Dynamic Nomogram Based on the Metastatic Number and Sites and Therapy Strategies Predicting the Prognosis of Patients with Metastatic Cervical Cancer

Introduction

Cervical cancer is the fourth most common cancer in women with an estimated 604,000 new cases and 342,000 deaths worldwide in 2020.¹ Although precancerous lesions screening and vaccination against human papilloma virus are becoming common in developed areas, cervical cancer still is the most commonly diagnosed cancer in 23 countries and is the leading cause of cancer death in 36 countries.^1,2 Despite much cervical cancer being localized due to vagina bleeding or abnormal leucorrhea, approximately 2–7.6% of patients are diagnosed with synchronous distant metastases with a 5-year prognosis of 16.5% compared to 91.5% for localized cervical cancer.^3,4 Due to the overall rarity and heterogeneity of metastatic cervical cancer (mCC), accurate survival prediction is crucial for risk stratification and individualized precision therapy strategy for the advanced tumor. Therefore, a new prognostic tool is needed in clinical practice.

In recent years, nomogram has been a frequently used and feasible tool that can incorporate multiple reliable factors and provide accurate and individualized prognostic information for clinicians in clinical practice.^5,6 Previous studies have reported that factors such as neutrophil-lymphocyte ratio, location of metastases, histological subtype and radiotherapy are associated with the prognosis of mCC.^7–9 Nevertheless, little attention has been focused on the concrete location and number of metastatic lesions of mCC patients at the initial diagnosis. A retrospective study demonstrated that patients with multiple organ metastases had a poorer prognosis compared to patients with single organ metastasis.¹⁰ Since knowledge of the prognosis of metastatic number and sites is crucial for pre-treatment evaluation, our study aimed to analyze the impact of metastases in different systems, such as lymph node (LN) and hematogenous (including lung, liver, bone and other organs) systems, and in specific locations on the prognosis of mCC patients. In addition, previous studies found that definitive pelvic radiotherapy and chemotherapy were associated with significantly longer survival compared with chemotherapy alone for cervical cancer patients,^8,11,12 but the effect of the appropriate timing of chemoradiotherapy (CRT) in patients with mCC has not been well studied. Until now, many studies related to mCC have contained patients with both synchronous and recurrent metastatic diseases.^7,13–15 But the reality is they have completely different treatment patterns that may affect the prognosis of patients.^16,17

Therefore, this retrospective study aimed to explore the risk factors associated with the prognosis of newly diagnosis mCC patients and to construct nomograms to facilitate clinical practice.

Materials and Methods

Ethics Statement

Patient consent was waived due to the retrospective nature of this study. We confirm that the privacy of participants will be kept in confidence. And this study was approved by the ethics committee of the First Affiliated Hospital of Xi’an Jiaotong University (approval No. NCT03257475).

Patients

A retrospective chart review was conducted on patients diagnosed with histopathology-confirmed cervical cancer from January 2008 to December 2018 at the First Affiliated Hospital of Xi’an Jiaotong University. Patients newly diagnosed with International Federation of Gynecology and Obstetrics (FIGO) 2018⁴ stage IVB were enrolled in this study. The distant metastases were confirmed through computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography-CT (PET-CT), or biopsy of metastatic sites if possible. The LN metastasis was defined based on the short axis >10 mm in CT. Exclusion criteria were as follows: cervical cancer was not the first primary malignancy, patients diagnosed with mCC without any treatment, recurrent distant metastases disease diagnosed with local cervical cancer at first, and patients with incomplete records or surveillance/follow-up.

Clinical Variables

Clinical variables captured from the electronic medical records encompassed age at first diagnosis, year of diagnosis, metabolic disease (hypertension, diabetes, and cardiovascular disease), symptom duration before diagnosis, pathological type, tumor differentiation, tumor maximum size, the presence of rectal/bladder invasion, pelvic LN metastases, pelvic effusion, and hydronephrosis/ureteral hydrops, laboratory results (the level of squamous cell carcinoma antigen (SCCAg), hemoglobin, albumin, the count of platelet and leukocyte), distant metastatic site (hematogenous: lung, bone, liver, and other (spleen and adrenal gland); LN: intraperitoneal, inguinal, and LN above diaphragm (mediastinal, axillary, and supraclavicular LN)), number of metastatic site, treatment strategy, vital status, and survival time. Definitive radiotherapy was administrated with external beam radiotherapy (EBRT) of 45–50 Gy in whole-pelvic and para-aortic (if they have para-aortic LN metastases) and brachytherapy of 20–30 Gy, and non-definitive radiotherapy was palliative therapy to relieve vaginal bleeding or other symptoms and could not be achieved at above doses. First-line chemotherapy regimens mainly include TP (Paclitaxel + Platinum), DP (Docetaxel + Platinum), and PF (Platinum + 5-Fluorouracil). The primary endpoint of the survival analysis was overall survival (OS), which was defined as the date from cancer diagnosis to death or last follow-up. The follow-up was terminated on 31 December 2021.

