Background
Worldwide, lung cancer remains an important public health concern affecting both men and women and the leading cause of cancer-associated mortality [
1]. In the United states, there were estimated 234,030 new diagnosed lung cancer cases in 2018 [
1]. Small-cell lung cancer (SCLC) is one of the easily aggressive pathology type and accounts for approximately 14–16% of all lung cancer cases [
2,
3].
SCLC is the mainly neuroendocrine tumor of lung which has poor prognosis for its high vascularity, rapid doubling time and early metastasis. Mainly treating choices of SCLC include surgery, chemotherapy and radiotherapy [
4]. Systemic platinum-based chemotherapy either alone or combined with concurrent radiotherapy is most commonly considered to be standard and potentially curative treatment for SCLC lesions, because most SCLC cases are highly sensitive to initial chemotherapy and radiotherapy [
5]. But patients often develop treatment-resistance quickly and subsequent relapse and eventual death.
The role of surgery in SCLC was reevaluate over and over again. Before 1970s, surgery was a common treating modality for SCLC, which was overturned by a Medical Research Council trial performed in 1973. This trial demonstrated the poor survival of SCLC patients with pulmonary resection than radiotherapy [
6]. Besides, the results of another prospective randomized trial in 1994 did not support the addition of pulmonary resection to the multimodality treatment of small cell lung cancer [
7]. These vital evidences led to abandonment of surgery as a standard treatment. But renewed studies advocated adopting surgery to increase local-control rate in certain early-stage SCLC. A study published in 2010 re-evaluated the role of surgery, and showed that lobectomy, in selected patients with limited-SCLC was associated with improved survival outcomes [
8]. And the research which focused on survival of patients with SCLC undergoing lung resection in 1998–2009 in England also suggested surgical resection for early stage SCLC [
9]. A Italian review published in 2015 summarized recent original researches and suggested that surgery should be offered (or at least considered) in intraoperative diagnosis of resectable SCLC or early-stage SCLC after chemotherapy [
10]. Therefore, more reasonable or proper staging and prognostic prediction is extremely important for surgical procedure and even following survival outcomes.
Most clinical guidelines for SCLC were based on the VALSG staging system in which SCLC patients were roughly distributed into extensive-stage and limited-stage. However, it has been recommended that the AJCC TNM staging system should replace the VALSG staging system because TNM system would allow for more proper treating selections (e.g. surgical resection) and more precise prognostic assessments [
5]. Nevertheless, except for TNM staging status, it was known that clinical characteristics like sex, age, location and treating modalities were also noteworthy factors influencing individual survival outcomes of cancer patients [
10‐
12]. For instance, lobectomy was demonstrated to have superior survival outcomes compared with sublobectomy or pneumonectomy [
13‐
15]. Above all, it is obvious that the TNM system is less sufficient for predicting outcomes of an individualized resectable SCLC patient. Therefore, a more refined model with better prognostic discrimination of is required, and a nomogram is an ideal tool to solve this problem [
16,
17].
Nomogram is a tool to predict individual prognosis of patients by regression analyses of the potential prognostic factors. Previous four nomograms were built by different institutions involving SCLC patients, but there still lack efficient nomogram that can predict survival outcomes of resectable SCLC patients specially [
18‐
21]. The objective of this study was to derive and externally validate a prognostic nomogram to predict overall survival (OS) for patients who did resection of SCLC in two independent cohorts, which would help clinical decision making and to assist ongoing efforts.
Materials and methods
Training cohort and data
The flow chart of this study was shown in Additional file
1.
The data of patients with SCLC diagnosed from 2004 to 2016 were retrieved from the SEER 18 database using the SEER*Stat program (v 8.3.5). The SEER program is a public national database which contains data on cancer occurrences in 18 areas of United States and covers approximately 26% of the population. Among these patients, there were 1485 patients conformed to our inclusion criteria: only one primary tumor; diagnosis confirmed by histology; histological type of small-cell carcinoma (ICD-O-3); surgery performed. SCLC, also named oat cell carcinoma, with histological codes included as follows: 8041/3, 8042/3, 8043/3, 8044/3, 8045/3. Variables with more than 10% missing values (Blanks or unknown or N/A are deemed as missing) were not eligible for analysis. Eventually, 1052 patients were included for analyses, after excluding the following ineligible cases: 411 patients with 8th TNM stage of M1/N3/Tx/Nx/Mx, 22 patients with unknown surgery details, 4 patients with no access to data of lymph nodes metastatic ratio (LNR). There were 4 patients that meet more than one of above exclusion criteria. LNR was the number of lymph nodes with metastasis divided by the total number of dissected lymph nodes [
22].
