Department of Obstetrics and Gynecology, Konkuk University School of Medicine, Seoul, Republic of Korea Department of Obstetrics and Gynecology, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Republic of Korea
H I G H L I G H T S • 245 consecutive LACC patients undergoing para-aortic lymphadenectomy before definitive treatment were analyzed. • Using tumor size and PET/CT features, a risk prediction model for predicting PALN metastasis was developed. • The model displayed good discrimination and calibration (concordance index = 0.886; 95% confidence interval = 0.825–0.947).
a r t i c l e
i n f o
Article history: Received 5 September 2016 Received in revised form 1 November 2016 Accepted 5 November 2016 Available online xxxx Keywords: Cervical cancer Likelihood functions Lymphatic metastasis Lymph node excision Morbidity
a b s t r a c t Objective. Concurrent chemoradiotherapy is usually administered to patients with locally advanced cervical cancer (LACC). Extended-field chemoradiotherapy is required if para-aortic lymph node (PALN) metastasis is detected. This study aimed to construct a prediction model for PALN metastasis in patients with LACC before definitive treatment. Methods. Between 2009 and 2016, all consecutive patients with LACC who underwent para-aortic lymphadenectomy at two tertiary centers were retrospectively analyzed. A multivariate logistic model was constructed, from which a prediction model for PALN metastasis was developed and internally validated. Before analysis, risk grouping was predefined based on the likelihood ratio. Results. In total, 245 patients satisfied the eligibility criteria. Thirty-four patients (13.9%) had pathologically proven PALN metastases. Additionally, 16/222 (7.2%) patients with negative PALNs on positron emission tomography/computed tomography (PET/CT) had PALN metastasis. Moreover, 11/105 (10.5%) patients with both negative PALNs and positive pelvic lymph nodes on PET/CT had PALN metastasis. Tumor size on magnetic resonance imaging and PALN status on PET/CT were independent predictors of PALN metastasis. The model incorporating these two predictors displayed good discrimination and calibration (bootstrap-corrected concordance index = 0.886; 95% confidence interval = 0.825–0.947). The model categorized 169 (69%), 52 (22%), and 23 (9%) patients into low-, intermediate-, and high-risk groups, respectively. The predicted probabilities of PALN metastasis for these groups were 2.9, 20.8, and 76.2%, respectively. Conclusion. We constructed a robust model predicting PALN metastasis in patients with LACC that may improve clinical trial design and help clinicians determine whether nodal-staging surgery should be performed. © 2016 Elsevier Inc. All rights reserved.
⁎ Correspondence to: Dae-Yeon Kim, MD, Department of Obstetrics and Gynecology, University of Ulsan College of Medicine, Asan Medical Center, 88 Olympic-ro 43 gil, Songpa-gu, Seoul 05505, Republic of Korea. ⁎⁎ Correspondence to: S.-B. Kang, Department of Obstetrics and Gynecology, Konkuk University School of Medicine, 263 Achasan-ro, Gwangjin-gu, Seoul 05030, Republic of Korea. E-mail addresses: [email protected] (D.-Y. Kim), [email protected] (S.-B. Kang). 1 Soon-Beom Kang and Dae-Yeon Kim contributed equally to this work and should be considered as co-corresponding authors.
1. Introduction Since cytological screening was introduced, the incidence of cervical cancer has decreased remarkably. However, this disease remains a sizeable health problem that accounted for an estimated 528,000 new cases and for 266,000 deaths worldwide in 2012 [1]. In Korea, it is the most common female genital malignancy, and the age-standardized incidence rate is 9.5 per 100,000 persons in 2013 [2].
http://dx.doi.org/10.1016/j.ygyno.2016.11.011 0090-8258/© 2016 Elsevier Inc. All rights reserved.
Please cite this article as: S.-H. Shim, et al., Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer, Gynecol Oncol (2016), http://dx.doi.org/10.1016/j.ygyno.2016.11.011
2
S.-H. Shim et al. / Gynecologic Oncology xxx (2016) xxx–xxx
Based on five phase III randomized trials demonstrating that concurrent chemoradiotherapy (CCRT) improves overall survival in patients with locally advanced cervical cancer (LACC), the current guidelines recommend CCRT as the standard treatment for these patients [3,4]. Nodal metastasis is among the most important prognostic factors for survival in patients with LACC. In patients with para-aortic lymph node (PALN) metastasis, extended-field radiotherapy should be considered [5]. Accurate pretreatment evaluation of PALN involvement is therefore of paramount importance in selecting radiation fields. The use of fluorine-18 fluorodeoxyglucose positron emission tomography/computed tomography (PET/CT) may aid in detecting extrapelvic disease compared with conventional imaging modalities such as pelvic magnetic resonance imaging (MRI) or CT [6,7]. Although PET is valuable for detecting extrapelvic disease, the reported false-negative rate of PALN metastasis on PET/CT ranged 9–22%, which is ascribable to small-volume metastases [5,7]. Consequently, nodal-staging surgery is possibly beneficial for patients with negative PALN involvement on PET/CT [8–10]. However, considering potential morbidity and the cost of staging surgery, nodal-staging surgery needs to be individualized in these patients [11]. Thus, it would be useful to develop an individualized prediction model for PALN metastasis before definitive treatment. Furthermore, individualized risk stratification of PALN metastasis may be advantageous in clinical trial design. Thus, this study aimed to develop and internally validate a prediction model for PALN metastasis in patients with LACC and identify potential candidates for PALN staging surgery using the model.
2.3. Para-aortic lymphadenectomy
2. Methods
In both institutions, the indications for para-aortic lymphadenectomy are as follows: a clinical diagnosis of FIGO stage IB2, IIA2, IIB, III, or IVA disease; age N 18 years and b80 years; and the absence of distant metastatic disease on imaging or physical examination. Para-aortic lymphadenectomy was performed as previously described [16]. Briefly, para-aortic lymphadenectomy was performed via laparoscopy or laparotomy. All of lymphatic tissue from the aorta, aortocaval space, and vena cava were completely removed. The boundaries of the para-aortic lymphadenectomy were defined as the bifurcation of the aorta caudally and the inferior mesenteric artery cranially. If the PALN at this level was enlarged or positive for metastasis, PALN dissection was extended to the left renal vein level. All harvested lymph nodes were grouped according to the name of the adjacent vessel (venacaval, aortocaval, aortic below the inferior mesenteric artery, and aortic above the inferior mesenteric artery). Each lymph node was sliced perpendicular to its long axis, stained with hematoxylin and eosin, and examined microscopically by a pathologist. Each pathology report included the number of lymph nodes retrieved from each area, the presence or absence of metastases, the largest size of lymph node metastases, and the presence or absence of extracapsular extension. Pelvic lymphadenectomy was not routinely performed as part of surgical nodal staging because the pelvic area is covered by the conventional radiation field. However, pelvic lymphadenectomy was performed to remove enlarged or suspicious pelvic nodes at the surgeon's discretion. In both institutions, for patients with negative results on frozen section of para-aortic lymphadenectomy, radical hysterectomy was considered in selected younger patients (especially those under 45 years old) and FIGO stage IB2 or IIA2.
2.1. Patients
2.4. Concurrent chemoradiotherapy
Two tertiary medical centers participated in this retrospective study. After obtaining institutional review board approval from both the participating institutions, patients were identified from a computerized database of patients with cervical cancer between March 2009 and February 2016. The inclusion criteria were as follows: pathologically proven cervical cancer; a clinical diagnosis of FIGO stage IB2, IIA2, IIB, III, or IVA disease; age N 18 years and b 80 years; para-aortic lymphadenectomy before definitive treatment; pelvic MRI and PET/CT performed within 3 weeks before lymphadenectomy; absence of distant metastatic disease on imaging or physical examination; and no history of chemotherapy or radiotherapy prior to para-aortic lymphadenectomy.
The uterine cervical cancer treatment protocol has been described previously [17]. A brief description is available in the Supplementary data (online only).
2.2. Preoperative assessment Patients received exact staging, including physical examination, chest X-ray, serum squamous cell carcinoma antigen (SCC Ag) analysis, urinalysis, intravenous pyelography, and sigmoidoscopy before paraaortic lymphadenectomy. To identify variables predicting PALN metastasis, the following factors were tested on the basis of previously reported results [8,12–14]: age, body mass index (BMI), parity, FIGO stage, histology, parametrial invasion, tumor size, pelvic lymph node (PLN) metastasis on PET/CT, PALN metastasis on PET/CT, and pretreatment serum SCC Ag content. Tumor size and parametrial invasion were determined by MRI. The settings and conditions of the PET/CT scanning process are described in the Supplementary data (online only). Scans were interpreted based on the criteria of the International Harmonization Project in Lymphoma [15]. The results of MRI and PET/CT were interpreted blindly without knowledge of lymph node metastasis by radiologists and nuclear medicine physicians, respectively. All clinicopathologic and radiologic data were obtained from the medical records of each institution, and no central review was performed.