Statistical Analysis

The distribution of demographic and clinicopathological variables was summarized with descriptive statistics. The categorical for different subgroups were compared using the chi-squared (χ²) test or Fisher’s exact test where appropriate, and continuous variables were compared using the Kruskal–Wallis test. The Cox proportional hazards regression analysis with hazard ratio (HR) and associated 95% confidence interval (CI) were utilized to investigate the effect of specific factors on the survival of mCC patients. Kaplan-Meier (KM) method was conducted to generate the survival curves, and the log-rank test was performed to investigate the differences between the curves based on the “survival” R package. And “randomForest” R package was used to evaluate the importance of clinicopathological parameters to therapy decision.

Nomogram and dynamic nomogram were established using the ‘rms’ and “DynNom” R package. Calibration curves were generated to verify the agreement between nomogram predicted survival and actual survival probabilities.¹⁸ Decision curve analysis (DCA) was further used to evaluate the applicability of the nomogram to clinical practice through the “rmda” R package.¹⁹ The multi-index receiver operating characteristic (ROC) curve was used to compare the predictive accuracy of the nomogram and other risk factors with the “survivalROC” R package.²⁰

The R software version 4.1.1 (The R Foundation for Statistical Computing, Vienna, Austria. http://www.r-project.org) and the IBM SPSS statistical software version 23.0 (SPSS, Armonk, New York, USA) were utilized for all data analyses. All statistical tests were two-tailed, and p < 0.05 was considered statistically significant.

Results

Clinicopathological Characteristics

Table 1 summarizes the clinicopathological characteristics of 98 mCC patients enrolled in this study. Histologically, 90 (91.8%) patients were diagnosed with squamous cell carcinoma, and 8 (8.2%) had adenocarcinoma. And 28 (28.6%) patients were diagnosed with single metastasis (SM), while 70 (71.4%) patients with multiple metastases (MM). In addition, 87 (88.8%) patients received chemotherapy, while 83 (84.7%) patients received definitive pelvic radiotherapy, with the median dose being 70.4 (10–75) Gy, and 55 (56.1%) and 23 (23.5%) patients received concurrent and sequential chemoradiotherapy, respectively (Table S1). And 25 (25.5%) patients received radiotherapy in metastatic lesions, mainly in bone and LN (inguinal and supraclavicular LN).

Table 1 The Clinicopathological Characteristics of mCC Patients

Exploration of Prognostic Factors of OS

By using univariate Cox regression analysis in the mCC cohort, the results indicated that characteristics including symptom duration before diagnosis, multiple metastases, external-beam radiation therapy (EBRT) dose, concurrent chemoradiotherapy (CCRT), postradiotherapy chemotherapy were all significantly associated with the OS of mCC patients (Figure 1A). Subsequently, multivariate Cox regression analysis indicated multiple metastases (p = 0.040) and CCRT (p = 0.002) as independent risk factors for OS (Figure 1B).

Figure 1 Forest plot of univariate and multivariate Cox regression analyses of risk factors related to overall survival in patients with metastatic cervical cancer. (A) The forest plot for univariate Cox analysis. (B) The forest plot for multivariate Cox analysis.

Characteristics of Metastases

Table 2 summarizes the distribution of metastatic sites of mCC patients. Amongst patients with distant metastases, we found that retroperitoneal LN was the most common site of distant metastases, followed by inguinal LN, lung, LN above diaphragm, bone, and liver. Subsequently, Kaplan–Meier survival analyses were used to evaluate the effects of metastases number and sites on the OS of mCC patients. Patients with single metastasis had better survival compared to patients with multiple metastases (median OS: less than 50% with death events vs 24 months, p = 0.019, Figure 2A). And for specific sites, metastases in liver (with vs without: 14 months vs 46 months, p = 0.029, Figure 2B) and LN above diaphragm (with vs without: 17 months vs 48 months, p = 0.016, Figure 2C) had significant impacts on OS. However, metastases in lymphatic and hematogenous systems, bone, lung, inguinal LN, and retroperitoneal LN were all not associated with the outcome of patients (p > 0.05, Figure 2D and Figure S1). Table 3 exhibits the metastatic sites and median OS of 10 metastatic forms involving the liver and above-diaphragm LN metastases (There are overlaps between these groups bacause too much metastatic patterns).