The data included clinical information of patients, histological characteristics, survival time (months) and vital status (the event of death). Continuous variables were transformed into categorical variables based on recognized cutoff values (for age). Clinical information of patients included sex (female v male), age (≤ 60 years, 60–70 years, > 70 years), marital status (unmarried, married), surgery (lobectomy, sublobectomy, pneumonectomy). Pathological characteristics of tumors include primary site (upper lobe, lower lobe, other), lateral (left, right), pathological grade (I–II, III, IV), T stage in 8th edition AJCC system (T1, T2, T3–T4), N stage in 8th edition AJCC system (N0, N1, N2), LNR (< 0.01, > 0.01, no resected lymph node), radiotherapy or not, chemotherapy or not. The acquisition of cutoff value of LNR was achieved by the receiver operating characteristic (ROC) analysis. The time of last follow-up was December 2016. The primary outcome was defined as overall survival (OS). Time of OS was counted from date of diagnosis to date of death or last contact.
External validation cohort and data
To further validate our new model in a responsible manner, we sought an external validation cohort from patients diagnosed from January 2004 to December 2016 in Shandong Provincial Hospital Affiliated to Shandong University, Shandong Provincial Hospital Affiliated to Shandong First Medical University. The validation cohort included 114 postoperative SCLC patients who were recruited according to inclusion and exclusion criteria same as the training cohort. We collected variables according the training cohort except for marital status and pathological grade. The time of last follow-up was July 2019. The outcome variable was OS too.
The ethical committee and institutional review board of Shandong Provincial Hospital c approved this study.
Construction and evaluation of the prognostic model
To identify independent prognostic factors to build our prognosis model, we performed univariate COX Proportional Hazard Regression analysis in a forward stepwise manner. Significant factors in univariate analysis (p value < 0.05) were carried into a multivariate COX Proportional Hazard Regression analysis to obtain the hazard ratio (HR) and corresponding 95% confidential interval (CI) for every independent prognostic variable. All the COX Regression analyses were performed by SPSS 25.0 (SPSS, Chicago, IL). The prognostic nomogram was built based on surgery methods and other independent prognostic variables by using the survival and rms packages of R 3.5.1.
Evaluation of a nomogram generally include two facets: discrimination and calibration accuracy. Discrimination means the efficiency of the model to distinguish patients with different survival outcomes. Usually, concordance index (C-index) is taken to be the tool to measure discrimination, which represents a concordance measure analogous to area under the receiver operating characteristic (ROC). Values of C-index range from 0.5 (no discrimination) to 1.0 (perfect discrimination). Calibration accuracy measures how the predicted probabilities are close to actual survival outcomes which is showed in the form of calibration curves. Calibration curves of the nomogram for 1-, 3- and 5-year OS were achieved by use of the “survival and rms” R package in the training and validation cohort. All the evaluation processes were performed by bootstrapping for 1000 times.
To further evaluate the benefits and advantages of our new predicting model, we adopted decision curve analysis (DCA). DCA is usually used to evaluate alternative diagnostic and prediction strategies that have advantages over other generally used measures and techniques. If the threshold probability of net benefits of the new prediction model is unpractical, the benefits of it will be less than the benefits of existing tools (for example, 8th edition AJCC TNM staging system), which means poor applicability. Integrated discrimination improvement (IDI) index were employed to assess whether the new model more accurate than 8th edition AJCC TNM staging system or not. A larger difference value between the death probability of individual patient predicted by the new model and TNM staging system demonstrates a better predicting accuracy. Z test was performed to examine the significance of IDI between the new model and the 8th edition AJCC staging.
Construction of a risk stratification model
Based on the aggregate score of every patient on the nomogram, the cohort was distributed into two different risk groups (low and high). We obtained the most propriate cutoff value by use of receiver operating characteristic (ROC) analysis. To test the application of the risk stratification model, we conducted Kaplan–Meier survival analyses in both the training cohort and the validation cohort with Chi square test. A two-sided P value < 0.05 was deemed significant.
RStudio software (version 1.1.463) was used to perform the survival and RMS R package. Details of all R code involving generation and further evaluations of the model were shown in Additional file
2. This study followed the TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis) statement (Additional file
3) and adhered to the Declaration of Helsinki for medical research involving human subjects.
Creation of a webserver for the nomogram
To facilitate clinicians’ usage of our nomogram, we created a user-friendly webserver. The webserver can calculate a survival probability as long as you input correct information of a SCLC with surgery performed and certain prediction time (months) such as 12 months. Meanwhile, it can also provide the corresponding survival plot of this case.