2.5. Statistical analysis Age, BMI, parity, tumor size, and serum SCC Ag levels were considered continuous variables. FIGO stage, histology, parametrial invasion, and lymph node metastasis determined by imaging studies were considered categorical variables. The histological subtype was classified as squamous cell or non-squamous cell carcinoma. The prediction model was developed as described previously [18]. To construct a robust and well-calibrated model predicting the risk of PALN metastasis, a logistic regression model was built using the entire cohort of 245 patients and internally validated using a bootstrap procedure. A multivariable logistic regression model with penalized maximum likelihood estimation was created to predict PALN metastasis. First, the bivariate relationship between risk factors and PALN metastasis was assessed via logistic regression in the enrolled cohort. Next, the candidate variables with a P-value b 0.5 on univariate analysis were tested by bootstrap resampling, in which a logistic regression model with a backward elimination procedure was performed with 1000 repetitions. The criterion for inclusion of predictors in the final logistic model was a 50% relative frequency of selection by bootstrap resampling. The final model was presented as a score chart to facilitate practical application. The score chart was derived from the multivariable regression coefficients. To score tumor size, the patients were categorized into three groups: ≤4 cm, 4.01–5 cm, and N5 cm. For simple application, the coefficients were divided by the effect of a 1-cm increase in tumor size and rounded. The performance of the model was assessed with respect to discrimination and calibration. To measure discrimination, the concordance index was determined by calculating the area under the receiver operating characteristics curve (AUC). To assess calibration, which means
Please cite this article as: S.-H. Shim, et al., Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer, Gynecol Oncol (2016), http://dx.doi.org/10.1016/j.ygyno.2016.11.011
S.-H. Shim et al. / Gynecologic Oncology xxx (2016) xxx–xxx Table 1 Characteristics of the enrolled patients. Characteristics Age, years BMI, kg/m2 Parity, n FIGO stage, n (%)
Histology, n (%)
Tumor size by MRI, cm Pretreatment SCC Ag, ng/mL PM involvement on MRI, n (%) PLN metastasis on PET/CT, n (%) PALN metastasis on PET/CT, n (%) Harvested PALNs, n PALN metastasis, n (%)
n = 245 Median (range) Median (range) Median (range) IB2 IIA2 IIB III IVA Squamous cell Adenocarcinoma Adenosquamous Median (range) Median (range)
Median (range)
49 (25–77) 22.6 (17.7–31.6) 2 (0–6) 117 (47.8) 23 (9.4) 96 (39.2) 2 (0.8) 7 (2.9) 172 (70.2) 54 (22.0) 19 (7.7) 4.3 (2–10.4) 2.6 (0.5–105) 122 (49.8) 128 (52.2) 23 (9.4) 5 (1–45) 34 (13.9)
BMI, body mass index; FIGO, International Federation of Gynecology and Obstetrics; MRI, magnetic resonance imaging; SCC Ag, squamous cell carcinoma antigen; PM, parametrial; PLN, pelvic lymph node; PET/CT, positron emission tomography/computed tomography; PALN, para-aortic lymph node.
agreement between the predicted and observed probabilities, the Hosmer-Lemeshow test was employed. For internal validation, the concordance index was calculated using 1000 bootstrapping resamples. Bootstrapping resamples were taken from the original data set with replacement of the same size as the original sample. Predictions from each bootstrap model were tested in the original data set. The difference in performance between the bootstrapping and original samples quantifies the optimism that may be expected when the score chart is applied to new, but similar, patients [19]. Finally, we stratified the patients into risk groups using a constructed prediction model. Before constructing the model, risk grouping was defined in accordance with the likelihood ratio. The cutoff for the low-risk group was determined using the highest probability with a negative likelihood ratio (LR−) b 0.2. The cutoff for the high-risk group was determined by the lowest probability with a positive likelihood ratio (LR +) N 5.0. Each value of the cutoff level was obtained by reference to the previous literatures [20,21]. The PALN metastasis rate of each risk-group was assessed. LR− was calculated using the following formula: LR− = (1 − sensitivity)/specificity. LR+ was calculated using the following formula: LR+ = sensitivity / (1 − specificity). Based on Bayes' theorem, LR− and LR+ were converted into negative (i.e., negative predictive value [NPV]) and positive post-test probabilities (i.e., positive predictive value [PPV]) at an appointed prevalence of LN
3
metastasis [22]. All analyses were performed using SPSS version 19.0 (SPSS, Chicago, IL) and R version 3.0.2 (http://cran.r-project.org/ mirrors.html). P b 0.05 was considered significant. 3. Results 3.1. Patient characteristics Among 1172 patients with cervical cancer, 245 patients fulfilled the inclusion criteria (Supplemental Fig. 1, online only). Of these cases, 126 patients received concomitant pelvic lymphadenectomy. Radical hysterectomy was performed in 51 patients with negative results on frozen section of para-aortic lymphadenectomy. Table 1 summarizes the characteristics of the enrolled patients. The median age at diagnosis was 49 years (range, 25–77). The frequency of PALN metastasis was 13.9% (34/245). The median number of metastatic PALN was 5 (range, 1– 36). In total, 16/222 (7.2%) patients with negative PALNs on PET/CT had actual PALN metastasis. Additionally, 11/105 (10.5%) patients with both negative PALNs and positive PLNs on PET/CT had actual PALN metastasis. 3.2. Prediction model for PALN metastasis The results of the logistic regression model for predicting PALN metastasis are shown in Table 2. Univariate analysis revealed that stage, SCC Ag, tumor size on MRI, PLN metastasis on PET/CT, and PALN metastasis on PET/CT were significantly associated with PALN metastasis. After a bootstrap resampling procedure with 1000 repetitions, the final model yielded two statistically significant predictors: tumor size and PALN metastasis on PET/CT. A prediction model was constructed based on this logistic regression model. Based on the multivariable effects presented in Table 2, a simple chart was used to assign a point value to each factor that was derived from its own beta coefficients by regression analysis. PALN metastasis on PET/CT was scored as 0 and 5 points for no and yes, respectively (Table 3). Tumor sizes of ≤ 4 cm, 4.01–5 cm, and N5 cm on MRI were scored as 0, 1, and 3, respectively. Internal validation was performed using the bootstrapping correction technique. After 1000 repetitions, the optimism-corrected concordance index for the score chart was 0.886 (95% confidence interval [CI] = 0.825–0.947, Fig. 1A). Fig. 1B shows the calibration plots of the model. When plotting the probabilities of PALN metastasis predicted by the model against the actual probabilities, the calibration curve lay close to the dashed line. The HosmerLemeshow test yielded a P-value of 0.770 for the model, illustrating that the score chart was well fitted (Supplemental Table 1, online only). In addition, the mean slope of the linear predictor was close to 1 in the bootstrap resampling procedure.
Table 2 Univariate and multivariate logistic regression model for predicting para-aortic lymph node metastasis. Variables
Age, yearsa BMI, kg/m2a Parity, na FIGO stage Histology Pretreatment SCC Ag, ng/mLa Tumor size on MRI, cma PM involvement on MRI PLN metastasis on PET/CT PALN metastasis on PET/CT
Univariate analysis
I, II III, IV Squamous cell Non-squamous
Yes Yes Yes
Multivariate analysis
Odds ratio (95% CI)
P
1.019 (0.987–1.051) 0.997 (0.870–1.142) 1.024 (0.753–1.394) 1 14.857 (3.517–62.768) 1 1.272 (0.701–2.309) 1.049 (1.027–1.072) 2.000 (1.515–2.640) 1.527 (0.732–3.182) 6.562 (2.446–17.603) 46.350 (15.218–141.171)
0.248 0.963 0.878
Odds ratio (95% CI)
P
2.411 (1.707–3.586)
b0.001
79.980 (22.752–346.894)
b0.001
b0.001 0.429 b0.001 b0.001 0.259 b0.001 b0.001
BMI, body mass index; FIGO, International Federation of Gynecology and Obstetrics; SCC Ag, squamous cell carcinoma antigen; MRI, magnetic resonance imaging; PM, parametrial; LN, lymph node; PET/CT, positron emission tomography/computed tomography; CI, confidence interval. a As continuous variable.
Please cite this article as: S.-H. Shim, et al., Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer, Gynecol Oncol (2016), http://dx.doi.org/10.1016/j.ygyno.2016.11.011
4
S.-H. Shim et al. / Gynecologic Oncology xxx (2016) xxx–xxx
Table 3 Scoring system to estimate para-aortic lymph node metastasis status in patients with locally advanced cervical cancer. Variables
Reference Beta value (W)
Tumor size on MRI, cm ≤4 4.01–5 N5 PALN metastasis on No PET/CT Yes
3.30 4.55 6.33 0 1
Beta (W − WREF) / B
0.880 0 1.252 3.032 4.382 0 4.980
Points 0 1 3 0 5
MRI, magnetic resonance imaging; PET/CT, positron emission tomography/computed tomography; Beta, regression coefficient; B, constant for the scoring system, 0.880.