Table 2 Distribution of Metastases Sites for mCC Patients

Table 3 Distribution and Overall Survival of Patients with Different Metastases Patterns

Figure 2 Kaplan–Meier curves and Log rank test for overall survival according to different metastatic sites: (A) Metastatic number (single metastasis vs multiple metastases), (B) liver metastases, (C) LN above diaphragm metastases, and (D) lymph node system metastases.

Prognostic Effects of Chemoradiotherapy

As shown in Table S1, 55 (56.1%) patients received CCRT. Of patients without CCRT, 32 (74.4%) patients received chemotherapy, and 28 (63.6%) received pelvic radiotherapy. Besides, patients with CCRT had relatively higher hemoglobin (p = 0.065) and albumin levels (p = 0.032). For newly diagnosed patients, it is essential to find out which patients are suitable for CCRT. The random forest algorithm was used to screen for factors associated with the administration of CCRT. Seven clinicopathological factors with importance greater than 2 were then included in the nomogram to predict the probability of a patient being suitable for CCRT (Figure S3A, B). Moreover, we deployed the nomogram on the shinyapps website (https://metastaticcervicalcancer.shinyapps.io/DynNomapp_CCRT/), which allows clinicians to conveniently utilize this model (Figure S3C).

Figure S4 and Table S2 exhibited the distribution and mOS of patients with different therapeutic strategies and revealed that patients with CCRT plus postradiotherapy chemotherapy have the longest OS. In further analyses, both CCRT (57 months vs 17 months, p = 0.0043, Figure 3A) and postradiotherapy chemotherapy (47 months vs 23 months, p = 0.037, Figure 3B) were predictors of better prognosis for mCC patients. And patients with CCRT had better survival in both the multiple metastases (46 months vs 12 months, p = 0.0026, Figure S2) and overall cohort. Although radiotherapy to metastatic sites had no prognostic impact on the overall cohort (Figure 1A), in subgroup analyses we found that radiotherapy to lymph node and bone prolonged the survival of mCC patients with LN and bone metastases, respectively (Figure 3C, D). However, year of diagnosis and first-line chemotherapy regimens have no effects on patient survival (Figure 3E, F).

Figure 3 Kaplan–Meier curves and Log rank test for overall survival according to different risk factors: (A) Concurrent chemoradiotherapy, (B) postradiotherapy chemotherapy, (C) radiotherapy in LN in patients with inguinal and above-diaphragm LN metastases, (D) radiotherapy in bone in patients with bone metastases, (E) year of diagnosis, and (F) first-line chemotherapy regimens.

Construction and Verification of the Prognostic Nomogram

Kaplan–Meier curves revealed that metastases number and sites (liver and LN above diaphragm) and therapy modifies (CCRT, postradiotherapy chemotherapy, and radiotherapy to metastatic site) were associated with the OS of mCC patients (Figures 2 and 3). Then, the above-mentioned 6 prognostic parameters were incorporated to develop the nomogram for predicting 1-, 2-, and 3-year OS (Figure 4A). The dynamic nomogram can be accessed via https://metastaticcervicalcancer.shinyapps.io/DynNomapp_OS_predict/ (Figure 4B). The calibration plots of the nomogram showed satisfactory agreement between the nomogram predicted and actual survival (Figure 4C). As shown in Figure 5A, the DCA curves indicated that the nomogram has gained significant benefits in predicting 1-, 2-, and 3-year OS. In the ROC curve analysis, the 1-, 2-, and 3-year AUC values of the nomogram were 0.796, 0.9, and 0.896, respectively, which were superior to other individual clinicopathological parameters (Figure 5B).

Figure 4 Construction and validation of nomogram associated with the overall survival in patients with metastatic cervical cancer. (A) Nomogram composed of multiple metastases, liver metastases, LN above diaphragm metastases, concurrent chemoradiotherapy, postradiotherapy chemotherapy, and radiotherapy to metastasis site was constructed to predict 1-, 2-, and 3- year survival rates. (B) Dynamic nomogram for predicting the survival of mCC patients in Shinyapps website. (C) The calibration plots of the nomogram for 1-, 2- and 3-year overall survival for mCC patients.