Discussion
SCLC is well recognized as an easily aggressive tumor which rarely amendable to surgical resection [
2]. But surgery was demonstrated to increase local-control rate in certain early-stage SCLC [
8,
9]. Besides, SCLC was definitely diagnosed until intraoperation occasionally, and resectable ones of which were considered to receive surgery [
10]. Survival outcomes of resectable SCLC varies from patient to patient. Existent VALSG or TNM staging system are not efficient in predicting individualized OS of resectable SCLC patient. Therefore, we constructed and externally validated a clinical prognostic model that assign predictions for OS of resectable SCLC based on surgery and other clinicopathological variables. When applied to the external validation cohort, the new model achieved considerable discrimination ability and calibration accuracy (Figs.
2,
3). The C-index in the validation cohort were 0.819, 0.656, 0.708, respectively for 1-, 3- and 5-year OS. And further DCA and IDI analyses testified its obvious clinical application benefit versus TNM staging system. The risk stratification model according to this nomogram can effectively stratify patients in training or validation cohort into two risk groups (high-risk and low-risk) with distinguish OS. Besides, we provided a webserver to clinicians for more facile individual survival prediction.
By COX regression analyses, we identified age, sex, T stage, N stage, LNR, surgery and chemotherapy as independent predictors of overall survival. Some of these variables have been studied in previous research for their influence on survival of SCLC [
12,
23‐
26]. Elder patients had worse survival than the younger ones might because degenerative changes in various aspects of organs function and increased prevalence of all types of comorbidities [
27]. The male had worse survival than the female, which was consistent with the studies of Wang et al. and Xiao et al. [
20,
21]. Lymph nodes metastatic ratio, as a new meaningful indicator for OS of SCLC, was also recognized as an independent predictor [
22,
25]. The ones with lymph nodes resection performed had better survival than who not, which suggested to conduct lymphadenectomy for resectable SCLC. As with the nomogram of Wang et al. AJCC eighth TNM staging system contributed the most to the final risk score [
21]. Notably, in addition to the common investigated factors, surgical procedure was a crucial independent predictor for OS, among which lobectomy posed the superior choice with better survival [
8,
10,
13,
15].
There were a certain number of existing nomograms that involved SCLC patients in. However, most of these models were designed for pan-stage of the diagnosed SCLC, and no one of them has included specific surgical procedure in. Xie et al. built two nomograms by simply divided patients into limited or extensive stage but failed to involve information of tumor pathology like definite tumor size and lymph nodes status. In contrast, we brought in data of 8th edition AJCC T-stage and N-stage of every case to get a better understanding of the tumors. But the work of incorporating hematological markers into the models was worth emulating for our further study. Although these two models of Xie et al. received a considerable C-index (0.730) in internal validation, externally validation of them in a larger number of patients at multiple institutions should be considered [
18]. Another single-institution study published by Xiao et al. built a nomogram to serve prediction of 3- and 5-year OS for SCLC. C-index of its nomogram (= 0.60) was not high enough might due to heterogeneity of the included patients which covered pan-stage SCLC, or due to lack of data of more detailed T, N, M information. What different from our study was that it didn’t regard surgery as a single factor in therapeutic regimen and didn’t mention the number of resected SCLC patients [
20]. Pan et al. built a nomogram for SCLC that included only a small sample size of patients with resected SCLC (n = 53 in primary cohort, n = 4 in validation cohort) [
19]. In 2018, a quality study constructed an update model based on a large sample size of cases and summarized previous three nomogram to a webserver. But its validation was performed by data of more recent years but from the same database as the training cohort, which would limit its generalizability. And it simply pointed out if surgery done or not, without mention of different surgical procedure, which means it might be not suitable for resectable SCLC patients [
21]. In contrary to previous researches, our new model was constructed specially for resectable SCLC based on a large-population database and included common surgical procedure. Besides, the new model received ideal C-index by independent external validation which proved its better predicting accuracy.
Limitations have to be admitted in this study. Firstly, all of the data were obtained retrospectively, which made it susceptible to the inherent weaknesses of retrospective data collection. Although SEER is a huge population-based database, it doesn’t have data of tumor marker associated with SCLC such as NSE, proGRP and inflammation-related hematological markers both of which are key determinants of tumor survival [
26]. The nomogram built by Xie et al. and Pan et al. both included such important information to efficiently increase model accuracy [
18,
19]. What’s more, the chemotherapy regimens were unavailable and heterogeneous because this retrospective research collected data of different institutions over a long period. Finally, the sample size of our validation cohort was not very large. Another external validation with larger sample size of the predictive model is still necessary.
Open AccessThis article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit
http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.