3.3. Identification of patients at low, intermediate, and high risk of PALN metastasis The LR − values with scores of ≤ 1 and ≤ 3 were 0.189 (95% CI = 0.084–0.426) and 0.482 (95% CI = 0.337–0.689), respectively. Thus, the patients with scores of 0 or 1 who were characterized by the absence of PALN metastasis on PET/CT and tumor size ≤ 5 cm on MRI were assigned to the low-risk group (Table 4). The model classified 169/245 patients (69%) into this group. In this group, the predicted probability of PALN metastasis was 2.9%, and the actual metastasis rate was 3.0% (5/169). This LR − can be converted into an NPV of 0.967 (95% CI = 0.930–0.985) based on Bayes' theorem using a PALN metastasis prevalence of 15%. The LR + values for patients with scores of ≥ 3 and ≥ 5 were 3.829 (95% CI = 2.871–5.108) and 22.341 (95% CI = 8.884– 56.186), respectively. Thus, patients with scores of ≥ 5 who were characterized by PALN metastasis on PET/CT regardless of tumor size were assigned to the high-risk group. The model classified 23/245 patients (9%) into this group. In this group, the predicted probability of LN metastasis was 76.2%, and the actual metastasis rate was 78.2% (18/23). This LR + can be converted into a PPV of 0.798 (95% CI = 0.611–0.908) using a PALN metastasis prevalence of 15%. The remaining patients (22%; 53/245) were assigned to the intermediate-risk group. In this group, the predicted probability of LN metastasis was 19.5%, and the actual metastasis rate was 20.8% (11/53).
4. Discussion Information regarding PALN metastasis is important for extending the radiation field in patients with LACC, but its presence is only predictable to a limited extent using conventional imaging modalities. Incorporating the PALN status assessed by PET/CT and tumor size measured by MRI, a prediction model for the individualized risk of PALN metastasis before definitive treatment was constructed and internally validated. The model performance was good in terms of discrimination and calibration. Moreover, it was designed as a user-friendly and simple scoring system, which would enhance its utilization in clinical practice and research design. Published data indicate that PET/CT provides better detection of PALN than MRI or CT [5,7]. However, the false-negative rate of PALN metastasis on PET/CT ranged 5–17% (overall mean, 12%) [5]. If the radiation field is determined solely based on this PET/CT finding (i.e., negative PALN metastasis on PET/CT) without histological analysis of PALN, 12% of patients are undertreated. Conversely, if the radiation field is extended to the para-aortic area without histological analysis, 22% of patients will be overtreated [5]. Therefore, surgical staging should be considered when negative PALN on PET/CT is reported [9,10]. However, the benefits of surgical staging should be balanced with potential morbidity and costs, although this procedure, when performed by trained teams through laparoscopy, appears feasible [23]. In this regard, risk group stratification using our score chart would be clinically and non-invasively helpful for identifying candidates who will benefit from PALN staging surgery. Fig. 2 shows the different clinical scenarios according to the risk group stratification using our current model. The low-risk group was characterized by the absence of PALN metastasis on PET/CT and tumor size ≤ 5 cm on MRI. It is interesting that more than two-thirds (69%; 169/245) of enrolled patients were classified into the low-risk group, and the actual rate of PALN metastasis was only 3.0% among them. Thus, it may be reasonable to omit para-aortic lymphadenectomy in these patients. For the high-risk group, which was characterized by PALN metastasis on PET/CT regardless of tumor size, the actual PALN metastasis rate was 78.2% (18/23). Published studies indicated that the true positive rate of PALN metastasis on PET/CT was 78% (50/64) [5], which is consistent with our findings. Therefore, high-risk patients (i.e., positive PALNs on PET/CT) can be treated with
Fig. 1. Performance of the prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer. (A) After 1000 repetitions, the bootstrap-corrected concordance index for the model was 0.886 (95% confidence interval = 0.825–0.947). (B) Calibration plots of the model. Dashed line: ideal reference value at which the predicted probabilities match the actual probabilities of lymph node metastasis; solid line: performance of the current model; triangular markers: calculations from a subcohort of the present database.
Please cite this article as: S.-H. Shim, et al., Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer, Gynecol Oncol (2016), http://dx.doi.org/10.1016/j.ygyno.2016.11.011
S.-H. Shim et al. / Gynecologic Oncology xxx (2016) xxx–xxx
5
Table 4 Risk estimate and stratification of patients into subgroups. Total points
Number of patients (n = 245)
Number of observed PALNMs (n = 34)
Actual PALNM rate
Predicted probability of PALNM
LR− (95% CI)
0
90
1
1.1%
1.7%
1
79
4
5.1%
4.0%
3
53
11
20.8%
19.5%
5
8
5
62.5%
58.5%
6
8
6
75.0%
77.3%
0.070 (0.010–0.484) 0.189 (0.084–0.426) 0.482 (0.337–0.689) 0.624 (0.478–0.813) 0.669 (0.669–0.942)
8
7
7
100%
95.2%
LR+ (95% CI)
Risk group
Number of patients (n = 245)
Number of observed PALNMs (n = 34)
Actual PALNM rate
Predicted probability of PALNM
Low
169
5
3.0%
2.9%
11
20.8%
19.5%
18
78.2%
76.2%
1.679 (1.475–1.910) 3.829 Intermediate 53 (2.871–5.108) 22.341 High 23 (8.884–56.186) 40.338 (9.519–170.932)
PALNM, para-aortic lymph node metastasis; LR−, negative likelihood ratio; LR+, positive likelihood ratio; CI, confidence interval.
extended-field radiotherapy including the para-aortic area without histological confirmation of PALN, as additional information is seldom provided by lymphadenectomy [5]. Meanwhile, in the intermediate-risk group, which was characterized by negative PALNs on PET/CT and tumor size N 5 cm on MRI, the false-negative rate of PALN metastasis was 21% (11/53). Therefore, staging surgery could be indicated in the intermediate-risk group to avert mismanagement owing to a high falsenegative rate. The survival effect of para-aortic lymphadenectomy in LACC patients treated with CCRT has been investigated previously in 685 patients from three phase 3 trials [24]. In that analysis, surgical staging was independently associated with better prognosis for both overall and progression free survival compared with radiological staging. These data suggested a survival benefit of surgical excision of metastatic PALNs in LACC patients treated with CCRT. In the era of PET/CT, a recent prospective multicenter study conducted survival analysis of 237 LACC patients with a negative PALN result on PET/CT who underwent PALN staging surgery followed by CCRT [8]. The event-free survival rates at 3 years in the patients without PALN metastasis and with pathologically proven PALN metastasis
measuring ≤5 mm were similar (74% vs 69%). This suggests a possible survival impact of staging surgery in a subset of patients. In this context, paraaortic lymphadenectomy may yield a survival benefit in a subset of intermediate risk patients (i.e., positive PALN in an intraoperative frozen section). As aforementioned, the reported false-negative rate of PALN metastasis on PET/CT was 12% [5]. Meanwhile, Ramirez et al. reported that in patients with positive PLNs but negative PALNs on PET/CT, the falsenegative rate (i.e., pathologically confirmed PALN metastasis) increased to 22% [10]. This means that considering both the PLN status and PALN status on PET/CT could identify more patients who would benefit from surgical staging compared to considering solely the PALN status on PET/CT. Thus, they suggested that laparoscopic para-aortic lymphadenectomy should be discussed for patients with positive PLNs but negative PALNs on PET/CT. In our present study, of 105 patients with positive PLNs but negative PALNs on PET/CT, 11 (11%) displayed pathologically confirmed PALN metastasis. Meanwhile, of 53 patients in the intermediate-risk group, 11 (21%) had pathologically confirmed PALN metastasis. Therefore, using our current classification, a larger number
Fig. 2. Clinical scenario according to risk-group stratification using the current prediction model. RT, radiation therapy; PALN, para-aortic lymph node.