Figure 5 Decision curve analysis (DCA) curves and receiver operating characteristic (ROC) curves for detecting the predictive value of the nomogram. (A) DCA of clinicopathological parameters including in nomogram in mCC patients. (B) The ROC curves of the nomogram and individual clinicopathological parameters for predicting 1-, 2-, and 3- year OS.

Discussion

Since distant metastasis remains a major cause of cervical-cancer-related death in women, in which complex biological profiles posed a huge challenge for clinical treatment, there is a need to identify clinical risk and prognostic factors of newly diagnosed mCC patients to improve their survival condition.^16,21 In this population-based study, we retrospectively evaluated the clinical characteristics and survival of mCC patients and constructed a nomogram to predict the prognosis of these patients. Several studies have constructed nomograms to predict the prognosis of cervical cancer patients,^22–24 but few studies built nomogram for stage IVB cervical cancer patients. A SEER-based nomogram can predict the survival of mCC patients, but lack more detailed information such as treatment strategies and specific metastasis number and locations.⁹ In the current study, Cox regression analysis showed significant impacts of multiple metastases and CCRT on prognosis. We further explored the prognostic value of metastatic sites and chemoradiotherapy strategies. Ultimately, six parameters (multiple metastases, liver, and LN above diaphragm metastases, concurrent chemoradiotherapy, postradiotherapy chemotherapy, and radiotherapy to metastatic site) were included in the nomogram with excellent predictive ability for the prognosis of mCC patients.

Cervical cancer metastasis to multiple distant sites is not rare, previous studies have reported that nearly one in four cases of distant organ metastasis had two or more site involvements, which is consistent with our study.²⁵ In addition, some studies have shown that cancer survival may be associated with different metastatic locations.^26–29 Previous studies have reported that cervical cancer patients with single metastasis have longer survival than those with multiple metastases, but these findings are based on the fact that metastases only involved organs.^10,25,30 In this study, we also found that multiple metastases in both the lymphatic and hematological systems can affect the prognosis of mCC patients. And lymphatic metastases are not rare, 62.3% (61/98) and 28.6% (28/98) of mCC patients in this study had LN and disseminated metastases, respectively.

Given the significant impact of metastases sites on survival, we further investigated the relationship between specific metastatic locations and cervical cancer survival in the current study. Kaplan–Meier plots revealed that metastases in liver and LN above diaphragm have significant impacts on the OS of mCC patients. However, there were no significant differences in survival for other metastatic sites. It has been reported that patients with hematogenous metastases had a 5.3-fold higher risk of death compared to those with lymphatic metastases,³¹ but there was no survival difference between those two groups in our findings, which might be due to the exclusion of patients with only retroperitoneal LN metastases in the 2018 FIGO stage IVB and the current study.⁴ Liver metastases are associated with a poorer prognosis of breast cancer, rectal cancer, and lung cancer.^32–34 In cervical cancer, the median OS for patients with liver metastases was 6.8–10.9 months.^10,35 Besides, we found that liver metastases patients with or without other metastatic lesions had similar prognoses (Table 3), indicating that liver metastasis is an independent prognostic risk factor for mCC patients. Patients with LN above diaphragm metastases are more likely to accompany other metastatic lesions, and these patients resulted in poorer survival rates than those with LN above diaphragm metastases only. Several studies have reported that 52–74% of patients with left supraclavicular LN metastases experienced pathological responses after chemoradiotherapy with acceptable toxicity.^36,37 But mCC patients with LN above diaphragm and organ metastases had dismal outcomes, and how to control these organ metastases may be the key to prolonging survival.