Please cite this article as: S.-H. Shim, et al., Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer, Gynecol Oncol (2016), http://dx.doi.org/10.1016/j.ygyno.2016.11.011
6
S.-H. Shim et al. / Gynecologic Oncology xxx (2016) xxx–xxx
of patients can be excluded from surgical staging safely and a larger proportion of patients could benefit from surgical staging compared with previous suggestions [10]. Moreover, in the low-risk group stratified by our model, the positive PLN rate on PET/CT was 43% (73/169). Of these 73 patients who were defined as low-risk by our model and also showed PLN positivity on PET/CT, true PALN metastasis was observed in only 4 patients (5%). Hence, if we performed PALN staging surgery for all low-risk patients with positive PLN on PET/CT, 95% (69 out of 73) would have undergone unnecessary surgery. We conclude that para-aortic lymphadenectomy is not required for such low-risk group patients, even those with a positive PLN on PET/CT. The current study has several limitations. First, our study excluded patients with LACC who did not undergo para-aortic lymphadenectomy before definite treatment because of its retrospective nature. Considering the tendency of physicians to omit staging surgery in patients with suspected PALN metastasis on PET/CT, our study population might have not had included patients with more advanced disease. This selection bias may have affected the performance of the model. Second, although we performed a rigorous internal validation via bootstrap resampling, the model requires external validation using independent data sets from other institutions to confirm its generalizability. Third, a central review was not performed in current study, thus causing measurement errors. However, this trade-off may have enhanced the generalizability of the model. In addition, known factors associated with LN metastasis, such as the genotype of human papillomavirus and lymphovascular space invasion, were not tested in the current study [25,26]. Finally, the current prediction model may not be useful in low-resource settings because PET/CT is not available as a pretreatment evaluation tool for cervical cancer. In conclusion, we constructed a robust model incorporating two preoperative variables to predict PALN metastasis in patients with LACC. Using the score chart, patients were stratified into low-, intermediate-, and high-risk groups regarding PALN metastasis. Our model may be useful in deciding whether nodal-staging surgery should be performed before definitive CCRT and in designing clinical trials. A prospective validation study using independent populations will be conducted to evaluate the performance of the model. Supplementary data to this article can be found online at doi:10. 1016/j.ygyno.2016.11.011. Disclosures The authors have no conflicts of interest or financial ties to disclose. Acknowledgments This work was supported by Konkuk University Medical Center Research Grant 2015 (Grant No. 201504). References [1] L.R. Medeiros, D.D. Rosa, M.C. Bozzetti, J.M. Fachel, S. Furness, R. Garry, et al., Laparoscopy versus laparotomy for benign ovarian tumour, Cochrane Database Syst. Rev. (2009), CD004751. [2] K.J. Min, Y.J. Lee, M. Suh, C.W. Yoo, M.C. Lim, J. Choi, et al., The Korean guideline for cervical cancer screening, J. Gynecol. Oncol. 26 (2015) 232–239. [3] H. Kim, J.Y. Kim, J. Kim, W. Park, Y.S. Kim, H.J. Kim, et al., Current status of brachytherapy in Korea: a national survey of radiation oncologists, J. Gynecol. Oncol. 27 (2016), e33. [4] N. Colombo, S. Carinelli, A. Colombo, C. Marini, D. Rollo, C. Sessa, et al., Cervical cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up, Ann. Oncol. 23 (Suppl. 7) (2012) vii27–vii32.
[5] S. Gouy, P. Morice, F. Narducci, C. Uzan, J. Gilmore, H. Kolesnikov-Gauthier, et al., Nodal-staging surgery for locally advanced cervical cancer in the era of PET, Lancet Oncol. 13 (2012) e212–e220. [6] H.J. Choi, W. Ju, S.K. Myung, Y. Kim, Diagnostic performance of computer tomography, magnetic resonance imaging, and positron emission tomography or positron emission tomography/computer tomography for detection of metastatic lymph nodes in patients with cervical cancer: meta-analysis, Cancer Sci. 101 (2010) 1471–1479. [7] S. Kang, S.K. Kim, D.C. Chung, S.S. Seo, J.Y. Kim, B.H. Nam, et al., Diagnostic value of (18)F-FDG PET for evaluation of paraaortic nodal metastasis in patients with cervical carcinoma: a metaanalysis, J. Nucl. Med. 51 (2010) 360–367. [8] S. Gouy, P. Morice, F. Narducci, C. Uzan, A. Martinez, A. Rey, et al., Prospective multicenter study evaluating the survival of patients with locally advanced cervical cancer undergoing laparoscopic para-aortic lymphadenectomy before chemoradiotherapy in the era of positron emission tomography imaging, J. Clin. Oncol. 31 (2013) 3026–3033. [9] E. Leblanc, H. Gauthier, D. Querleu, G. Ferron, S. Zerdoud, P. Morice, et al., Accuracy of 18-fluoro-2-deoxy-D-glucose positron emission tomography in the pretherapeutic detection of occult para-aortic node involvement in patients with a locally advanced cervical carcinoma, Ann. Surg. Oncol. 18 (2011) 2302–2309. [10] P.T. Ramirez, A. Jhingran, H.A. Macapinlac, E.D. Euscher, M.F. Munsell, R.L. Coleman, et al., Laparoscopic extraperitoneal para-aortic lymphadenectomy in locally advanced cervical cancer: a prospective correlation of surgical findings with positron emission tomography/computed tomography findings, Cancer 117 (2011) 1928–1934. [11] E. Brockbank, F. Kokka, A. Bryant, C. Pomel, K. Reynolds, Pre-treatment surgical paraaortic lymph node assessment in locally advanced cervical cancer, Cochrane Database Syst. Rev. 3 (2013), CD008217. [12] S. Kodama, K. Kanazawa, S. Honma, K. Tanaka, Age as a prognostic factor in patients with squamous cell carcinoma of the uterine cervix, Cancer 68 (1991) 2481–2485. [13] M. Takeda, N. Sakuragi, K. Okamoto, Y. Todo, S. Minobe, E. Nomura, et al., Preoperative serum SCC, CA125, and CA19-9 levels and lymph node status in squamous cell carcinoma of the uterine cervix, Acta Obstet. Gynecol. Scand. 81 (2002) 451–457. [14] R. Grigiene, K.P. Valuckas, E. Aleknavicius, J. Kurtinaitis, S.R. Letautiene, The value of prognostic factors for uterine cervical cancer patients treated with irradiation alone, BMC Cancer 7 (2007) 234. [15] M.E. Juweid, S. Stroobants, O.S. Hoekstra, F.M. Mottaghy, M. Dietlein, A. Guermazi, et al., Use of positron emission tomography for response assessment of lymphoma: consensus of the Imaging Subcommittee of International Harmonization Project in Lymphoma, J. Clin. Oncol. 25 (2007) 571–578. [16] J.Y. Park, J.H. Nam, Laparotomy conversion rate of laparoscopic radical hysterectomy for early-stage cervical cancer in a consecutive series without case selection, Ann. Surg. Oncol. 21 (2014) 3030–3035. [17] Y.S. Kim, S.S. Shin, J.H. Nam, Y.T. Kim, Y.M. Kim, J.H. Kim, et al., Prospective randomized comparison of monthly fluorouracil and cisplatin versus weekly cisplatin concurrent with pelvic radiotherapy and high-dose rate brachytherapy for locally advanced cervical cancer, Gynecol. Oncol. 108 (2008) 195–200. [18] S.H. Shim, S.J. Lee, S.O. Kim, S.N. Kim, D.Y. Kim, J.J. Lee, et al., Nomogram for predicting incomplete cytoreduction in advanced ovarian cancer patients, Gynecol. Oncol. 136 (2015) 30–36. [19] E.W. Steyerberg, B.A. Neville, L.B. Koppert, V.E. Lemmens, H.W. Tilanus, J.W. Coebergh, et al., Surgical mortality in patients with esophageal cancer: development and validation of a simple risk score, J. Clin. Oncol. 24 (2006) 4277–4284. [20] J.J. Deeks, D.G. Altman, Diagnostic tests 4: likelihood ratios, BMJ 329 (2004) 168–169. [21] S. Kang, J.M. Lee, J.K. Lee, J.W. Kim, C.H. Cho, S.M. Kim, et al., A Web-based nomogram predicting para-aortic nodal metastasis in incompletely staged patients with endometrial cancer: a Korean Multicenter Study, Int. J. Gynecol. Cancer 24 (2014) 513–519. [22] T. Bayes, An essay towards solving a problem in the doctrine of chances. 1763, MD Comput. 8 (1991) 157–171. [23] V. Benito, A. Lubrano, O. Arencibia, M. Andujar, B. Pinar, N. Medina, et al., Laparoscopic extraperitoneal para-aortic lymphadenectomy in the staging of locally advanced cervical cancer: is it a feasible procedure at a peripheral center? Int. J. Gynecol. Cancer 22 (2012) 332–336. [24] M.A. Gold, C. Tian, C.W. Whitney, P.G. Rose, R. Lanciano, Surgical versus radiographic determination of para-aortic lymph node metastases before chemoradiation for locally advanced cervical carcinoma: a Gynecologic Oncology Group Study, Cancer 112 (2008) 1954–1963. [25] G.G. Garzetti, A. Ciavattini, G. Lucarini, G. Goteri, S. Menso, M. De Nictolis, et al., The role of human papillomavirus DNAs in cervical carcinoma and risk of lymph node metastasis: association with 72-kilodalton metalloproteinase immunostaining, Cancer 82 (1998) 886–892. [26] J.R. van Nagell Jr., E.S. Donaldson, E.G. Wood, J.C. Parker Jr., The significance of vascular invasion and lymphocytic infiltration in invasive cervical cancer, Cancer 41 (1978) 228–234.
Please cite this article as: S.-H. Shim, et al., Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer, Gynecol Oncol (2016), http://dx.doi.org/10.1016/j.ygyno.2016.11.011
Contents lists available at ScienceDirect
Gynecologic Oncology journal homepage: www.elsevier.com/locate/ygyno
Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer Seung-Hyuk Shim, MD a, Dae-Yeon Kim, MD, PhD b,⁎,1, Sun Joo Lee, MD, PhD a, Soo-Nyung Kim, MD, PhD a, Soon-Beom Kang, MD, PhD a,⁎⁎,1, Shin-Wha Lee, MD, PhD b, Jeong-Yeol Park, MD, PhD b, Dae-Shik Suh, MD, PhD b, Jong-Hyeok Kim, MD, PhD b, Yong-Man Kim, MD, PhD b, Young-Tak Kim, MD, PhD b, Joo-Hyun Nam, MD, PhD b a b
Department of Obstetrics and Gynecology, Konkuk University School of Medicine, Seoul, Republic of Korea Department of Obstetrics and Gynecology, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Republic of Korea
H I G H L I G H T S • 245 consecutive LACC patients undergoing para-aortic lymphadenectomy before definitive treatment were analyzed. • Using tumor size and PET/CT features, a risk prediction model for predicting PALN metastasis was developed. • The model displayed good discrimination and calibration (concordance index = 0.886; 95% confidence interval = 0.825–0.947).
a r t i c l e
i n f o
Article history: Received 5 September 2016 Received in revised form 1 November 2016 Accepted 5 November 2016 Available online xxxx Keywords: Cervical cancer Likelihood functions Lymphatic metastasis Lymph node excision Morbidity
a b s t r a c t Objective. Concurrent chemoradiotherapy is usually administered to patients with locally advanced cervical cancer (LACC). Extended-field chemoradiotherapy is required if para-aortic lymph node (PALN) metastasis is detected. This study aimed to construct a prediction model for PALN metastasis in patients with LACC before definitive treatment. Methods. Between 2009 and 2016, all consecutive patients with LACC who underwent para-aortic lymphadenectomy at two tertiary centers were retrospectively analyzed. A multivariate logistic model was constructed, from which a prediction model for PALN metastasis was developed and internally validated. Before analysis, risk grouping was predefined based on the likelihood ratio. Results. In total, 245 patients satisfied the eligibility criteria. Thirty-four patients (13.9%) had pathologically proven PALN metastases. Additionally, 16/222 (7.2%) patients with negative PALNs on positron emission tomography/computed tomography (PET/CT) had PALN metastasis. Moreover, 11/105 (10.5%) patients with both negative PALNs and positive pelvic lymph nodes on PET/CT had PALN metastasis. Tumor size on magnetic resonance imaging and PALN status on PET/CT were independent predictors of PALN metastasis. The model incorporating these two predictors displayed good discrimination and calibration (bootstrap-corrected concordance index = 0.886; 95% confidence interval = 0.825–0.947). The model categorized 169 (69%), 52 (22%), and 23 (9%) patients into low-, intermediate-, and high-risk groups, respectively. The predicted probabilities of PALN metastasis for these groups were 2.9, 20.8, and 76.2%, respectively. Conclusion. We constructed a robust model predicting PALN metastasis in patients with LACC that may improve clinical trial design and help clinicians determine whether nodal-staging surgery should be performed. © 2016 Elsevier Inc. All rights reserved.
⁎ Correspondence to: Dae-Yeon Kim, MD, Department of Obstetrics and Gynecology, University of Ulsan College of Medicine, Asan Medical Center, 88 Olympic-ro 43 gil, Songpa-gu, Seoul 05505, Republic of Korea. ⁎⁎ Correspondence to: S.-B. Kang, Department of Obstetrics and Gynecology, Konkuk University School of Medicine, 263 Achasan-ro, Gwangjin-gu, Seoul 05030, Republic of Korea. E-mail addresses: [email protected] (D.-Y. Kim), [email protected] (S.-B. Kang). 1 Soon-Beom Kang and Dae-Yeon Kim contributed equally to this work and should be considered as co-corresponding authors.
1. Introduction Since cytological screening was introduced, the incidence of cervical cancer has decreased remarkably. However, this disease remains a sizeable health problem that accounted for an estimated 528,000 new cases and for 266,000 deaths worldwide in 2012 [1]. In Korea, it is the most common female genital malignancy, and the age-standardized incidence rate is 9.5 per 100,000 persons in 2013 [2].
http://dx.doi.org/10.1016/j.ygyno.2016.11.011 0090-8258/© 2016 Elsevier Inc. All rights reserved.
Please cite this article as: S.-H. Shim, et al., Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer, Gynecol Oncol (2016), http://dx.doi.org/10.1016/j.ygyno.2016.11.011
2
S.-H. Shim et al. / Gynecologic Oncology xxx (2016) xxx–xxx
Based on five phase III randomized trials demonstrating that concurrent chemoradiotherapy (CCRT) improves overall survival in patients with locally advanced cervical cancer (LACC), the current guidelines recommend CCRT as the standard treatment for these patients [3,4]. Nodal metastasis is among the most important prognostic factors for survival in patients with LACC. In patients with para-aortic lymph node (PALN) metastasis, extended-field radiotherapy should be considered [5]. Accurate pretreatment evaluation of PALN involvement is therefore of paramount importance in selecting radiation fields. The use of fluorine-18 fluorodeoxyglucose positron emission tomography/computed tomography (PET/CT) may aid in detecting extrapelvic disease compared with conventional imaging modalities such as pelvic magnetic resonance imaging (MRI) or CT [6,7]. Although PET is valuable for detecting extrapelvic disease, the reported false-negative rate of PALN metastasis on PET/CT ranged 9–22%, which is ascribable to small-volume metastases [5,7]. Consequently, nodal-staging surgery is possibly beneficial for patients with negative PALN involvement on PET/CT [8–10]. However, considering potential morbidity and the cost of staging surgery, nodal-staging surgery needs to be individualized in these patients [11]. Thus, it would be useful to develop an individualized prediction model for PALN metastasis before definitive treatment. Furthermore, individualized risk stratification of PALN metastasis may be advantageous in clinical trial design. Thus, this study aimed to develop and internally validate a prediction model for PALN metastasis in patients with LACC and identify potential candidates for PALN staging surgery using the model.
2.3. Para-aortic lymphadenectomy
2. Methods
In both institutions, the indications for para-aortic lymphadenectomy are as follows: a clinical diagnosis of FIGO stage IB2, IIA2, IIB, III, or IVA disease; age N 18 years and b80 years; and the absence of distant metastatic disease on imaging or physical examination. Para-aortic lymphadenectomy was performed as previously described [16]. Briefly, para-aortic lymphadenectomy was performed via laparoscopy or laparotomy. All of lymphatic tissue from the aorta, aortocaval space, and vena cava were completely removed. The boundaries of the para-aortic lymphadenectomy were defined as the bifurcation of the aorta caudally and the inferior mesenteric artery cranially. If the PALN at this level was enlarged or positive for metastasis, PALN dissection was extended to the left renal vein level. All harvested lymph nodes were grouped according to the name of the adjacent vessel (venacaval, aortocaval, aortic below the inferior mesenteric artery, and aortic above the inferior mesenteric artery). Each lymph node was sliced perpendicular to its long axis, stained with hematoxylin and eosin, and examined microscopically by a pathologist. Each pathology report included the number of lymph nodes retrieved from each area, the presence or absence of metastases, the largest size of lymph node metastases, and the presence or absence of extracapsular extension. Pelvic lymphadenectomy was not routinely performed as part of surgical nodal staging because the pelvic area is covered by the conventional radiation field. However, pelvic lymphadenectomy was performed to remove enlarged or suspicious pelvic nodes at the surgeon's discretion. In both institutions, for patients with negative results on frozen section of para-aortic lymphadenectomy, radical hysterectomy was considered in selected younger patients (especially those under 45 years old) and FIGO stage IB2 or IIA2.
2.1. Patients
2.4. Concurrent chemoradiotherapy
Two tertiary medical centers participated in this retrospective study. After obtaining institutional review board approval from both the participating institutions, patients were identified from a computerized database of patients with cervical cancer between March 2009 and February 2016. The inclusion criteria were as follows: pathologically proven cervical cancer; a clinical diagnosis of FIGO stage IB2, IIA2, IIB, III, or IVA disease; age N 18 years and b 80 years; para-aortic lymphadenectomy before definitive treatment; pelvic MRI and PET/CT performed within 3 weeks before lymphadenectomy; absence of distant metastatic disease on imaging or physical examination; and no history of chemotherapy or radiotherapy prior to para-aortic lymphadenectomy.
The uterine cervical cancer treatment protocol has been described previously [17]. A brief description is available in the Supplementary data (online only).
2.2. Preoperative assessment Patients received exact staging, including physical examination, chest X-ray, serum squamous cell carcinoma antigen (SCC Ag) analysis, urinalysis, intravenous pyelography, and sigmoidoscopy before paraaortic lymphadenectomy. To identify variables predicting PALN metastasis, the following factors were tested on the basis of previously reported results [8,12–14]: age, body mass index (BMI), parity, FIGO stage, histology, parametrial invasion, tumor size, pelvic lymph node (PLN) metastasis on PET/CT, PALN metastasis on PET/CT, and pretreatment serum SCC Ag content. Tumor size and parametrial invasion were determined by MRI. The settings and conditions of the PET/CT scanning process are described in the Supplementary data (online only). Scans were interpreted based on the criteria of the International Harmonization Project in Lymphoma [15]. The results of MRI and PET/CT were interpreted blindly without knowledge of lymph node metastasis by radiologists and nuclear medicine physicians, respectively. All clinicopathologic and radiologic data were obtained from the medical records of each institution, and no central review was performed.