Growing studies have demonstrated that radiotherapy and chemotherapy are administrated for survival improvement in cervical cancer patients with metastatic disease.^8,38–41 Besides, our results depicted longer survival in patients developing metastases after CCRT, which was compatible with the results in the previous study.⁴² Furthermore, the effects remained significant and strong in the multiple metastases subgroup. This finding is in line with previous studies that demonstrated an increased sensitivity of radiotherapy when combined with concurrent chemotherapy.¹⁶ Up till now, CCRT was generally considered the primary treatment choice for stages IB2, IIA2, and IIB to IVA cervical cancer.^43,44 Kim et al reported that 52% patients with left supraclavicular and paraaortic LN metastases could achieve complete response from CCRT.³⁶ A prospective observational study revealed that CCRT induces increased infiltration of CD8⁺ T cells and activation of IFN signaling, indicating great promise for the combination of immunotherapy and radiotherapy.⁴⁵ A multi-institutional retrospective study showed that adjuvant chemotherapy after CCRT has a potential role in improving disease control for locally advanced cervical cancer patients.⁴⁶ In Table S2, patients receiving CCRT plus postradiotherapy chemotherapy had the longest OS, but there was overlap between patients treated with CCRT and postradiotherapy chemotherapy, reminding that postradiotherapy chemotherapy may be a confounder for CCRT, resulting in postradiotherapy chemotherapy not being statistically significant in multiple Cox regression analysis. In addition, there were survival benefits for patients receiving sequential chemoradiotherapy. Radiotherapy in metastatic lesions (bone and LN) prolonged the OS of patients in subgroup analysis, which is consistent with previous research.³⁷ There is a consensus that chemoradiotherapy has significant effect on the survival of cervical cancer patients.^4,16 The ROC curves showed that the nomogram without chemoradiotherapy factors had relative lower AUC values compared to the prognostic nomogram. As shown in Table S2, patients with different therapeutic regimens had completely different survival. Collectively, we included these three therapy-related factors into the nomogram for better predictive ability.

Due to the poor prognosis of patients receiving only chemotherapy or radiotherapy, we strongly recommend the combination of chemotherapy and definitive radiotherapy for newly diagnosed mCC patients, but the sequence of these treatments should be prudently decided to prevent intolerable toxic effects and treatment interruptions. A randomized clinical trial suggested sequential chemoradiation, rather than CCRT, improved disease-free survival and reduced the risk of cancer death in women with early-stage cervical cancer.⁴⁷ Many previous studies and our results indicated that patients with CCRT have the best prognosis.^44,48 However, in clinical practice, it remains a huge challenge for oncologists to find the optimal treatment for individual patients to prevent catastrophic toxic effects. Because patients in this study had positive therapy with manageable side effects, we further analyzed the factors associated with the implementation of CCRT. Random forest and nomogram revealed that patients with younger age, shorter symptom duration, and better laboratory test results were intended to receive CCRT. Then, a dynamic nomogram was built that can be used as a support system for clinical decision. There are studies reporting that pretreatment hemoglobin level, leukocyte, and platelet counts are associated with the prognosis of locally advanced cervical cancer patients undergoing CCRT.^48,49 After using the CCRT-nomogram to predict the probability that individual patient is suitable to receive CCRT, the prognostic nomogram could effectively predict the prognosis of newly diagnosed mCC patients. However, the administration of CCRT in this study was affected by human subjective factors, and the model should be validated with larger samples and in prospective study.

In summary, previous studies support that these metastatic sites and therapy regimens are prognostic risk factors for mCC patients. To our knowledge, there are some SEER-based nomograms for cervical cancer, but few studies focused on stage IVB disease. Compared to one nomogram constructed for mCC patients,⁹ our prognostic model consists of more detailed metastatic and therapeutic information and presents high predictive ability, which can be proven by the relative higher AUCs value of ROC curves. Besides, previous studies related to mCC have mainly focused on the effect of singular factors on patient survival, the current study analyzed factors including metastatic status and therapy strategies and put forward a nomogram that can predict individual survival. Besides, previous study for metastatic patients including synchronous and recurrent metastatic diseases.^13–15 The survival and subsequent therapy of recurrent mCC patients may be influenced by previous treatment, so we only included newly diagnosed mCC patients to prevent the impact of other confounders. This model also gives us a clue that mCC patients can benefit from chemoradiotherapy, which can be administrated for newly diagnosed mCC patients. The sequence of definitive radiotherapy and chemotherapy can be decided using the CCRT nomogram.

Despite the valuable findings above, we have to admit there are several limitations in our study due to its retrospective nature and long duration. The first is the uncontrolled, non-randomized retrospective design. Secondly, the clinical target volumes of radiotherapy in this study are potential confounders. However, all treatments implemented according to standard treatment guidelines of cervical cancer. Thirdly, not all prognostic factors were identified due to the limited sample size in the study, and more studies with larger sample sizes were needed in the future.

Conclusions

We constructed a prognostic nomogram including the number and sites of metastases (multiple metastases, liver and LN above diaphragm metastases) and chemoradiotherapy strategies (concurrent chemoradiotherapy, postradiotherapy chemotherapy, and radiotherapy to metastatic sites) for predicting the prognosis of mCC patients. Newly diagnosed patients with younger age, shorter symptom duration, and better laboratory test results are suitable for CCRT according to the CCRT-nomogram, which can facilitate treatment decisions and help the use of prognostic nomogram for untreated mCC patients.