2.5. Statistical analysis Age, BMI, parity, tumor size, and serum SCC Ag levels were considered continuous variables. FIGO stage, histology, parametrial invasion, and lymph node metastasis determined by imaging studies were considered categorical variables. The histological subtype was classified as squamous cell or non-squamous cell carcinoma. The prediction model was developed as described previously [18]. To construct a robust and well-calibrated model predicting the risk of PALN metastasis, a logistic regression model was built using the entire cohort of 245 patients and internally validated using a bootstrap procedure. A multivariable logistic regression model with penalized maximum likelihood estimation was created to predict PALN metastasis. First, the bivariate relationship between risk factors and PALN metastasis was assessed via logistic regression in the enrolled cohort. Next, the candidate variables with a P-value b 0.5 on univariate analysis were tested by bootstrap resampling, in which a logistic regression model with a backward elimination procedure was performed with 1000 repetitions. The criterion for inclusion of predictors in the final logistic model was a 50% relative frequency of selection by bootstrap resampling. The final model was presented as a score chart to facilitate practical application. The score chart was derived from the multivariable regression coefficients. To score tumor size, the patients were categorized into three groups: ≤4 cm, 4.01–5 cm, and N5 cm. For simple application, the coefficients were divided by the effect of a 1-cm increase in tumor size and rounded. The performance of the model was assessed with respect to discrimination and calibration. To measure discrimination, the concordance index was determined by calculating the area under the receiver operating characteristics curve (AUC). To assess calibration, which means
Please cite this article as: S.-H. Shim, et al., Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer, Gynecol Oncol (2016), http://dx.doi.org/10.1016/j.ygyno.2016.11.011
S.-H. Shim et al. / Gynecologic Oncology xxx (2016) xxx–xxx Table 1 Characteristics of the enrolled patients. Characteristics Age, years BMI, kg/m2 Parity, n FIGO stage, n (%)
Histology, n (%)
Tumor size by MRI, cm Pretreatment SCC Ag, ng/mL PM involvement on MRI, n (%) PLN metastasis on PET/CT, n (%) PALN metastasis on PET/CT, n (%) Harvested PALNs, n PALN metastasis, n (%)
n = 245 Median (range) Median (range) Median (range) IB2 IIA2 IIB III IVA Squamous cell Adenocarcinoma Adenosquamous Median (range) Median (range)
Median (range)
49 (25–77) 22.6 (17.7–31.6) 2 (0–6) 117 (47.8) 23 (9.4) 96 (39.2) 2 (0.8) 7 (2.9) 172 (70.2) 54 (22.0) 19 (7.7) 4.3 (2–10.4) 2.6 (0.5–105) 122 (49.8) 128 (52.2) 23 (9.4) 5 (1–45) 34 (13.9)
BMI, body mass index; FIGO, International Federation of Gynecology and Obstetrics; MRI, magnetic resonance imaging; SCC Ag, squamous cell carcinoma antigen; PM, parametrial; PLN, pelvic lymph node; PET/CT, positron emission tomography/computed tomography; PALN, para-aortic lymph node.
agreement between the predicted and observed probabilities, the Hosmer-Lemeshow test was employed. For internal validation, the concordance index was calculated using 1000 bootstrapping resamples. Bootstrapping resamples were taken from the original data set with replacement of the same size as the original sample. Predictions from each bootstrap model were tested in the original data set. The difference in performance between the bootstrapping and original samples quantifies the optimism that may be expected when the score chart is applied to new, but similar, patients [19]. Finally, we stratified the patients into risk groups using a constructed prediction model. Before constructing the model, risk grouping was defined in accordance with the likelihood ratio. The cutoff for the low-risk group was determined using the highest probability with a negative likelihood ratio (LR−) b 0.2. The cutoff for the high-risk group was determined by the lowest probability with a positive likelihood ratio (LR +) N 5.0. Each value of the cutoff level was obtained by reference to the previous literatures [20,21]. The PALN metastasis rate of each risk-group was assessed. LR− was calculated using the following formula: LR− = (1 − sensitivity)/specificity. LR+ was calculated using the following formula: LR+ = sensitivity / (1 − specificity). Based on Bayes' theorem, LR− and LR+ were converted into negative (i.e., negative predictive value [NPV]) and positive post-test probabilities (i.e., positive predictive value [PPV]) at an appointed prevalence of LN
3
metastasis [22]. All analyses were performed using SPSS version 19.0 (SPSS, Chicago, IL) and R version 3.0.2 (http://cran.r-project.org/ mirrors.html). P b 0.05 was considered significant. 3. Results 3.1. Patient characteristics Among 1172 patients with cervical cancer, 245 patients fulfilled the inclusion criteria (Supplemental Fig. 1, online only). Of these cases, 126 patients received concomitant pelvic lymphadenectomy. Radical hysterectomy was performed in 51 patients with negative results on frozen section of para-aortic lymphadenectomy. Table 1 summarizes the characteristics of the enrolled patients. The median age at diagnosis was 49 years (range, 25–77). The frequency of PALN metastasis was 13.9% (34/245). The median number of metastatic PALN was 5 (range, 1– 36). In total, 16/222 (7.2%) patients with negative PALNs on PET/CT had actual PALN metastasis. Additionally, 11/105 (10.5%) patients with both negative PALNs and positive PLNs on PET/CT had actual PALN metastasis. 3.2. Prediction model for PALN metastasis The results of the logistic regression model for predicting PALN metastasis are shown in Table 2. Univariate analysis revealed that stage, SCC Ag, tumor size on MRI, PLN metastasis on PET/CT, and PALN metastasis on PET/CT were significantly associated with PALN metastasis. After a bootstrap resampling procedure with 1000 repetitions, the final model yielded two statistically significant predictors: tumor size and PALN metastasis on PET/CT. A prediction model was constructed based on this logistic regression model. Based on the multivariable effects presented in Table 2, a simple chart was used to assign a point value to each factor that was derived from its own beta coefficients by regression analysis. PALN metastasis on PET/CT was scored as 0 and 5 points for no and yes, respectively (Table 3). Tumor sizes of ≤ 4 cm, 4.01–5 cm, and N5 cm on MRI were scored as 0, 1, and 3, respectively. Internal validation was performed using the bootstrapping correction technique. After 1000 repetitions, the optimism-corrected concordance index for the score chart was 0.886 (95% confidence interval [CI] = 0.825–0.947, Fig. 1A). Fig. 1B shows the calibration plots of the model. When plotting the probabilities of PALN metastasis predicted by the model against the actual probabilities, the calibration curve lay close to the dashed line. The HosmerLemeshow test yielded a P-value of 0.770 for the model, illustrating that the score chart was well fitted (Supplemental Table 1, online only). In addition, the mean slope of the linear predictor was close to 1 in the bootstrap resampling procedure.
Table 2 Univariate and multivariate logistic regression model for predicting para-aortic lymph node metastasis. Variables
Age, yearsa BMI, kg/m2a Parity, na FIGO stage Histology Pretreatment SCC Ag, ng/mLa Tumor size on MRI, cma PM involvement on MRI PLN metastasis on PET/CT PALN metastasis on PET/CT
Univariate analysis
I, II III, IV Squamous cell Non-squamous
Yes Yes Yes
Multivariate analysis
Odds ratio (95% CI)
P
1.019 (0.987–1.051) 0.997 (0.870–1.142) 1.024 (0.753–1.394) 1 14.857 (3.517–62.768) 1 1.272 (0.701–2.309) 1.049 (1.027–1.072) 2.000 (1.515–2.640) 1.527 (0.732–3.182) 6.562 (2.446–17.603) 46.350 (15.218–141.171)
0.248 0.963 0.878
Odds ratio (95% CI)
P
2.411 (1.707–3.586)
b0.001
79.980 (22.752–346.894)
b0.001
b0.001 0.429 b0.001 b0.001 0.259 b0.001 b0.001
BMI, body mass index; FIGO, International Federation of Gynecology and Obstetrics; SCC Ag, squamous cell carcinoma antigen; MRI, magnetic resonance imaging; PM, parametrial; LN, lymph node; PET/CT, positron emission tomography/computed tomography; CI, confidence interval. a As continuous variable.
Please cite this article as: S.-H. Shim, et al., Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer, Gynecol Oncol (2016), http://dx.doi.org/10.1016/j.ygyno.2016.11.011
4
S.-H. Shim et al. / Gynecologic Oncology xxx (2016) xxx–xxx
Table 3 Scoring system to estimate para-aortic lymph node metastasis status in patients with locally advanced cervical cancer. Variables
Reference Beta value (W)
Tumor size on MRI, cm ≤4 4.01–5 N5 PALN metastasis on No PET/CT Yes
3.30 4.55 6.33 0 1
Beta (W − WREF) / B
0.880 0 1.252 3.032 4.382 0 4.980
Points 0 1 3 0 5
MRI, magnetic resonance imaging; PET/CT, positron emission tomography/computed tomography; Beta, regression coefficient; B, constant for the scoring system, 0.880.
3.3. Identification of patients at low, intermediate, and high risk of PALN metastasis The LR − values with scores of ≤ 1 and ≤ 3 were 0.189 (95% CI = 0.084–0.426) and 0.482 (95% CI = 0.337–0.689), respectively. Thus, the patients with scores of 0 or 1 who were characterized by the absence of PALN metastasis on PET/CT and tumor size ≤ 5 cm on MRI were assigned to the low-risk group (Table 4). The model classified 169/245 patients (69%) into this group. In this group, the predicted probability of PALN metastasis was 2.9%, and the actual metastasis rate was 3.0% (5/169). This LR − can be converted into an NPV of 0.967 (95% CI = 0.930–0.985) based on Bayes' theorem using a PALN metastasis prevalence of 15%. The LR + values for patients with scores of ≥ 3 and ≥ 5 were 3.829 (95% CI = 2.871–5.108) and 22.341 (95% CI = 8.884– 56.186), respectively. Thus, patients with scores of ≥ 5 who were characterized by PALN metastasis on PET/CT regardless of tumor size were assigned to the high-risk group. The model classified 23/245 patients (9%) into this group. In this group, the predicted probability of LN metastasis was 76.2%, and the actual metastasis rate was 78.2% (18/23). This LR + can be converted into a PPV of 0.798 (95% CI = 0.611–0.908) using a PALN metastasis prevalence of 15%. The remaining patients (22%; 53/245) were assigned to the intermediate-risk group. In this group, the predicted probability of LN metastasis was 19.5%, and the actual metastasis rate was 20.8% (11/53).
4. Discussion Information regarding PALN metastasis is important for extending the radiation field in patients with LACC, but its presence is only predictable to a limited extent using conventional imaging modalities. Incorporating the PALN status assessed by PET/CT and tumor size measured by MRI, a prediction model for the individualized risk of PALN metastasis before definitive treatment was constructed and internally validated. The model performance was good in terms of discrimination and calibration. Moreover, it was designed as a user-friendly and simple scoring system, which would enhance its utilization in clinical practice and research design. Published data indicate that PET/CT provides better detection of PALN than MRI or CT [5,7]. However, the false-negative rate of PALN metastasis on PET/CT ranged 5–17% (overall mean, 12%) [5]. If the radiation field is determined solely based on this PET/CT finding (i.e., negative PALN metastasis on PET/CT) without histological analysis of PALN, 12% of patients are undertreated. Conversely, if the radiation field is extended to the para-aortic area without histological analysis, 22% of patients will be overtreated [5]. Therefore, surgical staging should be considered when negative PALN on PET/CT is reported [9,10]. However, the benefits of surgical staging should be balanced with potential morbidity and costs, although this procedure, when performed by trained teams through laparoscopy, appears feasible [23]. In this regard, risk group stratification using our score chart would be clinically and non-invasively helpful for identifying candidates who will benefit from PALN staging surgery. Fig. 2 shows the different clinical scenarios according to the risk group stratification using our current model. The low-risk group was characterized by the absence of PALN metastasis on PET/CT and tumor size ≤ 5 cm on MRI. It is interesting that more than two-thirds (69%; 169/245) of enrolled patients were classified into the low-risk group, and the actual rate of PALN metastasis was only 3.0% among them. Thus, it may be reasonable to omit para-aortic lymphadenectomy in these patients. For the high-risk group, which was characterized by PALN metastasis on PET/CT regardless of tumor size, the actual PALN metastasis rate was 78.2% (18/23). Published studies indicated that the true positive rate of PALN metastasis on PET/CT was 78% (50/64) [5], which is consistent with our findings. Therefore, high-risk patients (i.e., positive PALNs on PET/CT) can be treated with
Fig. 1. Performance of the prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer. (A) After 1000 repetitions, the bootstrap-corrected concordance index for the model was 0.886 (95% confidence interval = 0.825–0.947). (B) Calibration plots of the model. Dashed line: ideal reference value at which the predicted probabilities match the actual probabilities of lymph node metastasis; solid line: performance of the current model; triangular markers: calculations from a subcohort of the present database.
Please cite this article as: S.-H. Shim, et al., Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer, Gynecol Oncol (2016), http://dx.doi.org/10.1016/j.ygyno.2016.11.011
S.-H. Shim et al. / Gynecologic Oncology xxx (2016) xxx–xxx
5
Table 4 Risk estimate and stratification of patients into subgroups. Total points
Number of patients (n = 245)
Number of observed PALNMs (n = 34)
Actual PALNM rate
Predicted probability of PALNM
LR− (95% CI)
0
90
1
1.1%
1.7%
1
79
4
5.1%
4.0%
3
53
11
20.8%
19.5%
5
8
5
62.5%
58.5%
6
8
6
75.0%
77.3%
0.070 (0.010–0.484) 0.189 (0.084–0.426) 0.482 (0.337–0.689) 0.624 (0.478–0.813) 0.669 (0.669–0.942)
8
7
7
100%
95.2%
LR+ (95% CI)
Risk group
Number of patients (n = 245)
Number of observed PALNMs (n = 34)
Actual PALNM rate
Predicted probability of PALNM
Low
169
5
3.0%
2.9%
11
20.8%
19.5%
18
78.2%
76.2%
1.679 (1.475–1.910) 3.829 Intermediate 53 (2.871–5.108) 22.341 High 23 (8.884–56.186) 40.338 (9.519–170.932)
PALNM, para-aortic lymph node metastasis; LR−, negative likelihood ratio; LR+, positive likelihood ratio; CI, confidence interval.
extended-field radiotherapy including the para-aortic area without histological confirmation of PALN, as additional information is seldom provided by lymphadenectomy [5]. Meanwhile, in the intermediate-risk group, which was characterized by negative PALNs on PET/CT and tumor size N 5 cm on MRI, the false-negative rate of PALN metastasis was 21% (11/53). Therefore, staging surgery could be indicated in the intermediate-risk group to avert mismanagement owing to a high falsenegative rate. The survival effect of para-aortic lymphadenectomy in LACC patients treated with CCRT has been investigated previously in 685 patients from three phase 3 trials [24]. In that analysis, surgical staging was independently associated with better prognosis for both overall and progression free survival compared with radiological staging. These data suggested a survival benefit of surgical excision of metastatic PALNs in LACC patients treated with CCRT. In the era of PET/CT, a recent prospective multicenter study conducted survival analysis of 237 LACC patients with a negative PALN result on PET/CT who underwent PALN staging surgery followed by CCRT [8]. The event-free survival rates at 3 years in the patients without PALN metastasis and with pathologically proven PALN metastasis
measuring ≤5 mm were similar (74% vs 69%). This suggests a possible survival impact of staging surgery in a subset of patients. In this context, paraaortic lymphadenectomy may yield a survival benefit in a subset of intermediate risk patients (i.e., positive PALN in an intraoperative frozen section). As aforementioned, the reported false-negative rate of PALN metastasis on PET/CT was 12% [5]. Meanwhile, Ramirez et al. reported that in patients with positive PLNs but negative PALNs on PET/CT, the falsenegative rate (i.e., pathologically confirmed PALN metastasis) increased to 22% [10]. This means that considering both the PLN status and PALN status on PET/CT could identify more patients who would benefit from surgical staging compared to considering solely the PALN status on PET/CT. Thus, they suggested that laparoscopic para-aortic lymphadenectomy should be discussed for patients with positive PLNs but negative PALNs on PET/CT. In our present study, of 105 patients with positive PLNs but negative PALNs on PET/CT, 11 (11%) displayed pathologically confirmed PALN metastasis. Meanwhile, of 53 patients in the intermediate-risk group, 11 (21%) had pathologically confirmed PALN metastasis. Therefore, using our current classification, a larger number
Fig. 2. Clinical scenario according to risk-group stratification using the current prediction model. RT, radiation therapy; PALN, para-aortic lymph node.
Please cite this article as: S.-H. Shim, et al., Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer, Gynecol Oncol (2016), http://dx.doi.org/10.1016/j.ygyno.2016.11.011
6
S.-H. Shim et al. / Gynecologic Oncology xxx (2016) xxx–xxx
of patients can be excluded from surgical staging safely and a larger proportion of patients could benefit from surgical staging compared with previous suggestions [10]. Moreover, in the low-risk group stratified by our model, the positive PLN rate on PET/CT was 43% (73/169). Of these 73 patients who were defined as low-risk by our model and also showed PLN positivity on PET/CT, true PALN metastasis was observed in only 4 patients (5%). Hence, if we performed PALN staging surgery for all low-risk patients with positive PLN on PET/CT, 95% (69 out of 73) would have undergone unnecessary surgery. We conclude that para-aortic lymphadenectomy is not required for such low-risk group patients, even those with a positive PLN on PET/CT. The current study has several limitations. First, our study excluded patients with LACC who did not undergo para-aortic lymphadenectomy before definite treatment because of its retrospective nature. Considering the tendency of physicians to omit staging surgery in patients with suspected PALN metastasis on PET/CT, our study population might have not had included patients with more advanced disease. This selection bias may have affected the performance of the model. Second, although we performed a rigorous internal validation via bootstrap resampling, the model requires external validation using independent data sets from other institutions to confirm its generalizability. Third, a central review was not performed in current study, thus causing measurement errors. However, this trade-off may have enhanced the generalizability of the model. In addition, known factors associated with LN metastasis, such as the genotype of human papillomavirus and lymphovascular space invasion, were not tested in the current study [25,26]. Finally, the current prediction model may not be useful in low-resource settings because PET/CT is not available as a pretreatment evaluation tool for cervical cancer. In conclusion, we constructed a robust model incorporating two preoperative variables to predict PALN metastasis in patients with LACC. Using the score chart, patients were stratified into low-, intermediate-, and high-risk groups regarding PALN metastasis. Our model may be useful in deciding whether nodal-staging surgery should be performed before definitive CCRT and in designing clinical trials. A prospective validation study using independent populations will be conducted to evaluate the performance of the model. Supplementary data to this article can be found online at doi:10. 1016/j.ygyno.2016.11.011. Disclosures The authors have no conflicts of interest or financial ties to disclose. Acknowledgments This work was supported by Konkuk University Medical Center Research Grant 2015 (Grant No. 201504). References [1] L.R. Medeiros, D.D. Rosa, M.C. Bozzetti, J.M. Fachel, S. Furness, R. Garry, et al., Laparoscopy versus laparotomy for benign ovarian tumour, Cochrane Database Syst. Rev. (2009), CD004751. [2] K.J. Min, Y.J. Lee, M. Suh, C.W. Yoo, M.C. Lim, J. Choi, et al., The Korean guideline for cervical cancer screening, J. Gynecol. Oncol. 26 (2015) 232–239. [3] H. Kim, J.Y. Kim, J. Kim, W. Park, Y.S. Kim, H.J. Kim, et al., Current status of brachytherapy in Korea: a national survey of radiation oncologists, J. Gynecol. Oncol. 27 (2016), e33. [4] N. Colombo, S. Carinelli, A. Colombo, C. Marini, D. Rollo, C. Sessa, et al., Cervical cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up, Ann. Oncol. 23 (Suppl. 7) (2012) vii27–vii32.
[5] S. Gouy, P. Morice, F. Narducci, C. Uzan, J. Gilmore, H. Kolesnikov-Gauthier, et al., Nodal-staging surgery for locally advanced cervical cancer in the era of PET, Lancet Oncol. 13 (2012) e212–e220. [6] H.J. Choi, W. Ju, S.K. Myung, Y. Kim, Diagnostic performance of computer tomography, magnetic resonance imaging, and positron emission tomography or positron emission tomography/computer tomography for detection of metastatic lymph nodes in patients with cervical cancer: meta-analysis, Cancer Sci. 101 (2010) 1471–1479. [7] S. Kang, S.K. Kim, D.C. Chung, S.S. Seo, J.Y. Kim, B.H. Nam, et al., Diagnostic value of (18)F-FDG PET for evaluation of paraaortic nodal metastasis in patients with cervical carcinoma: a metaanalysis, J. Nucl. Med. 51 (2010) 360–367. [8] S. Gouy, P. Morice, F. Narducci, C. Uzan, A. Martinez, A. Rey, et al., Prospective multicenter study evaluating the survival of patients with locally advanced cervical cancer undergoing laparoscopic para-aortic lymphadenectomy before chemoradiotherapy in the era of positron emission tomography imaging, J. Clin. Oncol. 31 (2013) 3026–3033. [9] E. Leblanc, H. Gauthier, D. Querleu, G. Ferron, S. Zerdoud, P. Morice, et al., Accuracy of 18-fluoro-2-deoxy-D-glucose positron emission tomography in the pretherapeutic detection of occult para-aortic node involvement in patients with a locally advanced cervical carcinoma, Ann. Surg. Oncol. 18 (2011) 2302–2309. [10] P.T. Ramirez, A. Jhingran, H.A. Macapinlac, E.D. Euscher, M.F. Munsell, R.L. Coleman, et al., Laparoscopic extraperitoneal para-aortic lymphadenectomy in locally advanced cervical cancer: a prospective correlation of surgical findings with positron emission tomography/computed tomography findings, Cancer 117 (2011) 1928–1934. [11] E. Brockbank, F. Kokka, A. Bryant, C. Pomel, K. Reynolds, Pre-treatment surgical paraaortic lymph node assessment in locally advanced cervical cancer, Cochrane Database Syst. Rev. 3 (2013), CD008217. [12] S. Kodama, K. Kanazawa, S. Honma, K. Tanaka, Age as a prognostic factor in patients with squamous cell carcinoma of the uterine cervix, Cancer 68 (1991) 2481–2485. [13] M. Takeda, N. Sakuragi, K. Okamoto, Y. Todo, S. Minobe, E. Nomura, et al., Preoperative serum SCC, CA125, and CA19-9 levels and lymph node status in squamous cell carcinoma of the uterine cervix, Acta Obstet. Gynecol. Scand. 81 (2002) 451–457. [14] R. Grigiene, K.P. Valuckas, E. Aleknavicius, J. Kurtinaitis, S.R. Letautiene, The value of prognostic factors for uterine cervical cancer patients treated with irradiation alone, BMC Cancer 7 (2007) 234. [15] M.E. Juweid, S. Stroobants, O.S. Hoekstra, F.M. Mottaghy, M. Dietlein, A. Guermazi, et al., Use of positron emission tomography for response assessment of lymphoma: consensus of the Imaging Subcommittee of International Harmonization Project in Lymphoma, J. Clin. Oncol. 25 (2007) 571–578. [16] J.Y. Park, J.H. Nam, Laparotomy conversion rate of laparoscopic radical hysterectomy for early-stage cervical cancer in a consecutive series without case selection, Ann. Surg. Oncol. 21 (2014) 3030–3035. [17] Y.S. Kim, S.S. Shin, J.H. Nam, Y.T. Kim, Y.M. Kim, J.H. Kim, et al., Prospective randomized comparison of monthly fluorouracil and cisplatin versus weekly cisplatin concurrent with pelvic radiotherapy and high-dose rate brachytherapy for locally advanced cervical cancer, Gynecol. Oncol. 108 (2008) 195–200. [18] S.H. Shim, S.J. Lee, S.O. Kim, S.N. Kim, D.Y. Kim, J.J. Lee, et al., Nomogram for predicting incomplete cytoreduction in advanced ovarian cancer patients, Gynecol. Oncol. 136 (2015) 30–36. [19] E.W. Steyerberg, B.A. Neville, L.B. Koppert, V.E. Lemmens, H.W. Tilanus, J.W. Coebergh, et al., Surgical mortality in patients with esophageal cancer: development and validation of a simple risk score, J. Clin. Oncol. 24 (2006) 4277–4284. [20] J.J. Deeks, D.G. Altman, Diagnostic tests 4: likelihood ratios, BMJ 329 (2004) 168–169. [21] S. Kang, J.M. Lee, J.K. Lee, J.W. Kim, C.H. Cho, S.M. Kim, et al., A Web-based nomogram predicting para-aortic nodal metastasis in incompletely staged patients with endometrial cancer: a Korean Multicenter Study, Int. J. Gynecol. Cancer 24 (2014) 513–519. [22] T. Bayes, An essay towards solving a problem in the doctrine of chances. 1763, MD Comput. 8 (1991) 157–171. [23] V. Benito, A. Lubrano, O. Arencibia, M. Andujar, B. Pinar, N. Medina, et al., Laparoscopic extraperitoneal para-aortic lymphadenectomy in the staging of locally advanced cervical cancer: is it a feasible procedure at a peripheral center? Int. J. Gynecol. Cancer 22 (2012) 332–336. [24] M.A. Gold, C. Tian, C.W. Whitney, P.G. Rose, R. Lanciano, Surgical versus radiographic determination of para-aortic lymph node metastases before chemoradiation for locally advanced cervical carcinoma: a Gynecologic Oncology Group Study, Cancer 112 (2008) 1954–1963. [25] G.G. Garzetti, A. Ciavattini, G. Lucarini, G. Goteri, S. Menso, M. De Nictolis, et al., The role of human papillomavirus DNAs in cervical carcinoma and risk of lymph node metastasis: association with 72-kilodalton metalloproteinase immunostaining, Cancer 82 (1998) 886–892. [26] J.R. van Nagell Jr., E.S. Donaldson, E.G. Wood, J.C. Parker Jr., The significance of vascular invasion and lymphocytic infiltration in invasive cervical cancer, Cancer 41 (1978) 228–234.
Please cite this article as: S.-H. Shim, et al., Prediction model for para-aortic lymph node metastasis in patients with locally advanced cervical cancer, Gynecol Oncol (2016), http://dx.doi.org/10.1016/j.ygyno.2016.11.011